Researchers at the universities of Heidelberg in Germany, and Zurich in Switzerland have developed an AI model to detect human motor impairments and determine underlying diseases.
The new software is also a test for how effective unsupervised AI behaviour analysis can be in discovering and determining complex disease states.
Conventional supervised instrumented movement analysis is time-consuming, potentially subjective, and cost-intensive. It requires prior knowledge of behaviours of interest, and typically a large amount of video frame annotation. There is scope for human annotator bias, with different annotators focusing on different behaviours, while ignoring or minimising others.
The researchers have developed uBAM ("unsupervised Behaviour Analysis and Magnification using Deep Learning"). This is a fully automatic, unsupervised diagnostic support system for behaviour analysis.
The new system can extract and classify behaviour automatically, with the ability to compare and quantify even small differences.
The system thus avoids the need for lengthy annotation.
The researchers claim their system is also more objective, avoiding user bias affecting the learning process.
An interface to the uBAM system is available for free online, together with interactive demos, at:
https://utabuechler.github.io/behaviourAnalysis/
Further reading:
Comments
Post a Comment