Unified framework for early stage status prediction of autism based on infant structural magnetic resonance imaging.
Kun Gao, Yue Sun, Sijie Niu and 1 others
PMID 34643325WHAT IT FOUND
A machine learning model predicted autism status at 24 months from infant brain MRI with 91.5% accuracy in one dataset.
It also classified subjects in a second dataset using a different scanner, but validation was small and not a clinical test.
Key findings
01The model reached 86.5% sensitivity, 92.8% specificity, and 91.5% accuracy in dataset A when sex information was included.
02When trained on dataset A and tested on dataset B, the model reported sensitivity 0.818, specificity 0.846, and accuracy 0.829.
03The proposed method showed more than 10% improvement in specificity and accuracy compared with prior extra-axial cerebrospinal fluid methods in dataset A.
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
The datasets were small, especially dataset B with 35 subjects. The study only tested infants scanned at about 24 months, so it does not show prediction at younger ages. The model was evaluated on research datasets with existing ASD and normal control labels, not on a new clinical population or prospective diagnosis. The study did not test whether the model changes diagnosis, intervention, or patient outcomes. The method depends on iBEAT V2.0 Cloud segmentation and parcellation, so performance may not transfer to other pipelines. The study did not include intelligence quotients, ages, or other non-imaging parameters.
Declared interests
The authors declared no conflict of interest. The work was funded by the National Institutes of Health.
The easy way to misread this
Do not treat the 91.5% accuracy as a diagnostic test ready for clinics. The study used small research datasets of infants already labelled ASD or normal control, and it did not test whether the model changes diagnosis or outcomes.