Aim: The aim of this project is to develop new predictive tools using artificial intelligence (AI) techniques to support the design of a personalized treatment in patients with clinical cardiac atherosclerotic disease, namely obstructive coronary artery disease (CAD), and severe aortic stenosis (AS).
Scientific rationale: Despite extensive clinical research efforts, mortality in CAD and severe AS remains high. The decision between medical treatment, interventional, or surgical treatment is often difficult. Moreover, numerous multivascular CAD patients have an equal indication for coronary artery bypass grafting / percutaneous coronary intervention and treatment decisions are at the discretion of the physician. In this context, our project will focus on developing new prognostic tools using AI techniques to support the design of the best patient-specific treatment. This will help to gain insight into the sub-classification of patients with CAD and severe AS, as well as the relationship of each disease sub-phenotype with the treatment outcome.
Methodology: All parameters will be one-hot encoded. All implementations will be performed in Python. The AI models will be developed using the XGBoost algorithm, a consecrated gradient boosting algorithm with proven high performance across a multitude of tasks, which also allows survival analysis with accelerated failure time. All models will be trained using 10-fold cross-validation: the dataset will be randomly split into 10 folds; one fold will serve as the validation dataset, one fold will serve as the testing dataset and the remaining 9 folds will serve as the training dataset. All models will use grid-search as hyperparameter tuning. The structure of the models will allow multimodality – the inclusion of parameters from different investigation methods. The structure of the models will also allow prediction-making even in the context of a missing investigation or parameter. Moreover, the models will be trained to predict survival (e.g., time to event instead of binary classification), allowing the evaluation of prediction performance even in the context of a few events’ occurrence during follow-up. The project duration will be 36 months.
Impact: Our research will contribute to a better understanding of the clinical factors associated with a favourable outcome after certain interventions, such as coronary artery bypass grafting or percutaneous coronary intervention.
