Title: LADD-Net: a lightweight deep learning approach for efficient classification of Alzheimer's disease progress: analysing the importance of addressing class imbalance issue

Authors: Sani John; M.G. Jibukumar

Addresses: School of Engineering, Cochin University of Science and Technology, Kochi-22, Kerala, India ' School of Engineering, Cochin University of Science and Technology, Kochi-22, Kerala, India

Abstract: The prognosis and diagnosis of medical anomalies are very relevant in deep learning. However, the complexity of deploying such deep networks is a major concern regarding total parameters. This work aims to enhance the classification model for Alzheimer's Disease progress by making it lighter by reducing the parameters. The authors suggest a lightweight network named LADD-Net that provides good classification parameters with lesser complexity, data volume, and training time. LADD-Net performs well in this classification task with a cross-validation accuracy score of 96.26% and Area Under the Curve, F1-score, precision, and recall of 98.35%, 94.28%, 94.5%, 94.5% respectively. The results demonstrate that LADD-Net exhibits decent performance matrices compared with other state-of-the-art models. Moreover, the parameter has reduced to 0.18M, representing only 35% in a recent lightweight network used for Alzheimer's classification. The LADD-Net surpassed iterations required to attain good performance matrices resulting in a relevant reduction in model training time.

Keywords: Alzheimer's disease; CNNs; convolutional neural networks; deep neural network models; shallow network models; SMOTE; synthetic minority oversampling technique; ADASYN; adaptive synthetic algorithm.

DOI: 10.1504/IJSISE.2025.150013

International Journal of Signal and Imaging Systems Engineering, 2025 Vol.14 No.1, pp.39 - 57

Received: 31 Aug 2024
Accepted: 27 Mar 2025
Published online: 21 Nov 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article