A Comparative Study of Non-Optimized, Semi-Optimized, and Fully Optimized Hyper parameters in Deep Learning

The rapid growth of deep learning in medical imaging has created both opportunities and challenges in diagnosing lung diseases from chest X-rays (CXR). While several studies have explored this area, questions remain about the optimal approach to model optimization. In this study, we analyze how different levels of hyperparameter optimization affect the performance of deep learning models in detecting pneumonia and COVID-19 from CXR. Using the NIH CXR Dataset, we evaluate a ResNet-50 architecture under three optimization scenarios: baseline (non-optimized), partially optimized, and fully optimized hyperparameters. Beyond reporting technical results, we synthesize recent developments in the field and provide practical recommendations for researchers, including an analysis of publicly available datasets and detailed methodological guidance. Our work reveals important trade-offs between model performance and computational demands, offering insights that help bridge the gap between theoretical advances and clinical implementation.