Multi-scale asynchronous correlation and 2D convolutional autoencoder for adolescent health risk prediction with...

Area of Science:

• Neuroscience
• Machine Learning
• Adolescent Health

Background:

• Adolescence involves critical physical, psychological, and behavioral changes.
• Accurate health risk assessment in adolescents is vital for timely interventions.
• Traditional methods struggle to predict mental/behavioral risks due to complex neural dynamics and limited fMRI data.

Purpose of the Study:

• To develop an innovative deep learning framework for adolescent health risk assessment.
• To enhance the accuracy of predicting mental and behavioral health risks using fMRI data.
• To leverage advanced techniques for analyzing spatial and temporal features in fMRI scans.

Main Methods:

• Utilized a two-dimensional convolutional autoencoder (2DCNN-AE) combined with multi-sequence learning.
• Employed multi-scale asynchronous correlation information extraction techniques.
• Analyzed spatial and temporal features within functional Magnetic Resonance Imaging (fMRI) data.

Main Results:

• The deep learning framework demonstrated superior performance on the Adolescent Risk Behavior (AHRB) dataset.
• Achieved a precision of 83.116%, recall of 84.784%, and F1-score of 83.942%.
• Outperformed conventional models in predicting adolescent health risks.

Conclusions:

• The deep learning approach significantly advances the precision of adolescent health risk assessments.
• This methodology offers a powerful tool for early detection and intervention in adolescents.
• Highlights the potential of AI in understanding and predicting health-related risks during this developmental stage.