Truace
Subscribe

Establishing Diagnosis Models for Autism Spectrum Disorder With Auditory Brainstem Response Using Machine Learning Methods

This study aimed to develop and validate models built using machine learning methods for the diagnosis of autism spectrum disorder (ASD). A total of 851 children with suspected ASD were enrolled. Related features were acquired using the Autism Behavior Checklist and Auditory Brainstem Response (ABR) outcomes. The Shapley additive explanation (SHAP) values were used to explain the feature impact used in the models with the best performance. In addition, different methods were used to depict group differences. Our…

By the Truvace desk2 min read1 sourcePeer-reviewedTRV-2026-1951· cite & verify
Establishing Diagnosis Models for Autism Spectrum Disorder With Auditory Brainstem Response Using Machine Learning Methods
Autism walk brings out supportive families DVIDS901298 by Staff Sgt. Antwaun Parrish. Public domain

In brief

Researchers built and validated machine learning diagnostic models for autism spectrum disorder using data from 851 children with suspected autism. Inputs were scores from the Autism Behavior Checklist and results from auditory brainstem response testing, and SHAP values were used to explain which features drove the best-performing models.

The work matters because it links measurable auditory pathway signals to autistic behaviors and suggests a physiological marker that could complement behavioral assessment. The source reports only moderate discrimination, with an AUROC of 0.779, and describes the pathway association as something that may exist, so clinical usefulness remains unproven.

Main points

  1. The study enrolled 851 children with suspected autism spectrum disorder.
  2. Features came from the Autism Behavior Checklist and auditory brainstem response outcomes.
  3. XGBoost outperformed the other candidate models, with 85.9% accuracy and an AUROC of 0.779.
  4. Waves III and V had the largest impact on model predictions.
  5. Group differences in ABR outcomes among phenotypes depended mainly on stimulation intensity.

The gain

Machine learning models combining Autism Behavior Checklist and auditory brainstem response features could classify children with suspected autism, with XGBoost reaching 85.9% accuracy and Waves III and V as the strongest features.

The rundown

The team compared several machine learning methods and used SHAP values to interpret the models that performed best. Additional methods were applied to depict differences between groups of children with different phenotypes.

Results showed that ABR differences between phenotypes depended mainly on stimulation intensity. The authors suggest signal conduction from the cochlear nerve to the inferior colliculus may be associated with autistic behaviors, with region-specific effects such as the cochlear nerve and olivary regions and intensity-dependent effects such as at 45 and 65 dB.

The authors describe the findings as highlighting atypical ABR patterns in autism and adding to literature on altered bioelectrical performance in affected children.

Sources

  1. Peer-reviewedAutism Research2026-10-16

The debate