Classification of celiac disease using novel approach of three-level DWT decomposition and linear support vector machine

In the medical field, the requirement for automated medical diagnosis software has risen as the application uses machine learning to facilitate health analysis from different data points from a patient comparing it with massive amounts of medical data to diagnose and prevent disease. The automated system can run multiple tests simultaneously and thus resulting to faster turnaround time and enables timely patient care along with higher accuracy and reliability. The irregularities in the small intestine villi’s structure cause various autoimmune disorders. So, this work aims to find a novel technique for the feature extraction of upper endoscopy images of celiac disease. We employed CLAHE (Contrast Limited Adaptive Histogram Equalization) for the preprocessing step and the Sobel operator with gradient magnitude for the segmentation of the enhanced image. In this manuscript, we have proposed a novel approach by calculating texture features through gray level co-occurrence matrix and frequency analysis using 3-level discrete wavelet transform decomposition on endoscopy images that renders novelty to the work. Thereafter, linear SVM (support vector machine) with PCA (Principal Component Analysis) is used for classification. The ensemble approach attains accuracy, sensitivity, and specificity of 78.49 %, 90.32 %, and 77.27% respectively. The outcomes achieved with the suggested approach are compared with some state-of-the-art methods using the same dataset. The results are promising due to high sensitivity for the treatment of untreated celiac disease and can prove boom to the medical industry by assisting clinicians to diagnose the disease at an early stage.