Search

Accurate and robust segmentation of lung cancers from CT, even those located close to mediastinum, is needed to more accurately plan and deliver radiotherapy and to measure treatment response. Therefore, we developed a new cross-modality educed distillation (CMEDL) approach, using unpaired CT and MRI scans, whereby an informative teacher MRI network guides a student CT network to extract features that signal the difference between foreground and background. Our contribution eliminates two requirements of distillation methods: (i) paired image sets by using an image to image (I2I) translation and (ii) pre-training of the teacher network with a large training set by using concurrent training of all networks. Our framework uses an end-to-end trained unpaired I2I translation, teacher, and student segmentation networks. Architectural flexibility of our framework is demonstrated using 3 segmentation and 2 I2I networks. Networks were trained with 377 CT and 82 T2w MRI from different sets of patients, with independent validation (N=209 tumors) and testing (N=609 tumors) datasets. Network design, methods to combine MRI with CT information, distillation learning under informative (MRI to CT), weak (CT to MRI) and equal teacher (MRI to MRI), and ablation tests were performed. Accuracy was measured using Dice similarity (DSC), surface Dice (sDSC), and Hausdorff distance at the 95$^{th}$ percentile (HD95). The CMEDL approach was significantly (p $, Comment: This manuscript is has been accepted by IEEE Transactions on Medical Imaging

PurposeRecent advances in computational resources, including software libraries and hardware, have enabled the use of high-dimensional, multi-modal datasets to build Artificial Intelligence (AI) models and workflows for radiation therapy and image analysis. The purpose of Software Toolbox for RAdioTherapy and Imaging analysiS (STRATIS) is to provide cloud-enabled, easy-to-share software workflows to train and deploy AI models for transparency and multi-institutional collaboration.MethodSTRATIS leverages open source medical image informatics software for application-specific analysis. Jupyter notebooks for AI modeling workflows are provided with Python language as the base kernel. In addition to Python, workflows use software written in other languages, such as MATLAB, GNU-Octave, R, and C++, with the help of bridge libraries. The workflows can be run on a cloud platform, local workstation, or an institutional HPC cluster. Computational environments are provided in the form of publicly available docker images -and build scripts for local Anaconda environments. Utilities provided with STRATIS simplify bookkeeping of associations between imaging objects and allow chaining data processing operations defined via a setting file for AI models.ResultsWorkflows available on STRATIS can be broadly categorized into image segmentation, deformable image registration, and outcomes modeling for radiotherapy toxicity and tumor control using radiomics and dosimetry features. The STRATIS-forge GitHub organizationhttps://www.github.com/stratis-forgehosts build-scripts for Docker and Anaconda as well as Jupyter notebooks for analysis workflows. The software for building environments and workflow notebooks has open source-GNU-GPL copyright, and AI models retain the copyright chosen by their original developers.ConclusionSTRATIS enables researchers to deploy and share AI modeling workflows for radiotherapy and image analysis. STRATIS is publicly available on Terra.bio’s FireCloud platform with a pre-deployed computational environment and on GitHub organization for users pursuing local deployment.

Cancer care increasingly relies on imaging for patient management. The two most common cross-sectional imaging modalities in oncology are computed tomography (CT) and magnetic resonance imaging (MRI), which provide high-resolution anatomic and physiological imaging. Herewith is a summary of recent applications of rapidly advancing artificial intelligence (AI) in CT and MRI oncological imaging that addresses the benefits and challenges of the resultant opportunities with examples. Major challenges remain, such as how best to integrate AI developments into clinical radiology practice, the vigorous assessment of quantitative CT and MR imaging data accuracy, and reliability for clinical utility and research integrity in oncology. Such challenges necessitate an evaluation of the robustness of imaging biomarkers to be included in AI developments, a culture of data sharing, and the cooperation of knowledgeable academics with vendor scientists and companies operating in radiology and oncology fields. Herein, we will illustrate a few challenges and solutions of these efforts using novel methods for synthesizing different contrast modality images, auto-segmentation, and image reconstruction with examples from lung CT as well as abdome, pelvis, and head and neck MRI. The imaging community must embrace the need for quantitative CT and MRI metrics beyond lesion size measurement. AI methods for the extraction and longitudinal tracking of imaging metrics from registered lesions and understanding the tumor environment will be invaluable for interpreting disease status and treatment efficacy. This is an exciting time to work together to move the imaging field forward with narrow AI-specific tasks. New AI developments using CT and MRI datasets will be used to improve the personalized management of cancer patients.

Fast and accurate multiorgans segmentation from computed tomography (CT) scans is essential for radiation treatment planning. Self-attention(SA)-based deep learning methodologies provide higher accuracies than standard methods but require memory and computationally intensive calculations, which restricts their use to relatively shallow networks.Our goal was to develop and test a new computationally fast and memory-efficient bidirectional SA method called nested block self-attention (NBSA), which is applicable to shallow and deep multiorgan segmentation networks.A new multiorgan segmentation method combining a deep multiple resolution residual network with computationally efficient SA called nested block SA (MRRN-NBSA) was developed and evaluated to segment 18 different organs from head and neck (HN) and abdomen organs. MRRN-NBSA combines features from multiple image resolutions and feature levels with SA to extract organ-specific contextual features. Computational efficiency is achieved by using memory blocks of fixed spatial extent for SA calculation combined with bidirectional attention flow. Separate models were trained for HN (n = 238) and abdomen (n = 30) and tested on set aside open-source grand challenge data sets for HN (n = 10) using a public domain database of computational anatomy and blinded testing on 20 cases from Beyond the Cranial Vault data set with overall accuracy provided by the grand challenge website for abdominal organs. Robustness to two-rater segmentations was also evaluated for HN cases using the open-source data set. Statistical comparison of MRRN-NBSA against Unet, convolutional network-based SA using criss-cross attention (CCA), dual SA, and transformer-based (UNETR) methods was done by measuring the differences in the average Dice similarity coefficient (DSC) accuracy for all HN organs using the Kruskall-Wallis test, followed by individual method comparisons using paired, two-sided Wilcoxon-signed rank tests at 95% confidence level with Bonferroni correction used for multiple comparisons.MRRN-NBSA produced an average high DSC of 0.88 for HN and 0.86 for the abdomen that exceeded current methods. MRRN-NBSA was more accurate than the computationally most efficient CCA (average DSC of 0.845 for HN, 0.727 for abdomen). Kruskal-Wallis test showed significant difference between evaluated methods (p=0.00025). Pair-wise comparisons showed significant differences between MRRN-NBSA than Unet (p=0.0003), CCA (p=0.030), dual (p=0.038), and UNETR methods (p=0.012) after Bonferroni correction. MRRN-NBSA produced less variable segmentations for submandibular glands (0.82 ± 0.06) compared to two raters (0.75 ± 0.31).MRRN-NBSA produced more accurate multiorgan segmentations than current methods on two different public data sets. Testing on larger institutional cohorts is required to establish feasibility for clinical use.

A barrier to the widespread adoption of watch-and-wait management for locally advanced rectal cancer is the inaccuracy and variability of identifying tumor response endoscopically in patients who have completed total neoadjuvant therapy (chemoradiotherapy and systemic chemotherapy).

Background For breast cancer patients undergoing neoadjuvant chemotherapy (NAC), pathologic complete response (pCR; no invasive or in situ) cannot be assessed non-invasively so all patients undergo surgery. The aim of our study was to develop and validate a radiomics classifier that classifies breast cancer pCR post-NAC on MRI prior to surgery. Methods This retrospective study included women treated with NAC for breast cancer from 2014 to 2016 with (1) pre- and post-NAC breast MRI and (2) post-NAC surgical pathology report assessing response. Automated radiomics analysis of pre- and post-NAC breast MRI involved image segmentation, radiomics feature extraction, feature pre-filtering, and classifier building through recursive feature elimination random forest (RFE-RF) machine learning. The RFE-RF classifier was trained with nested five-fold cross-validation using (a) radiomics only (model 1) and (b) radiomics and molecular subtype (model 2). Class imbalance was addressed using the synthetic minority oversampling technique. Results Two hundred seventy-three women with 278 invasive breast cancers were included; the training set consisted of 222 cancers (61 pCR, 161 no-pCR; mean age 51.8 years, SD 11.8), and the independent test set consisted of 56 cancers (13 pCR, 43 no-pCR; mean age 51.3 years, SD 11.8). There was no significant difference in pCR or molecular subtype between the training and test sets. Model 1 achieved a cross-validation AUROC of 0.72 (95% CI 0.64, 0.79) and a similarly accurate (P = 0.1) AUROC of 0.83 (95% CI 0.71, 0.94) in both the training and test sets. Model 2 achieved a cross-validation AUROC of 0.80 (95% CI 0.72, 0.87) and a similar (P = 0.9) AUROC of 0.78 (95% CI 0.62, 0.94) in both the training and test sets. Conclusions This study validated a radiomics classifier combining radiomics with molecular subtypes that accurately classifies pCR on MRI post-NAC.

Background Lower-grade gliomas (LGGs) with isocitrate dehydrogenase 1 and/or 2 (IDH1/2) mutations have long survival times, making evaluation of treatment efficacy difficult. We investigated the volumetric growth rate of IDH mutant gliomas before and after treatment with established glioma therapies to determine whether a significant change in growth rate could be documented and perhaps be used in the future to evaluate treatment response to investigational agents in LGG trials. Methods In this multicenter retrospective study, 230 adult patients with IDH1/2 mutated LGGs (World Health Organization grade II or III) undergoing surgery, radiation, or chemotherapy for progressive non-enhancing tumor were identified. Subjects were required to have 3 MRI scans containing T2/fluid attenuated inversion recovery imaging spanning a minimum of 6 months prior to treatment. A mixed-effect model was used to estimate tumor growth prior to treatment. A subset of 95 patients who received chemotherapy, radiotherapy, or chemoradiotherapy and had 2 posttreatment imaging time points available were evaluated for change in pre- and posttreatment volumetric growth rates using a piecewise mixed model. Results The pretreatment volumetric growth rate across all 230 patients was 27.37%/180 days (95% CI: [23.36%, 31.51%]). In the 95 patients with both pre- and posttreatment scans available, there was a significant difference in volumetric growth rates before (26.63%/180 days, 95% CI: [19.31%, 34.40%]) and after treatment (−15.24% /180 days, 95% CI: [−21.37%, −8.62%]) (P < 0.0001). The growth rates for patient subgroup with 1p/19q codeletion (N = 118) was significantly slower than the rate of the 1p/19q non-codeleted group (N = 68) (22.84% vs 35.49%, P = 0.0108). Conclusion In this study, we evaluated the growth rates of IDH mutant gliomas before and after standard therapy. Further study is needed to establish whether a change in growth rate is associated with patient survival and its use as a surrogate endpoint in clinical trials for IDH mutant LGGs.

PURPOSE Automatic segmentation offers many benefits for radiotherapy treatment planning; however, the lack of publicly available benchmark datasets limits the clinical use of automatic segmentation. In this work, we present a well-curated computed tomography (CT) dataset of high-quality manually drawn contours from patients with thoracic cancer that can be used to evaluate the accuracy of thoracic normal tissue auto-segmentation systems. ACQUISITION AND VALIDATION METHODS Computed tomography scans of 60 patients undergoing treatment simulation for thoracic radiotherapy were acquired from three institutions: MD Anderson Cancer Center, Memorial Sloan Kettering Cancer Center, and the MAASTRO clinic. Each institution provided CT scans from 20 patients, including mean intensity projection four-dimensional CT (4D CT), exhale phase (4D CT), or free-breathing CT scans depending on their clinical practice. All CT scans covered the entire thoracic region with a 50-cm field of view and slice spacing of 1, 2.5, or 3 mm. Manual contours of left/right lungs, esophagus, heart, and spinal cord were retrieved from the clinical treatment plans. These contours were checked for quality and edited if necessary to ensure adherence to RTOG 1106 contouring guidelines. DATA FORMAT AND USAGE NOTES The CT images and RTSTRUCT files are available in DICOM format. The regions of interest were named according to the nomenclature recommended by American Association of Physicists in Medicine Task Group 263 as Lung_L, Lung_R, Esophagus, Heart, and SpinalCord. This dataset is available on The Cancer Imaging Archive (funded by the National Cancer Institute) under Lung CT Segmentation Challenge 2017 (http://doi.org/10.7937/K9/TCIA.2017.3r3fvz08). POTENTIAL APPLICATIONS This dataset provides CT scans with well-delineated manually drawn contours from patients with thoracic cancer that can be used to evaluate auto-segmentation systems. Additional anatomies could be supplied in the future to enhance the existing library of contours.

Vision transformers efficiently model long-range context and thus have demonstrated impressive accuracy gains in several image analysis tasks including segmentation. However, such methods need large labeled datasets for training, which is hard to obtain for medical image analysis. Self-supervised learning (SSL) has demonstrated success in medical image segmentation using convolutional networks. In this work, we developed a self-distillation learning with masked image modeling method to perform SSL for vision transformers (SMIT) applied to 3D multi-organ segmentation from CT and MRI. Our contribution combines a dense pixel-wise regression pretext task performed within masked patches called masked image prediction with masked patch token distillation to pre-train vision transformers. Our approach is more accurate and requires fewer fine tuning datasets than other pretext tasks. Unlike prior methods, which typically used image sets arising from disease sites and imaging modalities corresponding to the target tasks, we used 3,643 CT scans (602,708 images) arising from head and neck, lung, and kidney cancers as well as COVID-19 for pre-training and applied it to abdominal organs segmentation from MRI pancreatic cancer patients as well as publicly available 13 different abdominal organs segmentation from CT. Our method showed clear accuracy improvement (average DSC of 0.875 from MRI and 0.878 from CT) with reduced requirement for fine-tuning datasets over commonly used pretext tasks. Extensive comparisons against multiple current SSL methods were done. Our code is available at: https://github.com/harveerar/SMIT.git.

Purpose To evaluate pix2pix and CycleGAN and to assess the effects of multiple combination strategies on accuracy for patch-based synthetic computed tomography (sCT) generation for magnetic resonance (MR)-only treatment planning in head and neck (HN) cancer patients. Materials and methods Twenty-three deformably registered pairs of CT and mDixon FFE MR datasets from HN cancer patients treated at our institution were retrospectively analyzed to evaluate patch-based sCT accuracy via the pix2pix and CycleGAN models. To test effects of overlapping sCT patches on estimations, we (a) trained the models for three orthogonal views to observe the effects of spatial context, (b) we increased effective set size by using per-epoch data augmentation, and (c) we evaluated the performance of three different approaches for combining overlapping Hounsfield unit (HU) estimations for varied patch overlap parameters. Twelve of twenty-three cases corresponded to a curated dataset previously used for atlas-based sCT generation and were used for training with leave-two-out cross-validation. Eight cases were used for independent testing and included previously unseen image features such as fused vertebrae, a small protruding bone, and tumors large enough to deform normal body contours. We analyzed the impact of MR image preprocessing including histogram standardization and intensity clipping on sCT generation accuracy. Effects of mDixon contrast (in-phase vs water) differences were tested with three additional cases. The sCT generation accuracy was evaluated using mean absolute error (MAE) and mean error (ME) in HU between the plan CT and sCT images. Dosimetric accuracy was evaluated for all clinically relevant structures in the independent testing set and digitally reconstructed radiographs (DRRs) were evaluated with respect to the plan CT images. Results The cross-validated MAEs for the whole-HN region using pix2pix and CycleGAN were 66.9 ± 7.3 vs 82.3 ± 6.4 HU, respectively. On the independent testing set with additional artifacts and previously unseen image features, whole-HN region MAEs were 94.0 ± 10.6 and 102.9 ± 14.7 HU for pix2pix and CycleGAN, respectively. For patients with different tissue contrast (water mDixon MR images), the MAEs increased to 122.1 ± 6.3 and 132.8 ± 5.5 HU for pix2pix and CycleGAN, respectively. Our results suggest that combining overlapping sCT estimations at each voxel reduced both MAE and ME compared to single-view non-overlapping patch results. Absolute percent mean/max dose errors were 2% or less for the PTV and all clinically relevant structures in our independent testing set, including structures with image artifacts. Quantitative DRR comparison between planning CTs and sCTs showed agreement of bony region positions to Conclusions The dosimetric and MAE based accuracy, along with the similarity between DRRs from sCTs, indicate that pix2pix and CycleGAN are promising methods for MR-only treatment planning for HN cancer. Our methods investigated for overlapping patch-based HU estimations also indicate that combining transformation estimations of overlapping patches is a potential method to reduce generation errors while also providing a tool to potentially estimate the MR to CT aleatoric model transformation uncertainty. However, because of small patient sample sizes, further studies are required.

PURPOSE To determine if radiomic measures of tumor heterogeneity derived from baseline contrast-enhanced computed tomography (CE-CT) are associated with durable clinical benefit and time to off-treatment in patients with recurrent ovarian cancer (OC) enrolled in prospective immunotherapeutic trials. MATERIALS AND METHODS This retrospective study included 75 patients with recurrent OC who were enrolled in prospective immunotherapeutic trials (n = 74) or treated off-label (n = 1) and had baseline CE-CT scans. Disease burden (total tumor volume, number of disease sites), radiomic measures of intertumor heterogeneity (cluster-site entropy, cluster-site dissimilarity), and intratumor heterogeneity of the largest lesion (Haralick texture features) were computed. Associations of clinical, conventional imaging, and radiomic measures with durable clinical benefit and time to off-treatment were examined. RESULTS In univariable analysis, fewer disease sites, lower intertumor heterogeneity (lower cluster-site entropy, lower cluster-site dissimilarity), and lower intratumor heterogeneity of the largest lesion (higher energy) were significantly associated with durable clinical benefit ( P ≤ .031). More disease sites, presence of pleural disease and/or distant metastases, higher intertumor heterogeneity (higher cluster-site entropy, higher cluster-site dissimilarity), and higher intratumor heterogeneity of the largest lesion (higher Contrastlargest-lesion) were significantly associated with shorter time to off-treatment ( P ≤ .034). In multivariable analysis, higher Energylargest-lesion (indicator of lower intratumor heterogeneity; P = .006; odds ratio, 1.41) and fewer disease sites ( P = .003; odds ratio, 1.64) remained significant indicators of durable clinical benefit (multivariable model C-index, 0.821). Higher cluster-site dissimilarity (indicator of higher intertumor heterogeneity) was a modest but single independent indicator of shorter time to off-treatment ( P = .004; hazard ratio, 1.19; C-index, 0.6). CONCLUSION Fewer disease sites and lower intra- and intertumor heterogeneity modeled from the baseline CE-CT may indicate better response of OC to immunotherapy.

An increasing number of patients with multiple brain metastases are being treated with stereotactic radiosurgery (SRS). Manually identifying and contouring all metastatic lesions is difficult and time-consuming, and a potential source of variability. Hence, we developed a 3D deep learning approach for segmenting brain metastases on MR and CT images. Five-hundred eleven patients treated with SRS were retrospectively identified for this study. Prior to radiotherapy, the patients were imaged with 3D T1 spoiled-gradient MR post-Gd (T1 + C) and contrast-enhanced CT (CECT), which were co-registered by a treatment planner. The gross tumor volume contours, authored by the attending radiation oncologist, were taken as the ground truth. There were 3 ± 4 metastases per patient, with volume up to 57 ml. We produced a multi-stage model that automatically performs brain extraction, followed by detection and segmentation of brain metastases using co-registered T1 + C and CECT. Augmented data from 80% of these patients were used to train modified 3D V-Net convolutional neural networks for this task. We combined a normalized boundary loss function with soft Dice loss to improve the model optimization, and employed gradient accumulation to stabilize the training. The average Dice similarity coefficient (DSC) for brain extraction was 0.975 ± 0.002 (95% CI). The detection sensitivity per metastasis was 90% (329/367), with moderate dependence on metastasis size. Averaged across 102 test patients, our approach had metastasis detection sensitivity 95 ± 3%, 2.4 ± 0.5 false positives, DSC of 0.76 ± 0.03, and 95th-percentile Hausdorff distance of 2.5 ± 0.3 mm (95% CIs). The volumes of automatic and manual segmentations were strongly correlated for metastases of volume up to 20 ml (r=0.97,p

Stereotactic body radiation therapy (SBRT) of locally advanced pancreatic cancer (LAPC) is challenging due to significant motion of gastrointestinal (GI) organs. The goal of our study was to quantify inter and intrafraction deformations and dose accumulation of upper GI organs in LAPC patients.Five LAPC patients undergoing five-fraction magnetic resonance-guided radiation therapy (MRgRT) using abdominal compression and daily online plan adaptation to 50 Gy were analyzed. A pre-treatment, verification, and post-treatment MR imaging (MRI) for each of the five fractions (75 total) were used to calculate intra and interfraction motion. The MRIs were registered using Large Deformation Diffeomorphic Metric Mapping (LDDMM) deformable image registration (DIR) method and total dose delivered to stomach_duodenum, small bowel (SB) and large bowel (LB) were accumulated. Deformations were quantified using gradient magnitude and Jacobian integral of the Deformation Vector Fields (DVF). Registration DVFs were geometrically assessed using Dice and 95th percentile Hausdorff distance (HD95) between the deformed and physician's contours. Accumulated doses were then calculated from the DVFs.Median Dice and HD95 were: Stomach_duodenum (0.9, 1.0 mm), SB (0.9, 3.6 mm), and LB (0.9, 2.0 mm). Median (max) interfraction deformation for stomach_duodenum, SB and LB was 6.4 (25.8) mm, 7.9 (40.5) mm and 7.6 (35.9) mm. Median intrafraction deformation was 5.5 (22.6) mm, 8.2 (37.8) mm and 7.2 (26.5) mm. Accumulated doses for two patients exceeded institutional constraints for stomach_duodenum, one of whom experienced Grade1 acute and late abdominal toxicity.LDDMM method indicates feasibility to measure large GI motion and accumulate dose. Further validation on larger cohort will allow quantitative dose accumulation to more reliably optimize online MRgRT.

Patients with high-grade serous ovarian cancer suffer poor prognosis and variable response to treatment. Known prognostic factors for this disease include homologous recombination deficiency status, age, pathological stage and residual disease status after debulking surgery. Recent work has highlighted important prognostic information captured in computed tomography and histopathological specimens, which can be exploited through machine learning. However, little is known about the capacity of combining features from these disparate sources to improve prediction of treatment response. Here, we assembled a multimodal dataset of 444 patients with primarily late-stage high-grade serous ovarian cancer and discovered quantitative features, such as tumor nuclear size on staining with hematoxylin and eosin and omental texture on contrast-enhanced computed tomography, associated with prognosis. We found that these features contributed complementary prognostic information relative to one another and clinicogenomic features. By fusing histopathological, radiologic and clinicogenomic machine-learning models, we demonstrate a promising path toward improved risk stratification of patients with cancer through multimodal data integration.

Objective To investigate associations between genetic mutations and qualitative as well as quantitative features on MRI in rectal adenocarcinoma at primary staging. Methods In this retrospective study, patients with rectal adenocarcinoma, genome sequencing, and pretreatment rectal MRI were included. Statistical analysis was performed to evaluate associations between qualitative features obtained from subjective evaluation of rectal MRI and gene mutations as well as between quantitative textural features and gene mutations. For the qualitative evaluation, Fisher’s Exact test was used to analyze categorical associations and Wilcoxon Rank Sum test was used for continuous clinical variables. For the quantitative evaluation, we performed manual segmentation of T2-weighted images for radiomics-based quantitative image analysis. Thirty-four texture features consisting of first order intensity histogram-based features (n = 4), second order Haralick textures (n = 5), and Gabor-edge based Haralick textures were computed at two different orientations. Consensus clustering was performed with 34 computed texture features using the K-means algorithm with Euclidean distance between the texture features. The clusters resulting from the algorithm were then used to enumerate the prevalence of gene mutations in those clusters. Results In 65 patients, 45 genes were mutated in more than 3/65 patients (5%) and were included in the statistical analysis. Regarding qualitative imaging features, on univariate analysis, tumor location was significantly associated with APC (p = 0.032) and RASA1 mutation (p = 0.032); CRM status was significantly associated with ATM mutation (p = 0.021); and lymph node metastasis was significantly associated with BRCA2 (p = 0.046) mutation. However, these associations were not significant after adjusting for multiple comparisons. Regarding quantitative imaging features, Cluster C1 had tumors with higher mean Gabor edge intensity compared with cluster C2 (θ = 0°, p = 0.018; θ = 45°, p = 0.047; θ = 90°, p = 0.037; cluster C3 (θ = 0°, p = 0.18; θ = 45°, p = 0.1; θ = 90°, p = 0.052), and cluster C4 (θ = 0°, p = 0.016; θ = 45°, p = 0.033; θ = 90°, p = 0.014) suggesting that the cluster C1 had tumors with more distinct edges or heterogeneous appearance compared with other clusters. Conclusions Although this preliminary study showed promising associations between quantitative features and genetic mutations, it did not show any correlation between qualitative features and genetic mutations. Further studies with larger sample size are warranted to validate our preliminary data.

Volumetric lung tumor segmentation and accurate longitudinal tracking of tumor volume changes from computed tomography images are essential for monitoring tumor response to therapy. Hence, we developed two multiple resolution residually connected network (MRRN) formulations called incremental-MRRN and dense-MRRN. Our networks simultaneously combine features across multiple image resolution and feature levels through residual connections to detect and segment the lung tumors. We evaluated our method on a total of 1210 non-small cell (NSCLC) lung tumors and nodules from three data sets consisting of 377 tumors from the open-source Cancer Imaging Archive (TCIA), 304 advanced stage NSCLC treated with anti- PD-1 checkpoint immunotherapy from internal institution MSKCC data set, and 529 lung nodules from the Lung Image Database Consortium (LIDC). The algorithm was trained using 377 tumors from the TCIA data set and validated on the MSKCC and tested on LIDC data sets. The segmentation accuracy compared to expert delineations was evaluated by computing the dice similarity coefficient, Hausdorff distances, sensitivity, and precision metrics. Our best performing incremental-MRRN method produced the highest DSC of 0.74 ± 0.13 for TCIA, 0.75±0.12 for MSKCC, and 0.68±0.23 for the LIDC data sets. There was no significant difference in the estimations of volumetric tumor changes computed using the incremental-MRRN method compared with the expert segmentation. In summary, we have developed a multi-scale CNN approach for volumetrically segmenting lung tumors which enables accurate, automated identification of and serial measurement of tumor volumes in the lung.

Purpose: The goal of this study is to develop innovative methods for identifying radiomic features that are reproducible over varying image acquisition settings. Approach: We propose a regularized partial correlation network to identify reliable and reproducible radiomic features. This approach was tested on two radiomic feature sets generated using two different reconstruction methods on computed tomography (CT) scans from a cohort of 47 lung cancer patients. The largest common network component between the two networks was tested on phantom data consisting of five cancer samples. To further investigate whether radiomic features found can identify phenotypes, we propose a k-means clustering algorithm coupled with the optimal mass transport theory. This approach following the regularized partial correlation network analysis was tested on CT scans from 77 head and neck squamous cell carcinoma (HNSCC) patients in the Cancer Imaging Archive (TCIA) and validated using an independent dataset. Results: A set of common radiomic features was found in relatively large network components between the resultant two partial correlation networks resulting from a cohort of lung cancer patients. The reliability and reproducibility of those radiomic features were further validated on phantom data using the Wasserstein distance. Further analysis using the network-based Wasserstein k-means algorithm on the TCIA HNSCC data showed that the resulting clusters separate tumor subsites as well as HPV status, and this was validated on an independent dataset. Conclusion: We showed that a network-based analysis enables identifying reproducible radiomic features and use of the selected set of features can enhance clustering results.

PurposeThis work presents a framework for deployment of deep learning image segmentation models for medical images across different operating systems and programming languages.MethodsComputational Environment for Radiological Research (CERR) platform was extended for deploying deep learning-based segmentation models to leverage CERR’s existing functionality for radiological data import, transformation, management, and visualization. The framework is compatible with MATLAB as well as GNU Octave and Python for license-free use. Pre and post processing configurations including parameters for pre-processing images, population of channels, and post-processing segmentations was standardized using JSON format. CPU and GPU implementations of pre-trained deep learning segmentation models were packaged using Singularity containers for use in Linux and Conda environment archives for Windows, macOS and Linux operating systems. The framework accepts images in various formats including DICOM and CERR’s planC and outputs segmentation in various formats including DICOM RTSTRUCT and planC objects. The ability to access the results readily in planC format enables visualization as well as radiomics and dosimetric analysis. The framework can be readily deployed in clinical software such as MIM via their extensions.ResultsThe open-source, GPL copyrighted framework developed in this work has been successfully used to deploy Deep Learning based segmentation models for five in-house developed and published models. These models span various treatment sites (H&N, Lung and Prostate) and modalities (CT, MR). Documentation for their usage and demo workflow is provided at https://github.com/cerr/CERR/wiki/Auto-Segmentation-models. The framework has also been used in clinical workflow for segmenting images for treatment planning and for segmenting publicly available large datasets for outcomes studies.ConclusionsThis work presented a comprehensive, open-source framework for deploying deep learning-based medical image segmentation models. The framework was used to translate the developed models to clinic as well as reproducible and consistent image segmentation across institutions, facilitating multi-institutional outcomes modeling studies.