A Neural Network Approach to Identify Left-Right Orientation of Anatomical Brain MRI

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Image Analysis

Background:

• Left-right orientation misidentification in brain MRIs is a significant challenge.
• Metadata loss or ambiguity arises from de-identification, format conversion, and software operations.
• Older imaging systems also stored orientation data differently, contributing to misidentification.

Purpose of the Study:

• To present a novel deep learning application for accurate left-right orientation identification in anatomical brain MRI scans.
• To address the challenges posed by metadata loss or ambiguity in brain MRI datasets.
• To improve the reliability of neuroimaging research through accurate image orientation.

Main Methods:

• A three-dimensional convolutional neural network (3D CNN) model was developed.
• The model was trained on 350 MRI scans.
• Performance was evaluated on eight diverse brain MRI databases, totaling 3056 scans, including those with neurodegenerative diseases.

Main Results:

• The deep learning framework achieved 99.8% accuracy in identifying left-right orientation.
• GradCAM visualization highlighted the right planum temporale as a key area for orientation determination.
• Biological validity was supported by the planum temporale's known asymmetry related to language functions.

Conclusions:

• The proposed deep learning approach offers a robust solution to persistent left-right misorientation in brain MRIs.
• Accurate orientation identification ensures reliable data interpretation, bolstering the integrity of neuroscientific research.
• This method has the potential to significantly improve the quality and usability of brain MRI data for research purposes.