PURPOSE: to evaluate the predictive quality of computational models to differentiate colic tissues, based on Cooccorrurence Matrices (MC) representation of Coloscopic Images (IC). MATERIALS AND METHODS: image analysis and artificial intelligence methods were employed to construct computational models. Sixty seven IC images, containing polyp, were considered in this work, from which a part containing a polypus and another without it were collected given origin to 134 images. For each one of these, different MC were constructed considering five distance parameters (D = 1 to 5) and the extraction of 11 texture characteristics. With this representation, five computational models were generated based on decision trees. These models were evaluated using two techniques: (a) cross-validation and (b) contingency tables. RESULTS: for the (a) analysis, the model with D = 3 presented the smaller average error (22.25% ± 11.85%). For the (b) analysis, models with D = 1 and 3 presented the best precision values. CONCLUSION: parameters D = 1 and 3 presented models with the best predictive qualities. Results showed that the constructed models were promising to be applied within decision making computational systems.
Colonoscopy; Intestinal Polyposis; Colonic Neoplasms; Artificial Intelligence; Image Interpretation