A Combined Model of Convolutional Neural Networks and Graph Attention Networks for Improved Classification of Mild Cognitive Impairment.
Authors
Affiliations (6)
Affiliations (6)
- Department of Medical and Digital Engineering, Hanyang University, 222 Wangsimni-ro, Seongdong-gu, Seoul 04763, Republic of Korea. Electronic address: [email protected].
- Department of Medical and Digital Engineering, Hanyang University, 222 Wangsimni-ro, Seongdong-gu, Seoul 04763, Republic of Korea. Electronic address: [email protected].
- Department of Medical and Digital Engineering, Hanyang University, 222 Wangsimni-ro, Seongdong-gu, Seoul 04763, Republic of Korea. Electronic address: [email protected].
- Department of Medical and Digital Engineering, Hanyang University, 222 Wangsimni-ro, Seongdong-gu, Seoul 04763, Republic of Korea. Electronic address: [email protected].
- Department of Neurology, College of Medicine, Hanyang University, 222 Wangsimni-ro, Seongdong-gu, Seoul 04763, Republic of Korea. Electronic address: [email protected].
- Clinical Research Institute, Konkuk University Medical Center, 120-1 Neungdong-ro, Gwangjin-gu, Seoul 05030, Republic of Korea; Institute of Biomedical Science and Technology, Konkuk University, 120-1 Neungdong-ro, Gwangjin-gu, Seoul 05030, Republic of Korea. Electronic address: [email protected].
Abstract
Mild cognitive impairment (MCI), a precursor of Alzheimer's disease (AD), underscores the importance of early diagnosis and treatment. With an aging global population, AD prevalence is rising, necessitating more precise diagnostic methods. Deep learning technology shows promise for MCI and AD classification, but existing convolutional neural network (CNN) and graph attention network (GAT) models have limitations in capturing brain structural features and detecting microlesions. To address these issues, we propose a novel approach combining a CNN and modified GAT model to improve MCI classification. Magnetic resonance imaging volume data were analyzed using a CNN, whereas cortical thickness data were modeled using a GAT, leveraging their complementary strengths. Preprocessing involved extracting brain's structural features via the CIVET pipeline, and t-SNE was used to visualize the data's high-dimensional distribution. Final classification was performed using a multilayer perceptron, integrating feature vectors from both models. Performance evaluation metrics included the area under the curve (AUC), F1-score, sensitivity, and specificity. The combined CNN-GAT model outperformed existing single-model approaches, particularly in MCI classification, effectively distinguishing subtle variations between normal aging and MCI. The combined CNN-GAT model improved MCI classification performance by addressing the limitations of existing approaches. By capturing brain structural features and inter-regional relationships, it offers significant potential for advancing early diagnosis and treatment strategies for neurodegenerative diseases. Future efforts will focus on enhancing performance through additional data optimization.