Early Diagnosis of Autism Disease by Multi-channel CNNs

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Developmental Pediatrics

Background:

• Early diagnosis of autism spectrum disorder (ASD) remains challenging, with current methods relying on behavioral observations typically made at 3-4 years of age.
• Delayed diagnosis can lead to missed intervention opportunities during critical developmental windows, potentially impacting long-term outcomes for affected children.
• Existing neuroimaging-based predictive methods for ASD are limited, particularly for early-age prediction.

Purpose of the Study:

• To develop and validate an automated method for identifying infants at risk of ASD using magnetic resonance imaging (MRI).
• To establish imaging-based biomarkers for earlier and more effective ASD diagnosis.
• To improve intervention timing by enabling detection before the typical diagnostic age.

Main Methods:

• Implementation of a patch-level data-expanding strategy inspired by deep multi-instance learning.
• Utilizing multi-channel convolutional neural networks (CNNs) for automated analysis of infant brain MRI data.
• Conducting experiments on the National Database for Autism Research (NDAR) dataset.

Main Results:

• The proposed method demonstrated a significant improvement in the performance of early ASD diagnosis.
• The patch-level data expansion strategy enhanced the CNN's ability to identify subtle imaging markers associated with ASD risk.
• Validation on the NDAR dataset confirmed the method's efficacy in predicting ASD risk in infants.

Conclusions:

• The developed AI-driven neuroimaging approach shows promise as a tool for early ASD risk identification in infants.
• This method has the potential to facilitate earlier interventions, addressing a critical gap in current diagnostic practices.
• Further research and validation are warranted to integrate this technique into clinical settings for improved early detection of ASD.