What is it about?

A new computer model detects Alzheimer's/dementia status in tens of thousands of routinely collected clinical brain MRIs. It uses advanced deep learning methods to both determine the uncertainty of diagnostic predictions, which aids in effective cross-institutional applications, as well as regression of confounding factors, such as patient age, from the prediction process.

Featured Image

Why is it important?

While prior work has been able to detect Alzheimer's/dementia in public benchmark MRI datasets, this shows the ability of advanced deep learning models to perform this same task on low-quality, heterogeneous clinical MRI, while minimizing the bias of the machine learning model and maximizing its ability to perform across institutions.


This publication is an essential step in making diagnostic health AI more clinically relevant. At the moment, much work exists on complex diagnostic projects, but these projects are rarely more than proofs-of-concept. When will we make more concrete steps towards addressing the day-to-day realities of what doctors deal with in clinical data? By applying these AI models to real-world data and adapting deep learning models to the unique considerations of these data, this work brings us one step closer to truly translatable diagnostic AI.

Matthew Leming
Massachusetts General Hospital

Read the Original

This page is a summary of: Adversarial confound regression and uncertainty measurements to classify heterogeneous clinical MRI in Mass General Brigham, PLoS ONE, March 2023, PLOS, DOI: 10.1371/journal.pone.0277572.
You can read the full text:



The following have contributed to this page