Segmentation of the Prostatic Gland and the Intraprostatic Lesions on Multiparametic Magnetic Resonance Imaging Using Mask Region-Based Convolutional Neural Networks

Abstract

Prostate cancer (PCa) is the most common cancer in men in the United States. Multiparametic magnetic resonance imaging (mp-MRI) has been explored by many researchers to targeted prostate biopsies and radiation therapy. However, assessment on mp-MRI can be subjective, development of computer-aided diagnosis systems to automatically delineate the prostate gland and the intraprostratic lesions (ILs) becomes important to facilitate with radiologists in clinical practice. In this paper, we first study the implementation of the Mask-RCNN model to segment the prostate and ILs. We trained and evaluated models on 120 patients from two different cohorts of patients. We also used 2D U-Net and 3D U-Net as benchmarks to segment the prostate and compared the model's performance. The contour variability of ILs using the algorithm was also benchmarked against the interobserver variability between two different radiation oncologists on 19 patients. Our results indicate that the Mask-RCNN model is able to reach state-of-art performance in the prostate segmentation and outperforms several competitive baselines in ILs segmentation.Automated segmentation of the prostate and screening of prostate cancer from MR images is critical for computer-aided clinical diagnosis, treatment planning and prognosis. However, the development of automatic algorithms remains challenging in several reasons. First of all, there are large variations in image quality caused by several factors at the time of image acquisition (e.g. patient motion, signal-to-noise ratio, use of an endorectal coil, Gadolinium enhancement, etc.). Second, the normal anatomy of the prostate is highly variable across patients and at different time points; and the boundaries between the prostate and surrounding structures (e.g. neurovascular bundles, bladder, rectum, seminal vessels and other soft tissues) are not always immediately clear. The prostate also shows a large variation in size and shape among different patients due to individual differences and physiological changes. Third, the presence of benign conditions such as benign prostatic hyperplasia and prostatitis may mimic the radiographic presentation of a malignancy. Contrast and pixel value of MRI also highlight a large variability in both tissue and texture information.Over the past few years, progress in image segmentation tasks has been exclusively driven by convolutional neural network (CNN) based models. Many segmentation models fall into two classes. The first class does not rely on the region proposal algorithm. Typical models in this class usually apply an encoder-decoder framework (Liou et al., 2014). The encoder network extracts representations of the image, and the decoder network reconstructs segmentation mask from the learned image representations produced by the encoder network. U-Net (Ronneberger et al., 2015), for instance, is a classic algorithm widely used in biomed-ical image segmentation tasks. Another class of models have their underlying fundamentals on region proposals such as the Mask-RCNN model,...

Automating the Analysis of Fish Abundance Using Object Detection: Optimizing Animal Ecology With Deep Learning

“…Mask R-CNN works by classifying and localizing the region of interest (RoI). It extends previous frameworks in that it can predict a segmentation mask on the RoI, and currently, it has the highest performance output for deep learning models (He et al, 2017;Dai et al, 2019). To develop our model, we used a ResNet50 configuration, pre-trained on the ImageNet-1k dataset.…”

Breast tumor segmentation in 3D automatic breast ultrasound using Mask scoring R‐CNN

“…Mask R‐CNN is a kind of R‐CNN and is recently used for prostate segmentation from MRI. Different from our proposed method, Mask R‐CNN does not include the attention gate to highlight the feature map and mask score head to re‐assign a score for detected tumor ROI 37 . The paired two‐tailed t‐test was used to compare the results between proposed method and comparing methods.…”

Applications of Artificial Intelligence to Prostate Multiparametric MRI (mpMRI): Current and Emerging Trends

“…A study by Liu et al [87] used fuzzy Markov random fields to achieve a Dice score of 0.62 with 11 patients. Two other groups, Kohl et al [88] and Dai et al [89], both employed DL algorithms and used U-Net and Mask R-CNN, respectively. Kohl's group used a dataset of 152 patients and implemented U-Net combined with an adversarial network.…”

“…Their architecture resulted in an average Dice score for prostate lesion segmentation of 0.41 [88]. Dai's group used a highly specialized DL algorithm, Mask R-CNN, and trained with 63 patients to achieve a prostate lesion Dice score of 0.46 [89]. To label the ground truth, Dai et al [89] used a clinician, Kohl et al [88] used a radiologist, and Liu et al [87] used a pathologist.…”