Cancer - Prostate

Prostate cancer is the second most commonly diagnosed cancer in men worldwide and the fifth leading cause of cancer death in men. Despite substantial research, the etiology of prostate cancer is not yet well understood. While age, race and family history have shown the strongest associations, no conclusive preventable risk factors have been identified. Therefore, work is ongoing to identify improved methods for accurate early diagnosis of prostate cancer.

Traditionally, the prostate-specific antigen (PSA) blood test has been used as a screening tool to determine which patients require confirmatory transrectal ultrasound-guided (TRUS) biopsy. However, PSA alone has a low specificity due to elevation under benign circumstances and TRUS biopsy misses clinically significant prostate cancer. These deficiencies can lead to incorrect diagnoses, inaccurate risk assessments and suboptimal therapy choices.

Multiparametric magnetic resonance imaging (mpMRI) is increasingly being incorporated into the prostate cancer diagnostic pathway to enable non-invasive cancer detection, targeted biopsy and targeted treatment planning. Whilst imaging protocols vary across centres, the most commonly collected sequences, and those recommended by the revised Prostate Imaging Reporting and Data System (PI-RADS v2) and Likert assessment system are, T2-weighted imaging (T2WI), diffusion-weighted imaging (DWI) and dynamic contrast-enhanced imaging (DCEI). Studies such as the “Prostate MRI Imaging study” (PROMIS) have shown that mpMRI prior to biopsy can identify men with elevated PSA who might safely avoid biopsy and that mpMRI followed by targeted biopsy can reduce the over-diagnosis of clinically insignificant cancer and improve the detection of clinically significant cancer.

However, reading mpMRI requires a high-level of expertise, can be prone to inter/intra-observer variability and is a time-consuming task. Computer-aided diagnosis (CAD) systems can help enable a more consistent assessment and reduce reading time. Our work focuses on developing CAD tools that use artificial intelligence to perform prostate cancer classification at the voxel, lesion and patient levels. Further, we hope to exploit uncertainty quantification methods to identify patient cases where CAD systems may fail due to factors such as poor registration between modalities or imaging artifacts.