Automated deep learning pipeline for callosal angle quantification

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Normal pressure hydrocephalus (NPH) is an underdiagnosed neurodegenerative disorder.
• Manual analysis of imaging biomarkers like callosal angle (CA) is labor-intensive and subjective.
• Automated analysis is needed to improve NPH diagnosis.

Purpose of the Study:

• To develop a fully automated deep learning framework for measuring the callosal angle (CA) from T1 MPRAGE MRI scans.
• To provide a robust and reproducible method for CA measurement, aiding NPH diagnosis.

Main Methods:

• A deep learning framework integrating BrainSignsNET for landmark detection (AC, PC) and a UNet-based network for lateral ventricle segmentation.
• Input: 3D MRI scans preprocessed and reoriented.
• Output: Automated measurement of the callosal angle (CA).

Main Results:

• The framework achieved high concordance with manual CA measurements (r=0.98, p<0.001).
• Mean absolute error (MAE) was 3.26 degrees.
• Performance was independent of patient age, gender, and Evans Index (EI), indicating broad applicability.

Conclusions:

• The automated CA measurement framework is reliable and reproducible, surpassing manual methods and interobserver variability.
• This tool has significant potential for enhancing early NPH detection and diagnosis in research and clinical settings.