Small bowel image classification using dual tree complex wavelet-based cross co-occurrence features and canonical discriminant analysis

In this paper, a modified algorithm is proposed for automatically classifying small bowel images into normal or abnormal class. Instead of the shift-invariant overcomplete wavelet transform, our modification performs the dual tree complex transform (DTCWT) to the small bowel images for three-decomposition scales and clamp and linearly scale the DTCWT subbands. As the original algorithm, we extracts cross co-occurrence matrix from each DTCWT subband, and calculate four textural features from each cross co-occurrence matrix. Unlike the original algorithm, we select a subset of the calculated texture features by means of minimum redundancy maximum relevance (mRMR) algorithm. We use canonical discriminant analysis as a classifier in order to classify a small bowel image into normal or abnormal class, just like the original algorithm. Experimental results show that our proposed modification outperforms the original algorithm for small bowel image classification by 5.3% in terms of correct classification rate for the same dataset.