Belief network for grading prostate lesions.

A Bayesian belief network for grading prostatic lesions into eight primary Gleason grades was developed and tested. The network employs 13 diagnostic clues, 8 based on tissue architectural features and 5 based on nuclear features. For every diagnostic clue, three to five different outcomes are specified by membership functions. The network works in a robust fashion and attained agreement with consensus visual grading in 241 of 256 microscopic fields.