Houmem Slimi ^1,2* Ala Balti ² Sabeur Abid ² Mounir Sayadi ²

• ¹ Higher National Engineering School of Tunis, Tunis, Tunisia
• ² Tunis University, Tunis, Tunisia

Alzheimer's disease (AD) is a progressive neurodegenerative disorder characterized by cognitive decline, memory loss, and impaired daily functioning. Despite extensive research, AD remains incurable, emphasizing the critical need for early diagnosis and intervention. Timely detection is crucial for effective management. Pretrained convolutional neural networks (CNNs), trained on large-scale datasets such as ImageNet, offer a head start for AD classification. In this paper, we propose a novel hybrid deep learning approach that merges the strengths of two specific pretrained architectures. The new proposed model leverages the feature extraction capabilities of both networks to enhance the representation of AD -related patterns in medical images. This new model is validated using a large dataset of AD MRI images, demonstrating significant performance gains over individual models. An accuracy classification rate of 99,85% is achieved. Furthermore, the performance of this new architecture is compared with several common models showing its superiority on classification rate, robustness against noise. This new hybrid model holds promise for early AD detection and monitoring, potentially aiding clinicians in making timely diagnoses and treatment decisions.