sequential predictor schematic
sequential predictor performances
predictive brain pattern
ML pipeline schematic

Project information

  • Scope:
    MRI image preprocessing, Data cleaning, Algorithm research and development, Model validation, Model deployment, Scientific publication production
  • Technologies:
    Matlab, Python, PHP, MySQL, JavaScript, D3.js, AWS, Pandas, Scikit-learn
  • Publication link: Translational Psychiatry

Sequential ML model increases rTMS treatment response in Schizophrenia from 50% to 94%

The challenge

Repetitive transcranial magnetic stimulation (rTMS) holds promise as a non-invasive treatment for patients with schizophrenia. However, its effectiveness varies significantly among individuals. Some patients experience substantial symptom reduction, while others show minimal improvement or no response at all. The overall treatment response rate is only around 50%. This variability poses a significant clinical challenge.

The solution

To address this challenge, we propose a sequential comprehensive Machine learning prediction system that leverages predictive models and diverse data sources: re-treatment sMRI, clinical, sociodemographic, and polygenic risk score (PRS) data. We meticulously trained and cross-validated our models. Rigorous validation ensures robust performance. Sequential modelling optimises model accuracies while minimises data acquisition costs.

The end result

Our multimodal sequential prediction workflow achieved impressive balanced accuracy (BAC) rates: 94% for active-treated patients (92% for non-responders, 95% for responders) and 50% for sham-treated patients. Notably, clinical + PRS and sMRI-based classifiers yielded BACs of 76% and 80%, respectively. A prospective clinical trial is recently completed to validate our model performances. With this model, clinicians can easily stratify patients before conducting rTMS treatment to maximise response rate.

Patients, healthcare providers, medical device manufacturers and insurances can all maximise their interest.