Improved Diagnosis and Prognosis of Neurological Diseases using Novel Training Tools and Decision Support Systems Based on a New Quantitative MRI Imaging Technique

Recently, we developed a quantitative MRI (qMRI) based technique, which can emulate pathologies on radiologic images at arbitrary levels of severity. This qMRI technique relies on the actual biological changes that occur during disease, thereby producing more faithful representation of how pathologies appear in radiologic images. The technique can also simulate a range of scanning protocols, scan settings, and types of MRI scanners.
The technology can be used to improve the training of radiologists, as well as the generation of artificial intelligence (AI) based computer vision algorithms – a valuable tool in research, for clinical diagnosis, and for medical imaging companies.
AI can be used for medical image interpretation. However, training of AI tools requires a large volume of tagged data, necessitating many (costly) scans, followed by extensive experts’ time and funds. A popular way to use the limited amount of available data is data augmentation. Current data augmentation methods in the medical area are mostly simplistic (including rotation, translation, scaling, resampling, editing qualitative images) or based on AI (e.g., generative adversarial networks (GANs)).
We propose a method to simulate pathologies based on qMRI measurements. Pathologies are simulated by adjusting the tissues’ physical parameters and simulating the resultant MR images. Simulations also incorporate the specific scanning protocols, thereby providing realistic behavior across different scan settings, scanners, and vendors.
1. The invention can be used as an augmentation technique for training of AI tools for a variety of clinical applications that use image segmentation or classification (e.g., detection of tumors, inflammation, radiation necrosis, MRI-guided interventional therapies, and more).
2. The invention can be employed as a tutorial platform for training radiologists on synthetically tagged images, e.g., (2.1) During their residence period; (2.2) While learning how to diagnose new pathologies; and (2.3) When integrating new MRI contrast mechanisms and protocols into clinical routine.
Working computer program is implemented in Python and is currently used for training a DNN.
Patent pending.

1. C. Solomon et al., “Quantitative Estimation of Visual Sensitivity to Early Pathological Changes in T­2-Weighted Images,” 2020, [Online]. Available: https://www.ismrm.org/20/program_files/DP08-01.htm.

Sign up for
our events

    Life Science