We are building a decision support system for pathologists to more accurately identify and separate in tissue samples slow-growing prostate cancer from more aggressive types, thereby increasing survival rates and reducing aggressive treatment that does not prolong life but often results in debilitating side effects. We have produced a large database consensus-graded by a group of international experts, with detailed examples of each morphological pattern occurring in prostate tissue. This database is the training dataset for a high-throughput cancer decision support system that will allow grading quality assessment. The support system uses image analysis to extract features in the tissue known to be linked to malignancy, and deep learning for automatic feature extraction.
UPPMAX will be used to validate the algorithm on a test database of 16 whole mounts of prostatic tissue from prostatectomies. Each whole mount is an image of around 100.000px x 100.000px.