Preprocessing and Optimizer impact on Image-Based Autoimmune Disease Diagnosis

The precise diagnosis of autoimmune diseases is challenging due to overlapping symptoms. Despite significant advancements in techniques based on Convolutional Neural Networks (CNNs), research opportunities remain in evaluating the impact of external factors on CNN performance. This study investigates CNN behavior by focusing on two factors: image preprocessing and optimizer choice. We assessed the effects of contrast stretching and histogram equalization on HEp-2 images and compared the Adam and SGD optimizers. Using a factorial experimental design, we trained and validated CNN models on a dataset from the International Conference on Pattern Recognition (ICPR-2014) competition, categorized into six classes of HEp-2 cells. This study contributes to understanding how different factors influence CNN models in medical image processing.