A deep feature fusion network with global context and cross-dimensional dependencies for classification of mild cognitive impairment from brain MRI

Abstract

Background and objectives:The accurate identification of people with Mild Cognitive Impairment (MCI) who may develop Alzheimer's disease (AD) holds significant importance in facilitating timely intervention and treatment. However, current classification methods have yet to be effective due to the subtle nature of the features involved. This research aims to improve the performance of the MCI classification by enhancing the feature representation in brain MRI. MethodsWe propose an integrated model that combines Group Shuffle Depth-wise Convolution (GSDW), Global Context Network (GCN), Hybrid Multi-Focus Attention Block (HMAB), and EfficientNet-B0 architecture for MCI classification. This model extracts fine-grained features, contextual information, and long-range dependencies and aggregates low-level details with high-level context information to learn discriminative features. ResultsOur comprehensive evaluation demonstrates significant improvements in the classification performance for both progressive MCI (pMCI) and stable MCI (sMCI), surpassing existing approaches. The proposed model achieved a notable accuracy of 77.2%. ConclusionsThis study introduces a novel feature fusion technique that combines global contextual representations and cross-dimensional dependencies to enhance classification results. These findings highlight the potential of the proposed framework for early identification and intervention in individuals at risk of cognitive decline.