Purpose: Accurate regional segmentation of the prostate boundaries on magnetic resonance (MR) images is a fundamental requirement before automated prostate cancer diagnosis can be achieved. In this paper, we describe a novel methodology to segment prostate whole gland (WG), central gland (CG), and peripheral zone (PZ), where PZ + CG = WG, from T2W and apparent diffusion coefficient (ADC) map prostate MR images. Methods: We designed two similar models each made up of two U-Nets to delineate the WG, CG, and PZ from T2W and ADC map MR images, separately. The U-Net, which is a modified version of a fully convolutional neural network, includes contracting and expanding paths with convolutional, pooling, and upsampling layers. Pooling and upsampling layers help to capture and localize image features with a high spatial consistency. We used a dataset consisting of 225 patients (combining 153 and 72 patients with and without clinically significant prostate cancer) imaged with multiparametric MRI at 3 Tesla. Results and conclusion: Our proposed model for prostate zonal segmentation from T2W was trained and tested using 1154 and 1587 slices of 100 and 125 patients, respectively. Median of Dice similarity coefficient (DSC) on test dataset for prostate WG, CG, and PZ were 95.33 AE 7.77%, 93.75 AE 8.91%, and 86.78 AE 3.72%, respectively. Designed model for regional prostate delineation from ADC map images was trained and validated using 812 and 917 slices from 100 and 125 patients. This model yielded a median DSC of 92.09 AE 8.89%, 89.89 AE 10.69%, and 86.1 AE 9.56% for prostate WG, CG, and PZ on test samples, respectively. Further investigation indicated that the proposed algorithm reported high DSC for prostate WG segmentation from both T2W and ADC map MR images irrespective of WG size. In addition, segmentation accuracy in terms of DSC does not significantly vary among patients with or without significant tumors. Significance: We describe a method for automated prostate zonal segmentation using T2W and ADC map MR images independent of prostate size and the presence or absence of tumor. Our results are important in terms of clinical perspective as fully automated methods for ADC map images, which are considered as one of the most important sequences for prostate cancer detection in the PZ and CG, have not been reported previously.

Prostate MRI is reported in clinical practice using the Prostate Imaging and Data Reporting System (PI‐RADS). PI‐RADS aims to standardize, as much as possible, the acquisition, interpretation, reporting, and ultimately the performance of prostate MRI. PI‐RADS relies upon mainly subjective analysis of MR imaging findings, with very few incorporated quantitative features. The shortcomings of PI‐RADS are mainly: low‐to‐moderate interobserver agreement and modest accuracy for detection of clinically significant tumors in the transition zone. The use of a more quantitative analysis of prostate MR imaging findings is therefore of interest. Quantitative MR imaging features including: tumor size and volume, tumor length of capsular contact, tumor apparent diffusion coefficient (ADC) metrics, tumor T1 and T2 relaxation times, tumor shape, and texture analyses have all shown value for improving characterization of observations detected on prostate MRI and for differentiating between tumors by their pathological grade and stage. Quantitative analysis may therefore improve diagnostic accuracy for detection of cancer and could be a noninvasive means to predict patient prognosis and guide management. Since quantitative analysis of prostate MRI is less dependent on an individual users' assessment, it could also improve interobserver agreement. Semi‐ and fully automated analysis of quantitative (radiomic) MRI features using artificial neural networks represent the next step in quantitative prostate MRI and are now being actively studied. Validation, through high‐quality multicenter studies assessing diagnostic accuracy for clinically significant prostate cancer detection, in the domain of quantitative prostate MRI is needed. This article reviews advances in quantitative prostate MRI, highlighting the strengths and limitations of existing and emerging techniques, as well as discussing opportunities and challenges for evaluation of prostate MRI in clinical practice when using quantitative assessment. Level of Evidence 5 Technical Efficacy Stage 2

No abstract

Purpose: Detection and accurate localization of renal masses (RM) are important steps toward future potential classification of benign vs malignant RM. A fully automated algorithm for detection and localization of RM may eliminate the observer variability in the clinical workflow. Method: In this paper, we describe a fully automated methodology for accurate detection and segmentation of RM from contrast-enhanced computed tomography (CECT) images. We first determine the boundaries of the kidneys on the CECT images utilizing a convolutional neural network-based method to be used as a region of interest to search for RM. We then employ a homogenous U-Netbased ensemble learning model to identify and delineate RM. We used an institutional dataset comprised of CECT images in 315 patients to train and evaluate the proposed method. We compared results of our method to those of three-dimensional (3D) U-Net for RM localization and further evaluated our algorithm using the kidney tumor segmentation (KiTS19) challenge dataset. Results: The developed algorithm reported a Dice similarity coefficient (DSC) of 95.79% AE 5.16% and 96.25 AE 3.37 (mean AE standard deviation) for segmentation accuracy of kidney boundary from 125 and 60 test images from institutional and KiTS19 datasets, respectively. Using our method, RM were detected in 125 and 52 test cases, which corresponds to 100% and 86.67% sensitivity at patient level in institutional and KiTS19 test images. Our ensemble method for RM localization yielded a mean DSC of 88.65% AE 7.31% and 87.91% AE 6.82% on the institutional and KiTS19 test datasets, respectively. The mean DSC for RM delineation from CECT institutional test images using 3D U-Net was 85.95% AE 1.46%. Conclusion: We describe a method for automated localization of RM using CECT images. Our results are important in terms of clinical perspective as fully automated detection of RM is a fundamental step for further diagnosis of cystic vs solid RM and eventually benign vs malignant solid RM, that has not been reported previously.

Purpose: Three-dimensional (3D) late gadolinium enhancement magnetic resonance (LGE-MR) imaging enables the quantification of myocardial scar at high resolution with unprecedented volumetric visualization. Automated segmentation of myocardial scar is critical for the potential clinical translation of this technique given the number of tomographic images acquired. Methods: In this paper, we describe the development of cascaded multi-planar U-Net (CMPU-Net) to efficiently segment the boundary of the left ventricle (LV) myocardium and scar from 3D LGE-MR images. In this approach, two subnets, each containing three U-Nets, were cascaded to first segment the LV myocardium and then segment the scar within the presegmented LV myocardium. The U-Nets were trained separately using two-dimensional (2D) slices extracted from axial, sagittal, and coronal slices of 3D LGE-MR images. We used 3D LGE-MR images from 34 subjects with chronic ischemic cardiomyopathy. The U-Nets were trained using 8430 slices, extracted in three orthogonal directions from 18 images. In the testing phase, the outputs of U-Nets of each subnet were combined using the majority voting system for final label prediction of each voxel in the image. The developed method was tested for accuracy by comparing its results to manual segmentations of LV myocardium and LV scar from 7250 slices extracted from 16 3D LGE-MR images. Our method was also compared to numerous alternative methods based on machine learning, energy minimization, and intensity-thresholds. Results: Our algorithm reported a mean dice similarity coefficient (DSC), absolute volume difference (AVD), and Hausdorff distance (HD) of 85.14% AE 3.36%, 43.72 AE 27.18 cm 3 , and 19.21 AE 4.74 mm for determining the boundaries of LV myocardium from LGE-MR images. Our method also yielded a mean DSC, AVD, and HD of 88.61% AE 2.54%, 9.33 AE 7.24 cm 3 , and 17.04 AE 9.93 mm for LV scar segmentation on the unobserved test dataset. Our method significantly outperformed the alternative techniques in segmentation accuracy (P < 0.05). Conclusions: The CMPU-Net method provided fully automated segmentation of LV scar from 3DLGE-MR images and outperformed the alternative techniques.