Explainable Machine Learning Approach for Ventriculomegaly Classification in Brain MRI Images

Ventriculomegaly (VM) is one of the most common brain abnormalities that may cause risks such as developmental delays in children. Automatic classification of VM cases helps physicians in delivering accurate diagnoses and offering evidence-based recommendations. This study introduces an explainable machine learning (ML) framework for classifying MRI images into VM and non-VM categories. We augment a publicly available dataset of normal transventricular MRI images to generate VM cases and extract the key features. These extracted features are trained using different ML models among which Random Forest achieved the highest accuracy. To ensure trust and transparency in the diagnostic tool, we employ explainable AI models, LIME and SHAP, to interpret the ML predictions and highlight the importance of each feature. The results show that the ventricular area is the most significant factor for accurate classification. This interpretability is essential for earning the trust of medical professionals and patients for a reliable and transparent diagnostic aid.