Synthetic Patient Simulator for Aphasia Rehabilitation Exercises
Developing and evaluating digital rehabilitation tools for aphasia (communication disorder related to stroke) requires realistic patient data — but access to real patients is limited. Synthetic patient simulation offers a principled solution: generating realistic patient personas, interaction sequences, linguistic metrics, and fatigue traces that can be used to test, train, and validate rehabilitation systems without requiring clinical access. This project continues the development of GazeTalk — an aphasia rehabilitation platform in pilot development at DTU — by expanding an existing synthetic conversation simulator into a comprehensive multi-exercise simulation framework. This project extends that architecture to cover the full breadth of evidence-based aphasia rehabilitation techniques, enabling realistic simulation across exercise types and aphasia subtypes.
The goal of this project is to design, implement, and validate a unified synthetic patient simulator covering multiple rehabilitation exercise paradigms. More specifically, the project should:
- Review existing literature on synthetic data generation in digital health, patient simulation, and aphasia rehabilitation techniques
- Extend the existing simulator architecture to cover one or multiple following exercise types: (1) Script Training — repetition-based rehearsal of personally relevant communicative scripts to automaticity, (2) VNeST — verb network strengthening through thematic role mapping, (3) SFA — semantic feature analysis with structured clinician cueing, and (4) PCA — phonological component analysis targeting sound-level retrieval
- Design a validation framework for synthetic data quality — assessing whether simulated sessions are clinically plausible against real session benchmarks
- Model each exercise's clinician-patient interaction as a finite state machine (FSM), use a large language model (LLM) with carefully engineered clinical prompts to generate aphasia-realistic patient responses conditioned on synthetic personas, and validate simulator output against real transcripts using statistical analysis
This project offers an opportunity to apply large language models to a clinically meaningful and technically novel challenge — generating realistic impaired language rather than optimising for fluency or correctness, a problem that sits at the cutting edge of both NLP and computational modelling of human cognition. For students interested in human-centred AI, the project grounds this technical challenge in real human impact — every improvement to the simulator directly accelerates the development of better rehabilitation tools for stroke survivors. The project will be supervised by postdoctoral researcher Ekky Tammarar Alfian and co-supervised by Prof. John Paulin Hansen. This project can be formulated into a Master thesis/Bachelor thesis/student project. The project is expected to start in the Fall semester 2026 (negotiable).
Thesis type: Computational modelling and system implementation
Technical skills: Software architecture; NLP and linguistic modelling; simulation and synthetic data generation; Python or JavaScript.
Research skills: Human-Computer Interaction; Clinical Informatics; Rehabilitation Science; Software Engineering.
Educational background: Computer Science, Artificial Intelligence, Software Engineering
ECTS credits: MSc thesis / BSc thesis / Special course (5 or 10 ECTS)
Contact: < email supprimé pour raison de sécurité >