Subtle changes in benign tissue adjacent to prostate neoplasia detected with a Bayesian belief network.

The aim of this paper was to test the usefulness of a Bayesian belief network (BBN) as a decision support system in the uncertainty assessment of benign prostatic tissue, either associated or not with inflammation or adjacent to prostatic adenocarcinoma (PAC) or prostatic intraepithelial neoplasia (PIN). A shallow network was used with eight first-level descendant nodes for the diagnostic clues, each independently linked by a conditional probability matrix to a root node containing the diagnostic alternatives. One diagnostic evidence node was based on the tissue architecture and the others were based on cell features. The efficacy of the network was tested on a series of 45 simple prostatectomy specimens, subdivided as follows; benign prostatic tissue not associated with other diseases (15 cases), associated with acute and/or chronic inflammation (15 cases), and adjacent to accidentally discovered PAC or PIN (15 cases). The highest belief values for the diagnostic alternative normal prostate (NP) were obtained in the 15 cases not associated with other diseases, the mean value being 0.996. The 15 cases evaluated in areas with inflammation showed the lowest belief values for NP (mean 0.774). For the 15 cases evaluated in specimens with PAC or PIN, the belief values for NP were intermediate between those from normal prostatic tissue associated with inflammation and those not associated (mean 0.925). Moreover, it was found that subtle changes were also present at a certain distance from the tumour. In conclusion, the network can be used as a decision support system to differentiate with high certainty benign prostate adjacent to PAC or PIN from benign prostatic tissue either associated or not with inflammation. The subtle morphological alteration detected with the BBN may be considered malignancy-associated changes.