Medical Image Segmentation And Classification On Efficiently Fused Image Using Fast Fuzzy C Means Algorithm And Convolutional Neural Network

Abstract

In the recent past, medical image processing plays significant role in diagnosis of disease using Computer Aided Diagnosis (CAD). In this research, we propose a novel approach for classification of medical images using Fast Fuzzy C-Means (FFCM) clustering and Convolutional Neural Networks (CNN). Initially, the images were pre-processed using filtering and enhancement techniques. Image filtering was performed using 2D Gabor Filter. This step helped to remove noise in the medical data. Then, image enhancement was performed using Edge Preservation-Contrast Limited Adaptive Histogram Equalization (EP-CLAHE) technique. Fusion of medical data belonging to different modality results in the generation of a single image that has extended information content and helps to increase the reliability of disease diagnosis. The images were fused using 2-Dimensional Double Density Wavelet Transform (2D-DDWT) and Empirical Principle Component Analysis (EPCA). Segmentation plays a crucial role in detecting tumor cells in medical images. Here, segmentation was performed using FFCM clustering algorithm. The FFCM clustering helps in achieving accurate segmentation results with reduced computational complexity. The efficiency and reliability of a classification algorithm depend on the type of features extracted from the classification data. In our research, Gray Level Co-Occurrence Matrix (GLCM) features were extracted from the segmented data. Deep Learning (DP) technique is widely used for classification of image using significant features with high accuracy and efficiency. Using these features, classification was performed using CNN. The images were classified into benign and malignant. The simulation was performed using publicly available datasets. The outcome of the research shows that the proposed scheme was very effective in the classification of tumor images. The classification performance of the proposed framework was validated using metrics like recall, precision, specificity, F-score and accuracy. Experimental results demonstrate the credibility of the proposed framework and prove that the proposed scheme outperforms state-of-the-art works in the research.