Analyzing the Effect of Contrast Enhancement Techniques on Chest X-Ray Classification Using Deep Learners

The accurate classification of medical images using deep learning techniques can significantly aid doctors in their medical diagnosis. However, medical datasets often have limited data and low image quality, affecting the detection of pathological conditions. This study evaluates deep learning models Convolutional Neural Networks (CNN) and pre-trained transfer learner VGG-16, to classify chest X-ray images for tuberculosis detection. We also evaluate the impact of different contrast enhancement techniques: Histogram Equalization (HE), gamma correction, and Contrast Limited Adaptive Histogram Equalization (CLAHE). Our results demonstrate that the pre-trained VGG-16 model outperforms CNN trained on the dataset. The choice of preprocessing method varies depending on the classification model. While gamma correction improves the classification performance of CNN, VGG-16 works well with CLAHE. On the other hand, HE fails to provide any improvement. The finding highlights the importance of optimal preprocessing and transfer learning in overcoming dataset limitations and enhancing diagnostic capabilities in medical imaging.