Related Experiment Video
Updated: Jun 2, 2025
A Cognitive Fusion-guided Prostate Biopsy Using Multiparametric Magnetic Resonance Imaging and Transrectal Ultrasound
Published on: March 21, 2025
Deep Learning to Simulate Contrast-Enhanced MRI for Evaluating Suspected Prostate Cancer
Hongyan Huang1, Junyang Mo1, Zhiguang Ding1
1From the Department of Radiology, Shenzhen Nanshan People's Hospital, Shenzhen University, Taoyuan Rd No. 89, Nanshan District, Shenzhen 518000, Guangdong, China (H.H., Z.D., Y.Q.); Medical AI Laboratory and Guangdong Key Laboratory of Biomedical Measurements and Ultrasound Imaging, School of Biomedical Engineering, Shenzhen University Medical School, Shenzhen University, Shenzhen, China (J.M., R.L., B.H.); Department of Medical Imaging, People's Hospital of Longhua, Shenzhen, Guangdong, China (X.P., Y.Z.); and Department of Radiology, Shenzhen People's Hospital, Shenzhen, Guangdong, China (D.Z., G.H.).
Deep learning can create simulated contrast-enhanced prostate MRI from noncontrast scans, offering a viable alternative to reduce contrast agent risks. This AI-generated imaging shows high similarity to real scans and aids in accurately assessing prostate cancer risk using PI-RADS scores.
Area of Science:
- Artificial Intelligence in Medical Imaging
- Radiology and Diagnostic Imaging
- Oncology and Cancer Diagnostics
Background:
- Multiparametric MRI (mpMRI) is standard for suspected prostate cancer, but contrast agents pose risks.
- Concerns exist regarding contrast agent accumulation and potential toxicity in patients.
- Need for alternative imaging methods to mitigate contrast-related risks.
Purpose of the Study:
- Evaluate feasibility of generating simulated contrast-enhanced MRI from noncontrast sequences using deep learning.
- Assess the utility of simulated contrast-enhanced MRI for prostate cancer assessment via PI-RADS v2.1.
- Explore potential to reduce contrast agent use in prostate MRI.
Main Methods:
- Retrospective study of 567 male patients with suspected prostate cancer undergoing mpMRI.
- Deep learning (pix2pix algorithm) trained to synthesize contrast-enhanced MRI from T1w, T2w, DWI, and ADC maps.
- Radiologists independently scored images using PI-RADS v2.1; agreement assessed with Cohen κ.
Main Results:
- Simulated and acquired contrast-enhanced images showed high similarity (MS-SSIM: 0.69–0.82).
- Excellent inter-reader agreement for PI-RADS scores (Cohen κ = 0.96) between simulated and acquired CE-MRI.
- Addition of simulated CE-MRI upgraded 10.5% of biparametric MRI cases to PI-RADS 4.
Conclusions:
- Feasible to generate simulated contrast-enhanced prostate MRI using deep learning.
- Simulated CE-MRI demonstrates high similarity and excellent agreement with acquired CE-MRI for PI-RADS scoring.
- Deep learning-based simulated CE-MRI is a promising tool for assessing clinically significant prostate cancer.
