Topic: ESOPRS 2021 ePoster sessions
Time: Sep 17, 2021 16:00 Amsterdam, Berlin, Rome, Stockholm, Vienna, 15:00 London
(plain text version here)
Validation of automated eyelid metrics using a novel Convolutional Neural Network model enabled headset
Author: Jameel Mushtaq
ePoster Number: 127
Purpose
Precise eyelid measurements are essential in diagnosing and managing a variety of eyelid disorders. However, traditional manual assessment of is time-consuming and prone to inter-observer variability, potentially impacting clinical consistency. Automated eyelid measurement systems using convolutional neural network (CNN) models aim to provide objective, reproducible metrics that can improve throughput, clinical decision-making and surgical planning. The present study evaluates the accuracy of a CNN-based system combined with an eye-tracking headset for automated measurements across different gaze positions compared to manual measurements.
Methods
A previously trained CNN model was used to automate measurement of marginal reflex difference (MRD) 1 and 2, in three positions: primary, elevation and depression. Images were captured using a near infrared eye tracking headset (BulbiCam, Bulbitech, Norway). 8mm markers were attached to the patient periocular area and recorded in the capture as a known reference point. Reference images in the cardinal positions were exported onto ImageJ (National Institutes of Health, US) and manual MRD1 and MRD2 values were computed using known markers as standardised reference by two observers. Bland-Altman analysis and Intraclass correlation coefficients (ICCs) were calculated to determine correlation of measurements.
Results
Ten patients (20 eyes) were recruited. Automated MRD1 and MRD2 measurements closely matched the manual measurements across all gaze positions. Accuracy was most affected on downgaze. Bland-Altman analysis demonstrated minimal bias with narrow limits of agreement, and ICCs indicated excellent correlation between the measures.
Conclusion
A CNN with dedicated image capture device is an accurate and precise tool for measuring eyelid metrics in multiple gaze positions, providing a non-invasive and efficient alternative to manual measurement. This automated approach could help standardise clinical assessments, reduce examiner variability, and streamline workflow in busy oculoplastic clinics.
Additional Authors
| First name | Last name | Base Hospital / Institution |
|---|---|---|
| Alyssa | Dunlop | Imperial College London |
| Omar | Alghaith | Institution of Ophthalmology, UCL |
| Oliver | Dawe | Moorfields Eye Hospital |
| Mohsan | Malik | Moorfields Eye Hospital |
| Jimmy | Uddin | Moorfields Eye Hospital |
Abstract ID: 25-481