“Interpreting machine learning models in neuroimaging: Towards a unified framework”
Machine Learning (ML) has rapidly increased in popularity in both basic and translational research. The use of ML in neuroimaging experiments has provided new answers to many enduring research questions. However, these models are complex and often hard to interpret, making it difficult to evaluate their neuroscientific validity and contribution to understanding the brain. Therefore, there is a pressing need for methods to help interpret and explain the model decisions and provide neuroscientific validation for neuroimaging ML models. In this protocol, we introduce a unified framework that consists of model-, feature- and biology-level assessments to provide complementary results that support the understanding of how and why a model works. We first propose a unified framework for interpreting ML models in neuroimaging based on model-level, feature-level, and neurobiology-level assessments. Then, we provide a workflow that illustrates how this framework can be employed to predictive models, along with practical examples of analyses for each level of assessment with a sample fMRI dataset (available for download at https://github.com/cocoanlab/interpret_ml_neuroimaging).
Access Article Here