Multi-task Joint Prediction of Infant Cortical Morphological and Cognitive Development

Area of Science:

• Neuroscience
• Developmental Biology
• Artificial Intelligence

Background:

• Infant brain development is rapid and complex, making individual-level prediction challenging.
• Longitudinal infant data is often irregular and incomplete, hindering accurate developmental trajectory analysis.
• Joint prediction of cognitive scores and cortical morphology is understudied despite their known relationship.

Purpose of the Study:

• To develop a flexible multi-task framework for the joint prediction of infant cognitive scores and cortical morphology.
• To address challenges posed by irregular and incomplete longitudinal data in developmental studies.
• To investigate the mutual benefits of jointly predicting cognitive and cortical development.

Main Methods:

• Developed a disentangled intensive triplet spherical adversarial autoencoder (DITSAA) framework.
• Utilized a triplet spherical ResNet and attention module to disentangle identity-related and age-related features.
• Incorporated identity recognition and age estimation tasks for supervision, adversarial learning for morphology prediction, and a cognitive module for score prediction.

Main Results:

• The DITSAA framework successfully performed joint prediction of cognitive scores and cortical morphology.
• The method demonstrated flexibility in handling incomplete longitudinal data.
• Experimental results validated the framework's ability to predict individual developmental trajectories.

Conclusions:

• The proposed DITSAA framework offers a novel and effective approach for joint prediction of infant cognitive and cortical development.
• This method advances the ability to analyze complex, incomplete longitudinal data in developmental neuroscience.
• The findings support the potential for mutual benefits in jointly modeling cognitive and morphological development.