Purpose: Develop a computer-aided detection (CAD) system for masses in digital breast tomosynthesis (DBT) volume using a deep convolutional neural network (DCNN) with transfer learning from mammograms. Methods: A data set containing 2282 digitized film and digital mammograms and 324 DBT volumes were collected with IRB approval. The mass of interest on the images was marked by an experienced breast radiologist as reference standard. The data set was partitioned into a training set (2282 mammograms with 2461 masses and 230 DBT views with 228 masses) and an independent test set (94 DBT views with 89 masses). For DCNN training, the region of interest (ROI) containing the mass (true positive) was extracted from each image. False positive (FP) ROIs were identified at prescreening by their previously developed CAD systems. After data augmentation, a total of 45 072 mammographic ROIs and 37 450 DBT ROIs were obtained. Data normalization and reduction of non-uniformity in the ROIs across heterogeneous data was achieved using a background correction method applied to each ROI. A DCNN with four convolutional layers and three fully connected (FC) layers was first trained on the mammography data. Jittering and dropout techniques were used to reduce overfitting. After training with the mammographic ROIs, all weights in the first three convolutional layers were frozen, and only the last convolution layer and the FC layers were randomly initialized again and trained using the DBT training ROIs. The authors compared the performances of two CAD systems for mass detection in DBT: one used the DCNN-based approach and the other used their previously developed feature-based approach for FP reduction. The prescreening stage was identical in both systems, passing the same set of mass candidates to the FP reduction stage. For the feature-based CAD system, 3D clustering and active contour method was used for segmentation; morphological, gray level, and texture features were extracted and merged with a linear discriminant classifier to score the detected masses. For the DCNN-based CAD system, ROIs from five consecutive slices centered at each candidate were passed through the trained DCNN and a mass likelihood score was generated. The performances of the CAD systems were evaluated using free-response ROC curves and the performance difference was analyzed using a non-parametric method. Results: Before transfer learning, the DCNN trained only on mammograms with an AUC of 0.99 classified DBT masses with an AUC of 0.81 in the DBT training set. After transfer learning with DBT, the AUC improved to 0.90. For breast-based CAD detection in the test set, the sensitivity for the feature-based and the DCNN-based CAD systems was 83% and 91%, respectively, at 1 FP/DBT volume. The difference between the performances for the two systems was statistically significant (p-value < 0.05). Conclusions: The image patterns learned from the mammograms were transferred to the mass detection on DBT slices through the DCNN. This study demonstrated that large data s...

Grand challenges stimulate advances within the medical imaging research community; within a competitive yet friendly environment, they allow for a direct comparison of algorithms through a well-defined, centralized infrastructure. The tasks of the two-part PROSTATEx Challenges (the PROSTATEx Challenge and the PROSTATEx-2 Challenge) are (1) the computerized classification of clinically significant prostate lesions and (2) the computerized determination of Gleason Grade Group in prostate cancer, both based on multiparametric magnetic resonance images. The challenges incorporate well-vetted cases for training and testing, a centralized performance assessment process to evaluate results, and an established infrastructure for case dissemination, communication, and result submission. In the PROSTATEx Challenge, 32 groups apply their computerized methods (71 methods total) to 208 prostate lesions in the test set. The area under the receiver operating characteristic curve for these methods in the task of differentiating between lesions that are and are not clinically significant ranged from 0.45 to 0.87; statistically significant differences in performance among the top-performing methods, however, are not observed. In the PROSTATEx-2 Challenge, 21 groups apply their computerized methods (43 methods total) to 70 prostate lesions in the test set. When compared with the reference standard, the quadratic-weighted kappa values for these methods in the task of assigning a five-point Gleason Grade Group to each lesion range from −0.24 to 0.27; superiority to random guessing can be established for only two methods. When approached with a sense of commitment and scientific rigor, challenges foster interest in the designated task and encourage innovation in the field.

Radiomics is to provide quantitative descriptors of normal and abnormal tissues during classification and prediction tasks in radiology and oncology. Quantitative Imaging Network members are developing radiomic “feature” sets to characterize tumors, in general, the size, shape, texture, intensity, margin, and other aspects of the imaging features of nodules and lesions. Efforts are ongoing for developing an ontology to describe radiomic features for lung nodules, with the main classes consisting of size, local and global shape descriptors, margin, intensity, and texture-based features, which are based on wavelets, Laplacian of Gaussians, Law’s features, gray-level co-occurrence matrices, and run-length features. The purpose of this study is to investigate the sensitivity of quantitative descriptors of pulmonary nodules to segmentations and to illustrate comparisons across different feature types and features computed by different implementations of feature extraction algorithms. We calculated the concordance correlation coefficients of the features as a measure of their stability with the underlying segmentation; 68% of the 830 features in this study had a concordance CC of ≥0.75. Pairwise correlation coefficients between pairs of features were used to uncover associations between features, particularly as measured by different participants. A graphical model approach was used to enumerate the number of uncorrelated feature groups at given thresholds of correlation. At a threshold of 0.75 and 0.95, there were 75 and 246 subgroups, respectively, providing a measure for the features’ redundancy.

Deep learning models are highly parameterized, resulting in difficulty in inference and transfer learning for image recognition tasks. In this work, we propose a layered pathway evolution method to compress a deep convolutional neural network (DCNN) for classification of masses in digital breast tomosynthesis (DBT). The objective is to prune the number of tunable parameters while preserving the classification accuracy. In the first stage transfer learning, 19 632 augmented regions-of-interest (ROIs) from 2454 mass lesions on mammograms were used to train a pre-trained DCNN on ImageNet. In the second stage transfer learning, the DCNN was used as a feature extractor followed by feature selection and random forest classification. The pathway evolution was performed using genetic algorithm in an iterative approach with tournament selection driven by count-preserving crossover and mutation. The second stage was trained with 9120 DBT ROIs from 228 mass lesions using leave-one-case-out cross-validation. The DCNN was reduced by 87% in the number of neurons, 34% in the number of parameters, and 95% in the number of multiply-and-add operations required in the convolutional layers. The test AUC on 89 mass lesions from 94 independent DBT cases before and after pruning were 0.88 and 0.90, respectively, and the difference was not statistically significant (p > 0.05). The proposed DCNN compression approach can reduce the number of required operations by 95% while maintaining the classification performance. The approach can be extended to other deep neural networks and imaging tasks where transfer learning is appropriate.

Rapid advances in artificial intelligence (AI) and machine learning, and specifically in deep learning (DL) techniques, have enabled broad application of these methods in health care. The promise of the DL approach has spurred further interest in computer‐aided diagnosis (CAD) development and applications using both “traditional” machine learning methods and newer DL‐based methods. We use the term CAD‐AI to refer to this expanded clinical decision support environment that uses traditional and DL‐based AI methods. Numerous studies have been published to date on the development of machine learning tools for computer‐aided, or AI‐assisted, clinical tasks. However, most of these machine learning models are not ready for clinical deployment. It is of paramount importance to ensure that a clinical decision support tool undergoes proper training and rigorous validation of its generalizability and robustness before adoption for patient care in the clinic. To address these important issues, the American Association of Physicists in Medicine (AAPM) Computer‐Aided Image Analysis Subcommittee (CADSC) is charged, in part, to develop recommendations on practices and standards for the development and performance assessment of computer‐aided decision support systems. The committee has previously published two opinion papers on the evaluation of CAD systems and issues associated with user training and quality assurance of these systems in the clinic. With machine learning techniques continuing to evolve and CAD applications expanding to new stages of the patient care process, the current task group report considers the broader issues common to the development of most, if not all, CAD‐AI applications and their translation from the bench to the clinic. The goal is to bring attention to the proper training and validation of machine learning algorithms that may improve their generalizability and reliability and accelerate the adoption of CAD‐AI systems for clinical decision support.

The results demonstrate the potential of CLASS for segmentation of the bladder.

We find there is a fairly high concordance in the size measurements for larger nodules (≥8 mm) than the lower sizes (<8 mm) across algorithms. We find the change in nodule volume (absolute and percent change) were consistent predictors of malignancy across institutions, despite using different segmentation algorithms. Using volume change estimates without corrections shows slightly lower predictability (for two teams).

Purpose We are developing a computerized system for bladder segmentation on CT urography (CTU), as a critical component for computer-aided detection of bladder cancer. Methods The presence of regions filled with intravenous contrast and without contrast presents a challenge for bladder segmentation. Previously, we proposed a Conjoint Level set Analysis and Segmentation System (CLASS). In case the bladder is partially filled with contrast, CLASS segments the non-contrast (NC) region and the contrast-filled (C) region separately and automatically conjoins the NC and C region contours; however, inaccuracies in the NC and C region contours may cause the conjoint contour to exclude portions of the bladder. To alleviate this problem, we implemented a local contour refinement (LCR) method that exploits model-guided refinement (MGR) and energy-driven wavefront propagation (EDWP). MGR propagates the C region contours if the level set propagation in the C region stops prematurely due to substantial non-uniformity of the contrast. EDWP with regularized energies further propagates the conjoint contours to the correct bladder boundary. EDWP uses changes in energies, smoothness criteria of the contour, and previous slice contour to determine when to stop the propagation, following decision rules derived from training. A data set of 173 cases was collected for this study: 81 cases in the training set (42 lesions, 21 wall thickenings, 18 normal bladders) and 92 cases in the test set (43 lesions, 36 wall thickenings, 13 normal bladders). For all cases, 3D hand segmented contours were obtained as reference standard and used for the evaluation of the computerized segmentation accuracy. Results For CLASS with LCR, the average volume intersection ratio, average volume error, absolute average volume error, average minimum distance and Jaccard index were 84.2±11.4%, 8.2±17.4%, 13.0±14.1%, 3.5±1.9 mm, 78.8±11.6%, respectively, for the training set and 78.0±14.7%, 16.4±16.9%, 18.2±15.0%, 3.8±2.3 mm, 73.8±13.4% respectively, for the test set. With CLASS only, the corresponding values were 75.1±13.2%, 18.7±19.5%, 22.5±14.9%, 4.3±2.2 mm, 71.0±12.6%, respectively, for the training set and 67.3±14.3%, 29.3±15.9%, 29.4±15.6%, 4.9±2.6 mm, 65.0±13.3%, respectively, for the test set. The differences between the two methods for all five measures were statistically significant (p<0.001) for both the training and test sets. Conclusions The results demonstrate the potential of CLASS with LCR for segmentation of the bladder.