Analyzing Cross-Population Domain Shift in Chest X-Ray Image Classification and Mitigating the Gap with Deep Supervised Domain Adaptation

Medical image analysis, empowered by artificial intelligence (AI), plays a crucial role in modern healthcare diagnostics. However, the effectiveness of machine learning models hinges on their ability to generalize to diverse patient populations, presenting domain shift challenges. This study investigates the domain shift problem within chest X-ray classification, with a particular emphasis on cross-population variations, especially within underrepresented groups. We examine the domain shift of a supervised version of Adversarial Domain Adaptation (ADA) across three distinct population datasets (sources), using a Nigerian chest X-ray dataset as the target dataset. By evaluating model performance, we quantify the disparities between the source and target populations. Our experiments revealed varying model performance when trained on the source domain and evaluated on the target domain. To address this variability, we propose a supervised domain adaptation technique that leverages labeled data from both domains for fine-tuning. The results demonstrate significant enhancements in model accuracy for chest X-ray classification in the Nigerian dataset. This research underscores the importance of domain-aware model development in AI-driven healthcare, contributing to addressing cross-population domain-shift challenges in medical imaging.