Study of Fusion of medical images and classification comparison using different kernels of SVM and K-NN classifiers

In this paper, the methodology of detecting cancer, has been presented. The database consists 200 samples of PET & CT scans each. 50% of the samples were non cancerous. Wavelet based decomposition is performed for each PET & CT scan images, using haar mother wavelet with depth 5. After decomposition, averaging fusion rule has been applied for fusing the four details of wavelet decomposition (approximation, horizontal, vertical & diagonal details) to get a fused image having both details of PET and CT images. Further, these 200 fused images are segmented manually by cropping method to extract the region of interests (ROIs). GLCM technique has been used to classify images in which 17 features have been extracted. Extracted features are examined further by using support vector machine (SVM) and k-nearest neighbors algorithm (k-NN). The accuracy percentage of SVM classifier vary from 95.5% - 98% and accuracy of k-NN classifier vary from 69.5% - 95.5%. This indicates that fused images can be a more powerful tool to diagnose the lung cancer.