A Block-based Approach for Malignancy Detection within the Prostate Peripheral Zone in T2-weighted MRI

In this paper, a computer-aided diagnosis method is proposed for the detection of prostate cancer within the peripheral zone. Firstly, the peripheral zone is modelled according to the generic 2D mathematical model from the literature. In the training phase, we captured 334 samples of malignant blocks from cancerous regions which were already defined by an expert radiologist. Subsequently, for every unknown block within the peripheral zone in the testing phase we compare its global, local and attribute similarities with training samples captured previously. Next we compare the similarity between subregions and find which of the subregion has the highest possibility of being malignant. An unknown block is considered to be malignant if it is similar in comparison to one of the malignant blocks, its location is within the subregion which has the highest possibility of being malignant and there is a significant difference in lower grey level distributions within the subregions. The initial evaluation of the proposed method is based on 260 MR images from 40 patients and we achieved 90% accuracy and sensitivity and 89% specificity with 5% and 6% false positives and false negatives, respectively.