Canonical text (the exact bytes fingerprinted)

TRUVACE RECORD VERSION
record: TRV-2026-2018
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-10-20T10:43:54.791788Z
status: published
lens: g_space
sector: health
headline: Are the Eyes a Mirror of Autism? Explainable AI-Based Biomarker Discovery for Diagnosis
dek: Autism spectrum disorder (ASD) presents significant difficulties in diagnosis due to its nature and diversity of symptoms. Traditional diagnostic methods rely on comprehensive clinical evaluations and complex diagnostic processes. Recent advances in machine learning (ML) and explainable artificial intelligence (XAI) offer promising models that help improve diagnostic accuracy and efficiency in ASD. XAI methods provide explainability by determining feature importance and indicating the extent to which features im…
gain_title: Using explainable AI to select eye-tracking features improved autism classification over traditional feature selection and surfaced four candidate biomarkers that traditional methods missed.
problem_title: (none)
trace_subject: (none)
gain_reading: Using explainable AI to select eye-tracking features improved autism classification over traditional feature selection and surfaced four candidate biomarkers that traditional methods missed.
gain_evidence: the XAI-based FS method improves model performance compared to traditional feature selection techniques | the XAI approach successfully identified four candidate biomarkers critical to ASD diagnosis
problem_reading: (none)
problem_evidence: (none)
quick_read: Researchers used eye-tracking data from children to test whether explainable AI could choose which features a diagnostic model relies on. They compared this approach with conventional feature selection and with a combination of the two.

The work matters because autism diagnosis currently depends on lengthy clinical evaluations, and interpretable models could give clinicians insight into which gaze measures separate autistic from typically developing children. The reported accuracy is moderate, and the biomarkers are only candidates that need further validation.
limitation: The biomarkers are described only as candidates, and the best AUC of 0.78 indicates moderate rather than definitive diagnostic performance.
tag: Evidence-backed gain
key_points: The study used children's eye-tracking data to distinguish children with ASD from typically developing peers. | XAI-based feature selection outperformed traditional feature selection methods when used individually. | Combining feature sets from both approaches gave the highest AUC score of 0.78. | The XAI approach identified four candidate biomarkers not selected by traditional methods.
rundown: The paper frames autism diagnosis as difficult because of the diversity of symptoms and the reliance on comprehensive clinical evaluations and complex diagnostic processes.

Explainable AI methods estimate feature importance, showing how much each feature affects a model's outcome. The authors used this property as a feature selection step, a process they link to biomarker discovery.

The authors state that the identified features could give clinicians useful insights and support clinical decision-making in pediatric neurodevelopmental disorders.
sources:
- peer_reviewed | Applied Sciences | https://doi.org/10.3390/app162010347 | 2026-10-19
prev: 0000000000000000000000000000000000000000000000000000000000000000

sha256

56d2ccc5179bba9428ed3df09896a62d6745bfd9aee953ddb06414c3510b8d40

previous

0000000000000000000000000000000000000000000000000000000000000000