FCMHW PROGRAM: Deep decision-making for prostate tumour Gleason Score 7 – Active surveillance or radical treatment?
Contact: Dr Gobert lee
Overview:
There are more than 18,110 new cases of prostate cancer diagnosed in Australian men each year. An estimated 60% of men with low risk, low grade, confined prostate cancers in Australia are managed with active surveillance whereby they do not undergo treatment, thus avoiding, or extending the time until they experience treatment related toxicities.
Decisions regarding treatment for moderate risk disease is made difficult by the reliability of current pathology grading of prostate tissue. The Gleason score, used to pathologically grade prostate tumours, is the total of two scores (1 to 5) assigned to the primary and secondary patterns of growth in a biopsy.
A Gleason score 6 (i.e. 3+3) is the lowest score for prostate cancer and active surveillance is increasingly being adopted as the standard of care for this score. Higher Gleason scores are indicative of more aggressive cancers requiring treatment.
A Gleason score 7 tumour where the primary pattern of growth is a score 3 (3+4) is slower growing and may be better treated by active surveillance, whereas a Gleason score 7 where the primary pattern of growth is a score 4 (4+3), is more aggressive and radical treatment is recommended yet is associated with higher risk of side effects including urinary incontinence, erectile dysfunction and bowel dysfunction, all of which may dramatically impact on men’s quality of life.
An accurate pathological grading, particularly for Gleason 7 influences treatment decisions and outcomes of treatment, yet grading, performed microscopically by a pathologist, can vary considerably with inter-observer agreement being only moderate (72%) and manifestations within a tumour pattern also impacting the score reliability.
An artificial intelligence approach may improve objectivity in tumour grading. This project, which brings together expertise in clinical pathology, machine learning, statistics and computer science, aims to use a Deep Learning approach to analyse prostate biopsy images to better distinguish between Gleason 3+4 and 4+3 more reliably.