Alzheimer’s disease (AD) classification utilizing 3D Convolutional Neural Networks (CNNs) on brain MRI poses significant challenges, including high computational overhead, the scarcity of large-scale training data, and the inherent difficulty in capturing subtle pathological features. Existing 3D-to-2D compression methods simplify the data but suffer from critical information loss. We propose Robust Rank Pooling (RRP), a novel compression technique that employs a sliding window strategy to transform a 3D MRI volume into multiple 2D dynamic images.
This approach effectively mitigates information loss by preserving robust structural features across different brain sections. We utilize these generated images to train an efficient, pre-trained 2D CNN architecture, which incorporates a model ensemble method to further enhance performance. The proposed method is validated on two T2-FLAIR MRI benchmarks, ADNI and BICWALZS.
Experimental results demonstrate that our approach significantly outperforms existing methods. Compared to the previous method, our model achieves improvements of approximately 2.3% in accuracy and 4.4% in Area Under the Curve (AUC). Relative to a standard 3D network, our method improves accuracy by 13.2% while demonstrating a 4.1\(\times \) and 2.1\(\times \) speedups in model inference and total processing time, respectively.
Furthermore, qualitative analysis confirms that our model focuses on clinically relevant biomarkers, consequently offering a computationally efficient and effective solution for Alzheimer’s disease classification. This study was facilitated by the biospecimens and data provided by the publicly available ADNI database and the Biobank of Ajou University Hospital, a member of the Korea Biobank Network. This research was supported by grants from the Korea Health Industry Development Institute(RS-2021-KH113821), ARPA-H Project (RS-2025-25455095), Ministry of Science and ICT(RS-2024-00356486, RS-2026-25519531) and National Institute of Health research project (2026-ER0605-00).
Department of Artificial Intelligence, Ajou University, Suwon, 16499, Republic of Korea Department of Psychiatry, School of Medicine, Ajou University, Suwon, 16499, Republic of Korea Hyun Woong Roh, Chang Hyung Hong & Sang Joon Son Department of Software and Computer Engineering, Ajou University, Suwon, 16499, Republic of Korea The authors declare no competing interests. Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material.
You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.
Extract — continue reading at the source.
