Deep Learning for Clinical Practice: Enhancing Chest X-ray Diagnostics

This thesis investigates the application of deep learning to enhance chest X-ray analysis, addressing key challenges in clinical diagnostics. Chapter 2 reviews around 300 studies on chest X-ray interpretation with deep learning, identifying critical research gaps and guiding future work. Chapter 3 introduces a segmentation-based approach for detecting cardiomegaly, demonstrating improved performance and explainability over commonly used classification-based approaches. Chapter 4 focuses on deep learning techniques for estimating total lung volume from chest X-rays, revealing the potential of new capabilities beyond standard visual assessments. Chapter 5 presents the NODE21 research challenge, designed to benchmark state-of-the-art methods for lung nodule detection and generation, highlighting the benefits of synthetic image generation when real data is limited. The thesis underscores the potential of automated systems to enhance diagnostic accuracy, reduce radiologists workloads, and support clinical decision-making, ultimately contributing to the development of clinically relevant AI tools for medical imaging.