Optimized ROI Extraction and Pixel Reduction Methods for Kidney Abnormality Detection

A global public health concern, Chronic Kidney Disease causes renal failure, cardiovascular disease, and premature death. Given the frequency of kidney diseases and their significant impact on public health, detection of any abnormalities in the kidney is crucial. Effective patient treatment and medical care planning depend on the early and precise diagnosis of a variety of kidney abnormalities. Early detection enable treatments that enhance patient outcomes, slow the progression of disease, and lessen the strain on healthcare systems. In order to detect anomalies in kidney, we utilized the Region of Interest (ROI) and Pixel Reduction techniques separately in this work. We have applied the Bicubic Interpolation method to reduce the size of individual pixels in an image while preserving a significant data of the original image data. A novel hybrid strategy that aims to preserve diagnostic data while reducing computing complexity uses the original images for ROI extraction. The deep learning method is then independently trained and tested using the pixel-reduced and localized dataset. In order to categorize the images, we use Inception V3 and EfficientNet B7 as Deep Learning models. Using the ROI-extracted data, we were able to achieve approximately 99.75% accuracy with the EfficientNet B7 model. Moreover, the same algorithm and dataset were used to obtain 98.45%, 99.02%, and 98.78% Precision, Recall, and F1-score, respectively. This framework provides medical professionals with a helpful prognostic tool to early detect identify various forms of kidney abnormalities.