Medical AI: From 3D Imaging and Vision-Language Models to Surgical Robotics and Clinical Agents

About the project

About the Project Applications are welcome from prospective PhD students, research master students, and visiting scholars who are interested in developing advanced artificial intelligence methods for healthcare. This broad research programme focuses on trustworthy, interpretable, and clinically relevant AI systems for medical imaging, surgical robotics, multimodal healthcare data, and intelligent clinical decision support. Modern healthcare generates increasingly complex data, including 3D medical images, surgical videos, clinical reports, electronic health records, physiological signals, and robotic sensor data. However, many existing AI systems remain limited by narrow task definitions, poor generalisation across clinical settings, limited interpretability, and insufficient integration into real-world healthcare workflows. This project aims to develop next-generation medical AI models that can understand, generate, reason over, and act upon multimodal healthcare data in a safe and clinically meaningful way. Depending on the applicant’s background and interests, the project can be tailored towards methodological AI research, applied medical image analysis, surgical video understanding, medical vision-language models, clinical large language models, or healthcare-focused embodied intelligence. Suitable projects may involve deep learning, generative AI, foundation models, multimodal learning, large language models, vision-language models, surgical robotics, digital twins, and trustworthy AI. Research Themes 1. Medical Image Analysis and Generative AI This theme focuses on advanced medical image analysis using 3D deep learning, multimodal representation learning, and generative AI. Potential projects may involve cardiac MRI, neuroimaging, retinal imaging, multi-organ imaging, image reconstruction, segmentation, missing data imputation, disease prediction, and interpretable biomarker discovery. The aim is to develop robust, scalable, and clinically meaningful AI models that can extract useful information from complex medical imaging datasets and support diagnosis, risk prediction, and treatment planning. 2. Medical Vision-Language Models and Large Language Models This theme explores large language models, medical vision-language models, and multimodal foundation models for healthcare applications. Potential projects may include medical image interpretation, report-guided image analysis, clinical text-image reasoning, automated research assistance, intelligent clinical documentation, and decision-support systems. Particular emphasis is placed on combining language, vision, and structured clinical data to build trustworthy AI systems for personalised and efficient healthcare. 3. Surgical Robot Intelligence and Surgical Video Understanding This theme focuses on intelligent surgical robotics, autonomous surgical systems, surgical video analysis, and vision-motor perception. Potential projects may involve spatio-temporal modelling of surgical procedures, surgical workflow recognition, tool-tissue interaction analysis, robotic perception, and safe decision-making for surgical assistance. The research aims to support precision surgery, surgical automation, and human-robot collaboration in complex clinical environments. 4. Embodied Intelligence and Clinical AI Agents This theme investigates embodied AI and intelligent agent systems for healthcare. Potential projects may explore how AI agents can perceive medical environments, reason over multimodal patient data, interact with clinicians or robotic systems, and support safe decision-making. Relevant topics include clinical decision-support agents, medical robotics, workflow-aware AI systems, safety mechanisms for healthcare AI, and interactive systems that integrate perception, reasoning, and action. Possible Project Directions Possible research directions include, but are not limited to: 3D deep learning and generative AI for cardiac MRI, neuroimaging, retinal imaging, and multi-organ imaging. Foundation models for robust and generalisable medical image segmentation, reconstruction, diagnosis, and prognosis. Vision-language models for medical image understanding, report-guided image analysis, and multimodal clinical reasoning. Large language models for clinical documentation, research automation, decision support, and healthcare knowledge integration. Multimodal learning from imaging, clinical text, electronic health records, genetics, ECG, and physiological signals. Surgical video understanding, surgical workflow recognition, and tool-tissue interaction modelling. AI models for surgical robot perception, planning, and safe automation. Embodied AI and clinical agents for safe, interactive, and workflow-aware healthcare applications. Trustworthy, explainable, uncertainty-aware, and clinically deployable AI systems. Opportunities for PhD and Research Master Students For PhD applicants, the project can be developed into a substantial research programme involving novel algorithm design, model development, experimental validation, clinical collaboration, and publication in leading conferences and journals in medical image analysis, machine learning, computer vision, robotics, and digital health. For research master students, the project can be shaped into a focused and achievable research topic, such as developing and evaluating a specific AI model, benchmarking medical foundation models, analysing a surgical video dataset, building a prototype medical vision-language system, or conducting a focused study on model robustness, interpretability, or clinical utility. Research Environment Students will join an interdisciplinary research environment at the University of Birmingham, working at the interface of artificial intelligence, medical imaging, computer vision, robotics, and digital healthcare. The research is connected to broader activities in digital healthcare and medical imaging, with opportunities to collaborate with clinicians, healthcare providers, and academic partners. The research environment provides opportunities to work on real-world healthcare problems, develop advanced computational methods, and engage with interdisciplinary collaborators. Existing research interests cover a wide range of medical imaging modalities and healthcare data types, including MRI, CT, ultrasound, microscopy, OCT, retinal imaging, clinical text, physiological signals, and electronic health records. Candidate Requirements Applicants should have a strong interest in AI for healthcare and a background in computer science, engineering, biomedical engineering, mathematics, physics, statistics, robotics, or a related discipline. Experience with machine learning, deep learning, computer vision, medical image analysis, natural language processing, robotics, or data science would be highly valuable. Strong programming skills in Python are expected. Experience with PyTorch, TensorFlow, or other deep learning frameworks is desirable. Applicants should be self-motivated, curious, and interested in conducting high-quality research. Good communication skills, problem-solving ability, and interest in publishing scientific work are also important. Expected Outcomes The project is expected to lead to new AI methods, prototype systems, benchmark studies, open-source research tools, and publications in leading venues in medical image analysis, machine learning, computer vision, robotics, and digital health. Students will gain experience in algorithm development, experimental design, scientific writing, interdisciplinary collaboration, and translation of AI methods towards real-world healthcare challenges. How to Apply Interested applicants are encouraged to contact Dr Le Zhang with a CV, academic transcripts, and a brief statement describing their research interests, technical background, and preferred project direction. Informal enquiries from prospective PhD students, research master students, and visiting scholars are welcome. Contact: l.zhang.16@bham.ac.uk Website: https://thisislezhang.github.io/

Research areas