Early screening of autism spectrum disorder using cry features

Area of Science:

• Developmental Pediatrics
• Computational Linguistics
• Machine Learning

Background:

• Rising prevalence of Autism Spectrum Disorder (ASD) necessitates early detection.
• Early intervention significantly improves outcomes for children with ASD.
• Existing screening methods may lack the sensitivity for early and automatic detection.

Purpose of the Study:

• To introduce a cry-based screening approach for early and automatic detection of ASD in children.
• To develop a novel classification method to identify ASD-specific vocal features and instances.
• To evaluate the performance of the proposed approach compared to existing methods.

Main Methods:

• A cry-based dataset was collected from children aged 18-53 months using high-quality recorders and smartphones.
• A new classification approach was developed to identify subtle ASD-specific vocal features.
• A classifier was trained on data from boys with ASD and typically developed (TD) boys, then tested on a mixed group of boys and girls with ASD and TD controls.
• The classifier was further piloted on infants aged 10-18 months.

Main Results:

• The cry-based approach demonstrated high sensitivity, specificity, and precision for screening boys (85.71%, 100%, 92.85%) and girls (71.42%, 100%, 85.71%) with ASD.
• The proposed method outperformed common classification techniques and previous voice-feature-based ASD screening studies.
• Pilot testing on infants aged 10-18 months yielded encouraging results for early ASD screening.

Conclusions:

• The developed cry-based approach offers a promising tool for early and automatic Autism Spectrum Disorder screening.
• The novel classification method effectively identifies ASD-specific vocal biomarkers, even when not consistently present.
• The approach shows significant potential for practical application in identifying infants at risk for ASD.