MSc Medical Physics in Cancer Radiation Therapy / Course details

The unit will provide the students with the following knowledge and practical skills:

• Introduction to the definition of AI, the classification of AI subfields and the guidelines for trustworthy AI proposed by the European Commission's High-Level Expert Group on AI.
• Overview of the role of artificial intelligence and data science within medical physics, with a focus on radiotherapy applications.
• Highlight the importance of AI and data science and where it fits in the patient pathway with a focus to its potential to benefit patients or clinical workflows.
• Considerations of the additional needs for using software in the clinical environment - regulatory framework, commissioning and quality assurance will be introduced highlighting the need for 'the human in the loop'.
• Introduction of key concepts in data science (primarily outcome modelling techniques) and in AI (classification and segmentation approaches). Best practice approaches for model creation, optimisation, assessment, and validation will be discussed.
• Knowledge and experience of key concepts in model performance and evaluation will be presented. A focus on the impact of selection bias will be assessed through examples and within the assessment.

Students will synthesis their knowledge and apply within the practical assessment:

• Working in Python, students will define and develop an optimised AI model for a classification task. They will experiment with introducing selection bias within the training data and assessing model performance.
• The future clinical use of AI in medical physics will be appraised and evaluated.

This unit aims to provide students with knowledge to appraise and evaluate the role of artificial intelligence and data science techniques within cancer radiation therapy and wider medical physics roles. At the end of this unit, students will be able to define artificial intelligence and its subfields according to the European Commission's definition. Students will distinguish the subfields covered by the AI umbrella including but not limited to machine learning. Students will be able to identify and categorise specific uses where AI has demonstrated effectiveness in medical physics with emphasis on radiotherapy (e.g., image analysis, treatment planning). Students will develop the ability to critically assess the benefits, challenges, and ethical considerations associated with implementing AI solutions in the field of medical physics.

The practical aspects of the unit will focus primarily on state-of-the-art outcome modelling, classification approaches, and segmentation enabled by machine learning. At the end of the unit, students will understand the complexities of implementing different AI approaches. Students will be able to critically compare and contrast different AI approaches for medical physics tasks. Students will be able to apply best practice approaches and their knowledge for optimisation and model assessment and validation. Students will be able to synthesis their knowledge to create an AI model for a given task, demonstrating understanding of the impact of selection bias within their network.

The following learning and teaching processes will be utilised: Classroom based teaching, podcasts, interactive computer simulation practical sessions and paired programming, formative assessments, interactive group based discussion and tutorial sessions, on-line resources, independent study, facility tours and demonstrations.

Students should/will be able to:

• Understand what is AI, the subfields are covered by this 'umbrella' term and their classification/hierarchy.
• Evaluate and recognise the state-of-the-art data science and AI within medical physics.
• Apprise the selection, development, training, validation and deployment of data science and AI modelling.
• Identify potential role(s) of AI within the patient pathway.

Students should/will be able to:

• Critically compare and distinguish appropriate data science or AI approaches and their application in medical physics.
• Evaluate approaches for optimisation of AI models and potential impacts of selection bias.
• Discuss and critic the future role of data science and AI within medical physics.

Students should/will be able to:

• Demonstrate application of knowledge to develop a model for prediction.
• Evaluation of the impact in model performance due to selection bias in training data.
• Critically appraise scientific publications of an AI model in medical physics.

Students should/will be able to:

• Synthesis and evaluation of data science and AI aspects in scientific literature.
• Technical: Practical experience of building an AI prediction model in the most-used language for this developments (Python), with detailed analysis of the impact of selection bias.

100% (70% notebook, 30% poster discussion)

Undertake an AI classification project using Juypter notebooks. Students will assess the impact of introducing selection bias in the training data used for model creation.

Students will work in pairs to support their development specially in programming. Students are required to justify and explain their work independently.

Students will submit their notebooks. Students will also create a poster describing the implementation, results, and impact of training bias, which will be assessed by the faculty and peers.

Paired feedback based on defined marking scheme for the notebook.

Peer (20% weighting) and academic (80% weighting) assessment of the poster and discussion.

Feedback will be provided within the required timeframes.