This paper presents EMIT-Diff, a new method for synthesizing medical image data to enhance medical image segmentation tasks. Leveraging diffusion probabilistic models, the proposed method generates realistic and diverse synthetic medical images that retain the key characteristics of the original images. The synthetic images adhere to medically relevant constraints and maintain the inherent structure of the imaging data. The proposed method has been tested extensively on various datasets, demonstrating significant improvements over the baseline segmentation methods. The study also explores the influence of different data augmentation ratios, hyper-parameter settings, and network architectures on the performance of the method.
Publication date: 20 Oct 2023
Project Page: Not Provided
Paper: https://arxiv.org/pdf/2310.12868
