Master Thesis AI Based Prediction for Anatomical Motion Modelling on Unity

Anatomical movement during radiotherapy can affect how accurately radiation reaches its target. Elekta Unity combines MRI with a linear accelerator to monitor this movement, but delays in imaging and processing require predictions of future target positions. This thesis will explore whether AI can improve motion prediction over longer time horizons.

The assignment

• Compile a dataset of target-centroid motion across anatomies affected by breathing, using public datasets and historical MR-Linac treatment data.
• Evaluate the existing prediction algorithm on the collected data to establish a baseline.
• Select, adapt and train an AI model to predict centroid motion at clinically relevant time horizons, with recurrent neural networks as a primary focus.
• Compare the AI approach with the current algorithm using suitable quantitative measures.
• Assess the benefits of AI-based prediction for the Automatic Position Management framework.
• Deliver a curated dataset, prototype software, reproducible processing pipelines, documentation, a thesis report and a final presentation.

Who we are looking for

• Master’s student in medical physics, computer science, applied mathematics, engineering physics or a related scientific field.
• Programming experience and the ability to investigate an open-ended research problem independently.
• Strong scientific or engineering foundations and confidence in quantitative analysis.
• Experience with medical image processing, preferably in Python, is advantageous.
• Knowledge of AI or deep learning, radiotherapy or MR-Linac systems is advantageous.

What we offer

• Thesis supervision by Ned Goodwin.
• An opportunity to investigate motion prediction for Elekta’s Unity radiotherapy platform.

Practical details

Location: Elekta’s Stockholm office. Flexible start date.