Explainable AI-based classification of asthma and COPD from 129Xe-MRI with regional interpretability

Motivation: Asthma and/or COPD diagnosis is challenging due to overlapping characteristics; hyperpolarized xenon-129 (129Xe)-MRI provides regional ventilation information. Classifying asthma and/or COPD using 129Xe-MRI could potentially inform diagnosis and patient management. Goal(s): Classify asthma and/or COPD from 129Xe-MRI using explainable AI to gain regional insights. Approach: 129Xe-MRI from 160 asthma and/or COPD patients were classified using a convolutional neural network. Occlusion sensitivity maps highlighted regional features associated with each diagnosis. Results: We demonstrate accurate classification of asthma and/or COPD patients. Occlusion sensitivity maps show 129Xe-MRI ventilation regions associated with disease grouping. Percentage COPD likelihood was moderately negatively correlated with TLCO. Impact: Novel use of xenon-129 (129Xe)-MRI to classify patients with asthma and/or COPD. Explainable AI techniques provide insights into regional patterns of ventilation associated with classifications; large-scale regional patterns that occur in 129Xe-MRI show qualitative differences in asthma and/or COPD phenotypes.