Addressing Class Imbalance for Improved Pneumonia Diagnosis: Comparative Analysis of GANs and Weighted Loss Function in Classification of the Chest X-Ray Images
Radhika V. Kulkarni, Ameya Paldewar, Sanket Paliwal, Palwe Amol Ajinath, Soham Panchal · 2024
Pneumonia detection using the chest X-ray images is genuinely, a critical task in the medical field of study and research, still there are multiple prevailing challenges, that include the overfitting and the lack of generalization in the existing models, because of scarcity in data and class imbalance. In response, the study on the problem statement introduces a comparative analysis that aims at significant enhancement of accuracy in pneumonia diagnosis. Our approach highlights on evaluating the effectiveness of two distinct methodologies: a weighted loss function and Generative Adversarial Networks (GANs). By putting the class imbalance within medical datasets into consideration, the weighted loss function targets the underrepresented pneumonia cases. By prioritizing these classes during model training, the weighted loss function seeks to give a boost and attention on sensitivity and consider false positives and negatives effectively, thus refining the diagnosis process. Along with this, our study holds the potential of GANs for data augmentation, aiming to almost nullify the impact of data scarcity. By generating synthetic but realistic chest X-ray images, GANs improve the training dataset, leading it to improvement in the model's robustness and generalization. The comparative analysis and study of both the models in enhancing pneumonia detection accuracy, provides critical points into the strengths and limitations of each approach. The outcomes of this comparative assessment not only offer valuable benchmarks for future advancements but also enlighten the way heading towards more reliable and effective pneumonia diagnosis from chest X-ray images.