Fully Automated Deep Learning-Based Pipeline for Evans Index Measurement from Raw 3D MRI

Area of Science:

• Neuroimaging
• Medical Imaging Analysis
• Artificial Intelligence in Medicine

Background:

• Ventriculomegaly is a key indicator in cerebrospinal fluid (CSF) disorders like normal pressure hydrocephalus (NPH).
• The Evans Index (EI) is a standard measure of ventricular enlargement, but manual calculation suffers from observer variability and alignment issues.
• Reproducibility challenges limit the utility of manual EI measurements in large-scale and multi-center studies.

Purpose of the Study:

• To develop and validate a fully automated deep learning pipeline for accurate and reproducible Evans Index calculation from T1-weighted MRI scans.
• To overcome the limitations of manual EI measurement, including observer variability and AC-PC plane dependency.
• To provide a scalable and reliable tool for assessing ventriculomegaly in clinical practice and research.

Main Methods:

• A deep learning pipeline integrating landmark detection (BrainSignsNet), AC-PC alignment, and lateral ventricle (LV) and intracranial volume (ICV) segmentation (nnU-Net) was developed.
• A custom nnU-Net model was trained on 1,300 annotated scans, specifically enriched for hydrocephalus cases.
• The Evans Index was derived from automated measurements of frontal horn width and inner skull diameter on standardized axial slices.

Main Results:

• Internal validation demonstrated high segmentation accuracy (Dice coefficient = 0.98) across multiple cohorts.
• External validation on the PENS trial showed excellent agreement with expert manual EI measurements (mean bias = 0.0068, r = 0.96).
• No significant association was found between measurement error and patient demographics or ventricular volume.

Conclusions:

• The automated pipeline provides accurate, reproducible, and orientation-standardized Evans Index measurements across diverse MRI datasets.
• Eliminating manual intervention enhances scalability for large neuroimaging cohorts and clinical applications.
• This method offers a reliable tool for screening, assessment, and monitoring of ventriculomegaly, especially in normal pressure hydrocephalus (NPH).