REAL-TIME PUPIL CENTER AND OPTICAL AXIS ESTIMATION FOR ROBOTIC OPHTHALMOSCOPY

Chow, Nathan

Permalink

https://hdl.handle.net/2142/124854 Copy

Description

Title

REAL-TIME PUPIL CENTER AND OPTICAL AXIS ESTIMATION FOR ROBOTIC OPHTHALMOSCOPY

Author(s)

Chow, Nathan

Issue Date

2023-05-01

Keyword(s)

pupil detection; gaze estimation; robotic telehealth

Date of Ingest

2024-10-14T11:10:25-05:00

Abstract

Advancements in telehealth technology are allowing medical professionals to reach patients in remote locations and serve populations without regular access to healthcare. Current telehealth services, however, are limited to basic surface examinations and consultations through telephone or video conferences. Expanding the utility of telehealth will require medical imaging systems that perform deeper examinations of patients. A promising candidate for telehealth examination is indirect ophthalmoscopy, which can not only diagnose eye health, but also indicate issues such as anemia, diabetes, or severe and chronic vascular diseases. Typically, performing an indirect ophthalmoscopic exam requires a skilled and trained ophthalmologist to inspect a patient’s fundus, or an expensive ultra-widefield fundus photography machine. Employing a robotic arm for indirect ophthalmoscopy has the potential to lower the cost of the exam in telehealth scenarios. An automated system will require multiple layers of patient tracking, including a precise eye tracker that helps the robot aim an imaging device through the pupil and decide whether or not the patient is looking in an acceptable direction for imaging. We propose using CNN-based regression to simultaneously predict pupil center coordinates and three-dimensional optical axis in real-time from a close-up eye image. We demonstrate the ability of the system, trained only on synthetic data, to make predictions on real eye image datasets as well as on images captured by an off-the-shelf USB camera.

Type of Resource

text

Language

eng

Owning Collections