RDGY41850

Learning Outcomes:

1. Design and implement advanced AI models for a chosen medical imaging task, selecting appropriate architectures, preprocessing techniques, and evaluation metrics based on the data and clinical context.
2. Critically evaluate the performance and limitations of AI models using domain specific criteria, including model interpretability, generalisation, data imbalance, and ethical implications in clinical deployment.
3. Implement reproducible workflows by incorporating best practices in model development, including experiment tracking, version control, and containerization, to ensure reproducible and reliable research outcomes.
4. Communicate technical findings and clinical relevance effectively, producing a well-structured written report aimed at interdisciplinary audiences.

Indicative Module Content:

Spring Trimester (Lectures 1–6)

Introduction to Medical Imaging AI and Challenge Workflows

Data Handling and Preprocessing for Medical Imaging

Task 1 – Radiology Report Generation

Task 2 – Multi-Abnormality Classification

Task 3 – Self-Supervised Multi-Abnormality Localization

Task 4 – Text-Conditional Medical Image Generation

Summer Trimester (Lectures 7–10)

Model Training, Optimization, and Evaluation

Reproducible Pipelines and Experiment Tracking

Scientific Advisory Panel & Clinical/Patient Insights

Writing and Communicating Research

Lab Sessions (5 × 2 hrs)

Spring Labs (1–2)

Lab 1: Environment setup, dataset exploration, Git workflow, task selection for all four tasks

Lab 2: Preprocessing pipelines and data augmentation for chosen task

Summer Labs (3–5)

Lab 3: Advanced model implementation and training for chosen task

Lab 4: Evaluation, hyperparameter tuning, experiment tracking

Lab 5: Submission preparation, reporting, integration of advisory panel feedback

Student Effort Hours:

Approaches to Teaching and Learning:

This module employs a blend of active, task-based, and reflective learning approaches designed to engage students with the full AI development cycle for medical imaging tasks. Key approaches include:

Lectures: Foundational and task-specific content to introduce AI methods, challenge workflows, preprocessing strategies, and model architectures.

Lab/Studio Work: Hands-on sessions focused on data exploration, preprocessing, model implementation, evaluation, and reproducible pipeline development. Labs emphasize continuous skill-building and reflection on methods used.

Task-Based Learning: Students undertake challenge-style projects, selecting one of four tasks and developing a full AI model, fostering practical problem-solving.

Peer and Group Work: Collaboration is encouraged in labs and advisory panel exercises, promoting discussion, feedback, and sharing of approaches to improve model design and implementation.

Critical Writing & Reflective Learning: Students produce MICCAI-style reports in both Spring and Summer trimesters. These include literature review, methodology, and reflection on model development decisions, encouraging critical thinking about AI methods and clinical relevance.

Enquiry & Problem-Based Learning: Students engage with real-world medical imaging datasets, encountering challenges such as data imbalance, multi-modal integration, and model interpretability. This encourages hypothesis-driven exploration and iterative problem-solving.

Advisory Panel Exercises: Insights from clinicians, patients, and technical experts guides students to consider ethical, clinical, and innovation perspectives, fostering reflective learning.

Use of AI/Generative AI in the Module

Students are encouraged to leverage Generative AI tools (e.g., large language models for coding assistance, documentation drafting, or literature synthesis) to support learning and project work. To ensure critical analysis and academic integrity, students must:

Document the prompts used for any AI assistance.

Explain how outputs were refined, verified, or critically assessed.

Indicate instances where AI suggestions were not trusted and alternative approaches were adopted.

Demonstrate understanding and decision-making rather than uncritical reliance on AI.

This approach ensures students benefit from AI tools while maintaining rigorous, critical engagement with their project work in alignment with the University Academic Integrity Policy.

Requirements, Exclusions and Recommendations

Learning Requirements:

Strong proficiency in Python and prior experience with deep learning frameworks (e.g., PyTorch, TensorFlow).

Module Requisites and Incompatibles

Not applicable to this module.
 

Assessment Strategy

 

Please see Student Jargon Buster for more information about remediation types and timing. 

Feedback Strategy/Strategies

• Feedback individually to students, on an activity or draft prior to summative assessment
• Feedback individually to students, post-assessment
• Group/class feedback, post-assessment
• Online automated feedback
• Peer review activities
• Self-assessment activities

How will my Feedback be Delivered?

Feedback will be provided based on an analysis of general weaknesses and strong points

Recommend reading the articles below prior to starting working on your challenge:
Maier-Hein, L., et al, (2020). BIAS: Transparent reporting of biomedical image analysis challenges. Medical Image Analysis, 66, 101796. https://doi.org/10.1016/j.media.2020.101796

Maier-Hein, L., Eisenmann, M., Reinke, A. et al. Why rankings of biomedical image analysis competitions should be interpreted with care. Nat Commun 9, 5217 (2018). https://doi.org/10.1038/s41467-018-07619-7

Understanding metrics in image analysis:
Reinke, A. et al., (2023). Understanding metric-related pitfalls in image analysis validation (arXiv:2302.01790). arXiv. http://arxiv.org/abs/2302.01790

Maier-Hein, L., et al., (2022). Metrics reloaded: Pitfalls and recommendations for image analysis validation (arXiv:2206.01653). arXiv. https://doi.org/10.48550/arXiv.2206.01653

Reinke, A., et al., (2022). Common Limitations of Image Processing Metrics: A Picture Story (arXiv:2104.05642). arXiv. https://doi.org/10.48550/arXiv.2104.05642