Infant Cognitive Scores Prediction with Multi-stream Attention-Based Temporal Path Signature Features
Xin Zhang1, Jiale Cheng1, Hao Ni2
1School of Electronic and Information Engineering, South China University of Technology, Guangzhou, China.
Medical Image Computing and Computer-Assisted Intervention : MICCAI ... International Conference on Medical Image Computing and Computer-Assisted Intervention
|February 17, 2021
Summary
This study introduces a novel path signature method (BrainPSNet) to predict cognitive scores from infant brain development data. The approach enhances understanding of brain morphology and cognitive abilities, achieving state-of-the-art results.
Area of Science:
- Neuroscience
- Developmental Biology
- Computational Biology
Background:
- Human brain development is rapid in the first year, with links between cognition and cortical morphology.
- Predicting cognitive scores from longitudinal brain data is challenging due to small sample sizes and missing data.
Purpose of the Study:
- To introduce the path signature method for analyzing longitudinal cortical morphology.
- To develop a novel neural network (BrainPSNet) for predicting cognitive scores.
Main Methods:
- Utilized a differentiable temporal path signature layer for feature representation.
- Employed a two-stream neural network combining raw and path signature features.
- Implemented a learning-based attention mask generator to weight brain regions.
Main Results:
- The proposed BrainPSNet method achieved state-of-the-art performance on an in-house longitudinal dataset.
- Demonstrated the ability to predict cognitive scores using brain morphological features.
- Analyzed the relationship between specific brain regions and cognitive abilities.
Conclusions:
- The path signature method offers a powerful tool for exploring hidden properties of longitudinal brain data.
- BrainPSNet effectively predicts cognitive scores, advancing our understanding of early brain development and cognition.
- Attention mechanisms improve the model's ability to identify key brain regions influencing cognitive function.
