Comparative Study of Deep and Hybrid Models for Alzheimer’s Disease Stage Classification Using MRI

This work presented a structured comparative study span-ning ten distinct models across four computational paradigms for four-class Alzheimer’s Disease stage classification from brain MRI. Our purpose-built CNN, designed around the input resolution and class structure of the target dataset, achieved 99.1% test accuracy and an AUC of 0.999 — outperforming every evaluated transfer learning backbone. Equally notewor-thy was the PCA-SVM pipeline’s 99.0% accuracy, demonstrat-ing that classical methods retain competitive power when an appropriate dimensionality reduction precedes classification. Together, these findings challenge the assumption that deep pretrained backbones are universally superior in medical imag-ing tasks and highlight the value of matching model capacity to dataset scale and domain. Future directions for this research include expanding to three-dimensional MRI volumes with 3D-CNN architectures, incorporating longitudinal data to track disease progression over time, and integrating complementary clinical biomarkers to build multimodal diagnostic systems. Crucially, any model intended for clinical deployment should be subjected to rigor-ous interpretability analysis — for instance through gradient-based attribution maps such as Grad-CAM [11] — to verify that predictions are grounded in radiologically meaningful image features rather than spurious correlations.

[1]     H. C. Shin et al., ”Deep convolutional neural networks for computer-aided detection: CNN architectures, dataset characteristics and transfer learning,” IEEE Trans. Med. Imag., vol. 35, no. 5, pp. 1285–1298, May 2016, doi: 10.1109/TMI.2016.2528162. [2]     K. Simonyan and A. Zisserman, ”Very deep convolutional networks for large-scale image recognition,” arXiv preprint arXiv:1409.1556, 2014. [3]     K. He, X. Zhang, S. Ren, and J. Sun, ”Deep residual learning for image recognition,” in Proc. IEEE CVPR, 2016, pp. 770–778, doi: 10.1109/CVPR.2016.90. [4]     C. Szegedy et al., ”Going deeper with convolutions,” in Proc. IEEE CVPR, 2015, pp. 1–9, doi: 10.1109/CVPR.2015.7298594. [5]     M. Tan and Q. V. Le, ”EfficientNet: Rethinking model scaling for convolutional neural networks,” in Proc. 36th ICML, 2019, vol. 97, pp. 6105–6114. [6]     L. Nanni, S. Ghidoni, and S. Brahnam, ”Handcrafted vs. non-handcrafted features for computer vision,” Pattern Recognition, vol. 71, pp. 158–178, 2017, doi: 10.1016/j.patcog.2017.05.025. [7]        J. Yosinski, J. Clune, Y. Bengio, and H. Lipson, ”How transferable are features in deep neural networks?,” in Adv. Neural Inf. Process. Syst. 27 (NIPS), 2014, pp. 3320–3328. [8]        R. R. Selvaraju et al., ”Grad-CAM: Visual explanations from deep networks via gradient-based localization,” in Proc. IEEE ICCV, 2017, pp. 618–626, doi: 10.1109/ICCV.2017.74. [9]        K. Kamnitsas et al., ”Efficient multi-scale 3D CNN with fully connected CRF for accurate brain lesion segmentation,” Med. Image Anal., vol. 36, pp. 61–78, 2017, doi: 10.1016/j.media.2016.10.004.