Encoding Clinical Priori in 3D Convolutional Neural Networks for Prostate Cancer Detection in bpMRI
This work addresses prostate cancer detection for medical imaging, offering incremental improvements by integrating clinical knowledge into existing neural network architectures.
The paper tackled the problem of detecting clinically significant prostate cancer in bi-parametric MRI by encoding anatomical priors into 3D convolutional neural networks, resulting in up to an 8.70% increase in AUROC for patient-based diagnosis and an average 1.08 pAUC increase for lesion-level detection.
We hypothesize that anatomical priors can be viable mediums to infuse domain-specific clinical knowledge into state-of-the-art convolutional neural networks (CNN) based on the U-Net architecture. We introduce a probabilistic population prior which captures the spatial prevalence and zonal distinction of clinically significant prostate cancer (csPCa), in order to improve its computer-aided detection (CAD) in bi-parametric MR imaging (bpMRI). To evaluate performance, we train 3D adaptations of the U-Net, U-SEResNet, UNet++ and Attention U-Net using 800 institutional training-validation scans, paired with radiologically-estimated annotations and our computed prior. For 200 independent testing bpMRI scans with histologically-confirmed delineations of csPCa, our proposed method of encoding clinical priori demonstrates a strong ability to improve patient-based diagnosis (upto 8.70% increase in AUROC) and lesion-level detection (average increase of 1.08 pAUC between 0.1-10 false positives per patient) across all four architectures.