Fast Accurate Detection and Classification of Kidney Diseases from CT Images using Hybrid Classifiers

Document Type : Original Article

Authors

1 Assistant professor at Egyptian Atomic Energy Authority (EAEA), National Center for Radiation Research and Technology (NCRRT), Radiation Engineering Dept.,

2 associate professor at Egyptian Atomic Energy Authority (EAEA), National Center for Radiation Research and Technology (NCRRT), Radiation Engineering Dept.,

3 assistant professor at Radiation Engineering Department, National Center for Radiation Research and Technology (NCRRT), Egyptian Atomic Energy Authority, Cairo 11787, Egypt.

Abstract

This research introduces an innovative method of Artificial Intelligence (AI) for improving the detection and classification of kidney
diseases using CT images. The proposed method includes image preprocessing to remove artifacts, noise, and other image quality issues
that can affect the accuracy of diagnosis. Then the area of interest in each image is segmented using Fractional Darwinian particle
swarm optimization. Different features including Local Binary Pattern, Hu Moments, and Gray level zone length matrix (GLZLM) are
extracted and fused using Canonical Correlation Analysis (CCA) and
reduced using Two Dimensional Principal Component Analysis (2DPCA) to maintain only dominant features. A two-level classification
approach is carried out to provide both fast and detailed diagnosis
using both Binary Support Vector Machine (BSVM) and Convolutional
Neural Network (CNN) in sequence. BSVM is used to initially discriminate between normal and kidney diseases categories. Afterwards, the
detected abnormal kidney images are classified using CNN to different kidney diseases such as stones, cysts, and tumors. This approach
aims to expedite the diagnostic procedure while also enhancing the efficiency and accuracy of classifying kidney disease in the clinical practice.
Obtained results validate the efficiency of our proposed in terms of
achieved accuracy when compared to alternative cutting-edge methods.

Keywords

Files

Share

How to cite

Statistics