Related Experiment Video
Updated: Oct 17, 2025
Neuroimaging-Guided TMS–EEG for Real-Time Cortical Network Mapping
Published on: June 13, 2025
Accelerating Whole-Body Diffusion-weighted MRI with Deep Learning-based Denoising Image Filters.
Konstantinos Zormpas-Petridis1, Nina Tunariu1, Andra Curcean1
1Division of Radiation Therapy and Imaging, The Institute of Cancer Research, 123 Old Brompton Rd, London SW7 3RP, England (K.Z.P., N.T., A.C., C.M., S.C., D.J.C., J.C.H., Y.J., D.M.K., M.D.B.); and Department of Radiology, The Royal Marsden National Health Service Foundation Trust, Surrey, England (N.T., A.C., C.M., S.C., J.C.H., D.M.K.).
Deep learning image filters enhance low-acquisition MRI scans, improving image quality for faster whole-body diffusion-weighted MRI (WBDWI). This method generates clinical-standard images from subsampled data, benefiting oncology and whole-body imaging.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Medicine
- Radiology
Background:
- Whole-body diffusion-weighted MRI (WBDWI) is crucial for staging and monitoring cancer.
- Reducing WBDWI acquisition time is essential for patient comfort and throughput.
- Subsampling images (e.g., number of acquisitions = 1 [NOA1]) significantly reduces scan time but compromises image quality.
Purpose of the Study:
- To develop and validate a deep learning-based denoising image filter (DNIF) to enhance the quality of subsampled WBDWI.
- To assess the feasibility of using DNIF to reduce WBDWI acquisition times while maintaining diagnostic image quality.
Main Methods:
- A deep learning model (DNIF) was trained and validated using retrospective WBDWI data from 17 patients with metastatic prostate cancer.
- Prospective data from 22 patients with advanced cancers (prostate, myeloma, breast) were used for testing, comparing DNIF-processed NOA1 images (NOA1-DNIF) with NOA1 and clinical NOA16 images.
- The model's performance was further evaluated in 28 patients with malignant pleural mesothelioma (MPM) undergoing lung MRI.
Main Results:
- The DNIF model visually improved the quality of NOA1 images across all tested patients.
- DNIF-processed NOA1 images (NOA1-DNIF) and clinical NOA16 images were predominantly graded as "average" or "good" for radiologic quality.
- Mean apparent diffusion coefficient (ADC) values from NOA1-DNIF images showed minimal deviation from higher acquisition images (1.9% for bone disease, 3.7% for MPM).
Conclusions:
- Deep learning-based image postprocessing can generate clinical-standard images from subsampled WBDWI.
- The DNIF model effectively enhances image quality, enabling reduced acquisition times for WBDWI.
- This approach demonstrates potential for improving efficiency in oncology and whole-body MRI applications.
