Exploring the Impact of CLAHE Processing on Disease Classes ‘Effusion,’ ‘Infiltration,’ ‘Atelectasis,’ and ‘Mass’ in the NIH Chest XRay Dataset Using VGG16 and ResNet50 Architectures

In this study, transfer learning was applied to VGG16 and ResNet50 models using the NIH Chest X-ray Dataset to classify chest X-ray images. The models were fine-tuned without altering their weights, leveraging previously learned knowledge from a different task for application in a new task. A distinctive aspect of this research involved the selection of images specific to the ‘Effusion,’ ‘Infiltration,’ ‘Atelectasis,’ and ‘Mass’ diseases, creating dedicated train, test, and validation datasets. Copies of the original images were subjected to Contrast Limited Adaptive Histogram Equalization (CLAHE) to create a separate dataset after applying the algorithm. To assess the impact of the CLAHE algorithm on classification results, separate models were run for processed and unprocessed images, resulting in four distinct models: CLAHE-applied VGG16, non-CLAHE-applied VGG16, CLAHE-applied ResNet50, and non-CLAHE-applied ResNet50. The performance of the generated models was thoroughly evaluated.