Ischemic Stroke Risk Assessment
The goal of this project was to develop an AI backend engine for an intelligent decision support system which asses an ischemic stroke risk. The system was developed in collaboration with a health insurer to allow preventive interventions for patients with high risk of a stroke.
We collected historical electronic health records, social-demographic data, and quality of life-related data to train and evaluate machine learning models.
The data was analyzed to find which aspects present in the collected datasets had the highest impact on the ischemic stroke risk. This was an iterative process in collaboration with multiple medical doctors and researchers, leading to the election of more than two hundred prospective features which were then engineered and gave a base for the feature selection process.
During a model selection phase, several models were evaluated, e.g., XGBoost, SVM, Random Forest, Neural Network. The area under the ROC curve was used as a performance metric because the data was unbalanced. The best result of 0.86 was achieved by a Neural Network consisting of three hidden layers 32 neurons each. The model was deployed and served as a REST API within a docker container, which enabled horizontal scalability of the solution.