Artificial neural networks distinguish among subtypes of neoplastic colorectal lesions

Background & Aims: There is a subtle distinction between sporadic colorectal adenomas and cancers (SAC) and inflammatory bowel disease (IBD)-associated dysplasias and cancers. However, this distinction is clinically important because sporadic adenomas are usually managed by polypectomy alone, whereas IBD-related high-grade dysplasias mandate subtotal colectomy. The current study evaluated the ability of artificial neural networks (ANNs) based on complementary DNA (cDNA) microarray data to discriminate between these 2 types of colorectal lesions. Methods: We hybridized cDNA microarrays, each containing 8064 cDNA clones, to RNAs derived from 39 colorectal neoplastic specimens. Hierarchical clustering was performed, and an ANN was constructed and trained on a set of 5 IBD-related dysplasia or cancer (IBDNs) and 22 SACS. Results: Hierarchical clustering based on all 8064 clones failed to correctly categorize the SACS and IBDNs. However, the ANN correctly diagnosed 12 of 12 blinded samples in a test set (3 IBDNs and 9 SACS). Furthermore, using an iterative process based on the computer programs GeneFinder, Cluster, and MATLAB, we reduced the number of clones used for diagnosis from 8064 to 97. Even with this reduced clone set, the ANN retained its capacity for correct diagnosis. Moreover, cluster analysis performed with these 97 clones now separated the 2 types of lesions. Conclusions: Our results suggest that ANNs have the potential to discriminate among subtly different clinical entities, such as IBDNs and SACS, as well as to identify gene subsets having the power to make these diagnostic distinctions.