Statistical Asymmetrical Histogram Stretching for Contrast Enhancement in Chest X-ray Images for Pneumonia Detection

Abstract

This paper presents an approach to enhance contrast in chest X-ray images which improves pneumonia detection when using several convolutional neural networks (CNNs). We introduce a Statistical Asymmetrical Histogram Stretching (SAHS) technique, which addresses the inherent asymmetry in chest X-ray image histograms. Our approach is compared with conventional techniques, like HE, HS and CLAHE, across various CNN architectures including AlexNet, Compact, Enhanced, ResNet-18, MobileNetV2, and ResNet-50. The SAHS method, combined with a Lung Finder Algorithm (LFA), significantly improved classification accuracy across all tested CNN models. SAHS consistently outperformed the conventional methods evaluated (HE, HS, CLAHE), demonstrating its effectiveness, particularly in preserving diagnostically relevant bright regions often altered by other techniques. Therefore, our results demonstrate SAHS is a valuable preprocessing technique for enhancing pneumonia recognition from chest X-ray images.