"PhD Studentship: Reasoning in Efficient Medical Vision–Language Models for Trustworthy AI in Healthcare"

Medical imaging is central to the diagnosis and management of lung and abdominal diseases, including lung cancer, pneumonia, liver disease, and gastrointestinal disorders. In real clinical workflows, radiologists interpret scans such as CT, MRI, and X-ray together with textual information, including patient history, laboratory findings, and prior reports. Although recent Vision–Language Models (VLMs) can jointly learn from images and text, most current systems are still limited in three important ways: they primarily rely on statistical pattern recognition rather than structured clinical reasoning, they are computationally expensive and difficult to deploy in resource-constrained healthcare environments, and they often function as “black boxes,” producing predictions that are hard for clinicians to interpret or trust.

This PhD project aims to develop efficient, reasoning-enhanced Vision–Language Models tailored to multimodal medical data. The main aim is to investigate how explicit clinical reasoning can be embedded into lightweight, deployable VLMs without sacrificing performance. The research will explore architectures that integrate medical images and clinical text while remaining parameter-efficient through modern techniques such as modular design and efficient fine-tuning. The project will also study the trade-offs between model efficiency, reasoning quality, and interpretability, developing methods that provide clinically meaningful explanations. Finally, the project will develop new evaluation frameworks to assess reasoning quality, robustness, computational efficiency, and real-world clinical usefulness, supporting the development of trustworthy and deployable medical AI systems.