Classification of primary and secondary malignant liver lesions using Laws' mask analysis and PNN classifier

As ultrasound images offers limited sensitivity for differential diagnosis of malignant liver lesions in the present work, an efficient computer aided classification system have been designed for this task using different ROI extraction protocols, i.e.: experiment 1) IROIs (multiple inner ROIs that lie within the boundary of the lesion) one NROI (neighbouring ROI from the region surrounding the lesion); experiment 2) LROI (a single largest ROI from the region within the lesion) and the corresponding NROI; experiment 3) GROI (a single global ROI which includes the complete lesion and the surrounding area). Texture feature extraction has been carried out using Laws' mask analysis. The probabilistic neural network has been used extensively for the classification task. From the results it can be concluded that concatenated feature vector consisting of texture features computed using Laws' mask of length 3 extracted from LROI and NROI combined with texture features computed using Laws' mask of length 7 from the corresponding GROI yields maximum classification accuracy of 93.3%.