Search

Deep two-dimensional (2-D) convolutional neural networks (CNNs) have been remarkably successful in producing record-breaking results in a variety of computer vision tasks. It is possible to extend CNNs to three dimensions using 3-D kernels to make them suitable for volumetric medical imaging data such as CT or MRI, but this increases the processing time as well as the required number of training samples (due to the higher number of parameters that need to be learned). In this paper, we address both of these issues for a 3-D CNN implementation through the development of a two-stage computer-aided detection system for automatic detection of pulmonary nodules. The first stage consists of a 3-D fully convolutional network for fast screening and generation of candidate suspicious regions. The second stage consists of an ensemble of 3-D CNNs trained using extensive transformations applied to both the positive and negative patches to augment the training set. To enable the second stage classifiers to learn differently, they are trained on false positive patches obtained from the screening model using different thresholds on their associated scores as well as different augmentation types. The networks in the second stage are averaged together to produce the final classification score for each candidate patch. Using this procedure, our overall nodule detection system called DeepMed is fast and can achieve 91% sensitivity at 2 false positives per scan on cases from the LIDC dataset.

Advances in machine learning in medical imaging are occurring at a rapid pace in research laboratories both at academic institutions and in industry. Important artificial intelligence (AI) tools for diagnostic imaging include algorithms for disease detection and classification, image optimization, radiation reduction, and workflow enhancement. Although advances in foundational research are occurring rapidly, translation to routine clinical practice has been slower. In August 2018, the National Institutes of Health assembled multiple relevant stakeholders at a public meeting to discuss the current state of knowledge, infrastructure gaps, and challenges to wider implementation. The conclusions of that meeting are summarized in two publications that identify and prioritize initiatives to accelerate foundational and translational research in AI for medical imaging. This publication summarizes key priorities for translational research developed at the workshop including: (1) creating structured AI use cases, defining and highlighting clinical challenges potentially solvable by AI; (2) establishing methods to encourage data sharing for training and testing AI algorithms to promote generalizability to widespread clinical practice and mitigate unintended bias; (3) establishing tools for validation and performance monitoring of AI algorithms to facilitate regulatory approval; and (4) developing standards and common data elements for seamless integration of AI tools into existing clinical workflows. An important goal of the resulting road map is to grow an ecosystem, facilitated by professional societies, industry, and government agencies, that will allow robust collaborations between practicing clinicians and AI researchers to advance foundational and translational research relevant to medical imaging.

RATIONALE AND OBJECTIVES To evaluate a new approach to establish compliance of segmentation tools with the computed tomography volumetry profile of the Quantitative Imaging Biomarker Alliance (QIBA); and determine the statistical exchangeability between real and simulated lesions through an international challenge. MATERIALS AND METHODS The study used an anthropomorphic phantom with 16 embedded physical lesions and 30 patient cases from the Reference Image Database to Evaluate Therapy Response with pathologically confirmed malignancies. Hybrid datasets were generated by virtually inserting simulated lesions corresponding to physical lesions into the phantom datasets using one projection-domain-based method (Method 1), two image-domain insertion methods (Methods 2 and 3), and simulated lesions corresponding to real lesions into the Reference Image Database to Evaluate Therapy Response dataset (using Method 2). The volumes of the real and simulated lesions were compared based on bias (measured mean volume differences between physical and virtually inserted lesions in phantoms as quantified by segmentation algorithms), repeatability, reproducibility, equivalence (phantom phase), and overall QIBA compliance (phantom and clinical phase). RESULTS For phantom phase, three of eight groups were fully QIBA compliant, and one was marginally compliant. For compliant groups, the estimated biases were -1.8 ± 1.4%, -2.5 ± 1.1%, -3 ± 1%, -1.8 ± 1.5% (±95% confidence interval). No virtual insertion method showed statistical equivalence to physical insertion in bias equivalence testing using Schuirmann's two one-sided test (±5% equivalence margin). Differences in repeatability and reproducibility across physical and simulated lesions were largely comparable (0.1%-16% and 7%-18% differences, respectively). For clinical phase, 7 of 16 groups were QIBA compliant. CONCLUSION Hybrid datasets yielded conclusions similar to real computed tomography datasets where phantom QIBA compliant was also compliant for hybrid datasets. Some groups deemed compliant for simulated methods, not for physical lesion measurements. The magnitude of this difference was small (

Purpose Most state-of-the-art automated medical image analysis methods for volumetric data rely on adaptations of two-dimensional (2D) and three-dimensional (3D) convolutional neural networks (CNNs). In this paper, we develop a novel unified CNN-based model that combines the benefits of 2D and 3D networks for analyzing volumetric medical images. Methods In our proposed framework, multiscale contextual information is first extracted from 2D slices inside a volume of interest (VOI). This is followed by dilated 1D convolutions across slices to aggregate in-plane features in a slice-wise manner and encode the information in the entire volume. Moreover, we formalize a curriculum learning strategy for a two-stage system (i.e., a system that consists of screening and false positive reduction), where the training samples are presented to the network in a meaningful order to further improve the performance. Results We evaluated the proposed approach by developing a computer-aided detection (CADe) system for lung nodules. Our results on 888 CT exams demonstrate that the proposed approach can effectively analyze volumetric data by achieving a sensitivity of > 0.99 in the screening stage and a sensitivity of > 0.96 at eight false positives per case in the false positive reduction stage. Conclusion Our experimental results show that the proposed method provides competitive results compared to state-of-the-art 3D frameworks. In addition, we illustrate the benefits of curriculum learning strategies in two-stage systems that are of common use in medical imaging applications.

A radiomics-based model was developed to predict bone fragility based on x-ray projection images. The model was trained using ROIs of MicroCT images, in which mean von Mises stress was computed using finite element modeling. The average normalized RMSE of the estimated von Mises stress of all folds was 14.7%. The model identifies GLSZM and GLDM features as important texture features for estimating stress. This work demonstrates radiomic features from projection images can be combined with nonlinear regression techniques to predict bone fragility and may result in a more accurate projection-domain estimate than areal bone mineral density.

Context.—Despite several studies focusing on the validation of whole slide imaging (WSI) across organ systems or subspecialties, the use of WSI for specific primary diagnosis tasks has been underexamined.Objective.—To assess pathologist performance for the histologic subtyping of individual sections of ovarian carcinomas using a light microscope and WSI.Design.—A panel of 3 experienced gynecologic pathologists provided reference subtype diagnoses for 212 histologic sections from 109 ovarian carcinomas based on optical microscopy review. Two additional attending pathologists provided diagnoses and also identified the presence of a set of 8 histologic features important for ovarian tumor subtyping. Two experienced gynecologic pathologists and 2 fellows reviewed the corresponding WSI images for subtype classification and feature identification.Results.—Across pathologists specialized in gynecologic pathology, concordance with the reference diagnosis for the 5 major ovarian carcinoma subtypes was significantly higher for a pathologist reading on a microscope than each of 2 pathologists reading on WSI. Differences were primarily due to more frequent classification of mucinous carcinomas as endometrioid with WSI. Pathologists had generally low agreement in identifying histologic features important to ovarian tumor subtype classification with either an optical microscopy or WSI. This result suggests the need for refined histologic criteria for identifying such features. Interobserver agreement was particularly low for identifying intracytoplasmic mucin with WSI. Inconsistencies in evaluating nuclear atypia and mitoses with WSI were also observed.Conclusions.—Further research is needed to specify the reasons for these diagnostic challenges and to inform users and manufacturers of WSI technology.

In this study, we show that when a training data set is supplemented by drawing samples from a distribution that is different from that of the target population, the differences in the distributions of the original and supplemental training populations should be considered to maximize the performance of the classifier in the target population. Depending on these distributions, drawing a large number of cases from the supplemental distribution may result in lower performance compared to limiting the number of added cases. This is relevant for medical images when synthetic data is used for training a machine learning algorithm, which may result in a mixed distribution for the training set. We simulated a twoclass classification problem and determined the performance of a linear classifier and a neural network classifier on test cases when trained with cases from only the target distribution, and when cases from a shifted, supplemental distribution are added to a limited number of cases from the target distribution. We show that adding data from a supplemental distribution for machine learning classifier training may improve the performance on the target test distribution. However, given the same number of training cases from a mixed distribution, the performance may not reach the performance of only training on data from the target distribution. In addition, the increase in performance will peak or plateau, depending on the shift in the distribution and the number of cases from the supplemental distribution.

Our work aims at developing a machine learning-based image conversion algorithm to adjust quantum noise, sharpness, scattering, and other characteristics of radiographic images acquired with a given imaging system as if they had been acquired with a different acquisition system. Purely physics-based methods which have previously been developed for image conversion rely on the measurement of the physical properties of the acquisition devices, which limit the range of their applicability. In this study, we focused on the conversion of mammographic images from a source acquisition system into a target system using a conditional Generative Adversarial Network (cGAN). This network penalizes any possible structural differences between network-generated and target images. The optimization process was enhanced by designing new reconstruction loss terms which emphasized the quality of high frequency image contents. We trained our cGAN model on a dataset of paired synthetic mammograms and slanted edge phantom images. We coupled one independent slanted edge phantom image with each anthropomorphic breast image and presented the pair as a combined input into the network. To improve network performance at high frequencies, we incorporated an edge-based loss function into the reconstruction loss. Qualitative results demonstrated the feasibility of our method to adjust the sharpness of mammograms acquired with a source system to appear as if the they were acquired with a different target system. Our method was validated by comparing the presampled modulation transfer function (MTF) of the network-generated edge image and the MTF of the source and target mammography acquisition systems at different spatial frequencies. This image conversion technique may help training of machine learning algorithms so that their applicability generalizes to a larger set of medical image acquisition devices. Our work may also facilitate performance assessment of computer-aided detection systems.

The performance of a classifier is largely dependent on the size and representativeness of data used for its training. In circumstances where accumulation and/or labeling of training samples is difficult or expensive, such as medical applications, data augmentation can potentially be used to alleviate the limitations of small datasets. We have previously developed an image blending tool that allows users to modify or supplement an existing CT or mammography dataset by seamlessly inserting a lesion extracted from a source image into a target image. This tool also provides the option to apply various types of transformations to different properties of the lesion prior to its insertion into a new location. In this study, we used this tool to create synthetic samples that appear realistic in chest CT. We then augmented different size training sets with these artificial samples, and investigated the effect of the augmentation on training various classifiers for the detection of lung nodules. Our results indicate that the proposed lesion insertion method can improve classifier performance for small training datasets, and thereby help reduce the need to acquire and label actual patient data.

Purpose Multiview two-dimensional (2D) convolutional neural networks (CNNs) and three-dimensional (3D) CNNs have been successfully used for analyzing volumetric data in many state-of-the-art medical imaging applications. We propose an alternative modular framework that analyzes volumetric data with an approach that is analogous to radiologists' interpretation, and apply the framework to reduce false positives that are generated in computer-aided detection (CADe) systems for pulmonary nodules in thoracic computed tomography (CT) scans. Methods In our approach, a deep network consisting of 2D CNNs first processes slices individually. The features extracted in this stage are then passed to a recurrent neural network (RNN), thereby modeling consecutive slices as a sequence of temporal data and capturing the contextual information across all three dimensions in the volume of interest. Outputs of the RNN layer are weighed before the final fully connected layer, enabling the network to scale the importance of different slices within a volume of interest in an end-to-end training framework. Results We validated the proposed architecture on the false positive reduction track of the lung nodule analysis (LUNA) challenge for pulmonary nodule detection in chest CT scans, and obtained competitive results compared to 3D CNNs. Our results show that the proposed approach can encode the 3D information in volumetric data effectively by achieving a sensitivity >0.8 with just 1/8 false positives per scan. Conclusions Our experimental results demonstrate the effectiveness of temporal analysis of volumetric images for the application of false positive reduction in chest CT scans and show that state-of-the-art 2D architectures from the literature can be directly applied to analyzing volumetric medical data. As newer and better 2D architectures are being developed at a much faster rate compared to 3D architectures, our approach makes it easy to obtain state-of-the-art performance on volumetric data using new 2D architectures.

Big data analysis can exhibit significant scaling problems when migrated to High Performance Computing (HPC) clusters and/or cloud computing platforms if traditional software parallelization techniques such as POSIX multi-threading and Message Passing Interface (MPI) are used. This paper introduces a novel scaling technique based on a-well-known array job mechanism to enable a team of FDA researchers to validate a method for identifying evidence of possible adverse events in very large sets of patient medical records. The analysis employed the widely-used basic Statistical Analysis Software (SAS) package, and the proposed parallelization approach dramatically increased the scaling and thus the speed of job completion for this application and is applicable to any similar software written in any other programming language. The new scaling technique offers O(T) theoretical speedup in comparison to multi-threading and MPI techniques. Here T is the number of array job tasks. The basis of the new approach is the segmentation of both (a) the big data set being analyzed and (b) the large number of SAS analysis types applied to each data segment. The large number of unique pairs of data set segment and analysis type segment are then each processed by a separate computing node (core) in pseudo-parallel manner. As a result, a SAS big data analysis which required more than 10 days to complete and consumed more than a terabyte of RAM on a single multi-core computing node completed in less than an hour parallelized over a large number of nodes, none of which needed more than 50 GB of RAM. The massive increase in the number of tasks when running an analysis job with this degree of segmentation reduces the size, scope and execution time of each task. Besides the significant speed improvement, additional benefits include fine-grained checkpointing and increased flexibility of job submission.

We evaluated whether using synthetic mammograms for training data augmentation may reduce the effects of overfitting and increase the performance of a deep learning algorithm for breast mass detection. Synthetic mammograms were generated using a combination of an in-silico random breast generation algorithm and x-ray transport simulation. In-silico breast phantoms containing masses were modeled across the four BI-RADS breast density categories, and the masses were modeled with different sizes, shapes and margins. A Monte Carlo-based xray transport simulation code, MC-GPU, was used to project the 3D phantoms into realistic synthetic mammograms. A training data set of 2,000 mammograms with 2,522 masses were generated and used for augmenting a data set of real mammograms for training. The data set of real mammograms included all the masses in the Curated Breast Imaging Subset of Digital Database for Screening Mammography (CBIS-DDSM) and consisted of 1,112 mammograms (1,198 masses) for training, 120 mammograms (120 masses) for validation, and 361 mammograms (378 masses) for testing. We used Faster R-CNN for our deep learning network with pre-training from ImageNet using Resnet-101 architecture. We compared the detection performance when the network was trained using only the CBIS-DDSM training images, and when subsets of the training set were augmented with 250, 500, 1,000 and 2,000 synthetic mammograms. FROC analysis was performed to compare performances with and without the synthetic mammograms. Our study showed that enlarging the training data with synthetic mammograms shows promise in reducing the overfitting, and that the inclusion of the synthetic images for training increased the performance of the deep learning algorithm for mass detection on mammograms.

We investigated effects of prevalence and case distribution on radiologist diagnostic performance as measured by area under the receiver operating characteristic curve (AUC) and sensitivity-specificity in lab-based reader studies evaluating imaging devices. Our retrospective reader studies compared full-field digital mammography (FFDM) to screen-film mammography (SFM) for women with dense breasts. Mammograms were acquired from the prospective Digital Mammographic Imaging Screening Trial. We performed five reader studies that differed in terms of cancer prevalence and the distribution of noncancers. Twenty radiologists participated in each reader study. Using split-plot study designs, we collected recall decisions and multilevel scores from the radiologists for calculating sensitivity, specificity, and AUC. Differences in reader-averaged AUCs slightly favored SFM over FFDM (biggest AUC difference: 0.047, [Formula: see text] , [Formula: see text]), where standard error accounts for reader and case variability. The differences were not significant at a level of 0.01 (0.05/5 reader studies). The differences in sensitivities and specificities were also indeterminate. Prevalence had little effect on AUC (largest difference: 0.02), whereas sensitivity increased and specificity decreased as prevalence increased. We found that AUC is robust to changes in prevalence, while radiologists were more aggressive with recall decisions as prevalence increased.

Rationale and Objectives Different computed tomography imaging protocols and patient characteristics can impact the accuracy and precision of the calcium score and may lead to inconsistent patient treatment recommendations. The aim of this work was to determine the impact of reconstruction algorithm and gender characteristics on coronary artery calcium scoring based on a phantom study using computed tomography. Materials and Methods Four synthetic heart vessels with vessel diameters corresponding to female and male left main and left circumflex arteries containing calcification-mimicking materials (200–1000 HU) were inserted into a thorax phantom and were scanned with and without female breast plates (male and female phantoms, respectively). Ten scans were acquired and were reconstructed at 3-mm slices using filtered-back projection (FBP) and iterative reconstruction with medium and strong denoising (IR3 and IR5) algorithms. Agatston and calcium volume scores were estimated for each vessel. Calcium scores for each vessel and the total calcium score (summation of all four vessels) were compared between the two phantoms to quantify the impact of the breast plates and reconstruction parameters. Calcium scores were also compared among vessels of different diameters to investigate the impact of the vessel size. Results The calcium scores were significantly larger for FBP reconstruction (FBP > IR3>IR5). Agatston scores (calcium volume score) for vessels in the male phantom scans were on average 4.8% (2.9%), 8.2% (7.1%), and 10.5% (9.4%) higher compared to those in the female phantom with FBP, IR3, and IR5, respectively, when exposure was conserved across phantoms. The total calcium scores from the male phantom were significantly larger than those from the female phantom (P

Purpose: To evaluate the performance of lesion volumetry in hepatic CT as a function of various imaging acquisition parameters. Methods: An anthropomorphic abdominal phantom with removable liver inserts was designed for this study. Two liver inserts, each containing 19 synthetic lesions with varying diameter (6–40 mm), shape, contrast (10–65 HU), and both homogenous and mixed-density were designed to have background and lesion CT values corresponding to arterial and portal-venous phase imaging, respectively. The two phantoms were scanned using two commercial CT scanners (GE 750 HD and Siemens Biograph mCT) across a set of imaging protocols (four slice thicknesses, three effective mAs, two convolution kernels, two pitches). Two repeated scans were collected for each imaging protocol. All scans were analyzed using a matched-filter estimator for volume estimation, resulting in 6080 volume measurements across all of the synthetic lesions in the two liver phantoms. A subset of portal venous phase scans was also analyzed using a semi-automatic segmentation algorithm, resulting in about 900 additional volume measurements. Lesions associated with large measurement error (quantified by root mean square error) for most imaging protocols were considered not measurable by the volume estimation tools and excluded for the statistical analyses. Imaging protocols were grouped into distinct imaging conditions based on ANOVA analysis of factors for repeatability testing. Statistical analyses, including overall linearity analysis, grouped bias analysis with standard deviation evaluation, and repeatability analysis, were performed to assess the accuracy and precision of the liver lesion volume biomarker. Results: Lesions with lower contrast and size ≤10 mm were associated with higher measurement error and were excluded from further analysis. Lesion size, contrast, imaging slice thickness, dose, and scanner were found to be factors substantially influencing volume estimation. Twenty-four distinct repeatable imaging conditions were determined as protocols for each scanner with a fixed slice thickness and dose. For the matched-filter estimation approach, strong linearity was observed for all imaging data for lesions ≥20 mm. For the Siemens scanner with 50 mAs effective dose at 0.6 mm slice thickness, grouped bias was about −10%. For all other repeatable imaging conditions with both scanners, grouped biases were low (−3%–3%). There was a trend of increasing standard deviation with decreasing dose. For each fixed dose, the standard deviations were similar among the three larger slice thicknesses (1.25, 2.5, 5 mm for GE, 1.5, 3, 5 mm for Siemens). Repeatability coefficients ranged from about 8% to 75% and showed similar trend to grouped standard deviation. For the segmentation approach, the results led to similar conclusions for both lesion characteristic factors and imaging factors but with increasing magnitude in all the error metrics assessed. Conclusions: Results showed that liver lesion volumetry was strongly dependent on lesion size, contrast, acquisition dose, and their interactions. The overall performances were similar for images reconstructed with larger slice thicknesses, clinically used pitches, kernels, and doses. Conditions that yielded repeatable measurements were identified and they agreed with the Quantitative Imaging Biomarker Alliance's (QIBA) profile requirements in general. The authors’ findings also suggest potential refinements to these guidelines for the tumor volume biomarker, especially for soft-tissue lesions.

Scores produced by statistical classifiers in many clinical decision support systems and other medical diagnostic devices are generally on an arbitrary scale, so the clinical meaning of these scores is unclear. Calibration of classifier scores to a meaningful scale such as the probability of disease is potentially useful when such scores are used by a physician. In this work, we investigated three methods (parametric, semi-parametric, and non-parametric) for calibrating classifier scores to the probability of disease scale and developed uncertainty estimation techniques for these methods. We showed that classifier scores on arbitrary scales can be calibrated to the probability of disease scale without affecting their discrimination performance. With a finite dataset to train the calibration function, it is important to accompany the probability estimate with its confidence interval. Our simulations indicate that, when a dataset used for finding the transformation for calibration is also used for estimating the performance of calibration, the resubstitution bias exists for a performance metric involving the truth states in evaluating the calibration performance. However, the bias is small for the parametric and semi-parametric methods when the sample size is moderate to large (>100 per class).

Purpose: Traditional ways to estimate 2D CT noise power spectrum (NPS) involve an ensemble average of the power spectrums of many noisy scans. When only a few scans are available, regions of interest are often extracted from different locations to obtain sufficient samples to estimate the NPS. Using image samples from different locations ignores the nonstationarity of CT noise and thus cannot accurately characterize its local properties. The purpose of this work is to develop a method to estimate local NPS using only a few fan-beam CT scans. Methods: As a result of FBP reconstruction, the CT NPS has the same radial profile shape for all projection angles, with the magnitude varying with the noise level in the raw data measurement. This allows a 2D CT NPS to be factored into products of a 1D angular and a 1D radial function in polar coordinates. The polar separability of CT NPS greatly reduces the data requirement for estimating the NPS. The authors use this property and derive a radial NPS estimation method: in brief, the radial profile shape is estimated from a traditional NPS based on image samples extracted at multiple locations. The amplitudes are estimated by fitting the traditional local NPS to the estimated radial profile shape. The estimated radial profile shape and amplitudes are then combined to form a final estimate of the local NPS. We evaluate the accuracy of the radial NPS method and compared it to traditional NPS methods in terms of normalized mean squared error (NMSE) and signal detectability index. Results: For both simulated and real CT data sets, the local NPS estimated with no more than six scans using the radial NPS method was very close to the reference NPS, according to the metrics of NMSE and detectability index. Even with only two scans, the radial NPS method was able to achieve a fairly good accuracy. Compared to those estimated using traditional NPS methods, the accuracy improvement was substantial when a few scans were available. Conclusions: The radial NPS method was shown to be accurate and efficient in estimating the local NPS of FBP-reconstructed 2D CT images. It presents strong advantages over traditional NPS methods when the number of scans is limited and can be extended to estimate the in-plane NPS of cone-beam CT and multislice helical CT scans.

Mammographic computer-aided detection (CADe) devices are typically first developed and assessed for a specific “original” acquisition system. When developers are ready to apply their CADe device to a mammographic acquisition system, they typically assess the device with images acquired using the system. Collecting large repositories of clinical images containing verified lesion locations acquired by a system is costly and time consuming. We previously developed an image blending technique that allows users to seamlessly insert regions of interest (ROIs) from one medical image into another image. Our goal is to assess the performance of this technique for inserting microcalcification clusters from one mammogram into another, with the idea that when fully developed, our technique may be useful for reducing the clinical data burden in the assessment of a CADe device for use with an image acquisition system. We first perform a reader study to assess whether experienced observers can distinguish between computationally inserted and native clusters. For this purpose, we apply our insertion technique to 55 clinical cases. ROIs containing microcalcification clusters from one breast of a patient are inserted into the contralateral breast of the same patient. The analysis of the reader ratings using receiver operating characteristic (ROC) methodology indicates that inserted clusters cannot be reliably distinguished from native clusters (area under the ROC [Formula: see text]). Furthermore, CADe sensitivity is evaluated on mammograms of 68 clinical cases with native and inserted microcalcification clusters using a commercial CADe system. The average by-case sensitivities for native and inserted clusters are equal, 85.3% (58/68). The average by-image sensitivities for native and inserted clusters are 72.3% and 67.6%, respectively, with a difference of 4.7% and a 95% confidence interval of [[Formula: see text] 11.6]. These results demonstrate the potential for using the inserted microcalcification clusters for assessing mammographic CADe devices.

Computer-aided detection (CADe) devices used for breast cancer detection on mammograms are typically first developed and assessed for a specific “original” acquisition system, e.g., a specific image detector. When CADe developers are ready to apply their CADe device to a new mammographic acquisition system, they typically assess the CADe device with images acquired using the new system. Collecting large repositories of clinical images containing verified cancer locations and acquired by the new image acquisition system is costly and time consuming. Our goal is to develop a methodology to reduce the clinical data burden in the assessment of a CADe device for use with a different image acquisition system. We are developing an image blending technique that allows users to seamlessly insert lesions imaged using an original acquisition system into normal images or regions acquired with a new system. In this study, we investigated the insertion of microcalcification clusters imaged using an original acquisition system into normal images acquired with that same system utilizing our previously-developed image blending technique. We first performed a reader study to assess whether experienced observers could distinguish between computationally inserted and native clusters. For this purpose, we applied our insertion technique to clinical cases taken from the University of South Florida Digital Database for Screening Mammography (DDSM) and the Breast Cancer Digital Repository (BCDR). Regions of interest containing microcalcification clusters from one breast of a patient were inserted into the contralateral breast of the same patient. The reader study included 55 native clusters and their 55 inserted counterparts. Analysis of the reader ratings using receiver operating characteristic (ROC) methodology indicated that inserted clusters cannot be reliably distinguished from native clusters (area under the ROC curve, AUC=0.58±0.04). Furthermore, CADe sensitivity was evaluated on mammograms with native and inserted microcalcification clusters using a commercial CADe system. For this purpose, we used full field digital mammograms (FFDMs) from 68 clinical cases, acquired at the University of Michigan Health System. The average sensitivities for native and inserted clusters were equal, 85.3% (58/68). These results demonstrate the feasibility of using the inserted microcalcification clusters for assessing mammographic CAD devices.

Quantitative assessment of tumor radiodensity is important for the clinical evaluation of contrast enhancement and treatment response, as well as for the extraction of texture-related features for image analysis or radiomics. Radiodensity estimation, Hounsfield Units (HU) in CT images, can be affected by patient factors such as tumor size, and by system factors such as acquisition and reconstruction protocols. In this project, we quantified the measurability of liver tumor HU using a 3D-printed phantom, imaged with two CT systems: Siemens Somatom Force and GE Lightspeed VCT. The phantom was printed by dithering two materials to create spherical tumors (10, 14 mm) with uniform densities (90, 95, 100, 105 HU). Image datasets were acquired at 120 kVp including 15 repeats using two matching exposures across the CT systems, and reconstructed using comparable algorithms. The radiodensity of each tumor was measured using an automated matched-filter method. We assessed the performance of each protocol using the area under the ROC curve (AUC) as the metric for distinguishing between tumors with different radiodensities. The AUC ranged from 0.8 to 1.0 and was affected by tumor size, radiodensity, and scanner; the lowest AUC values corresponded to low dose measurements of 10 mm tumors with less than 5 HU difference. The two scanners exhibited similar performance >0.9 AUC for large lesions with contrast above 7 HU, though differences were observed for the smallest and lowest contrast tumors. These results show that HU estimation should be carefully examined, considering that uncertainty in the tumor radiodensity may propagate to quantification of other characteristics, such as size and texture.

Grand challenges stimulate advances within the medical imaging research community; within a competitive yet friendly environment, they allow for a direct comparison of algorithms through a well-defined, centralized infrastructure. The tasks of the two-part PROSTATEx Challenges (the PROSTATEx Challenge and the PROSTATEx-2 Challenge) are (1) the computerized classification of clinically significant prostate lesions and (2) the computerized determination of Gleason Grade Group in prostate cancer, both based on multiparametric magnetic resonance images. The challenges incorporate well-vetted cases for training and testing, a centralized performance assessment process to evaluate results, and an established infrastructure for case dissemination, communication, and result submission. In the PROSTATEx Challenge, 32 groups apply their computerized methods (71 methods total) to 208 prostate lesions in the test set. The area under the receiver operating characteristic curve for these methods in the task of differentiating between lesions that are and are not clinically significant ranged from 0.45 to 0.87; statistically significant differences in performance among the top-performing methods, however, are not observed. In the PROSTATEx-2 Challenge, 21 groups apply their computerized methods (43 methods total) to 70 prostate lesions in the test set. When compared with the reference standard, the quadratic-weighted kappa values for these methods in the task of assigning a five-point Gleason Grade Group to each lesion range from [Formula: see text] to 0.27; superiority to random guessing can be established for only two methods. When approached with a sense of commitment and scientific rigor, challenges foster interest in the designated task and encourage innovation in the field.

The availability of large medical image datasets is critical in many applications such as training and testing of computer aided diagnosis (CAD) systems, evaluation of segmentation algorithms, and conducting perceptual studies. However, collection of data and establishment of ground truth for medical images are both costly and difficult. To address this problem, we are developing an image blending tool that allows users to modify or supplement existing datasets by seamlessly inserting a lesion extracted from a source image into a target image. In this study we focus on the application of this tool to pulmonary nodules in chest CT exams. We minimize the impact of user skill on the perceived quality of the composite image by limiting user involvement to two simple steps: the user first draws a casual boundary around a nodule in the source, and then selects the center of desired insertion area in the target. We demonstrate the performance of our system on clinical samples, and report the results of a reader study evaluating the realism of inserted nodules compared to clinical nodules. We further evaluate our image blending techniques using phantoms simulated under different noise levels and reconstruction filters. Specifically, we compute the area under the ROC curve (AUC) of the Hotelling observer (HO) and noise power spectrum (NPS) of regions of interest enclosing native and inserted nodules, and compare the detectability, noise texture, and noise magnitude of inserted and native nodules. Our results indicate the viability of our approach for insertion of pulmonary nodules in clinical CT images.

Purpose: To determine inter-related factors that contribute substantially to measurement error of pulmonary nodule measurements with CT by assessing a large-scale dataset of phantom scans and to quantitatively validate the repeatability and reproducibility of a subset containing nodules and CT acquisitions consistent with the Quantitative Imaging Biomarker Alliance (QIBA) metrology recommendations. Methods: The dataset has about 40 000 volume measurements of 48 nodules (5–20 mm, four shapes, three radiodensities) estimated by a matched-filter estimator from CT images involving 72 imaging protocols. Technical assessment was performed under a framework suggested by QIBA, which aimed to minimize the inconsistency of terminologies and techniques used in the literature. Accuracy and precision of lung nodule volume measurements were examined by analyzing the linearity, bias, variance, root mean square error (RMSE), repeatability, reproducibility, and significant and substantial factors that contribute to the measurement error. Statistical methodologies including linear regression, analysis of variance, and restricted maximum likelihood were applied to estimate the aforementioned metrics. The analysis was performed on both the whole dataset and a subset meeting the criteria proposed in the QIBA Profile document. Results: Strong linearity was observed for all data. Size, slice thickness × collimation, and randomness in attachment to vessels or chest wall were the main sources of measurement error. Grouping the data by nodule size and slice thickness × collimation, the standard deviation (3.9%–28%), and RMSE (4.4%–68%) tended to increase with smaller nodule size and larger slice thickness. For 5, 8, 10, and 20 mm nodules with reconstruction slice thickness ≤0.8, 3, 3, and 5 mm, respectively, the measurements were almost unbiased (−3.0% to 3.0%). Repeatability coefficients (RCs) were from 6.2% to 40%. Pitch of 0.9, detail kernel, and smaller slice thicknesses yielded better (smaller) RCs than those from pitch of 1.2, medium kernel, and larger slice thicknesses. Exposure showed no impact on RC. The overall reproducibility coefficient (RDC) was 45%, and reduced to about 20%–30% when the slice thickness and collimation were fixed. For nodules and CT imaging complying with the QIBA Profile (QIBA Profile subset), the measurements were highly repeatable and reproducible in spite of variations in nodule characteristics and imaging protocols. The overall measurement error was small and mostly due to the randomness in attachment. The bias, standard deviation, and RMSE grouped by nodule size and slice thickness × collimation in the QIBA Profile subset were within ±3%, 4%, and 5%, respectively. RCs are within 11% and the overall RDC is equal to 11%. Conclusions: The authors have performed a comprehensive technical assessment of lung nodule volumetry with a matched-filter estimator from CT scans of synthetic nodules and identified the main sources of measurement error among various nodule characteristics and imaging parameters. The results confirm that the QIBA Profile set is highly repeatable and reproducible. These phantom study results can serve as a bound on the clinical performance achievable with volumetric CT measurements of pulmonary nodules.

© 2015 The Authors. PURPOSE: To determine the variability of lesion size measurements in computed tomography data sets of patients imaged under a “no change” (“coffee break”) condition and to determine the impact of two reading paradigms on measurement variability. METHOD AND MATERIALS: Using data sets from 32 non-small cell lung cancer patients scanned twice within 15 minutes (“no change”), measurements were performed by five radiologists in two phases: (1) independent reading of each computed tomography dataset (timepoint): (2) a locked, sequential reading of datasets. Readers performed measurements using several sizing methods, including one-dimensional (1D) longest in-slice dimension and 3D semi-automated segmented volume. Change in size was estimated by comparing measurements performed on both timepoints for the same lesion, for each reader and each measurement method. For each reading paradigm, results were pooled across lesions, across readers, and across both readers and lesions, for each measurement method. RESULTS: The mean percent difference (±SD) when pooled across both readers and lesions for 1D and 3D measurements extracted from contours was 2.8 ± 22.2% and 23.4 ± 105.0%, respectively, for the independent reads. For the locked, sequential reads, the mean percent differences (±SD) reduced to 2.52 ± 14.2% and 7.4 ± 44.2% for the 1D and 3D measurements, respectively. CONCLUSION: Even under a “no change” condition between scans, there is variation in lesion size measurements due to repeat scans and variations in reader, lesion, and measurement method. This variation is reduced when using a locked, sequential reading paradigm compared to an independent reading paradigm.

Novel medical imaging devices are often evaluated with multi-reader multi-case (MRMC) studies in which radiologists read images of patient cases for a specified clinical task (e.g., cancer detection). A pilot study is often used to measure the effect size and variance parameters that are necessary for sizing a pivotal study (including sizing readers, non-diseased and diseased cases). Due to the practical difficulty of collecting patient cases or recruiting clinical readers, some investigators attempt to include the pilot data as part of their pivotal study. In other situations, some investigators attempt to perform an interim analysis of their pivotal study data based upon which the sample sizes may be re-estimated. Re-use of the pilot data or interim analyses of the pivotal data may inflate the type I error of the pivotal study. In this work, we use the Roe and Metz model to simulate MRMC data under the null hypothesis (i.e., two devices have equal diagnostic performance) and investigate the type I error rate for several practical designs involving re-use of pilot data or interim analysis of pivotal data. Our preliminary simulation results indicate that, under the simulation conditions we investigated, the inflation of type I error is none or only marginal for some design strategies (e.g., re-use of patient data without re-using readers, and size re-estimation without using the effect-size estimated in the interim analysis). Upon further verifications, these are potentially useful design methods in that they may help make a study less burdensome and have a better chance to succeed without substantial loss of the statistical rigor.

Tumor volume measured from computed tomography images is considered a biomarker for disease progression or treatment response. The estimation of the tumor volume depends on the imaging system parameters selected, as well as lesion characteristics. In this study, we examined how different image reconstruction methods affect the measurement of lesions in an anthropomorphic liver phantom with a non-uniform background. Iterative statistics-based and model-based reconstructions, as well as filtered back-projection, were evaluated and compared in this study. Statistics-based and filtered back-projection yielded similar estimation performance, while model-based yielded higher precision but lower accuracy in the case of small lesions. Iterative reconstructions exhibited higher signal-to-noise ratio but slightly lower contrast of the lesion relative to the background. A better understanding of lesion volumetry performance as a function of acquisition parameters and lesion characteristics can lead to its incorporation as a routine sizing tool.

Deep convolutional neural networks (CNNs) form the backbone of many state-of-the-art computer vision systems for classification and segmentation of 2D images. The same principles and architectures can be extended to three dimensions to obtain 3D CNNs that are suitable for volumetric data such as CT scans. In this work, we train a 3D CNN for automatic detection of pulmonary nodules in chest CT images using volumes of interest extracted from the LIDC dataset. We then convert the 3D CNN which has a fixed field of view to a 3D fully convolutional network (FCN) which can generate the score map for the entire volume efficiently in a single pass. Compared to the sliding window approach for applying a CNN across the entire input volume, the FCN leads to a nearly 800-fold speed-up, and thereby fast generation of output scores for a single case. This screening FCN is used to generate difficult negative examples that are used to train a new discriminant CNN. The overall system consists of the screening FCN for fast generation of candidate regions of interest, followed by the discrimination CNN.

Technological developments and greater rigor in the quantitative measurement of biological features in medical images have given rise to an increased interest in using quantitative imaging biomarkers to measure changes in these features. Critical to the performance of a quantitative imaging biomarker in preclinical or clinical settings are three primary metrology areas of interest: measurement linearity and bias, repeatability, and the ability to consistently reproduce equivalent results when conditions change, as would be expected in any clinical trial. Unfortunately, performance studies to date differ greatly in designs, analysis method, and metrics used to assess a quantitative imaging biomarker for clinical use. It is therefore difficult or not possible to integrate results from different studies or to use reported results to design studies. The Radiological Society of North America and the Quantitative Imaging Biomarker Alliance with technical, radiological, and statistical experts developed a set of technical performance analysis methods, metrics, and study designs that provide terminology, metrics, and methods consistent with widely accepted metrological standards. This document provides a consistent framework for the conduct and evaluation of quantitative imaging biomarker performance studies so that results from multiple studies can be compared, contrasted, or combined.

We evaluated whether using synthetic mammograms for training data augmentation may reduce the effects of overfitting and increase the performance of a deep learning algorithm for breast mass detection. Synthetic mammograms were generated using in silico procedural analytic breast and breast mass modeling algorithms followed by simulated x-ray projections of the breast models into mammographic images. In silico breast phantoms containing masses were modeled across the four BI-RADS breast density categories, and the masses were modeled with different sizes, shapes, and margins. A Monte Carlo-based x-ray transport simulation code, MC-GPU, was used to project the three-dimensional phantoms into realistic synthetic mammograms. 2000 mammograms with 2522 masses were generated to augment a real data set during training. From the Curated Breast Imaging Subset of the Digital Database for Screening Mammography (CBIS-DDSM) data set, we used 1111 mammograms (1198 masses) for training, 120 mammograms (120 masses) for validation, and 361 mammograms (378 masses) for testing. We used faster R-CNN for our deep learning network with pretraining from ImageNet using the Resnet-101 architecture. We compared the detection performance when the network was trained using different percentages of the real CBIS-DDSM training set (100%, 50%, and 25%), and when these subsets of the training set were augmented with 250, 500, 1000, and 2000 synthetic mammograms. Free-response receiver operating characteristic (FROC) analysis was performed to compare performance with and without the synthetic mammograms. We generally observed an improved test FROC curve when training with the synthetic images compared to training without them, and the amount of improvement depended on the number of real and synthetic images used in training. Our study shows that enlarging the training data with synthetic samples can increase the performance of deep learning systems.

Rationale and Objectives The change in volume of lung nodules is being examined as a measure of response to treatment. The aim of this study was to determine the minimum detectable change in nodule volume with the use of computed tomography. Materials and Methods Four different layouts of synthetic nodules with different shapes but with the same size (5, 8, 9, or 10 mm) for each layout were placed within an anthropomorphic phantom and scanned with a 16-detector-row computed tomography scanner using multiple imaging parameters. Nodule volume estimates were determined using a previously developed matched-filter estimator. Analysis of volume change was then conducted as a detection problem. For each nodule size, the pooled distribution of volume estimates was shifted by a percentage c to simulate a changing nodule, while accounting for standard deviation. The value of c resulting in a prespecified area under the receiver operating characteristic curve (AUC) was deemed the minimum detectable change for that AUC value. Results Both nodule size at baseline and choice of slice collimation protocol had an effect on the value of minimum detectable growth. For AUC = 0.95, the minimum detectable nodule growth in volume when using the thin-slice collimation protocol (16 × 0.75 mm) was 17%, 19%, and 15% for nodule sizes of 5, 8, and 9 mm, respectively. Conclusions Our results indicate that an approximate bound for detectable nodule growth in subcentimeter nodules may be relatively small, on the order of 20% or less in volume for a thin-slice CT acquisition protocol.

Purpose Colitis refers to inflammation of the inner lining of the colon that is frequently associated with infection and allergic reactions. In this paper, we propose deep convolutional neural networks methods for lesion-level colitis detection and a support vector machine (SVM) classifier for patient-level colitis diagnosis on routine abdominal CT scans. Methods The recently-developed Faster Region-based Convolutional Neural Network (Faster RCNN) is utilized for lesion-level colitis detection. For each 2D slice, rectangular region proposals are generated by region proposal networks (RPN). Then, each region proposal is jointly classified and refined by a softmax classifier and bounding box regressor. Two convolutional neural networks, 8 layers of ZF net and 16 layers of VGG net are compared for colitis detection. Finally, for each patient, the detections on all 2D slices are collected and a support vector machine (SVM) classifier is applied to develop a patient-level diagnosis. We trained and evaluated our method with 80 colitis patients and 80 normal cases using 4×4-fold cross validation. Results For lesion-level colitis detection, with ZF net, the mean of average precisions (mAP) were 48.7% and 50.9% for RCNN and Faster RCNN, respectively. The detection system achieved sensitivities of 51.4% and 54.0% at 2 false positives per patient for RCNN and Faster RCNN, respectively. With VGG net, Faster RCNN increased the AP to 56.9% and increased the sensitivity to 58.4% at 2 false positive per patient. For patient-level colitis diagnosis, with ZF net, the average areas under the ROC curve (AUC) were 0.978±0.009 and 0.984±0.008 for RCNN and Faster RCNN method, respectively. The difference was not statistically significant with p=0.18. At the optimal operating point, the RCNN method correctly identified 90.4% (72.3/80) of the colitis patients and 94.0% (75.2/80) of normal cases. The sensitivity improved to 91.6% (73.3/80) and the specificity improved to 95.0% (76.0/80) for the Faster RCNN method. With VGG net, Faster RCNN increased the AUC to 0.986±0.007 and increased the diagnosis sensitivity to 93.7% (75.0/80) and specificity was unchanged at 95.0% (76.0/80). Conclusion Colitis detection and diagnosis by deep neural networks is accurate and promising for future clinical application. This article is protected by copyright. All rights reserved.

This work aimed to quantify and compare the potential impact of gender differences on coronary artery calcium scoring with dual-energy CT. An anthropomorphic thorax phantom with four synthetic heart vessels (diameter 3-4.5 mm: female/male left main and left circumflex artery) were scanned with and without female breast plates. Ten repeat scans were acquired in both single- and dual-energy modes and reconstructed at six reconstruction settings: two slice thicknesses (3 mm, 0.6 mm) and three reconstruction algorithms (FBP, IR3, IR5). Agatston and calcium volume scores were estimated from the reconstructed data using a segmentation-based approach. Total calcium score (summation of four vessels), and male/female calcium scores (summation of male/female vessels scanned in phantom without/with breast plates) were calculated accordingly. Both Agatston and calcium volume scores were found comparable between single- and dual-energy scans (Pearson r= 0.99, p

A major initiative of the Quantitative Imaging Biomarker Alliance (QIBA) is to develop standards-based documents called “Profiles”, which describe one or more technical performance claims for a given imaging modality. The term “actor” denotes any entity (device, software, person) whose performance must meet certain specifications in order for the claim to be met. The objective of this paper is to present the statistical issues in testing actors’ conformance with the specifications. In particular, we present the general rationale and interpretation of the claims, the minimum requirements for testing whether an actor achieves the performance requirements, the study designs used for testing conformity, and the statistical analysis plan. We use three examples to illustrate the process: apparent diffusion coefficient (ADC) in solid tumors measured by MRI, change in Perc 15 as a biomarker for the progression of emphysema, and percent change in solid tumor volume by CT as a biomarker for lung cancer progression.

Rationale and Objectives Quantifying changes in lung tumor volume is important for diagnosis, therapy planning, and evaluation of response to therapy. The aim of this study was to assess the performance of multiple algorithms on a reference data set. The study was organized by the Quantitative Imaging Biomarker Alliance (QIBA). Materials and Methods The study was organized as a public challenge. Computed tomography scans of synthetic lung tumors in an anthropomorphic phantom were acquired by the Food and Drug Administration. Tumors varied in size, shape, and radiodensity. Participants applied their own semi-automated volume estimation algorithms that either did not allow or allowed post-segmentation correction (type 1 or 2, respectively). Statistical analysis of accuracy (percent bias) and precision (repeatability and reproducibility) was conducted across algorithms, as well as across nodule characteristics, slice thickness, and algorithm type. Results Eighty-four percent of volume measurements of QIBA-compliant tumors were within 15% of the true volume, ranging from 66% to 93% across algorithms, compared to 61% of volume measurements for all tumors (ranging from 37% to 84%). Algorithm type did not affect bias substantially; however, it was an important factor in measurement precision. Algorithm precision was notably better as tumor size increased, worse for irregularly shaped tumors, and on the average better for type 1 algorithms. Over all nodules meeting the QIBA Profile, precision, as measured by the repeatability coefficient, was 9.0% compared to 18.4% overall. Conclusion The results achieved in this study, using a heterogeneous set of measurement algorithms, support QIBA quantitative performance claims in terms of volume measurement repeatability for nodules meeting the QIBA Profile criteria.

Rationale and Objectives The aims of this study were to define volumetric nomograms for identifying hepatomegaly and to retrospectively evaluate the performance of radiologists in assessing hepatomegaly. Materials and Methods Livers were automatically segmented from 148 abdominal contrast-enhanced computed tomographic scans: 77 normal livers and 71 cases of hepatomegaly (diagnosed by visual inspection and/or linear liver height by radiologists). Quantified liver volumes were compared to manual measurements using volume overlap and error. Liver volumes were normalized to body surface area, from which hepatomegaly nomograms were defined (H scores) by analyzing the distribution of liver sizes in the healthy population. H scores were validated against consensus reports. The performance of radiologists in diagnosing hepatomegaly was retrospectively evaluated. Results The automated segmentation of livers was robust, with volume overlap and error of 96.2% and 2.2%, respectively. There were no significant differences (P > .10) between manual and automated segmentation for either the normal or the hepatomegaly subgroup. The average volumes of normal and enlarged livers were 1.51 ± 0.25 and 2.32 ± 0.75 L, respectively. One-way analysis of variance found that body surface area (P = .004) and gender (P = .02), but not age, significantly affected normal liver volume. No significant effects were observed for two-way and three-way interactions among the three variables (P > .18). H-score cutoffs of 0.92 and 1.08 L/m2 were used to define mild and massive hepatomegaly (95% confidence interval, ±0.02 L/m2). Using the H score as the reference standard, the sensitivity of radiologists in detecting all, mild, and massive hepatomegaly was 84.4%, 56.7%, and 100.0% at 90.1% specificity, respectively. Radiologists disagreed on 20.9% of the diagnosed cases (n = 31). The area under the receiver-operating characteristic curve of the H-score criterion for hepatomegaly detection was 0.98. Conclusions Nomograms for the identification and grading of hepatomegaly from automatic volumetric liver assessment normalized to body surface area (H scores) are introduced. H scores match well with clinical interpretations for hepatomegaly and may improve hepatomegaly detection compared with height measurements or visual inspection, commonly used in current clinical practice.

Purpose: Surface curvatures are important geometric features for the computer-aided analysis and detection of polyps in CT colonography (CTC). However, the general kernel approach for curvature computation can yield erroneous results for small polyps and for polyps that lie on haustral folds. Those erroneous curvatures will reduce the performance of polyp detection. This paper presents an analysis of interpolation’s effect on curvature estimation for thin structures and its application on computer-aided detection of small polyps in CTC. Methods: The authors demonstrated that a simple technique, image interpolation, can improve the accuracy of curvature estimation for thin structures and thus significantly improve the sensitivity of small polyp detection in CTC. Results: Our experiments showed that the merits of interpolating included more accurate curvature values for simulated data, and isolation of polyps near folds for clinical data. After testing on a large clinical data set, it was observed that sensitivities with linear, quadratic B-spline and cubic B-spline interpolations significantly improved the sensitivity for small polyp detection. Conclusions: The image interpolation can improve the accuracy of curvature estimation for thin structures and thus improve the computer-aided detection of small polyps in CTC.

Purpose: The development of computer-aided diagnostic (CAD) methods for lung nodule detection, classification, and quantitative assessment can be facilitated through a well-characterized repository of computed tomography (CT) scans. The Lung Image Database Consortium (LIDC) and Image Database Resource Initiative (IDRI) completed such a database, establishing a publicly available reference for the medical imaging research community. Initiated by the National Cancer Institute (NCI), further advanced by the Foundation for the National Institutes of Health (FNIH), and accompanied by the Food and Drug Administration (FDA) through active participation, this public-private partnership demonstrates the success of a consortium founded on a consensus-based process. Methods: Seven academic centers and eight medical imaging companies collaborated to identify, address, and resolve challenging organizational, technical, and clinical issues to provide a solid foundation for a robust database. The LIDC/IDRI Database contains 1018 cases, each of which includes images from a clinical thoracic CT scan and an associated XML file that records the results of a two-phase image annotation process performed by four experienced thoracic radiologists. In the initial blinded-read phase, each radiologist independently reviewed each CT scan and marked lesions belonging to one of three categories (" nodule�3 mm," " nodule

Extracting coronary artery calcium (CAC) scores from contrast-enhanced computed tomography (CT) images using dual-energy (DE) based material decomposition has been shown feasible, mainly through patient studies. However, the quantitative performance of such DE-based CAC scores, particularly per stenosis, is underexamined due to lack of reference standard and repeated scans. In this work we conducted a comprehensive quantitative comparative analysis of CAC scores obtained with DE and compare to conventional unenhanced single-energy (SE) CT scans through phantom studies. Synthetic vessels filled with iodinated blood mimicking material and containing calcium stenoses of different sizes and densities were scanned with a third generation dual-source CT scanner in a chest phantom using a DE coronary CT angiography protocol with three exposures/CTDIvol: auto-mAs/8 mGy (automatic exposure), 160 mAs/20 mGy and 260 mAs/34 mGy and 10 repeats. As a control, a set of vessel phantoms without iodine was scanned using a standard SE CAC score protocol (3 mGy). Calcium volume, mass and Agatston scores were estimated for each stenosis. For DE dataset, image-based three-material decomposition was applied to remove iodine before scoring. Performance of DE-based calcium scores were analyzed on a per-stenosis level and compared to SE-based scores. There was excellent correlation between the DE- and SE-based scores (correlation coefficient r: 0.92-0.98). Percent bias for the calcium volume and mass scores varied as a function of stenosis size and density for both modalities. Precision (coefficient of variation) improved with larger and denser stenoses for both DE- and SE-based calcium scores. DE-based scores (20 mGy and 34 mGy) provided comparable per-stenosis precision to SE-based (3 mGy). Our findings suggest that on a per-stenosis level, DE-based CAC scores from contrast-enhanced CT images can achieve comparable quantification performance to conventional SE-based scores. However, DE-based CAC scoring required more dose compared with SE for high per-stenosis precision so some caution is necessary with clinical DE-based CAC scoring.

Rationale and Objectives To determine whether the display of computer-aided detection (CAD) marks on individual polyps on both the supine and prone scans leads to improved polyp detection by radiologists compared to the display of CAD marks on individual polyps on either the supine or the prone scan, but not both. Materials and Methods The acquisition of patient data for this study was approved by the Institutional Review Board and was Health Insurance Portability and Accountability Act–compliant. Subsequently, the use of the data was declared exempt from further institutional review board review. Four radiologists interpreted 33 computed tomography colonography cases, 21 of which had one adenoma 6–9 mm in size, with the assistance of a CAD system in the first reader mode (ie, the radiologists reviewed only the CAD marks). The radiologists were shown each case twice, with different sets of CAD marks for each of the two readings. In one reading, a true-positive CAD mark for the same polyp was displayed on both the supine and prone scans (a double-mark reading). In the other reading, a true-positive CAD mark was displayed either on the supine or prone scan, but not both (a single-mark reading). True-positive marks were randomized between readings and there was at least a 1-month delay between readings to minimize recall bias. Sensitivity and specificity were determined and receiver operating characteristic (ROC) and multiple-reader multiple-case analyses were performed. Results The average per polyp sensitivities were 60% (38%–81%) versus 71% (52%–91%) (P = .03) for single-mark and double-mark readings, respectively. The areas (95% confidence intervals) under the ROC curves were 0.76 (0.62–0.88) and 0.79 (0.58–0.96), respectively (P = NS). Specificities were similar for the single-mark compared with the double-mark readings. Conclusion The display of CAD marks on a polyp on both the supine and prone scans led to more frequent detection of polyps by radiologists without adversely affecting specificity for detecting 6–9 mm adenomas.

Rationale and Objectives New ways to understand biology as well as increasing interest in personalized treatments requires new capabilities for the assessment of therapy response. The lack of consensus methods and qualification evidence needed for large-scale multicenter trials, and in turn the standardization that allows them, are widely acknowledged to be the limiting factor in the deployment of qualified imaging biomarkers. Materials and Methods The Quantitative Imaging Biomarker Alliance is organized to establish a methodology whereby multiple stakeholders collaborate. It has charged the Volumetric Computed Tomography (CT) Technical Subcommittee with investigating the technical feasibility and clinical value of quantifying changes over time in either volume or other parameters as biomarkers. The group selected solid tumors of the chest in subjects with lung cancer as its first case in point. Success is defined as sufficiently rigorous improvements in CT-based outcome measures to allow individual patients in clinical settings to switch treatments sooner if they are no longer responding to their current regimens, and reduce the costs of evaluating investigational new drugs to treat lung cancer. Results The team has completed a systems engineering analysis, has begun a roadmap of experimental groundwork, documented profile claims and protocols, and documented a process for imaging biomarker qualification as a general paradigm for qualifying other imaging biomarkers as well. Conclusion This report addresses a procedural template for the qualification of quantitative imaging biomarkers. This mechanism is cost-effective for stakeholders while simultaneously advancing the public health by promoting the use of measures that prove effective.

RATIONALE AND OBJECTIVES: This article describes issues and methods that are specific to the measurement of change in tumor volume as measured from computed tomographic (CT) images and how these would relate to the establishment of CT tumor volumetrics as a biomarker of patient response to therapy. The primary focus is on the measurement of lung tumors, but the approach should be generalizable to other anatomic regions. MATERIALS AND METHODS: The first issues addressed are the various sources of bias and variance in the measurement of tumor volumes, which are discussed in the context of measurement variation and its impact on the early detection of response to therapy. RESULTS AND RESOURCES: Research that seeks to identify the magnitude of some of these sources of error is ongoing, and several of these efforts are described herein. In addition, several resources for these investigations are being made available through the National Institutes of Health-funded Reference Image Database to Evaluate Response to therapy in cancer project, and these are described as well. Other measures derived from CT image data that might be predictive of patient response are described briefly, as well as the additional issues that each of these metrics may encounter in real-life applications. CONCLUSIONS: The article concludes with a brief discussion of moving from the assessment of measurement variation to the steps necessary to establish the efficacy of a metric as a biomarker for response.

Purpose: In computed tomographic colonography (CTC), a patient will be scanned twice—Once supine and once prone—to improve the sensitivity for polyp detection. To assist radiologists in CTC reading, in this paper we propose an automated method for colon registration from supine and prone CTC scans. Methods: We propose a new colon centerline registration method for prone and supine CTC scans using correlation optimized warping (COW) and canonical correlation analysis (CCA) based on the anatomical structure of the colon. Four anatomical salient points on the colon are first automatically distinguished. Then correlation optimized warping is applied to the segments defined by the anatomical landmarks to improve the global registration based on local correlation of segments. The COW method was modified by embedding canonical correlation analysis to allow multiple features along the colon centerline to be used in our implementation. Results: We tested the COW algorithm on a CTC data set of 39 patients with 39 polyps (19 training and 20 test cases) to verify the effectiveness of the proposed COW registration method. Experimental results on the test set show that the COW method significantly reduces the average estimation error in a polyp location between supine and prone scans by 67.6%, from 46.27±52.97 to 14.98 mm±11.41 mm, compared to the normalized distance along the colon centerline algorithm (p

Rationale and Objectives The factors that influence the conspicuity of polyps on computed tomographic (CT) colonography (CTC) are poorly understood. The aim of this study is to compare radiologists' visual assessment of polyp conspicuity to quantitative image features and show the relationship between visual conspicuity and the detection of colonic polyps by computer-aided detection (CAD) on CTC. Methods One polyp (size range 6—10 mm) was selected from the CTC examination of each of 29 patients from a larger cohort. All patients underwent oral contrast-enhanced CTC with same-day optical colonoscopy with segmental unblinding. The polyps were analyzed by a previously validated CAD system and placed into one of two groups (detected [n = 12] or not detected [n = 17] by CAD). The study population was intentionally enriched with polyps that were not detected by the CAD system. Four board-certified radiologists, blinded to the CAD results, reviewed two- and three-dimensional CTC images of the polyps and scored the conspicuity of the polyps using a 4-point scale (0 = least conspicuous, 3 = most conspicuous). Polyp height and width were measured by a trained observer. A t-test (two-tailed, unpaired equal variance) was done to determine statistical significance. Intra- and interobserver variabilities of the conspicuity scores were assessed using the weighted κ test. Regression analysis was used to investigate the relationship of conspicuity to polyp height and width. Results A statistically significant difference was found between the average conspicuity scores for polyps that were detected by CAD compared to those that were not (2.3 ± 0.6 vs. 1.4 ± 0.8) (P = .004). There was moderate intraobserver agreement of the conspicuity scores (weighted κ 0.57 ± 0.09). Interobserver agreement was fair (average weighted κ for six pair-wise comparisons, 0.38 ± 0.15). Conspicuity was correlated with manual measurement of polyp height (r2 = 0.38–0.56, P Conclusions This CAD system tends to detect 6—10 mm polyps that are more visually conspicuous. Polyp height is a major determinant of visual conspicuity. The generalizability of these findings to other CAD systems is currently unknown. Nevertheless, CAD developers may need to specifically target flatter and less conspicuous polyps for CAD to better assist the radiologist to find polyps in this clinically important size category.

To evaluate the effect of computer-aided detection (CAD) as second reader on radiologists' diagnostic performance in interpreting computed tomographic (CT) colonographic examinations by using a primary two-dimensional (2D) approach, with segmental, unblinded optical colonoscopy as the reference standard.This HIPAA-compliant study was IRB-approved with written informed consent. Four board-certified radiologists analyzed 60 CT examinations with a commercially available review system. Two-dimensional transverse views were used for initial polyp detection, while three-dimensional (3D) endoluminal and 2D multiplanar views were available for problem solving. After initial review without CAD, the reader was shown CAD-identified polyp candidates. The readers were then allowed to add to or modify their original diagnoses. Polyp location, CT Colonography Reporting and Data System categorization, and reader confidence as to the likelihood of a candidate being a polyp were recorded before and after CAD reading. The area under the receiver operating characteristic (ROC) curve (AUC), sensitivity, and specificity were estimated for CT examinations with and without CAD readings by using multireader multicase analysis.Use of CAD led to nonsignificant average reader AUC increases of 0.03, 0.03, and 0.04 for patients with adenomatous polyps 6 mm or larger, 6-9 mm, and 10 mm or larger, respectively (Por = .25); likewise, CAD increased average reader sensitivity by 0.15, 0.16, and 0.14 for those respective groups, with a corresponding decrease in specificity of 0.14. These changes achieved significance for the 6 mm or larger group (P.01), 6-9 mm group (P.02), and for specificity (P.01), but not for the 10 mm or larger group (P.16). The average reading time was 5.1 minutes +/- 3.4 (standard deviation) without CAD. CAD added an average of 3.1 minutes +/- 4.3 (62%) to each reading (supine and prone positions combined); average total reading time, 8.2 minutes +/- 5.8.Use of CAD led to a significant increase in sensitivity for detecting polyps in the 6 mm or larger and 6-9 mm groups at the expense of a similar significant reduction in specificity.

This HIPAA-compliant study, with institutional review board approval and informed patient consent, was conducted to retrospectively develop a teniae coli-based circumferential localization method for guiding virtual colon navigation and colonic polyp registration. Colonic surfaces (n = 72) were depicted at computed tomographic (CT) colonography performed in 36 patients (26 men, 10 women; age range, 47-72 years) in the supine and prone positions. For 70 (97%) colonic surfaces, the tenia omentalis (TO), the most visible of the three teniae coli on a well-distended colonic surface, was manually extracted from the cecum to the descending colon. By virtually dissecting and flattening the colon along the TO, the authors developed a localization system involving 12 grid lines to estimate the circumferential positions of polyps. A sessile polyp would most likely (at 95% confidence level) be found within +/-1.2 grid lines (one grid line equals 1/12 the circumference) with use of the proposed method. By orienting and positioning the virtual cameras with use of the new localization system, synchronized prone and supine navigation was achieved. The teniae coli are extractable landmarks, and the teniae coli-based circumferential localization system helps guide virtual navigation and polyp registration at CT colonography.

In computer-aided diagnosis (CAD) systems for prostate cancer, dynamic contrast enhanced (DCE) magnetic resonance imaging is useful for distinguishing cancerous and benign tissue. The Tofts physiological model is a commonly used representation of the DCE image data, but the parameters require extensive computation. Hence, we developed an alternative representation based on the Hilbert transform of the DCE images. The time maximum of the Hilbert transform, a binary metric of early enhancement, and a pre-DCE value was assigned to each voxel and appended to a standard feature set derived from T2-weighted images and apparent diffusion coefficient maps. A cohort of 40 patients was used for training the classifier, and 20 patients were used for testing. The AUC was calculated by pooling the voxel-wise prediction values and comparing with the ground truth. The resulting AUC of 0.92 (95% CI [0.87 0.97]) is not significantly different from an AUC calculated using Tofts physiological models of 0.92 (95% CI [0.87 0.97]), as validated by a Wilcoxon signed rank test on each patient’s AUC (p = 0.19). The time required for calculation and feature extraction is 11.39 seconds (95% CI [10.95 11.82]) per patient using the Hilbert-based feature set, two orders of magnitude faster than the 1319 seconds (95% CI [1233 1404]) required for the Tofts parameter-based feature set (p

The availability of large medical image datasets is critical in training and testing of computer aided diagnosis (CAD) systems. However, collection of data and establishment of ground truth for medical images are both costly and difficult. To address this problem, we have developed an image composition tool that allows users to modify or supplement existing datasets by seamlessly inserting a clinical lesion extracted from a source image into a different location on a target image. In this study we focus on the application of this tool to the training of a CAD system designed to detect pulmonary nodules in chest CT. To compare the performance of a CAD system without and with the use of our image composition tool, we trained the system on two sets of data. The first training set was obtained from original CT cases, while the second set consisted of the first set plus nodules in the first set inserted into new locations. We then compared the performance of the two CAD systems in differentiating nodules from normal areas by testing each trained system against a fixed dataset containing natural nodules, and using the area under the ROC curve (AUC) as the figure of merit. The performance of the system trained with the augmented dataset was found to be significantly better than that trained with the original dataset under several training scenarios.