Towards Confident Detection of Prostate Cancer using High Resolution Micro-ultrasound

Abstract

MOTIVATION: Detection of prostate cancer during transrectal ultrasound-guided biopsy is challenging. The highly heterogeneous appearance of cancer, presence of ultrasound artefacts, and noise all contribute to these difficulties. Recent advancements in high-frequency ultrasound imaging -micro-ultrasound -have drastically increased the capability of tissue imaging at high resolution. Our aim is to investigate the development of a robust deep learning model specifically for micro-ultrasound-guided prostate cancer biopsy. For the model to be clinically adopted, a key challenge is to design a solution that can confidently identify the cancer, while learning from coarse histopathology measurements of biopsy samples that introduce weak labels. METH-ODS: We use a dataset of micro-ultrasound images acquired from 194 patients, who underwent prostate biopsy. We train a deep model using a co-teaching paradigm to handle noise in labels, together with an evidential deep learning method for uncertainty estimation. We evaluate the performance of our model using the clinically relevant metric of accuracy vs. confidence. RESULTS: Our model achieves a well-calibrated estimation of predictive uncertainty with area under the curve of 88%. The use of co-teaching and evidential deep learning in combination yields significantly better uncertainty estimation than either alone. We also provide a detailed comparison against state-of-the-art in uncertainty estimation.