FiboNeXt: Investigations for Alzheimer's Disease detection using MRI

Abstract

Background: Deep learning models are currently at the forefront of machine learning. Researchers have proposed and used various deep-learning models. In this research, our primary objective is to introduce a next-generation convolutional neural network inspired by the Fibonacci sequence. Materials and Methods: We utilized a public Alzheimer's disorder (AD) magnetic resonance imaging (MRI) dataset for this model. This dataset is divided into four categories and includes both augmented and original versions. To detect the AD type, we proposed a new lightweight Fibonacci network, incorporating the structure of ConvNeXt. We also integrated attention and concatenation layers. As a result, we named the proposed convolutional neural network FiboNeXt. The primary goal of FiboNeXt is to achieve high classification capability with fewer trainable parameters, making it a competitive CNN. Results: The proposed FiboNeXt model was tested on two open-access MRI image datasets comprising both augmented and original versions. The augmented versions were utilized for training, while the original dataset was used for testing. The model achieved 95.40% and 95.93% validation accuracies for the first and second datasets, respectively. Furthermore, it attained test accuracies of 99.66% and 99.63% on the two utilized AD MR image datasets, respectively. Conclusions: The results and findings unequivocally demonstrate that FiboNeXt is a potent deep-learning model. It holds the potential for addressing other computer vision challenges.