Researchconvolutional autoencoderlstmimu
Deep Learning Classifies ASD From IMU Movements
7.0
In a 2025 JMIR Medical Informatics report, researchers recorded IMU hand-tracking data from 41 school-aged children with and without ASD to classify diagnoses using deep learning. A convolutional autoencoder combined with LSTM achieved 90.2% accuracy in k-fold testing and 91.9% accuracy with 93.7% F1 in patient-separated validation. The results suggest wearable IMU-based models could support objective, low-cost ASD screening.
Key Points
- 1Demonstrates convolutional autoencoder+LSTM classifies ASD with ~91.9% accuracy and 93.7% F1.
- 2Highlights motor-control differences in ASD detectable via low-cost wrist-mounted IMU tracking.
- 3Enables clinically feasible, small-scale diagnostic models without massive datasets for early screening.
Scoring Rationale
Credible peer-reviewed results with practical diagnostic promise, but small sample and need for wider validation limit impact.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
