IMPROVING 3D LESION SEGMENTATION ROBUSTNESS AGAINST IMAGE COMPRESSION IN MULTIPLE SCLEROSIS
(2024) 21st IEEE International Symposium on Biomedical Imaging (ISBI 2024) ā Location: Athens, Greece (27.May.2024)
Files
Details
- Authors
- Abstract
- Accurate segmentation of white matter lesions (WMLs) in MRI is crucial for the diagnosis and prognosis of multiple sclerosis (MS). Given that this manual process is laborious and time consuming, several deep learning (DL)-based methods have been proposed to automate the process. Data transfer and storage are key for DL methods due to their requirement for large amounts of data, making image compression (IC) a powerful tool to enhance data management efficiency. Nonetheless, lossy IC can impact image quality, potentially affecting the accuracy of DL tasks. While related works have studied the effect of IC on DL-based tasks in medical imaging, no work has studied the impact of IC on DL-based segmentation of WMLs yet. In this work, we propose an iterative fine-tuning approach to improve the robustness of 3D U-Net segmentation against IC of WMLs in MS. We show that our proposed approach can handle compression ratios (CRs) up to 64:1 while only slightly decreasing segmentation performance.
- Affiliations
Citations
El Khoury, K., Wynen, M., Maggi, P., Meritxell Bach Cuadra, Macq, B., & et al. (2024). IMPROVING 3D LESION SEGMENTATION ROBUSTNESS AGAINST IMAGE COMPRESSION IN MULTIPLE SCLEROSIS. 21st IEEE International Symposium on Biomedical Imaging (ISBI 2024), Athens, Greece. https://hdl.handle.net/2078.5/247687