CAMM: Confidence-Aligned Multiview Multimodal Fusion for Brain Disorders Prediction With Imaging Transcriptomics.

Area of Science:

• Neuroscience
• Genomics
• Computational Biology

Background:

• Neuroimaging offers structural and functional brain insights but lacks molecular context.
• Transcriptomic data provides molecular details but is difficult to obtain at the individual level.
• Integrating imaging and molecular data is crucial for understanding brain disorders.

Purpose of the Study:

• To develop a confidence-aware multi-modal framework (CAMM) for brain disorder prediction.
• To embed molecular context from transcriptomic data into neuroimaging features.
• To enhance model robustness and interpretability in precision medicine.

Main Methods:

• CAMM integrates transcriptomic priors with imaging features for multi-modal fusion.
• A confidence calibration-regularization strategy adapts modality contributions per sample.
• High-confidence samples inform predictions for low-confidence samples, improving robustness.

Main Results:

• CAMM consistently outperformed state-of-the-art baselines on large neuroimaging cohorts.
• The framework identified biologically meaningful biomarkers for brain disorders.
• CAMM demonstrated effective integration of molecular and imaging data.

Conclusions:

• CAMM successfully bridges molecular mechanisms and imaging data for brain disorder prediction.
• The confidence-aware approach enhances model robustness and interpretability.
• This framework advances precision modeling for neurological conditions.