Sparse Learning and Hybrid Probabilistic Oversampling for Alzheimer's Disease Diagnosis

Alzheimers Disease (AD) is the most common neurodegenerative disorder associated with aging. Early diagnosis of AD is key to the development, assessment, and monitoring of new treatments for AD. Machine learning approaches are increasingly being applied on the diagnosis of AD from structural MRI. However, the high feature-dimension and imbalanced data learning problem is two major challenges in the study of computer aided AD diagnosis. To circumvent this problem, we propose a novel formulation with hinge loss and sparse group lasso to select the discriminative features since features exhibit certain intrinsic group structures, then we propose a hybrid probabilistic oversampling to alleviate the class imbalanced distribution. Extensive experiments were conducted to compare this method against the baseline and the state-of-the-art methods, and the results illustrated that this proposed method is more effective for diagnosis of AD compared to commonly used techniques.