Alara group | [VEILLE TECHNOLOGIQUE IRM] Towards prioritizing brain MRI images with Artificial Intelligence

Monitoring

[VEILLE TECHNOLOGIQUE IRM] Towards prioritizing brain MRI images with Artificial Intelligence

Towards prioritizing brain MRI images with Artificial Intelligence

MRI is an essential tool for assessing brain pathologies. However, the urgency of care differs from patient to patient, and it’s important to help practitioners quickly identify patients requiring the most immediate care.

Artificial Intelligence could potentially provide a solution to this problem, not only by optimizing and accelerating clinical procedures, but also by identifying incidental findings: for the first time, a team of researchers1 has used a very large multi-site clinical dataset to train a convolutional neural network to recognize abnormalities in brain MRI images.

3 datasets (for training, validation and testing) from over 9,000 clinical brain MRI scans acquired at various institutions
Images annotated by 4 radiologists according to 10 categories: “probably normal”, “hemorrhage”, “inflammation”,
Images (FLAIR sequences) categorized as “probably normal” or “probably abnormal

The model performs well in distinguishing “probably normal” from “probably abnormal” images
Capacity for generalization to be improved: model tested on a validation set acquired at a different institution from the training set.
Challenge: variability of pathologies and images
Future developments: ability to accurately classify “probably abnormal” images into different categories

A very promising work, to be continued!

Gauriau et al. A Deep Learning-Based Model for Detecting Abnormalities on Brain MRI for Triaging: Preliminary Results from a Multi-Site Experience. Radiology: Artificial Intelligence. April 2021.

[VEILLE TECHNOLOGIQUE IRM] Towards prioritizing brain MRI images with Artificial Intelligence

Share this article

Let's talk together!

Are you a health care professional?

Want to become a partner?

Our brands