Impact of Image Filtering Techniques on Deep Learning-Based Kidney Disease Classification Using CT scans

Medical images are one of the most effective tools that is required for automated diagnosis because of its ability to capture pictorial view of an infected area or region. In any classification task that has to do with medical diseases, medical images are very useful to achieve such task. The more the region of interest is visible to the deep learning model or tool, it increases the chances of accurate classification and accuracy of detection. This study evaluates the impact of image enhancement strategy such has the application of image processing filters such has Gaussian blur, Median filter and bilateral filters to different kidney diseases such as Cyst, Tumor, and stone to evaluate the impact of image filters on deep learning models specifically in this case VGG 19. The result for indicated that image filtering methods has some influence on the deep learning performance by enhancing feature extraction and noise reduction. Gaussian blur filter has a 97% accuracy which outperforms others, bilateral filter had an accuracy of 96%, median filter and no filter resulted in a 95% accuracy of the deep learning model. This suggest that noise in CT imaging has the ability to negatively impact classification and that not all filtering techniques has the ability to equally enhance performance.