DIFFUSION MODELS FOR DATA AUGMENTATION OF MEDICAL IMAGES

Abstract

Recently, there has been a substantial surge in interest surrounding diffusion probabilistic models, which are considered a prominent class of generative models, particularly in the realm of deep learning. These models have garnered significant attention due to their po tential applications in a range of deep-learning problems. The primary objective of this thesis is to assess the effectiveness of Diffusion models as a data augmentation technique in the context of medical image analysis. Furthermore, it aims to conduct a comparative analysis of the performance exhibited by deep-based classi fiers trained on two distinct datasets. One dataset was augmented using diffusion models, while the other dataset underwent traditional data augmentation techniques. Utilizing the IDRID dataset for the purpose of diabetic retinopathy diagnosis, the acquired outcomes substantiate the efficacy of Diffusion models as a data augmentation methodol ogy for medical images, in contrast to the traditional data augmentation technique which is predominantly employed. The integration of diffusion model augmented data yielded su perior performance for both classifiers, namely the Fine-tuned Resnet50 and the proposed CNN, surpassing the performance of classifiers trained using traditional data augmentation