Related Experiment Video
Updated: Jun 24, 2025
06:08
A Cognitive Fusion-guided Prostate Biopsy Using Multiparametric Magnetic Resonance Imaging and Transrectal Ultrasound
Published on: March 21, 2025
155
ssVERDICT: Self-supervised VERDICT-MRI for enhanced prostate tumor characterization
Snigdha Sen1, Saurabh Singh2, Hayley Pye3
1Center for Medical Image Computing, Department of Computer Science, University College London, London, UK.
Magnetic Resonance in Medicine
|June 9, 2024
Summary
Self-supervised machine learning for the VERDICT model (ssVERDICT) accurately fits prostate cancer diffusion MRI data without training labels. This new method outperforms existing techniques in simulations and patient data analysis.
Area of Science:
- Medical Imaging
- Machine Learning
- Oncology
Background:
- Diffusion MRI models like VERDICT (vascular, extracellular and restricted diffusion for cytometry in tumors) are crucial for prostate cancer characterization.
- Accurate model fitting is essential for reliable quantitative analysis and clinical decision-making.
Purpose of the Study:
- To develop and evaluate a self-supervised machine learning approach for fitting the VERDICT model in prostate cancer.
- To assess the performance of this self-supervised method against conventional and supervised deep learning techniques.
Main Methods:
- A novel self-supervised neural network (ssVERDICT) was derived to estimate VERDICT parameter maps without requiring training data.
- Performance was quantitatively assessed using simulated data (Pearson's correlation coefficient, MSE, bias, variance) and in vivo data from 20 prostate cancer patients.
Main Results:
- ssVERDICT demonstrated superior performance over nonlinear least squares and supervised deep learning in simulations, achieving better parameter estimation.
- In vivo analysis revealed enhanced lesion conspicuity and improved discrimination between benign and cancerous prostate tissue using ssVERDICT.
Conclusions:
- Self-supervised VERDICT model fitting (ssVERDICT) significantly outperforms current state-of-the-art methods.
- This study presents the first successful application of machine learning for fitting a complex biophysical diffusion MRI model without labeled training data.
