Deep Convolutional Generative Adversarial Networks (DCGANs) and transfer learning for Alzheimer’s disease classification

Alzheimer’s disease (AD) presents a pressing global health challenge, with projections indicating its impact on 139 million people by 2050. However, access to MRI scans for AD diagnosis remains limited and ethically complex, compounded by the imbalance often found in open-access datasets. This study addresses these challenges by employing Deep Convolutional Generative Adversarial Networks (DCGANs) to generate synthetic brain magnetic resonance images reflecting various stages of AD. The primary objective of this work is to utilize DCGANs alongside a classifier (MobileNetV2) to enhance AD stage recognition via MRI. Our results demonstrate a noteworthy 11% increase in accuracy, achieving 88% accuracy in classifying AD stages with the assistance of DCGANs, compared to using the original dataset alone (77% accuracy). Additionally, image quality indexes such as SSIM and MSE were computed to assess the synthetic images. Despite limitations, such as incorrect pattern matching between test and training datasets, this research underscores the considerable potential of DCGANs in advancing computer-aided AD diagnosis. Future research avenues may explore refining model training methodologies and expanding the scope to encompass additional AD-related metrics for comprehensive diagnostic support.

• Autore principale: Aguilar Martínez, Jaylenne Salomé (author)
• Natura: bachelorThesis
• Lingua: eng
• Pubblicazione: 2024
• Soggetti: Aprendizaje por transferencia; Clasificación de Alzheimer; Procesamiento de imágenes médicas; Transfer learning; Alzheimer’s classification; Medical image processing
• Accesso online: http://repositorio.yachaytech.edu.ec/handle/123456789/773