Search

Performance changes in a binary environment when using additional information is affected only when changes in recommendations are made due to the additional information in question. In a recent study, we have shown that, contrary to general expectation, introducing prior examinations improved recall rates, but not sensitivity. In this study, we assessed cancer detection differences when prior examinations and/or digital breast tomosynthesis (DBT) were made available to the radiologist. We identified a subset of 21 cancer cases with differences in the number of radiologists who recalled these cases after reviewing either a prior examination or DBT. For the cases with differences in recommendations after viewing either priors or DBT, separately, we evaluated the total number of readers that changed their recommendations, regardless of the specific radiologist in question. Confidence intervals for the number of readers and a test for the hypothesis of no difference was performed using the non-parameteric bootstrap approach addressing both case and reader-related sources of variability by resampling cases and readers. With the addition of priors, there were 14 cancer cases (out of 15) where the number of “recalling radiologists” decreased. With the addition of DBT, the number of “recalling radiologists” decreased in only five cases (out of 15) while increasing in the remaining 9 cases. Unlike most new approaches to breast imaging DBT seems to improve both recall rates and cancer detection rates. Changes in recommendations were noted by all radiologists for all cancers by type, size, and breast density.

Breast magnetic resonance imaging (MRI) is recommended as an adjunct to mammography for women who are considered at elevated risk of developing breast cancer. As a key component of breast MRI, dynamic contrast-enhanced MRI (DCE-MRI) uses a contrast agent to provide high intensity contrast between breast tissues, making it sensitive to tissue composition and vascularity. Breast DCE-MRI characterizes certain physiologic properties of breast tissue that are potentially related to breast cancer risk. Studies have shown that increased background parenchymal enhancement (BPE), which is the contrast enhancement occurring in normal cancer-unaffected breast tissues in post-contrast sequences, predicts increased breast cancer risk. Signal enhancement ratio (SER) computed from pre-contrast and post-contrast sequences in DCE-MRI measures change in signal intensity due to contrast uptake over time and is a measure of contrast enhancement kinetics. SER quantified in breast tumor has been shown potential as a biomarker for characterizing tumor response to treatments. In this work we investigated the relationship between quantitative measures of SER and breast cancer risk. A pilot retrospective case-control study was performed using a cohort of 102 women, consisting of 51 women who had diagnosed with unilateral breast cancer and 51 matched controls (by age and MRI date) with a unilateral biopsy-proven benign lesion. SER was quantified using fully-automated computerized algorithms and three SER-derived quantitative volume measures were compared between the cancer cases and controls using logistic regression analysis. Our preliminary results showed that SER is associated with breast cancer risk, after adjustment for the Breast Imaging Reporting and Data System (BI-RADS)-based mammographic breast density measures. This pilot study indicated that SER has potential for use as a risk factor for breast cancer risk assessment in women at elevated risk of developing breast cancer.

Full-field digital mammography (FFDM) is the most common screening procedure for detecting early breast cancer. However, due to complications such as overlapping breast tissue in projection images, the efficacy of FFDM reading is reduced. Recent studies have shown that digital breast tomosynthesis (DBT), in combination with FFDM, increases detection sensitivity considerably while decreasing false-positive, recall rates. There is a huge interest in creating diagnostically accurate 2-D interpretations from the DBT slices. Most of the 2-D syntheses rely on visualizing the maximum intensities (brightness) from each slice through different methods. We propose a wavelet based fusion method, where we focus on preserving holistic information from larger structures such as masses while adding high frequency information that is relevant and helpful for diagnosis. This method enables the spatial generation of a 2D image from a series of DBT images, each of which contains both smooth and coarse structures distributed in the wavelet domain. We believe that the wavelet-synthesized images, generated from their DBT image datasets, provide radiologists with improved lesion and micro-calcification conspicuity as compared with FFDM images. The potential impact of this fusion method is (1) Conception of a device-independent, data-driven modality that increases the conspicuity of lesions, thereby facilitating early detection and potentially reducing recall rates; (2) Reduction of the accompanying radiation dose to the patient.

In this paper we present a simple technique that can be employed to filter the output of the computerized mass detection schemes. The sensitivity of computer-aided detection (CAD) systems is high; nevertheless specificity is not due to high false positive (FP) detection rates. Our approach is based on Histogram of Oriented Gradients (HOG) descriptor for filtering the mass and normal tissue regions. After the descriptors are computed, Support Vector Machines (SVM) are applied to classify the identified masses. The devised technique was tested on 1881 regions of interest (ROIs) acquired using a previously proposed CAD system. Extensive simulations are conducted to illustrate the capacity of the HOG descriptor to improve the performances of mass detection systems.

This study was motivated by anecdotal reports from our clinicians that the lung parenchyma appears “different” (more heterogeneous) in asthmatics compared to non-asthmatics. We investigated whether traditional texture features were different between severe asthmatics and non-asthmatics. CT examinations from 76 subjects classified as “severe asthma” ( n = 51) and “normal control” ( n = 25) based on Severe Asthma Research Program (SARP) criteria were used in this study. The CT exams were performed on a 64-detector or 16-detector GE scanner at a radiation exposure of 96.6 (±30.7) mAs with the subjects holding their breath at end-normal-expiration (functional residual capacity). The CT images were reconstructed at 0.625 or 1.25 mm thickness using either GE’s “standard” or “detail” kernels. Air trapping was computed as the percentage of voxels with a value less than -856 HU. Gray level co-occurrence matrices (GLCM) were computed from the CT images, and 15 Haralick texture descriptors were computed from the GLCM. Air trapping was significantly greater in the severe asthma subjects compared to the normal control subjects. Seven of the 15 texture features were significantly different between the severe asthma and normal control subjects. Our findings provide some validity to anecdotal reports of differences between the parenchyma of asthmatic and non-asthmatics. The significant texture features may ultimately be used to classify individuals as asthmatic or non-asthmatic, which should improve the limited performance of air trapping alone.

In this study we present a computational method of CT examination classification into visual assessed emphysema severity. The visual severity categories ranged from 0 to 5 and were rated by an experienced radiologist. The six categories were none, trace, mild, moderate, severe and very severe. Lung segmentation was performed for every input image and all image features are extracted from the segmented lung only. We adopted a two-level feature representation method for the classification. Five gray level distribution statistics, six gray level co-occurrence matrix (GLCM), and eleven gray level run-length (GLRL) features were computed for each CT image depicted segment lung. Then we used wavelets decomposition to obtain the low- and high-frequency components of the input image, and again extract from the lung region six GLCM features and eleven GLRL features. Therefore our feature vector length is 56. The CT examinations were classified using the support vector machine (SVM) and k-nearest neighbors (KNN) and the traditional threshold (density mask) approach. The SVM classifier had the highest classification performance of all the methods with an overall sensitivity of 54.4% and a 69.6% sensitivity to discriminate "no" and "trace visually assessed emphysema. We believe this work may lead to an automated, objective method to categorically classify emphysema severity on CT exam.

The FDA recently approved Digital Breast Tomosynthesis (DBT) for use in screening for the early detection of breast cancer. However, MQSA qualification for interpreting DBT through training was noted as important. Performance issues related to training are largely unknown. Therefore, we assembled a unique computerized training module to assess radiologists' performances before and after using the training module. Seventy-one actual baseline mammograms (no priors) with FFDM and DBT images were assembled to be read before and after training with the developed module. Fifty examinations of FFDM and DBT images enriched with positive findings were assembled for the training module. Depicted findings were carefully reviewed, summarized, and entered into a specially designed training database where findings were identified by case number and synchronized to the display of the related FFDM plus DBT examinations on a clinical workstation. Readers reported any findings using screening BIRADS (0, 1, or 2) followed by instantaneous feedback of the verified truth. Six radiologists participated in the study and reader average sensitivity and specificity were compared before and after training. Average sensitivity improved and specificity remained relatively the same after training. Performance changes may be affected by disease prevalence in the training set.

The incidence of thyroid cancer is rising faster than other malignancies and has nearly doubled in the United States (U.S.) in the last 30 years. However, classifying between malignant and benign thyroid nodules is often difficult. Although ultrasound guided Fine Needle Aspiration Biopsy (FNAB) is considered an excellent tool for triaging patients, up to 25% of FNABs are inconclusive. As a result, definitive diagnosis requires an exploratory surgery and a large number of these are performed in the U.S. annually. It would be extremely beneficial to develop a non-invasive tool or procedure that could assist in assessing the likelihood of malignancy of otherwise indeterminate thyroid nodules, thereby reducing the number of exploratory thyroidectomies that are performed under general anesthesia. In this preliminary study we demonstrate a unique hand-held Resonance-frequency based Electrical Impedance Spectroscopy (REIS) device with six pairs of detection probes to detect and classify thyroid nodules using multi-channel EIS output signal sweeps. Under an Institutional Review Board (IRB)-approved case collection protocol, this REIS device is being tested in our clinical facility and we have been collecting an initial patient data set since March of this year. Between March and August of 2011, 65 EIS tests were conducted on 65 patients. Among these cases, six depicted pathology-verified malignant cells. Our initial assessment indicates the feasibility of easily applying this REIS device and measurement approach in a very busy clinical setting. The measured resonance frequency differences between malignant and benign nodules could potentially make it possible to accurately classify indeterminate thyroid nodules.

In order to establish a personalized breast cancer screening program, it is important to develop risk models that have high discriminatory power in predicting the likelihood of a woman developing an imaging detectable breast cancer in near-term (e.g.

The incidence of thyroid cancer has risen faster than many malignancies and has nearly doubled in the USA over the past 30 years. Palpable nodules and subclinical nodules detected by imaging are found in a large percentage of the USA population. Most of these (.>95%) are fortunately benign. This vast reservoir of nodules makes the detection and diagnosis of thyroid cancer a diagnostic dilemma. Ultrasound guided Fine Needle Aspiration Biopsy (FNAB) is excellent for triaging patients but up to 25% of FNABs are inconclusive. As a result, definitive diagnosis is often only possible with a diagnostic lobectomy; many thousands of these are performed in the USA annually for ultimately benign disease. It would be extremely beneficial if we could develop a non-invasive procedure that could assist the diagnostician in reliably predicting the likelihood of malignancy of otherwise indeterminate thyroid nodules, thereby reducing the number of these "exploratory/diagnostic" lobectomies performed under general anesthesia. Electrical Impedance Spectroscopy (EIS) was considered as a possible approach to address this problem. However, the diagnostic accuracy of EIS is too low for routine clinical use to date. In our group, we developed a substantially modified technology termed Resonance-frequency Electrical Impedance Spectroscopy (REIS), which yields usable information for classifying risk of having breast abnormalities. We preliminarily applied REIS to measure signals on participants having thyroid nodules aiming to assess whether we can assist in improving diagnosis of indeterminate thyroid nodules. In this study we present a new multi-probe based REIS device specifically designed for the assessment of indeterminate thyroid nodules. Our preliminary assessment presented here demonstrates the feasibility of using this proposed REIS device in a busy tertiary care center.

We have developed a multi-probe resonance-frequency electrical impedance spectroscope (REIS) system to detect breast abnormalities. Based on assessing asymmetry in REIS signals acquired between left and right breasts, we developed several machine learning classifiers to classify younger women (i.e., under 50YO) into two groups of having high and low risk for developing breast cancer. In this study, we investigated a new method to optimize performance based on the area under a selected partial receiver operating characteristic (ROC) curve when optimizing an artificial neural network (ANN), and tested whether it could improve classification performance. From an ongoing prospective study, we selected a dataset of 174 cases for whom we have both REIS signals and diagnostic status verification. The dataset includes 66 “positive” cases recommended for biopsy due to detection of highly suspicious breast lesions and 108 “negative” cases determined by imaging based examinations. A set of REIS-based feature differences, extracted from the two breasts using a mirror-matched approach, was computed and constituted an initial feature pool. Using a leave-one-case-out cross-validation method, we applied a genetic algorithm (GA) to train the ANN with an optimal subset of features. Two optimization criteria were separately used in GA optimizat ion, namely the area under th e entire ROC curve (AUC) and the partial area under the ROC curve, up to a predetermined threshold (i.e., • 90% specificity). The results showed that although the ANN optimized using the entire AUC yielded higher overall performance (AUC = 0.83 versus 0.76), the ANN optimized using the partial ROC area criterion achieved substantially higher operational performance (i.e., increasing sensitivity level from 28% to 48% at 95% specificity and/ or from 48% to 58% at 90% specificity). Keywords: Electrical impedance spectroscope (EIS), breast cancer, artificial neural network, technology assessment, ROC, AUC, partial AUC.

After developing a multi-probe resonance-frequency electrical impedance spectroscopy (REIS) system aimed at detecting women with breast abnormalities that may indicate a developing breast cancer, we have been conducting a prospective clinical study to explore the feasibility of applying this REIS system to classify younger women (< 50 years old) into two groups of "higher-than-average risk" and "average risk" of having or developing breast cancer. The system comprises one central probe placed in contact with the nipple, and six additional probes uniformly distributed along an outside circle to be placed in contact with six points on the outer breast skin surface. In this preliminary study, we selected an initial set of 174 examinations on participants that have completed REIS examinations and have clinical status verification. Among these, 66 examinations were recommended for biopsy due to findings of a highly suspicious breast lesion ("positives"), and 108 were determined as negative during imaging based procedures ("negatives"). A set of REIS-based features, extracted using a mirror-matched approach, was computed and fed into five machine learning classifiers. A genetic algorithm was used to select an optimal subset of features for each of the five classifiers. Three fusion rules, namely sum rule, weighted sum rule and weighted median rule, were used to combine the results of the classifiers. Performance evaluation was performed using a leave-one-case-out cross-validation method. The results indicated that REIS may provide a new technology to identify younger women with higher than average risk of having or developing breast cancer. Furthermore, it was shown that fusion rule, such as a weighted median fusion rule and a weighted sum fusion rule may improve performance as compared with the highest performing single classifier.

Airways tree segmentation is an important step in quantitatively assessing the severity of and changes in several lung diseases such as chronic obstructive pulmonary disease (COPD), asthma, and cystic fibrosis. It can also be used in guiding bronchoscopy. The purpose of this study is to develop an automated scheme for segmenting the airways tree structure depicted on chest CT examinations. After lung volume segmentation, the scheme defines the first cylinder-like volume of interest (VOI) using a series of images depicting the trachea. The scheme then iteratively defines and adds subsequent VOIs using a region growing algorithm combined with adaptively determined thresholds in order to trace possible sections of airways located inside the combined VOI in question. The airway tree segmentation process is automatically terminated after the scheme assesses all defined VOIs in the iteratively assembled VOI list. In this preliminary study, ten CT examinations with 1.25mm section thickness and two different CT image reconstruction kernels ("bone" and "standard") were selected and used to test the proposed airways tree segmentation scheme. The experiment results showed that (1) adopting this approach affectively prevented the scheme from infiltrating into the parenchyma, (2) the proposed method reasonably accurately segmented the airways trees with lower false positive identification rate as compared with other previously reported schemes that are based on 2-D image segmentation and data analyses, and (3) the proposed adaptive, iterative threshold selection method for the region growing step in each identified VOI enables the scheme to segment the airways trees reliably to the 4th generation in this limited dataset with successful segmentation up to the 5th generation in a fraction of the airways tree branches.

In our previous study, we reported on the development and preliminary testing of a prototype resonance electrical impedance spectroscopy (REIS) system with a pair of probes. Although our pilot study on 150 young women ranging from 30 to 50 years old indicated the feasibility of using REIS output sweep signals to classify between the women who had negative examinations and those who would ultimately be recommended for biopsy, the detection sensitivity was relatively low. To improve performance when using REIS technology, we recently developed a new multi-probe based REIS system. The system consists of a sensor module box that can be easily lifted along a vertical support device to fit women of different height. Two user selectable breast placement "cups" with different curvatures are included in the system. Seven probes are mounted on each of the cups on opposing sides of the sensor box. By rotating the sensor box, the technologist can select the detection sensor cup that better fits the breast size of the woman being examined. One probe is mounted in the cup center for direct contact with the nipple and the other six probes are uniformly distributed along an outside circle to enable contact with six points on the outer and inner breast skin surfaces. The outer probes are located at a distance of 60mm away from the center (nipple) probe. The system automatically monitors the quality of the contact between the breast surface and each of the seven probes and data acquisition can only be initiated when adequate contact is confirmed. The measurement time for each breast is approximately 15 seconds during which time the system records 121 REIS signal sweep outputs generated from 200 KHz to 800 KHz at 5 KHz increments for all preselected probe pairs. Currently we are measuring 6 pairs between the center probe and each of six probes located on the outer circle as well as two pairs between probe pairs on the outer circle. This new REIS system has been installed in our clinical breast imaging facility. We are conducting a prospective study to assess performance when using this REIS system under an approved IRB protocol. Over 200 examinations have been conducted to date. Our experience showed that this new REIS system was easy to operate and the REIS examination was fast and considered "comfortable" by examinees since the women presses her breast into the cup herself without any need for forced breast compression, and all but a few highly sensitive women have any sensation of an electrical current during the measurement.

Digital breast tomosynthesis (DBT) has emerged as a promising imaging modality for screening mammography. However, visually detecting micro-calcification clusters depicted on DBT images is a difficult task. Computer-aided detection (CAD) schemes for detecting micro-calcification clusters depicted on mammograms can achieve high performance and the use of CAD results can assist radiologists in detecting subtle micro-calcification clusters. In this study, we compared the performance of an available 2D based CAD scheme with one that includes a new grouping and scoring method when applied to both projection and reconstructed DBT images. We selected a dataset involving 96 DBT examinations acquired on 45 women. Each DBT image set included 11 low dose projection images and a varying number of reconstructed image slices ranging from18 to 87. In this dataset 20 true-positive micro-calcification clusters were visually det ected on the projection images and 40 were visually detected on the reconstructed images, respectively. We first applied the CAD scheme that was previously developed in our laboratory to the DBT dataset. We then tested a new grouping method that defines an independent cluster by grouping the same cluster detected on different projection or reconstructed images. We then compared four scoring methods to assess the CAD performance. The maximum sensitivity level observed for the different grouping and scoring methods were 70% and 88% for the projection and reconstructed images with a maxi mum false-positive rate of 4.0 and 15.9 per examination, respectively. This preliminary study demonstrates that (1) among the maximum, the minimum or the average CAD generated scores, using the maximum score of the grouped cluster regions achieved the highest performance level, (2) the histogram based scoring method is r easonably effective in reducing false-pos itive detections on the projection images but the overall CAD sensitivity is lower due to lower signal- to-noise ratio, and (3) CAD achieved higher sensitivity and higher false-positive rate (per examination) on the reconstructed images. We concluded that without changing the detection threshold or performing pre-filtering to possibly increase detection sensitivity, current CAD schemes developed and optimized for 2D mammograms perform relatively poorly and need to be re-optimized using DBT datasets and new grouping and scoring methods need to be inco rporated into the schemes if these are to be used on the DBT examinations

In this study we randomly select 250 malignant and 250 benign mass regions as a training dataset. The boundary contours of these regions were manually identified and marked. Twelve image features were computed for each region. An artificial neural network (ANN) was trained as a classifier. To select a specific testing dataset, we applied a topographic multi-layer region growth algorithm to detect boundary contours of 1,903 mass regions in an initial pool of testing regions. All processed regions are sorted based on a size difference ratio between manual and automated segmentation. We selected a testing dataset involving 250 malignant and 250 benign mass regions with larger size difference ratios. Using the area under ROC curve (A Z value) as performance index we investigated the relationship between the accuracy of mass segmentation and the performance of a computer-aided diagnosis (CAD) scheme. CAD performance degrades as the size difference ratio increases. Then, we developed and tested a hybrid region growth algorithm that combined the topographic region growth with an active contour approach. In this hybrid algorithm, the boundary contour detected by the topographic region growth is used as the initial contour of the active contour algorithm. The algorithm iteratively searches for the optimal region boundaries. A CAD likelihood score of the growth region being a true-positive mass is computed in each iteration. The region growth is automatically terminated once the first maximum CAD score is reached. This hybrid region growth algorithm reduces the size difference ratios between two areas segmented automatically and manually to less than ±15% for all testing regions and the testing A Z value increases to from 0.63 to 0.90. The results indicate that CAD performance heavily depends on the accuracy of mass segmentation. In order to achieve robust CAD performance, reducing lesion segmentation error is important.

Using electrical impedance spectroscopy (EIS) technology to detect breast abnormalities in general and cancer in particular has been attracting research interests for decades. Large clinical te sts suggest that current EIS systems can achieve high specificity ( • 90%) at a relatively low sensitivity ranging from 15% to 35%. In this study, we explore a new resonance frequency based electrical impedance spectroscopy (REIS) technology to measure breast tissue EIS signals in vivo, which aims to be more sensitive to small tissue changes. Through collaboration between our imaging research group and a commercial company, a unique prototype REIS system has been assembled and preliminary signal acquisition has commenced. This REIS system has two detec tion probes mounted in the two ends of a Y-shape support device with probe separation of 60 mm. During REIS measurement, one probe touches the nipple and the other touches to an outer point of the breast. The electronic system con tinuously generates sweeps of mu lti-frequency electrical pulses ranging from 100 to 4100 kHz. The maximum electric voltage and the current applied to the probes are 1.5V and 30mA, respectively. Once a “record” command is entered, multi-frequency sweeps are recorded every 12 seconds until the program receives a “stop recording” command. In our imagin g center, we have collected REIS measurements from 150 women under an IRB approved protocol. The database includes 58 biopsy cases, 78 screening negative cases, and other “recalled” cases (for additional imaging pr ocedures). We measured eight signal f eatures from the effective REIS sweep of each breast. We applied a multi-feature based artificial neural network (ANN) to classify between “biopsy” and normal “non-biopsy” breasts. The ANN performance is evaluated using a leave-one-out validation method and ROC analysis. We conducted two experiments. The first experiment attempted to classify 58 “biopsy” breasts and 58 “non-biopsy” breasts acquired on 58 women each having one breast recommended for biopsy. The second experiment attempted to classify 58 “biopsy” breasts and 58 negative breasts from the set of screening negative cases. The areas under ROC curves are 0.679 ± 0.033 and 0.606 ± 0.035 for the first and the second experiment, respectively. The preliminary results demonstrate (1) even with this rudimentary system with only one paired probes there is a measurable signal of changes in breast tissue demonstrating the feasibility of applying REIS technology for identifying at least some women with highly suspicious breast abnormalities and (2) the electromagnetic asymmetry between two breasts may be more sensitive in detecting changes in the abnormal breast. To further improve the REIS system performance, we are currently designing a new REIS system with mult iple electrical probes and a more sophisticated analysis scheme.

Chronic obstructive pulmonary disease may cause airway remodeling, and small airways are the mostly likely site of associated airway flow obstruction. Detecting and quantifying airways depicted on a typical computed tomography (CT) images is limited by spatial resolution. In this study, we examined the association between lung function and airway size. CT examinations and spirometry measurement of forced expiratory volume in one second as a percent predicted (FEV1%) from 240 subjects were used in this study. Airway sections depicted in axial CT section were automatically detected and quantified. Pearson correlation coefficients (PCC) were computed to compare lung function across three size categories: (1) all detected airways, (2) the smallest 50% of detected airways, and (3) the largest 50% of detected airways using the CORANOVA test. The mean number of all airways detected per subject was 117.4 (± 40.1) with mean size ranging from 20.2 to 50.0 mm2. The correlation between lung function (i.e., FEV1) and airway morphometry associated with airway remodeling and airflow obstruction (i.e., lumen perimeter and wall area as a percent of total airway area) was significantly stronger for smaller compared to larger airways (p < 0.05). The PCCs between FEV1 and all airways, the smallest 50%, and the largest 50% were 0.583, 0.617, 0.523, respectively, for lumen perimeter and -0.560, -0.584, and -0.514, respectively, for wall area percent. In conclusion, analyzing a set of smaller airways compared to larger airways may improve detection of an association between lung function and airway morphology change.

This study investigated the agreement between breast mass spiculation levels as rated subjectively by observers and a computer scheme. An image dataset with 1,263 mass regions was selected. First, three experienced observers independently rated subjectively the visualized spiculation levels of these mass regions and classified them into three categories (none/minimal, moderate, and severe/significant). We then developed a computerized scheme to detect mass margins and classify the spiculation levels of the suspected mass regions. The scheme applied a hybrid region growth algorithm to segment the mass regions. An edge map was computed inside a 30-pixel-wide band surrounding the mass boundary contour. The scheme then applied a threshold to convert the edge map into a binary image followed by labeling and detecting line orientation. In the original edge map the scheme computed the average local pixel value fluctuation. In the binary edge map, the scheme computed the ratio between the number of "spiculated" pixels and the number of total pixels inside the band. The scheme also computed mass region conspicuity using the original image. Using these three features, a Bayesian Belief Network (BBN) was built to classify mass regions into one of the three spiculation categories. We compared the inter-observer variation as well as agreement levels between the subjective and computerized ratings. Agreement rates between paired observers ranged from 41.3% to 58.8% (Kappa = 0.136 to 0.309). The agreement between the computer scheme and observers' average rating was 49.2% (Kappa = 0.218). This study demonstrated a large inter-observer variability in subjective rating of mass speculation levels as well as a large difference between the rating results of a computerized scheme and observers. As a result, in an Interactive Computer-Aided Diagnosis (ICAD) environment, CAD-selected reference regions may be considered "very similar" by some observers and "not similar" by others. Hence, improving the selection of actually visually similar reference regions by a computerized scheme remains an important yet unsolved task for ICAD development.

Many diagnostic problems involve the assessment of vascular structures or bronchial trees depicted in volumetric datasets, but previous algorithms for segmenting cylindrical structures are not sufficiently robust for them to be widely applied clinically. Local geometric information that is of importance in segmentation consists of voxel values and their first and second derivatives. First derivatives can be generalized to the gradient and more generally the structure tensor, while the second derivatives can be represented by Hessian matrices. It is desirable to exploit both kinds of information, at the same time, in any voxel classification process, but few segmentation algorithms have attempted to do this. This project compares segmentation based on the structure tensor to that based on the Hessian matrix, and attempts to determine whether some combination of the two can demonstrate better performance than either individually. To compare performance in a situation where a gold standard exists, the methods were tested on simulated tree structures. We generated 3D tree structures with varying amounts of added noise, and processed them with algorithms based on the structure tensor, the Hessian matrix, and a combination of the two. We applied an orientation-sensitive filter to smooth the tensor fields. The results suggest that the structure tensor by itself is more effective in detecting cylindrical structures than the Hessian tensor, and the combined tensor is better than either of the other tensors.

This study measured the accuracy of our airway quantification scheme using phantoms airway under different CT protocols. Airway remodeling is associated with several thoracic diseases (e.g., chronic bronchitis, asthma, and bronchiectasis), and, therefore, quantification of airway remodeling may have wide clinical application. Our scheme assigns pixels partial membership in the airway wall and lumen based on the pixel's HU value, which is intended to account for partial volume averaging inherent in CT image reconstruction. Twenty-four phantom airways with an outer diameter from 2.6 to 14.0 mm and wall thicknesses from 0.5 to 2.0 mm were analyzed. The absolute differences between measurements supplied by the manufacture and computed from CT images acquired at 40 mAs and reconstructed at 1.25 mm thickness using GE's "soft" and "lung" reconstruction kernels for lumen area ranged from 1.4% to 49.3% and 0.4% to 33.0%, respectively, and for wall area ranged from 0.3% to 118.0% and 2.1 to 92.9%, respectively. Accuracy typically improved as the kernel's spatial frequency increased. Airways whose wall thickness was close to the pixels dimensions were challenging to quantify. The partial membership assignment of our airway quantification accurately computed airway morphometry across a range of phantom airway sizes.

Although computer-aided detection (CAD) systems were designed and approved for and are assumed to be used as a "second-reader", namely radiologists are expected to interpret mammograms and detect suspected abnormalities (i.e., micro-calcification clusters and masses) independently before viewing CAD results, it is not clear whether radiologists in a busy clinical environment follow the intended use. In this study, we observed ten experienced radiologists during the clinical reading of 635 mammography examinations and recorded their workflow pattern in terms of the use of CAD. The observations suggest that for detecting micro-calcification clusters only a few radiologists actually used a magnifying glass to carefully and systematically scan all images. Areas in which no CAD cues were identified for micro-calcifications were largely discarded. The majority of radiologists used CAD for identification micro-calcifications clusters almost as a "pre-screening" tool. In less than 15% of cases with CAD cues for microcalcifications cluster the radiologists actually scanned the complete set of images for possible additional clusters. The majority of more careful searches were performed by only three radiologists who voluntarily admitted they knew they were an exception in regards to their reading style and admitted they personally believed they were also "slower". CAD marks of possible masses were often discarded by the majority of the radiologists in particular when cues appeared only on one view. We found that there was a large difference between the use of CAD for detection of micro-calcifications clusters and masses. In addition, radiologists frequently use CAD in a manner that is substantially different than originally intended.

Although CAD schemes can detect a high percentage of subtle cancers depicted on "prior" and false-negative cases, radiologists frequently discard most of CAD-cued subtle masses in the clinical environment. As a result of the relatively high false-positive detection rate and the fact that a large number of subtle masses are typically cued only on one view cause radiologists to rely less on (and often ignore) CAD-cued masses. In this study, we present a multi-view based method to increase the number of actual masses that are cued by the CAD scheme on both (ipso-lateral) views and at the same time limiting the overall "case-based" false-positive detection rate. The new scheme includes a traditional single-image based CAD scheme and a multi-view processing module. An image database from 435 examinations (or a total of 1,740 images) consisting 235 examinations depicting a verified malignant mass each and 200 negative examinations was used in this study. The single-image based CAD scheme with a fixed operating threshold (i.e. 0.55) was applied to all images. For each CAD identified region (i.e. with detection score ≥ 0.55), the multi-view processing module defined a matched strip on the corresponding ipso-lateral image and identified all "matched" regions located inside the strip (including those with detection score < 0.55). All matched regions are cued on both views and unmatched regions were discarded. CAD scheme initially detected 172 true masses and 576 false-positive regions. Of the 172 masses, 90 were detected on two views (52%) and 82 were detected only on one view. Of the 576 false-positive detections, only 72 pairs (14%) were considered "matched" and 432 were not, resulting in 504 case-based ("independent") cues. Case-based sensitivity and false-positive rate of the CAD scheme were 73.2% and 1.16 per case. When only matched region pairs were cued, 160 masses (68.1%) and 308 false-positive detections (0.71 per case) were identified on two views. Reducing the original operating threshold from 0.55 to 0.49 resulted in 172 masses and 413 false-positive detections (or 0.95 per case) being cued on two views. By increasing sensitivity through the reduction of the detection threshold only inside the matched strip regions in ipso-lateral images the scheme successfully cued masses on two views by maintaining detection sensitivity while reducing the "case-based" false-positive detection rate.

The purpose of the study was to decrease the variability of computed tomographic airway measurements. We to developed and evaluated a novel computer scheme to automatically segment airways depicted on chest CT examinations at the level of the lobar and segmental bronchi and to decrease. The computer scheme begins with manual selection of a seed point within the airway from which the airway wall and lumen are automatically segmented and airway pixels were assigned full or partial membership to the lumen or wall. Airway pixels not assigned full membership to the lumen ( 0 HU) were assigned partial membership to the lumen and wall. In fifteen subjects with no visible signs of emphysema and a range of pulmonary obstruction from none to severe, airway measures were compared to pulmonary function parameters in a rank order analysis to evaluate measuring a single airway versus multiple airways. The quality of the automated airway segmentation was visually acceptable. The Pearson Correlation coefficients for the ranking of FEV 1 versus wall area percent (percent of total airway size) and FVC versus wall area percent were 0.164 and 0.175 for a single measurement, respectively, and were 0.243 and 0.239 for multiple measurements, respectively. Our preliminary results suggest that averaging the measurements from multiple airways may improve the relation between airway measures and lung function compared to measurement from a single airway, which improve quantification of airway remodeling in COPD patients.

The fraction of lung voxels below a pixel value "cut-off" has been correlated with pathologic estimates of emphysema. We performed a "standard" quantitative CT (QCT) lung analysis using a -950 HU cut-off to determine the volume fraction of emphysema (below the cut-off) and a "corrected" QCT analysis after removing small group (5 and 10 pixels) of connected pixels ("blobs") below the cut-off. CT examinations two dataset of 15 subjects each with a range of visible emphysema and pulmonary obstruction were acquired at "low-dose and conventional dose reconstructed using a high-spatial frequency kernel at 2.5 mm section thickness for the same subject. The "blob" size (i.e., connected-pixels) removed was inversely related to the computed fraction of emphysema. The slopes of emphysema fraction versus blob size were 0.013, 0.009, and 0.005 for subjects with both no emphysema and no pulmonary obstruction, moderate emphysema and pulmonary obstruction, and severe emphysema and severe pulmonary obstruction, respectively. The slopes of emphysema fraction versus blob size were 0.008 and 0.006 for low-dose and conventional CT examinations, respectively. The small blobs of pixels removed are most likely CT image artifacts and do not represent actual emphysema. The magnitude of the blob correction was appropriately associated with COPD severity. The blob correction appears to be applicable to QCT analysis in low-dose and conventional CT exams.

Radiologists' performance reviewing and rating breast cancer screening mammography exams using a telemammography system was evaluated and compared with the actual clinical interpretations of the same interpretations. Mammography technologists from three remote imaging sites transmitted 245 exams to a central site (radiologists), which they (the technologists) believed needed additional procedures (termed "recall"). Current exam image data and non-image data (i.e., technologist's text message, technologist's graphic marks, patient's prior report, and Computer Aided Detection (CAD) results) were transmitted to the central site and displayed on three high-resolution, portrait monitors. Seven radiologists interpreted ("recall" or "no recall") the exams using the telemammography workstation in three separate multi-mode studies. The mean telemammography recall rates ranged from 72.3% to 82.5% while the actual clinical recall rates ranged from 38.4% to 42.3% across the three studies. Mean Kappa of agreement ranged from 0.102 to 0.213 and mean percent agreement ranged from 48.7% to 57.4% across the three studies. Eighty-seven percent of the disagreement interpretations occurred when the telemammography interpretation resulted in a recommendation to recall and the clinical interpretation resulted in a recommendation not to recall. The poor agreement between the telemammography and clinical interpretations may indicate a critical dependence on images from prior screening exams rather than any text based information. The technologists were sensitive, if not specific, to the mammography features and changes that may lead to recall. Using the telemammography system the radiologists were able to reduce the recommended recalls by the technologist by approximately 25 percent.© (2005) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

We developed a computer-aided diagnosis (CAD) scheme to detect and quantitatively assess interstitial lung diseases (ILD) depicted on low-dose and multi-slice helical high-resolution computed tomography (CT) examinations. Eighteen CT cases acquired from patients who underwent routine low-dose whole-lung screening examinations for the detection of lung cancer were used to test the scheme. ILD was identified in all of these cases. The CAD scheme involves multiple steps to segment lung areas, identify suspicious ILD regions depicted on each CT slice, and generate volumetric ILD lesions by grouping and matching ILD regions detected on multiple adjacent slices. The scheme computes five “global” features for each identified ILD region, which include size (or volume), contrast, average local pixel value fluctuation, mean of stochastic fractal dimension, and geometric fractal dimension. Two sets of classification rules are applied to remove false-positive detections. The severity of ILD in each case was rated by one experienced chest radiologist into one of the three categories (mild, moderate, and severe). A distance-weighted k-nearest neighbor algorithm and round-robin validation method was applied to classify each testing case into one of the three categories of severity. In this experiment, the CAD scheme classified 78% (14 out of 18) cases into the same categories as rated by the radiologist.

The widespread adoption of chest CT for lung cancer screening will greatly increase the workload of chest radiologists. Contributing to this effort is the need for radiologists to differentiate between localized nodules and slices through linear structures such as blood vessels, in each of a large number of slices acquired for each subject. To increase efficiency and accuracy, thin slices can be combined to provide thicker slabs for presentation, but the resulting superposition of tissues can make it more difficult to detect and characterize smaller nodules. The stereo display of a stack of thin CT slices may be able to clarify three-dimensional structures, while avoiding the loss of resolution and ambiguities due to tissue superposition. The current work focuses on the development and evaluation of stereo projection models that are appropriate for chest CT. As slices are combined into a three dimensional structure, maximum image intensity, which is limited by the display, must be preserved. But, compositing methods that effectively average slices together typically reduce contrast of subtle nodules. For monoscopic viewing, orthographic maximum-intensity projection (MIP), of thick slabs, has been employed to overcome this effect, but this method provides no information of depth or of the geometrical relationships between structures. Our comparison of various rendering options indicates that a stereographic perspective transformation, used in conjunction with a compositing model that combines maximum-intensity projection with an appropriate brightness weighting function, shows promise for this application. The main drawback uncovered was that, for the images used in this study, the lung volume was undersampled in the z-direction, resulting in certain unavoidable image artifacts.

This study was designed to evaluate radiologists’ ability to identify highly-compressed, digitized mammographic images displayed on high-resolution, monitors. Mammography films were digitized at 50 micron pixel dimensions using a high-resolution laser film digitizer. Image data were compressed using the irreversible (lossy), wavelet-based JPEG 2000 method. Twenty images were randomly presented in pairs (one image per monitor) in three modes: mode 1, non-compressed versus 50:1 compression; mode 2, non-compressed versus 75:1 compression; and mode 3, 50:1 versus 75:1 compression with 20 random pairs presented twice (80 pairs total). Six radiologists were forced to choose which image had the lower level of data compression in a two-alternative forced choice paradigm. The average percent correct across the six radiologists for modes 1, 2 and 3 were 52.5% (+/-11.3), 58.3% (+/-14.7), and 58.3% (+/-7.5), respectively. Intra-reader agreement ranged from 10 to 50% and Kappa from -0.78 to -0.19. Kappa for inter-reader agreement ranged from -0.47 to 0.37. The “monitor effect” (left/right) was of the same order of magnitude as the radiologists’ ability to identify the lower level of image compression. In this controlled evaluation, radiologists did not accurately discriminate non-compressed and highly-compressed images. Therefore, 75:1 image compression should be acceptable for review of digitized mammograms in a telemammography system.

Our goal was to develop an inexpensive, high-quality, multi-site telemammography system, implemented with low-level data connections that provided a communication link for an “almost real-time” response from a radiologist (central site) to remote “underserved” sites. The remote sites digitize mammographic films using high-resolution, laser digitizers. Images are automatically cropped, compressed (wavelet-based), and encrypted prior to transmission. At the central site images are decrypted, decompressed, unsharp masked, and displayed using automatically determined LUTs. The sites communicate instantly via a “chat box.” Remote sites 1, 2, and 3 are 15, 20, and 90 miles from the central site, respectively, and connected by POTS (sites 1 and 2) and LAN (site 3). Only minimal noticeable difference at compression levels of 50:1 and 75:1 could be identified unless magnified to extreme levels. Two experienced observers rated the LUTs for 200 images as “acceptable” to “excellent.” Average cycle times to digitize, transmit and receive cases (four films each) at 75:1 compression were 5.97, 6.85, and 5.77 min/case from sites 1, 2, and 3, respectively. Unique data-handling schemes significantly decrease the image file size and allow successful transmission in a reliable, timely manner. Over 1000 cases have been transmitted to date. Messaging was found to be easy to use.

We investigated a new approach to improve the performance of a computer-aided detection (CAD) scheme in identifying masses depicted on images acquired earlier ("prior"). The scheme was trained using a dataset with simulated mass features. From a database with images acquired during two consecutive examinations, 100 locations matched pairs of malignant mass regions were selected in both the “current” and the most recent “prior” images. While reviewing the current images, mass regions were identified and as a result biopsies were ultimately performed. Prior images were not identified as suspicious by radiologists during the original interpretation. The same number of false-positive regions was also selected in both current and prior images. The selected regions were then randomly divided into training and testing datasets with 50 true-positive and 50 false-positive regions in each. For each selected region, five features; area, contrast, circularity, normalized standard deviation of radial length, and conspicuity; were computed. The ratios of the average difference of five feature values between current and prior mass regions in the training datasets were also computed. Multiplying these ratios by the computed values in current mass regions, we generated a new dataset of simulated features of “prior” mass regions. Three artificial neural networks (ANN) were trained. ANN-1 and ANN-2 were trained using training datasets of current and prior regions, respectively. ANN-3 was trained using simulated “prior” dataset. The performance of three ANNs was then evaluated using the testing dataset of prior images. Areas under ROC curves (A z ) were 0.613 ± 0.026 for ANN-1, 0.678 ± 0.029 for ANN-2, and 0.667 ± 0.029 for ANN-3, respectively. This preliminary study demonstrated that one could estimate an average change of feature values over time and "adjust" CAD performance for better detection of masses at an earlier stage.

We developed and tested an automated scheme to segment lung areas depicted in CT images. The scheme includes a series of six steps. 1) Filtering and removing pixels outside the scanned anatomic structures. 2) Segmenting the potential lung areas using an adaptive threshold based on pixel value distribution in each CT slice. 3) Labeling all selected pixels ingo segmented regions and deleting isolated regions in non-lung area. 4) Labeling and filling interior cavities (e.g., pleural nodules, airway wall, and major blood vessels) inside lung areas. 5) Detecting and deleting the main airways (e.g., trachea and central bronchi) connected to the segmented lung areas. 6) Detecting and separating possible anterior or posterior junctions between the lungs. Five lung CT cases (7-10 mm in slice thickness) with variety of disease patterns were used to train or set up the classification rules in the scheme. Fifty examinations of emphysema patients were then used to test the scheme. The results were compared with the results generated from a semi-automated method with manual interaction by an expert observer. The experimental results showed that the average difference in estimated lung volumes between the automated scheme and manually corrected approach was 2.91%±0.88%. Visual examination of segmentation results indicated that the difference of the two methods was larger in the areas near the apices and the diaphragm. This preliminary study demonstrated that a simple multi-stage scheme had potential of eliminating the need for manual interaction during lunch segmentation. Hence, it can ultimately be integrated into computer schemes for quantitative analysis and diagnosis of lung diseases.

As the number of mammographic examinations increases, it becomes clear that in many underserved locations, there is a lack of expertise that is required for consistent, highly accurate, and timely diagnosis. Hence, mammograms are frequently sent to other medical facilities, and a significant fraction of women (typically 3 - 10%) are recalled for additional examinations. It is the purpose of this project to develop, test, and clinically evaluate a telemammography system that will operate between several remote locations and a large breast cancer center. In this manuscript we describe the design considerations, implementation, and initial testing that were undertaken, to date. The system digitizes a mammogram at 50 micrometers pixel size, compresses the resulting image file (approximately 75:1), and transmits it over a telephone line to the central site where the data received are decompressed and displayed on a high-resolution workstation in approximately 4 minutes per image. Initial testing of the system indicates that a relatively inexpensive system for 'almost real-time' telemammography can be employed in any geographic area that possesses standard telephone lines, and this approach to enhance communication may make it possible to offer better mammographic services at remote locations.© (2002) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

In this study, we test a new method to automatically search for matched regions in bilateral digitized mammograms and to compute differences in region conspicuities in pairs of matched regions. One hundred pairs of bilateral images of the same view were selected for the experiment. Each pair of images depicted one verified mass. These 100 mass regions, along with 356 suspicious but actually negative mass regions, were first detected by a single-image-based CAD scheme. To find the matched regions in the corresponding bilateral images, a Procrustean-type technique was used to register the two images, which corrects the deformation of tissue structure between images by guaranteeing the registration of nipples, skin lines, and chest walls. Then, a region growth algorithm was applied to generate a growth region in the matched area, which has the same effective size as the suspicious region in the abnormal image. The conspicuities in the two matched regions, as well as their differences, were computed. Using the conspicuity in the original mass regions and the difference of conspicuities in the two matched regions as two identification indices to classify this set of 456 suspicious regions, the computed areas under the ROC curves (A z ) were 0.77 and 0.75, respectively. This preliminary study indicates that by comparing the difference of conspicuities in two matched regions that a very useful feature for the CAD schemes can be extracted.

The purpose was to evaluate the effect of incorporating negative but suspicious regions into a knowledge-based computer-aided detection (CAD) scheme of masses depicted in mammograms. To determine if a suspicious region is positive for a mass, the region was compared not only with actually positive regions (masses), but also with known negative regions. A set of quantitative measures (i.e., a positive, a negative, and a combined likelihood measure) was computed. In addition, a process was developed to integrate two likelihood measures that were derived using two selected features. An initial evaluation with 300 positive and 300 negative regions was performed to determine the parameters associated with the likelihood measures. Then, an independent set of 500 positive and 500 negative regions was used to test the performance of the CAD scheme. During the training phase, the performance was improved from Az=0.83 to 0.87 with the incorporation of negative regions and the integration process. During the independent test, the performance was improved from Az=0.80 to 0.83. The incorporation of negative regions and the integration process was found to add information to the scheme. Hence, it may offer a relatively robust solution to differentiate masses from normal tissue in mammograms.

One way to increase the efficiency of large-scale observer performance studies that utilize receiver-operator- characteristic (ROC) analysis would be to identify a group of observers whose ROC results consistently reflect a strong sensitivity to imaging mode differences. We reviewed the data of four large ROC studies performed by our group. The primary index used to classify observers was defined as the absolute difference between readers' Azs for two modes divided by the standard deviation, appropriately corrected for correlation. Using this index, observers were ranked as to their ability to detect differences between modes. We performed four different analyses on data that included 119,550 observations in four studies and 14 different modes performed by 20 readers. With the exception of one data set, there was no correlation between the level of performance of observers, as measured by Az, and their ability to discriminate between modes. Analyses designed to identify observers who were consistent in their rankings as to their ability to discriminated between modes was also unsuccessful in showing differences. This study suggests that without the use of special procedures or training, it is not likely that most research facilities will be able to select a group of highly experienced observers who consistently detect small differences among modes.© (2002) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

To assess the performance levels of a radiologist in detecting non-cued masses and microcalcification clusters depicted on digitized mammograms, 120 mammograms depicting 57 verified masses and 38 microcalcification clusters were selected. During an observer performance study, the images were displayed on a computer monitor. Except for the first mode where no regions were cued, the images were cued in the other four modes using a combination of two cueing sensitivities (90% and 50%) and two false-positive rates (0.5 and 2 per image). One reader ignored all cued regions and identified suspicious regions only in non-cued areas. We examined how the performance of this observer was affected using the different cueing modes. Detection sensitivities of non-cued mammographic abnormalities ranged from 43% to 60%, which were lower (P

Small, efficient observer performance study methods, such as multi-point rank order and forced choice, can be used to either determine the necessity to perform a large scale, receiver operator characteristic (ROC) study or to set the boundary conditions at which it makes sense to perform an ROC study. Because these studies often require observers to discriminate between (among) modes having small visual differences, we decided to address the issue of observers' abilities to make distinctions between modes in these types of non-ROC studies. In this project we reviewed the data of six different non-ROC studies that included a total of 24 observers to determine whether some behave as better mode discriminators. Because of the actual small differences, most observers performed poorly in identifying differences between or among modes. At the same time, at least one and sometimes more observers could identify extremely small differences between modes. These differences were statistically significant. Our results indicate that a good mode discriminator in one study may not perform as well in another such study. Non-ROC studies can be highly sensitive to differences between modes. However, large differences in observer performance combined with observer inconsistency across studies necessitates that these studies include multiple observers.

A constrained ROC formulation from probability summation is proposed for measuring observer performance in detecting abnormal findings on medical images. This assumes the observer's detection or rating decision on each image is determined by a latent variable that characterizes the specific finding (type and location) considered most likely to be a target abnormality. For positive cases, this 'maximum- suspicion' variable is assumed to be either the value for the actual target or for the most suspicious non-target finding, whichever is the greater (more suspicious). Unlike the usual ROC formulation, this constrained formulation guarantees a 'well-behaved' ROC curve that always equals or exceeds chance- level decisions and cannot exhibit an upward 'hook.' Its estimated parameters specify the accuracy for separating positive from negative cases, and they also predict accuracy in locating or identifying the actual abnormal findings. The present maximum-likelihood procedure (runs on PC with Windows 95 or NT) fits this constrained formulation to rating-ROC data using normal distributions with two free parameters. Fits of the conventional and constrained ROC formulations are compared for continuous and discrete-scale ratings of chest films in a variety of detection problems, both for localized lesions (nodules, rib fractures) and for diffuse abnormalities (interstitial disease, infiltrates or pnumothorax). The two fitted ROC curves are nearly identical unless the conventional ROC has an ill behaved 'hook,' below the constrained ROC.

We tested new analytic procedures for combining an observer's image-ratings of lesion-likelihood with localization reports that are incomplete (unavailable on images rated as 'normal') and/or imprecise (possibly scored as 'correct' by chance), and for fitting a constrained ROC formulation to the rating data alone. Eight radiologist readers in a previous study had rated the likelihood of nodular lesions on each of 250 chest-film cases (39 with subtle nodules, 36 with 'typical' nodules and 175 normal cases) that were presented in two display modes (original films or on video workstation). Ratings in the four positive categories (2 to 5) were accompanied by reports that grossly localized the suspected nodules into one of 7 film- regions (upper, middle or lower portions of left or right lung field, or retrocardiac), but there was no localization for the cases rated as 'normal' (category 1). In each of 29 sets of data, we estimated the area below the ROC curve (Az) and its standard error using three different fits: (1) the usual ROC formulation, (2) the constrained ROC formulation and (3) the new procedure that included incomplete and imprecise localization data (I&I). Estimates of Az from the usual and constrained ROC fits were quite similar unless the standard ROC exhibited an upward 'hook,' but standard errors of Az were always the same or smaller for the constrained ROC fit. The I&I fit that included localization data often estimated Az to be either larger or smaller than the usual or constrained ROC fits that considered only the rating data, but its Az had substantially smaller standard errors in 28 of the 29 sets of observer data.© (1999) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

Six radiologists used continuous scales to rate 529 chest-film cases for likelihood of five separate types of abnormalities (interstitial disease, nodules, pneumothorax, alveolar infiltrates and rib fractures) in each of six replicated readings, yielding 36 separate ratings of each case for the five abnormalities. Analyses for each type of abnormality estimated the relative gains in accuracy (area below the ROC curve) obtained by averaging the case-ratings across: (1) six independent replications by each reader (30% gain), (2) six different readers within each replication (39% gain) or (3) all 36 readings (58% gain). Although accuracy differed among both readers and abnormalities, ROC curves for the median ratings showed similar relative gains in accuracy. From a latent-variable model for these gains, we estimate that about 51% of a reader's total decision variance consisted of random (within-reader) errors that were uncorrelated between replications, another 14% came from that reader's consistent (but idiosyncratic) responses to different cases, and only about 35% could be attributed to systematic variations among the sampled cases that were consistent across different readers.

The study is to investigate the use of a Bayesian belief network (BBN) in a computer-assisted diagnosis (CAD) scheme for mass detection in digitized mammograms. Two independent image sets were used in the experiments. After initial processing of image segmentation and adaptive topographic region growth in our CAD scheme, 288 true-positive mass regions and 2,204 false-positive regions were identified in the training image set. In the testing set, 304 true-positive and 1,586 false-positive regions were identified. Fifty features were computed for each region. After using a genetic algorithm search, a BBN was constructed based on 12 local and four global features in order to classify these regions as positive or negative for mass. The performance of the BBN was evaluated using an ROC methodology. The BBN achieved an area under the ROC curve of 0.873 plus or minus 0.009 in classifying the 304 positive and 1,586 negative regions in the testing set. This result was better than using an artificial neural network with the same set of input features. After incorporating the BBN into our CAD scheme as the last classification stage, we detected 80% of 189 positive mass cases (in 433 testing images) with an average detection rate of 0.76 false-positive regions per image. Therefore, this study demonstrated that a BBN approach could yield a comparable performance to that using other classifiers. Using a probabilistic learning concept and interpretable topology, the BBN provides a flexible approach to improving CAD schemes.

On mammograms, certain kinds of features related to masses (e.g., location, texture, degree of spiculation, and integrated density difference) tend to be relatively invariant, or at last predictable, with respect to breast compression. Thus, ipsilateral pairs of mammograms may contain information not available from analyzing single views separately. To demonstrate the feasibility of incorporating multi-view features into CAD algorithm, `single-image' CAD was applied to each individual image in a set of 60 ipsilateral studies, after which all possible pairs of suspicious regions, consisting of one from each view, were formed. For these 402 pairs we defined and evaluated `multi-view' features such as: (1) relative position of centers of regions; (2) ratio of lengths of region projections parallel to nipple axis lines; (3) ratio of integrated contrast difference; (4) ratio of the sizes of the suspicious regions; and (5) measure of relative complexity of region boundaries. Each pair was identified as either a `true positive/true positive' (T) pair (i.e., two regions which are projections of the same actual mass), or as a falsely associated pair (F). Distributions for each feature were calculated. A Bayesian network was trained and tested to classify pairs of suspicious regions based exclusively on the multi-view features described above. Distributions for all features were significantly difference for T versus F pairs as indicated by likelihood ratios. Performance of the Bayesian network, which was measured by ROC analysis, indicates a significant ability to distinguish between T pairs and F pairs (Az equals 0.82 +/- 0.03), using information that is attributed to the multi-view content. This study is the first demonstration that there is a significant amount of spatial information that can be derived from ipsilateral pairs of mammograms.© (1999) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

The purpose of this study is to explore the potential application of region conspicuity as an index of difficulty for mass detection using computer-assisted diagnosis (CAD) schemes on mammograms and to assess the performance improvement of our own CAD scheme by incorporation of conspicuity as well as other features related to tissue background.

The relationship between radiologists' perception of image quality and their actual performance was assessed. If the two variables are strongly correlated, the easier obtained perception of quality index might be used as a prerequisite test to determine if a ROC study is justified. One-hundred seventy cases were evaluated for the presence or absence of interstitial disease and nodules by nine readers using seven display modes. Each reader also assigned each image a mode- specific perceived quality rating using a 5-category ordinal scale. Average perceived quality was highest for conventional film. It was slightly poorer for the subsets of cases with interstitial disease and for cases classified independently as 'subtle.' Trend tests indicated a relationship between area under the ROC curves (Az) and perceived image quality for nodules. For interstitial disease, the relationship was weaker and of borderline statistical significance. The subjective image quality index was related to the area under the ROC curve, but the average difference between pairs of display modes was not a good predictor of difference in actual observer performances. A subjective quality index may have limited usefulness in screening differences between modalities prior to the performance of a ROC study.© (1997) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

Sample size in ROC studies may be significantly reduced by including only difficult cases in the image set, but variability across studies may be a possible obstacle in the development of such a methodology due to case selection. To assess this situation, 300 cases used in a previous large ROC study, which included nine observers, were independently classified as subtle or typical by two experienced readers. Data from the previous study were reanalyzed using data sets consisting only subtle or typical images as designated by each classifier. Results showed a marked decrease in observer performance from the original study's results when only the subtle cases of either classifier were included in the analyses. For 12 of 15 possible comparisons (3 imaging modes and 3 diseases for subtle cases of either classifier were included in the analyses. For 12 of 15 possible comparisons (3 imaging modes and 3 diseases for subtle cases, 3 modes and 2 diseases for typical cases), the Spearman rho correlation coefficient between the performance indices computed for each reader for the subsets classified as subtle and typical was high and significant (P < 0.05). The results obtained in this preliminary study are encouraging and point out areas that warrant further investigation.

In order to subjectively determine acceptable dose levels for portable CR, two blind experiments were performed while maintaining conventional screen-film radiography as a reference quality. In the first experiment, a team of two technologists and two radiologists were trained to expose and to quality-assure portable CR images obtained with standard screens at conventional exposures (400 speed). After providing them with a new set of cassettes and informing them that these were 'better and faster,: they were allowed to practice clinically, using the system for bedside AP chest radiography. After four weeks, exposure factors used in those cases were reviewed, and the experiment was repeated with two different teams and two CR systems. In the second experiment, portable CR was used for a large number of cases. The number of radiologists who complained about CR image quality was monitored for cased were alternating technologies (CR and conventional) were used and routinely viewed side by side. during a two-month period, radiation dose gradually increased to a level where radiologists' complaints were significantly reduced. In both experiments, exposure levels gradually migrated to and stabilized at 40-60% higher levels than that routinely used with conventional 400-speed film screen techniques. The perceived need for high exposure ratios between Cr and conventional radiography was related to body size. When the reference quality 'gold standard' remains unchanged, Cr requires higher exposures to yield acceptable image quality, particularly in large patients.

To assess the type and frequency of data entry errors, we implemented an interface between a radiology information system and a Kodak image-management system. During a patient verification process that included 2,550 cases transmitted from 15 acquisition devices and archived on our system, 195 cases (7.6%) had mismatches that could not be corrected without manual intervention. These included multiple misspellings, errors in the patient identification numbers, and errors due to the incomplete entry of information from the emergency room (trauma cases). Our results clearly demonstrate the need for a comprehensive patient identification and verification function prior to permanent archiving of imaging information.© (1994) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

Availability of initial radiographic images acquired in the Emergency Department (ED) for patients admitted to an Intensive Care Unit (ICU) has been a problem in our operations. It is not uncommon that images from the ED are delivered to the appropriate ICU several hours after admission, and this problem is typically magnified `off hours'. We installed a film digitizer in the ED and required technologists to digitize all chest radiographs as they came out of the film processor. These images are archived and transmitted to a workstation located near one of our busier medical ICUs. The system has been operational for eight months, and it provides reliable timely access to such images. Careful review of a large number of cases clearly demonstrated that such a system is not only feasible, but extremely effective in improving both perceptions and actual quality of radiology services in this difficult environment. Image quality was found to be acceptable for this purpose.© (1994) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

A preliminary study was conducted to investigate the feasibility of using high resolution computed radiography techniques for dentomaxillofacial imaging. Storage phosphors were cut into various sizes and used with an experimental laser scanning reader for three different imaging procedures: intraoral, cephalometric and panoramic. Both phantom and patient images were obtained for comparing the computed radiography technique with the conventional screen/film or dental film techniques. It has been found that current computed radiography techniques are largely adequate for cephalometric and panoramic imaging but need further improvement on their spatial resolution capability for intraoral imaging. In this paper, the methods of applying the computer radiography techniques to dentomaxillofacial imaging are described and discussed. Images of phantoms, resolution bar patterns and patients are presented and compared. Issues on image quality and cost are discussed.