Feed-forward hierarchical model of the ventral visual stream applied to functional brain image classification

Area of Science:

• Neuroimaging
• Computer Vision
• Medical Diagnostics

Background:

• Functional brain imaging is crucial for tracking neurodegenerative diseases.
• Computer vision excels at object recognition but its application to medical imaging is evolving.
• Accurate disease detection from brain scans is a significant medical challenge.

Purpose of the Study:

• To extend a 2D computer vision model to 3D functional brain imaging for disease detection.
• To evaluate the model's effectiveness on Alzheimer's disease and NFL player datasets.
• To identify key components of the filtering pipeline for improved classification accuracy.

Main Methods:

• Adapted a biophysically inspired filtering method (Serre et al., 2005) for 3D volumetric data.
• Applied the filter to Positron Emission Tomography (PET) and Single-Photon Emission Computed Tomography (SPECT) scans.
• Trained neural network and logistic regression classifiers using filter outputs on ADNI and NFL datasets.

Main Results:

• The 3D filtering approach demonstrated effective signal detection in functional brain imaging.
• Classification performance compared favorably with existing methods and outperformed a human expert.
• Analysis identified critical steps within the filtering pipeline influencing classification accuracy.

Conclusions:

• The extended 3D computer vision model offers a valuable tool for neurological disease detection in functional imaging.
• This approach shows potential for enhancing diagnostic capabilities in conditions like Alzheimer's disease.
• The method's performance suggests its utility in clinical settings and for assessing neurological health.