Diagnostic Performance of Deep Learning for Angle Closure

Diagnostic Performance of Deep Convolutional Neural Networks for Angle Closure Glaucoma: an International Multicenter Study

Primary angle closure diseases (PACD) are commonly seen in Asia. In clinical practice, gonioscopy is the gold standard for angle width classification in PACD patietns. However, gonioscopy is a contact examination and needs a long learning curve. Anterior segment optical coherence tomography (AS-OCT) is a non-contact test which can obtain three dimensional images of the anterior segment within seconds. Therefore, the investigators designed the study to verify if AS-OCT based deep learning algorithm is able to detect the PACD subjects diagnosed by gonioscopy.

Study Overview

• Status: Active, not recruiting
• Conditions: Angle Closure Glaucoma
• Intervention / Treatment: Diagnostic test: Deep learning algorithm based on AS-OCT scans

• Study Type: Observational
• Enrollment (Anticipated): 3000

Contacts and Locations

Study Locations

• China
  • Guangdong
    • Guangzhou, Guangdong, China, 51000
      • Zhongshan Ophthalmic Center

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: 18 years and older (Adult, Older Adult)
• Accepts Healthy Volunteers: No
• Genders Eligible for Study: All

• Sampling Method: Non-Probability Sample

Study Population

The training and primary validation datasets were collected from the databases of electronic medical and research records at Zhongshan Ophthalmic Center from September 1, 2016, to September 1, 2019. The external test dataset was obtained from the Singapore Eye Research Institute (SERI), Singapore during June 2008 to November 2019, and the Chulalongkorn University and King Chulalongkorn Memorial Hospital (KCMH, Bangkok, Thailand) from October, 2019 to April, 2020.

Description

The inclusion criteria in the study were as follows: (1) All participants must be ≥ 18 years old; (2) Study subjects had a previous diagnosis of the ACA status (narrow or open, PAS or non-PAS) based on gonioscopy, SS-OCT scans and medical history records. Exclusion criteria of the data include: (1) poor compliance in receiving gonioscopy examination; (2) unclear AS-OCT scans due to blinking or out of focus; (3) recent use of miotics within a month; 4) secondary angle closure sue to subluxation or dislocation, uveitis, neovascular glaucoma, et al.; 5) history of ocular surgery or laser iridotomy; 6) patients who previously had an episode of primary angle closure (which was obtained on history by asking the patients).

Study Plan

How is the study designed?

Design Details

• Observational Models: Other
• Time Perspectives: Retrospective

Number of groups / cohorts

4

Cohorts and Interventions

Group / Cohort	Intervention / Treatment
Angle closure group	Diagnostic test: Deep learning algorithm based on AS-OCT scans; The OCT scans of study subjects would be imported into the algorithm. Automated classfication of angle width and detection of synechia would be performed by the algorithm. The diagnostic performance of the algorithm would be compared with gonioscopy records.
Open angle group	Diagnostic test: Deep learning algorithm based on AS-OCT scans; The OCT scans of study subjects would be imported into the algorithm. Automated classfication of angle width and detection of synechia would be performed by the algorithm. The diagnostic performance of the algorithm would be compared with gonioscopy records.
Peripheral synechia (PAS) group	Diagnostic test: Deep learning algorithm based on AS-OCT scans; The OCT scans of study subjects would be imported into the algorithm. Automated classfication of angle width and detection of synechia would be performed by the algorithm. The diagnostic performance of the algorithm would be compared with gonioscopy records.
Non-peripheral synechia (PAS) group	Diagnostic test: Deep learning algorithm based on AS-OCT scans; The OCT scans of study subjects would be imported into the algorithm. Automated classfication of angle width and detection of synechia would be performed by the algorithm. The diagnostic performance of the algorithm would be compared with gonioscopy records.

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Area under receiver operating curve (AUC)	AUC value of the deep learning algorithm in angle width classfication and synechia detection	Immediately after obtaining the AS-OCT images

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Sensitivity and specificity	Sensitivity and specificity of the automated algorithm in angle width classfication and synechia detection	Immediately after obtaining the AS-OCT images

Collaborators and Investigators

• Sponsor: Sun Yat-sen University

Study record dates

Study Major Dates

• Study Start (Actual): January 15, 2019
• Primary Completion (Anticipated): December 1, 2021
• Study Completion (Anticipated): March 1, 2022

Study Registration Dates

• First Submitted: January 23, 2020
• First Submitted That Met QC Criteria: January 23, 2020
• First Posted (Actual): January 27, 2020

Study Record Updates

• Last Update Posted (Actual): April 8, 2021
• Last Update Submitted That Met QC Criteria: April 3, 2021
• Last Verified: April 1, 2021

More Information

Terms related to this study

• Additional Relevant MeSH Terms: Eye Diseases; Ocular Hypertension; Glaucoma; Glaucoma, Angle-Closure
• Other Study ID Numbers: 2018KYPJ074

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: Yes
• IPD Plan Description: The imaging data of study subjects would be available to other researchers upon reasonable request. Part of the data would be open as public datasets after the related article is published.
• IPD Sharing Time Frame: Part of the data would be open as public datasets after the related article is published.
• IPD Sharing Supporting Information Type: Study Protocol; Analytic Code

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No