• Light
  • Dark
  • Auto
Select Page

Prostate MRI can be a game-changer for many men with elevated prostate-specific antigen (PSA). For decades these many men underwent biopsies while never developing prostate cancer. Expert prostate MRI can help avoid these unnecessary biopsies and better target any biopsies. Unfortunately, reading prostate MRI is challenging and time-consuming. Like other medical imaging modalities, AI is explored for helping read prostate MRI. A requirement is that AI achieves expert performance. MRI is challenging because it comes with relatively many artifacts. Moreover, prostate MRI is multi-parametric, requiring AI to simultaneously read highly differing images.

This study investigated if deep learning (DL) can predict expert radiologists that use the Prostate Imaging Reporting and Data System (PI-RADS). AI and radiologists were then evaluated on an independent pathology confirmed data set. We show that training data size is hugely important on the most extensive data set to date (2,734). Expert performance is feasible but requires many more cases. Therefore, we are extending our dataset to include more subjects and multiple centers.

Key points

  • AI for prostate MRI analysis depends strongly on data size and prior zonal knowledge.
  • AI needs substantially more than 2000 training cases to achieve expert performance.
  • Be careful when AI claims to achieve ‘expert’ performance. Our experts achieve an AUC of 0.90, allowing our center to avoid many unnecessary biopsies without missing prostate cancer. Many AI papers show ”experts” with an AUC of 0.70 – 0.80, likely insufficient to reduce biopsies. One article showed AI ‘beating’ radiologists operating at AUC=0.50!

Article: Deep learning–assisted prostate cancer detection on bi-parametric MRI: minimum training data size requirements and effect of prior knowledge

Authors: Matin Hosseinzadeh, Anindo Saha, Patrick Brand, Ilse Slootweg, Maarten de Rooij & Henkjan Huisman

Latest posts