Early Identification of Mild Cognitive Impairment with Graph Convolutional Networks

Early identification of Mild Cognitive Impairment (MCI) is crucial for treatment to be efficient and effective. The convergence of detailed medical imaging and patient background selected from the Alzheimer's Disease Neuroimaging Initiative has provided abundant data, serving as the bedrock for MCI early diagnosis with machine learning models. In search of enhanced early detection, this paper aims to outline an effective and more efficient methodology, leveraging insights gleaned from high-definition brain scans and data extraction. The process begins with Freesurfer, a software that processes brain imaging data, obtaining intricate information from these high-resolution scans into a myriad of metrics and variables. This initial step sets the stage for the application of Confidence-based Graph Convolutional Networks (GCNs), which operate on the multimodal graph derived from the amalgamation of brain imaging data. Compared to pre-existing models, this model proposes several key enhancements. These GCNs, utilizing confidence-based mechanisms recognize subtle patterns and deviations indicative of cognitive impairment. This model selects for the optimal activation function, or a convex activation that combines multiple functions such as Sigmoid, tanh, Relu, and others. This nuanced approach iteratively refines the model, assimilates insights from previous diagnostic instances, and progressively hones its ability to discern early signs of MCI.