Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Next-Generation Teleophthalmology: AI-enabled Quality Assessment Aiding Remote Smartphone-based Consultation

Dhruv Srikanth, Jayang Gurung, N Satya Deepika, Vineet Joshi, Lopamudra Giri, Pravin Vaddavalli, Soumya Jana

TL;DR

This study tackles AI-assisted quality assessment for smartphone-based teleophthalmology using the Grabi adapter. It envisions a hierarchical two-detector system (presence of eye and suitable lighting) built on a ResNet-18 backbone, trained on user-captured data to emulate clinician judgments and provide instant feedback. The approach yields strong end-to-end performance (≈91% accuracy, P4 ≈ 0.904, Custom ≈ 0.824) and demonstrates feasibility for reducing clinician vetting and re-capture delays in remote eye care, with Tier 3 focus/clarity considerations slated for future work. Overall, the work presents a practical pathway to reliable, real-time image quality assessment in low-resource teleophthalmology settings, potentially improving access and responsiveness in LMICs.

Abstract

Blindness and other eye diseases are a global health concern, particularly in low- and middle-income countries like India. In this regard, during the COVID-19 pandemic, teleophthalmology became a lifeline, and the Grabi attachment for smartphone-based eye imaging gained in use. However, quality of user-captured image often remained inadequate, requiring clinician vetting and delays. In this backdrop, we propose an AI-based quality assessment system with instant feedback mimicking clinicians' judgments and tested on patient-captured images. Dividing the complex problem hierarchically, here we tackle a nontrivial part, and demonstrate a proof of the concept.

Next-Generation Teleophthalmology: AI-enabled Quality Assessment Aiding Remote Smartphone-based Consultation

TL;DR

This study tackles AI-assisted quality assessment for smartphone-based teleophthalmology using the Grabi adapter. It envisions a hierarchical two-detector system (presence of eye and suitable lighting) built on a ResNet-18 backbone, trained on user-captured data to emulate clinician judgments and provide instant feedback. The approach yields strong end-to-end performance (≈91% accuracy, P4 ≈ 0.904, Custom ≈ 0.824) and demonstrates feasibility for reducing clinician vetting and re-capture delays in remote eye care, with Tier 3 focus/clarity considerations slated for future work. Overall, the work presents a practical pathway to reliable, real-time image quality assessment in low-resource teleophthalmology settings, potentially improving access and responsiveness in LMICs.

Abstract

Blindness and other eye diseases are a global health concern, particularly in low- and middle-income countries like India. In this regard, during the COVID-19 pandemic, teleophthalmology became a lifeline, and the Grabi attachment for smartphone-based eye imaging gained in use. However, quality of user-captured image often remained inadequate, requiring clinician vetting and delays. In this backdrop, we propose an AI-based quality assessment system with instant feedback mimicking clinicians' judgments and tested on patient-captured images. Dividing the complex problem hierarchically, here we tackle a nontrivial part, and demonstrate a proof of the concept.
Paper Structure (10 sections, 2 equations, 5 figures, 2 tables)

This paper contains 10 sections, 2 equations, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Materials and Methods
  3. Problem Statement and Approach
  4. Data Collection and Data Preparation
  5. Proposed AI-based Tools and Solutions
  6. General engineering principle
  7. Specific considerations and variants
  8. Performance Metrics
  9. Training, validation and testing
  10. Experimental Results & Discussion

Figures (5)

  • Figure 1: (a) Labelled slitlmap image of anterior segment of human eye, (b) Grabi universal attachment fixed on dummy smartphone, (c) Grabi assisted eye image capture, (d) and (e) Slitlamp and Grabi images, respectively, of the same eye.
  • Figure 2: Ten representative Grabi images of the anterior segment with effect of specific factors illustrated columnwise: (a) presence of eye, (b) focus, (c) illumination, (d) magnification, (e) completeness of cornea; top panel -- unsatisfactory images, bottom panel -- satisfactory images.
  • Figure 3: Quality assessment of self-captured Grabi images: Hierarchical solution approach.
  • Figure 4: Flowchart of the envisaged solution for end-to-end image quality assessment.
  • Figure 5: Cross-validation loss plot for ResNet-18 (all layers trained without transform): (a) Presence-of-eye detector; (b) Suitable-lighting detector. Dotted lines bounding each loss curve denote $\textit{mean} \pm \textit{standard deviation}$.