SliceMap: a binary classification-driven 2D pipeline for detecting discriminative candidate regions in brain MRI

Area of Science:

• Neuroimaging
• Radiology
• Artificial Intelligence in Medicine

Background:

• Detecting subtle, localized neuroanatomical signals in brain MRI is crucial for clinical diagnosis and understanding disease mechanisms.
• Challenges persist in reliably identifying these signals within high-dimensional MRI data.

Purpose of the Study:

• To develop and validate a performance-guided pipeline for identifying candidate spatial regions in brain MRI.
• To leverage 2D convolutional neural networks for efficient slice-level analysis and 3D region localization.

Main Methods:

• A 2D slice-based pipeline using convolutional neural networks for binary classification on MRI slices.
• Employing occlusion-based attribution analysis on the best-performing slices to generate localization maps.
• Joint examination of attribution maps to identify a candidate 3D brain region.

Main Results:

• The pipeline successfully identified a spatially localized candidate region associated with sex classification.
• The identified region corresponded to the anterior cingulate cortex and adjacent medial structures.
• Findings align with existing neuroanatomical knowledge of sex differences in these brain areas.

Conclusions:

• The proposed pipeline offers an effective method for identifying localized neuroanatomical signals in brain MRI.
• This approach can aid in medical diagnosis and the investigation of neurological conditions.
• The technique demonstrates potential for discovering subtle, spatially specific biomarkers in neuroimaging.