Deep-learning approaches for Gleason grading of prostate biopsies
Ce dossier présente deux études évaluant la performance de systèmes automatisés, basés sur des technologies de l'intelligence artificielle, pour détecter, à partir d'images histologiques d'échantillons prostatiques, des lésions cancéreuses et déterminer leur grade
Gleason grades are assigned by pathologists based on prostate cancer morphology to describe the loss of tissue structure and order and are strongly correlated with disease aggressiveness and patient outcome. Gleason scoring categorises tumour tissue into patterns from 1 (low risk) to 5 (high risk). Although Gleason grade has long been recognised as being strongly associated with risk of prostate cancer recurrence and metastasis, substantial inter-observer disagreement exists, especially with respect to the intermediate grades. Deep learning, an automated approach using labelled images to train a network with no other assumptions, has proven to be useful in a wide variety of similar areas in digital pathology. Two papers in The Lancet Oncology by Ström and colleagues and Bulten and colleagues use deep learning for detection and Gleason grading of prostate cancer in digital images of biopsies.
The Lancet Oncology , commentaire, 2019