Embedding Explainable Artificial Intelligence in Clinical Decision Support Systems: The Brain Age Prediction Case Study

The efficiency of the predictions produced by clinical decision support systems is essential to ensure their applications in the diagnostic domains, but it is equally important to ensure other characteristics such as transparency and explain ability of the decisions. In this work, we formalized the main findings of a deep learning framework embedding explainable techniques (XAI) to predict the brain age of a healthy cohort of subjects by using several morphological features extracted from their MRI scans. We evaluated each step of the framework by analyzing some critical aspects concerning both stability and accuracy of the algorithms. We also showed how XAI algorithms can be used to explain the effect of the marginal variables related to the imaging quality of MRI scans