An Image Enhancement and Data augmentation of Alzheimer's MRI Data using modified SRGAN

• Published in: 2023 14th International Conference on Computing Communication and Networking Technologies (ICCCNT) pp. 1 - 5
• Main Authors: Rashmi, Uppin, B M, Beena, A, Preethi, Ambesange, Sateesh
• Format: Conference Proceeding
• Language: English
• Published: IEEE 06.07.2023
• Subjects: Alzheimer's disease; Computational modeling; Data augmentation; Data models; Image Enhancement; Magnetic resonance imaging; MRI; Predictive models; SRGAN; Superresolution; Transfer learning; VGG-19
• ISSN: 2473-7674
• DOI: 10.1109/ICCCNT56998.2023.10307164

Machine learning (ML) models, Deep Learning (DL) Models rely on labeled data for classification and prediction. Automated or computer-aided medical diagnosis requires a large dataset to maintain clinical accuracy and precision. High-resolution images of different clinical and symptomatic stages are essential for implementing ML /DL algorithms for medical diagnosis. High resolution MRI image aids in better performance of DL Models. Here we propose a model based on Super Resolution Generative Adversarial Networks (SRGAN) with Transfer learning to enhance the MR images for Alzheimer's disease (AD). These images can also be used as data augmentation and increase dataset size for better model performance. AD, an irreversible cognitive impairment disorder, is one of the significant causes of death in the elderly. No cure is yet developed, and diagnosis at an early stage will enable the elderly to make informed decisions. The model enhance low resolution image of lower size to high resolution higher size. The model trained for 5000 epochs resulted in a final G loss of 0.001 and a D loss of 1.006, which is a significantly good result. The high-resolution images generated by the proposed model are validated by the pixel-wise loss function, with a comparable average MSE of 0.009.