Using resting state functional MRI to build a personalized autism diagnosis system

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Autism spectrum disorder (ASD) is a complex neurodevelopmental condition with no known cause.
• Understanding brain functional connectivity offers new diagnostic and research avenues for ASD.
• Current diagnostic methods for ASD can be subjective and time-consuming.

Purpose of the Study:

• To develop and validate a machine learning-based automated system for diagnosing ASD.
• To investigate the utility of resting-state functional MRI (fMRI) for ASD diagnosis.
• To generate personalized diagnostic reports for individuals with ASD.

Main Methods:

• Utilized resting-state fMRI data from 283 subjects obtained from the National Database of Autism Research (NDAR).
• Developed a machine learning model employing stacked autoencoders and probabilistic support vector machines.
• Input features were derived from power spectral densities (PSDs) of fMRI time courses.

Main Results:

• The automated system achieved approximately 90% sensitivity, specificity, and accuracy on the dataset.
• The system demonstrated strong generalizability across different population prevalence rates, particularly in high-risk groups.
• Personalized reports identified global brain differences and impacted areas in individuals with ASD.

Conclusions:

• The developed machine learning system shows high efficacy in diagnosing ASD using fMRI data.
• The system's ability to provide personalized reports aids in understanding ASD and planning tailored treatments.
• This work represents a significant step towards personalized medicine for autism.