Kidney Abnormalities Detection and Classification Using CNN-based Feature Extraction
Mino George, Anita HB · 2022
The presents of noises degrade the quality of ultrasound images and diminishes the disease diagnosis accuracy. Thus, an effective automatic stone and cyst detection system is beneficial to both the medical practitioners and patients. In this paper, an automatic detection and classification system for kidney stone and cyst image is proposed. The Gaussian filtering and Contrast Limited Adaptive Histogram Equalization (CLHE) techniques are applied to improve the quality of the images. In the next step, segmentation has been done based on the entropy of the image. The gamma correction technique has been applied to improve the overall brightness and an optimal global threshold value is selected to extract the region. The CNN model has attained much attention in medical image recognition and classification. In this paper, the pre-trained model ResNet-50 is utilized as a feature-extractor and Support Vector Machine as classifier to categorize the normal, cyst and stone images. The CNN model is analyzed with various other classification models such as k-nearest neighbor, decision tree and Naïve Bayes. The results demonstrate that the ResNet-50 with supervised classification algorithm SVM is an optimal solution for analyzing kidney diseases.