Enhancing cerebral infarct classification by automatically extracting relevant fMRI features.

Area of Science:

• Neuroimaging
• Machine Learning
• Neurology

Background:

• Accurate detection of cortical infarcts is crucial for effective stroke treatment and patient outcomes.
• Current brain imaging methods are often invasive and focus on vascular or white matter damage, not neuronal viability.
• There is a need for non-invasive functional MRI (fMRI) techniques to assess neuronal function in infarct detection.

Purpose of the Study:

• To utilize automated machine learning (auto-ML) to discover novel infarct-specific fMRI biomarkers for chronic cortical infarcts.
• To evaluate the performance of auto-generated fMRI biomarkers against existing metrics for infarct detection.
• To develop a robust, non-invasive method for enhanced infarct detection using fMRI.

Main Methods:

• Analysis of resting-state fMRI data from the multi-center Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset.
• Application of surface-based registration to mitigate partial-volume effects in fMRI data.
• Evaluation of 7 known and 107 auto-generated fMRI biomarkers across 33 classification models.

Main Results:

• Identification of 6 novel fMRI biomarkers that significantly improved infarct detection performance.
• The best biomarker-classifier combination achieved a cross-validation ROC score of 0.791, comparable to acute stroke imaging methods.
• The auto-ML fMRI technique demonstrated robustness across different imaging sites and scanner types.

Conclusions:

• Automated feature extraction using auto-ML can significantly enhance non-invasive infarct detection via fMRI.
• The identified novel fMRI biomarkers offer a promising tool for improved diagnosis of chronic cortical infarcts.
• This approach has the potential to improve patient outcomes through earlier and more accurate detection.