Prostatic intraepithelial neoplasia. Development of a Bayesian belief network for diagnosis and grading

The diagnosis and grading of prostatic intraepithelial neoplasia (PIN) are affected by uncertainties that arise from the fact that almost all our knowledge of PIN histopathology is not expressed in numeric form but rather in descriptive linguistic terms, words and concepts. A Bayesian belief network (BBN) was used to reduce the problem of uncertainty in diagnostic clue assessment while considering the dependencies between elements in the reasoning sequence. A shallow network was developed with an open-tree topology, with a root node containing the diagnostic alternatives and seven first-level descendant nodes for the diagnostic clues. One of these nodes was based on tissue architecture and the others on cell features. The results obtained with prototypes of relative likelihood ratios showed that beliefs for the diagnostic alternatives are very high. The network can grade and differentiate PIN lesions from other prostate lesions with certainty. A number of diagnostic clues greater than seven did not significantly improve network performance, whereas a reduced number of clues resulted in decreased beliefs. A BBN for PIN diagnosis and grading offers a descriptive classifier that is readily implemented and allows the use of linguistic, fuzzy variables. A BBN allows the accumulation of evidence presented by diagnostic clues, each offering only weak evidence.