Abstract
A new automated algorithm for accurate and reliable decision-making in the discrimination of normal and cancerous colon mucosa is proposed. Quantitative texture features such as entropy, angular second moment, contrast, inverse angular moment, correlation, homogeneity, were extracted from the co-occurrence matrix whilst other features are based on morphology such as Euler number, convex area, nuclear contour index, elongation, shape factor (BE) and fractal dimension. 46 samples from different patients consisting of 22 normal microscopic specimens and 24 adenocarcinoma images (512×512×3) were analyzed. Extracted features from both dimensions were able to identify abnormalities (P < 0.0001) between colon tissue types. A parametric approach using a linear discrimination method was implemented for the classification stage. Combining texture and morphological features shows that a ratio of 98.3 % and 97.7 % is obtained for the sensitivity and specificity, respectively. Only one case from each class was wrongly misclassified. The proposed algorithm achieves a very significant result with an overall accuracy of 98 % for the identification of colon microscopic images