AI for precision mental health: Data-driven healthcare solutions

Developing a cost-effective clinical decision support system to help clinicians

Project status



Predicting disorders of mental health early and developing personalised interventions has major implications for clinical management and treatment. Yet, progress in early diagnosis and personalised treatment is compromised by heterogeneity in patient populations. This project aims to establish a cross-disciplinary team of researchers and research engineers that will work together to deliver and sustain digital data-driven, cost-effective healthcare solutions for mental health.

This work is related to a previous project, AI for precision mental health.

Explaining the science

This work will build human capital for delivering digital healthcare solutions for early prediction and precision stratification of patients with mental health disorders. The approach is unique in bringing together interdisciplinary know-how from machine learning, neuroscience, clinical practice and industry to tackle the challenge of early detection and prediction of personalised trajectories for mental health disorders.

The competitive advantage of the work rests on using machine learning to develop predictive models and mine large-scale multimodal data to reveal biologically-relevant predictors invisible to the clinician’s eye. In particular, the proposed artificial intelligence systems are:

  • Optimised through the lens of biology to deliver interpretable predictors that link non-invasive data to pathology markers and generalise across populations
  • Tailored to target the challenge of large-scale clinical data that are often sparse, incomplete, non-standardised and unlabelled
  • Scalable to deliver a clinical decision support system that will provide intelligent guidance to clinicians informed by diverse data types

Project aims

Previous research with The Alan Turing Institute on this topic has so far focused on dementia and has developed predictive and prognostic machines (PPMs) that are trained and validated to reliably predict individual rate of cognitive dysfunction at early dementia stages from low-cost, less invasive data (e.g. cognitive testing).

This project will translate PPMs into a fully deployable clinical decision support system that will:

  • Help clinicians assign the right patient at the right time to the right diagnostic or treatment pathway
  • Improve patient well-being and reduce healthcare costs, as patients undergo fewer, less invasive, less expensive diagnostic tests
  • Guide patient selection for clinical trials to enhance their efficacy and pave the way to drug discovery

To deliver this system, the research will validate:

  • PPM interoperability across diverse populations and multimodal data types
  • PPM scalability to operate on healthcare digital data
  • PPM reproducibility and transparency that adheres to medical device software standards

In the longer term, the research team will test the impact of the system on patent well-being and healthcare costs by integrating PPMs in clinical trials, paving the way to novel biomarker and drug discovery for neurodegenerative disease.


Despite the lack of curative treatments for mental health disorders, different disease subtypes benefit from different care management strategies. Treating mood-related disorders or slowing subtype-specific dementia progression has major implications for patient well-being (i.e. patients have better quality of life for longer) and health economies (e.g. patients and their carers remain in the workforce and out of the healthcare system for longer).

Further, predicting progression of mental health disorders based on low-cost measures that can be incorporated in routine clinical testing will result in:

  • Personalised screening involving fewer, less invasive and less time consuming tests
  • Reduced healthcare costs incurred by costly screening (e.g. PET)
  • Improved patient outcome from early interventions and stratification to treatment pathways

Finally, AI-guided patient selection has strong potential to:

  • Enhance clinical trial efficacy and reinvigorate pharma investment in drug discovery for mental health disorders
  • Guide the development of critical tests and assays to improve diagnosis and predictability of patient outcome
  • Inform policy and trust to AI and digital data for personalised healthcare (BIES Digital Data & Technology Strategy)

Recent updates

August 2021

The BBC ran an exclusive news story featuring The Alan Turing Institute’s work on “an AI system capable of diagnosing dementia after a single brain scan.” Turing Fellow Zoe Kourtzi was interviewed by Pallab Ghosh, BBC Science Correspondent, about her work on memory clinics (seed-funded by the Turing), and EDoN (Early Detection of Neurodegenerative diseases, the initiative with Alzheimer’s Research UK). Artificial Intelligence may diagnose dementia in a day.


Professor Zoe Kourtzi

Turing Fellow, Professor of Cognitive Computational Neuroscience at the University of Cambridge and Turing University Lead - Cambridge

Researchers and collaborators