Imperial Prostate 6 - Cancer Histology Artificial Intelligence Reliability Study.

In the UK, about 80-100,000 men every year undergo prostate biopsy to diagnose prostate cancer. This equates to approximately 4 million histology slides; this is estimated to increase to 160,000-200,000 men and up to 6 million slides by 2030 due to rising numbers of men being tested for prostate cancer. Health Education England and the Royal College of Pathology point to a significant pathology work-force shortage with only 3% of departments having adequate staffing levels and a 10% vacancy rate filled by locums costing £26M every year. By 2021, there will be a 3% decrease of the pathology consultant workforce (40 full-time pathologists); a period of time in which other specialties are expected to see a 13% increase. However, to meet the rising numbers of referrals to pathology departments, it is projected that there will need to be a 3-5% annual growth in the number of pathologists. Inter-observer variability can occur between pathologists in terms of reporting a diagnosis of clinically important and clinically unimportant prostate cancer by as much as 20% although the differences are smaller when highly expert uro-pathologists are compared. This can lead to inappropriate management of cases. Galen Prostate AI is a CE-marked deep learning AI-algorithm for prostate needle biopsies that can identify cell types, tissue structures and morphological features for cancer diagnosis. The technology is based on multi-layered convolutional neural networks (CNNs) designed for image classification in which whole-slide imaging is analysed for the detection of tissue areas and then benign versus cancer versus other pathology classification. Compared to almost all competitors, Galen Prostate AI has been tested in \~10 times more tissue samples. Further, Galen Prostate AI is the only algorithm that extends beyond cancer detection/grading to other clinically relevant features (e.g., perineural invasion, high-grade prostatic intraepithelial neoplasia \[PIN\], inflammation). This AI-algorithm is believed to be the only one in routine clinical deployment - demonstrating technical feasibility and with proven clinical utility. The proposed study will perform validation in the NHS, for the first time. It is important to stress that this type of algorithm has never been tested on a UK-based population, and in particular, a population that includes a cohort of MRI targeted biopsies, which is now the new diagnostic strategy as it detects clinically relevant prostate cancer in higher percentages than the routine systematic biopsy. The study is the first and only to address the performance of the AI-based prostate algorithm that extends beyond cancer detection and Gleason grading, by measuring amount of cancer and detecting clinically meaningful features such as perineural invasion in addition to multiple benign structures (e.g. HGPIN, atrophy, inflammation). Given the clinical relevance for such features in the diagnosis process, a study addressing their validation and performance is not only novel, but critical for implementation in routine clinical use.