A Fusion-Based Machine Learning Approach for Autism Detection in Young Children Using Magnetoencephalography Signals.
Kasturi Barik, Katsumi Watanabe, Joydeep Bhattacharya and 1 others
PMID 36192669WHAT IT FOUND
Resting MEG brain signals from 60 children aged 4 to 7 years could separate autistic from typically developing children, with the best model reaching 98% correct among the study children.
It was a small research study.
Key findings
01Combining power and phase features in a score-level fusion model gave an accuracy of 98.33 ± 0.74%.
02The phase-based preferred phase angle feature classified children better than the power spectral density feature, with whole-brain accuracy of 88.20 ± 3.87% versus 82.13 ± 2.11%.
03Most discriminating phase features came from the theta frequency band.
STILL TO COME
How it was doneWhat they found
Read the rest of this summary
You get three full summaries a month, free, and we do not ask for a card. Search, the TL;DRs and your library stay unlimited either way.
What it does not show
Only 60 children were studied, with 4 females in each group. All training, validation and testing data came from the same 60 children, split into fivefold cross-validation folds. The analysis used MEG sensor space only, so it cannot identify the precise brain sources. The recording was a brief 3-minute period. Children watched cartoons, so the paper says intrinsic and task-driven contributions cannot be separated. The authors say additional studies are needed to substantiate complementarity between the phase and power features.
The easy way to misread this
Do not read the 98.33 ± 0.74% accuracy as evidence that MEG can diagnose autism in a clinic. The paper reports a machine-learning result from cross-validation within 60 children.