Feature Extraction Techniques for Classifying Chest X-Ray Abnormalities

Chest X-Ray (CXR) is commonly utilized to detect abnormalities in the heart and lung region. Accurately classifying and detecting these problems might significantly improve real-world diagnostic procedures. Automated algorithms and technologies can enhance the reading process and increase productivity. CXR imaging is a cost-effective, accessible, user-friendly imaging option for diagnosing and screening lung disorders caused by infectious illnesses such as Covid19, Pneumonia, and Normal. This research suggests a technique for categorizing and identifying chest irregularities by utilizing image processing with CXR. A set of procedures are applied, including resize the image, gray-scale conversion, then noise removal, thresholding, morphological operations and canny edge detection for preprocessing, LBP and HOG feature extraction and classification done through different deep learning (DL) classifiers such as CNN and Transfer learning methods ResNet50, DenseNet. InceptionV3, MobileNetV2, VGG16. The evaluation result shows high diagnosis accuracy of 98.95% with InceptionV3 classifier and HOG feature extraction technique. We evaluate the efficacy of our suggested work by comparing the experimental outcome to the existing literature.