Construction of a Repeatable Framework for Prostate Cancer Lesion Binary Semantic Segmentation using Convolutional Neural Networks

Prostate cancer is the 3rd most diagnosed cancer overall. Current screening methods such as the prostate-specific antigen test could result in overdiagonosis and overtreatment while other methods such as a transrectal ultrasonography are invasive. Recent medical advancements have allowed the use of multiparametric MRI — a noninvasive and reliable screening process for prostate cancer. However, assessment would still vary from different professionals introducing subjectivity. While con-volutional neural network has been used in multiple studies to ob-jectively segment prostate lesions, due to the sensitivity of datasets and varying ground-truth established used in these studies, it is not possible to reproduce and validate the results. In this study, we executed a repeatable framework for segmenting prostate cancer lesions using annotated apparent diffusion coefficient maps from the QIN-PROSTATE-Repeatability dataset — a publicly available dataset that includes multiparametric MRI images of 15 patients that are confirmed or suspected of prostate cancer with two studies each. We used a main architecture of U-Net with batch normalization tested with different encoders, varying data image augmentation combinations, and hyperparameters adopted from various published frameworks to validate which combination of parameters work best for this dataset. The best performing framework was able to achieve a Dice score of 0.47 (0.44-0.49) which is comparable to previously published studies. The results from this study can be objectively compared and improved with further studies whereas this was previously not possible