ADoptDeepN: A novel optimized deep learning model for early diagnosis of Alzheimer’s disease

Résumé

Abstract Alzheimer's disease (AD) is a neurological condition that impairs memory, movement, and brain. Early detection using non-invasive techniques has become critical to reduce complications and improve patient outcomes due to the severity of the disease and its continuous and rapid progression. Magnetic resonance imaging (MRI) is the most widely used method for AD diagnosis. However, several studies using MRI images were conducted on imbalance/ limited datasets, thus provided limited accuracy. To address this issue, the study presents a novel deep learning system, called ADoptDeepN, which is a CNN-based model that helps classify different stages of AD. Two MRI datasets, namely OASIS and Kaggle's were used. To resolve the imbalanced dataset, the synthetic minority over-sampling technique (SMOTE) was carried out on the training dataset. Additionally, the performance was improved by using particle swarm optimization (PSO) to fine-tune the key hyperparameters, including the learning rate, batch size, the number of units in the dense layer, the dropout rate, number of filters, kernel size and pool size. Finally, hybrid loss function was proposed by combining cross-entropy and focal loss to improve the training stability and reliability. The ADoptDeepN that has been proposed showed great performance in classifying the stages of AD, both four- and five-stage classification tasks, and went beyond the performance of the existing methods thus validating its prospect for Alzheimer's disease early detection.

Citer ce document

Accès au document

Texte intégral en lecture en ligne, réservé aux abonnés SPHAERO et aux membres de l'institution. Se connecter

Auteur(s)

Statistiques

Consultations : 2

Téléchargements : 0