Related Experiment Video
Updated: Sep 5, 2025
13:44
Detection of Architectural Distortion in Prior Mammograms via Analysis of Oriented Patterns
Published on: August 30, 2013
43.0K
Improved differentiation between hypo/hypertelorism and normal fetuses based on MRI using automatic ocular biometric
Netanell Avisdris1,2, Daphna Link Sourani3, Liat Ben-Sira4,5,6
1Sagol Brain Institute, Tel Aviv Sourasky Medical Center, Tel Aviv, Israel. netana03@cs.huji.ac.il.
European Radiology
|July 13, 2022
Summary
A new automatic method using fetal MRI accurately measures ocular biometry. This approach, along with novel BOD and IOD ratios and machine learning, improves detection of fetal hypo-/hypertelorism.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Fetal Medicine
Background:
- Accurate differentiation of fetal hypo-/hypertelorism is crucial for diagnosing craniofacial abnormalities.
- Current diagnostic criteria rely on specific inter-ocular distance (IOD) percentiles, which may have limitations.
- Automated analysis of fetal magnetic resonance imaging (MRI) offers potential for objective biometric assessment.
Purpose of the Study:
- To develop and validate a fully automatic method for fetal ocular biometry using MRI.
- To introduce and evaluate novel biometric ratios (BOD-ratio, IOD-ratio) for assessing ocular development.
- To enhance the classification accuracy of fetal hypo-/hypertelorism using machine learning (ML) with automated measurements.
Main Methods:
- A dataset of 244 normal and 52 abnormal fetal MRI scans (22-40 weeks gestational age) was analyzed.
- A deep learning and geometric algorithm-based method automatically measured binocular (BOD), inter-ocular (IOD), ocular (OD) diameters, and ocular volume (OV).
- Eight ML classifiers were trained and tested using measured parameters and two novel ratios (BOD-ratio, IOD-ratio) to detect abnormalities.
Main Results:
- The automatic method demonstrated high accuracy in ocular biometry measurements (e.g., 3D-Dice score of 93.7% for OV).
- Novel BOD-ratio and IOD-ratio remained constant across gestational age in normal fetuses and aligned with existing reference data.
- A random forest classifier using the automated measurements and ratios achieved superior performance (AUC-ROC = 0.941) compared to traditional IOD percentile criteria (AUC-ROC = 0.650).
Conclusions:
- The developed fully automatic MRI-based method provides accurate fetal ocular biometry.
- The novel BOD and IOD ratios are reliable, constant parameters for assessing normal ocular development.
- Multi-parametric ML classification incorporating these new ratios significantly improves the detection of fetal hypo-/hypertelorism.
