BPEN: Brain Posterior Evidential Network for trustworthy brain imaging analysis

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Deep learning advances brain functional magnetic resonance imaging (fMRI) analysis for identifying neurological biomarkers.
• Prediction uncertainty in fMRI analysis remains underexplored, hindering trustworthy AI in diagnostics.
• Accurate uncertainty estimation is crucial for reliable clinical decision-making with fMRI data.

Purpose of the Study:

• To introduce a novel deep learning model, the Brain Posterior Evidential Network (BPEN), for quantifying uncertainty in fMRI data analysis.
• To capture both aleatoric and epistemic uncertainty within brain fMRI predictions.
• To enhance the reliability and trustworthiness of AI-driven diagnostic tools in neurology.

Main Methods:

• Developed the Brain Posterior Evidential Network (BPEN), a novel deep learning architecture.
• Applied BPEN to analyze functional magnetic resonance imaging (fMRI) data from Alzheimer's Disease Neuroimaging Initiative (ADNI) and ADNI-depression (ADNI-D) cohorts.
• Evaluated BPEN's performance in predicting mild cognitive impairment (MCI) and depression, comparing it against state-of-the-art methods.

Main Results:

• The BPEN model demonstrated superior predictive performance compared to existing state-of-the-art methods in fMRI analysis.
• BPEN effectively captured and quantified both aleatoric and epistemic uncertainty in predictions for MCI and depression.
• Experiments highlighted the critical role of uncertainty estimation for robust diagnostic predictions.

Conclusions:

• The Brain Posterior Evidential Network (BPEN) offers a significant advancement in analyzing brain fMRI data by incorporating uncertainty estimation.
• BPEN enhances the trustworthiness of AI models in clinical applications, particularly for neurological conditions like MCI and depression.
• This work emphasizes the necessity of addressing prediction uncertainty for reliable AI-powered diagnostics in neuroimaging.