Lung Disease Diagnoses Using Hybrid Multi-Wavelet Transform and Deep Convolution Features with Support Vector Classifier

Artificial intelligence (AI) has demonstrated its utility in different industries along with its growth, particularly in the past few years with the massive rise in accessible data.Its primary job is to assist in achieving better, quicker, and more trustworthy decisions.AI and machine learning (ML) are progressively discovering applications in medicine.The situation is particularly relevant in the medical sector, which employs many forms of medical images, and diagnostic processes depend on gathering and analyzing a vast amount of data.Implementing machine learning (ML) in medical data analysis supports uniformity and enhances efficiency in early diagnosis.COVID-19, pneumonia, tuberculosis, lung opacity, fibrosis, mass, infiltration, Effusion, pneumothorax, and healthy cases are spread in all developing countries.The prediction of lung illness is generally provided with picture observation in X-ray photos of the chest, particularly with minimal resource regulation.This study presented a novel model for automated categorization and detection of various lung illnesses from chest X-ray pictures utilizing a hybrid convolution neural network and support vector machine method.The model started from a pre-processing step that enhanced the input Xray image resolution using contrast-limited adaptive histogram equalization (CLAHE).All the data of images were pre-processed: resizing 256x256 pixels and augmentation of the pictures (rotating, cropping, and reflecting).This work proposed a novel feature extraction method that integrates the two-dimensional discrete multi-wavelet transform (2D-DMWT) and principal component analysis (PCA).2D-DMWT was used to get discriminant characteristics from datasets (images) and reduce dimensions in the feature selection phase by principal component analysis (PCA).A deep neural network model predicated on a Convolution Neural network (CNN) is employed for deep feature extraction.Finally, support vector machine (SVM) is utilized for multi-class classification.An accuracy, sensitivity, and specificity of 99.3%, 99.8%, and 99.93% were obtained for multi-class classification (12 classes).The model also facilitates the acquisition of images from radiographic pictures, thereby broadening its application in regions where only traditional X-ray images are accessible.The number of images we used in this paper was 20400, 1700 images for each disease.