June 2023
Volume 64, Issue 8
Open Access
ARVO Annual Meeting Abstract  |   June 2023
Combination of infrared reflectance and OCT imaging for age-related macular degeneration screening via deep learning
Author Affiliations & Notes
  • Felipe Salgado
    LEAOS, Faculty of Medicine/Clinical Hospital, Universidad de Chile, Santiago, Chile
  • LORETO CUITINO
    LEAOS, Faculty of Medicine/Clinical Hospital, Universidad de Chile, Santiago, Chile
  • Carla Vairetti
    Faculty of Engineering and Applied Sciences, Universidad de los Andes, Santiago, Chile
    Instituto de Sistemas Complejos de Ingenieria, Santiago, Chile
  • Fabian Vega-Tapia
    LEAOS, Faculty of Medicine/Clinical Hospital, Universidad de Chile, Santiago, Chile
  • Sofia Valdivieso
    Faculty of Engineering and Applied Sciences, Universidad de los Andes, Santiago, Chile
  • Sebastián Maldonado
    Instituto de Sistemas Complejos de Ingenieria, Santiago, Chile
    Department of Management, Control and Information Systems, Faculty of Economics and Business, University of Chile, Santiago, Chile, Chile
  • Felipe Rojas
    LEAOS, Faculty of Medicine/Clinical Hospital, Universidad de Chile, Santiago, Chile
    Escuela de Tecnología Médica, Facultad de Medicina, Universidad Andres Bello, Santiago, Chile
  • Alvaro Olate-Perez
    LEAOS, Faculty of Medicine/Clinical Hospital, Universidad de Chile, Santiago, Chile
    Institut Clínic d’Oftalmología, Hospital Clinic de Barcelona, Barcelona, Catalunya, Spain
  • Rodrigo A Valenzuela
    Department of Health Sciences, Universidad de Aysen, Coyhaique, Aysén, Chile
    LEAOS, Faculty of Medicine/Clinical Hospital, Universidad de Chile, Santiago, Chile
  • Cristhian A Urzua
    LEAOS, Faculty of Medicine/Clinical Hospital, Universidad de Chile, Santiago, Chile
    Faculty of Medicine, Universidad del Desarrollo, Clinica Alemana de Santiago SA, Vitacura, Metropolitan Region, Chile
  • Footnotes
    Commercial Relationships   Felipe Salgado None; LORETO CUITINO None; Carla Vairetti None; Fabian Vega-Tapia None; Sofia Valdivieso None; Sebastián Maldonado None; Felipe Rojas None; Alvaro Olate-Perez None; Rodrigo Valenzuela None; Cristhian Urzua None
  • Footnotes
    Support  ANID FONDECYT 1212038, 11191215, 12200007, 1200221; ANID PIA/BASAL AFB180003, ANID FONDEF IT17I0087
Investigative Ophthalmology & Visual Science June 2023, Vol.64, 326. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Felipe Salgado, LORETO CUITINO, Carla Vairetti, Fabian Vega-Tapia, Sofia Valdivieso, Sebastián Maldonado, Felipe Rojas, Alvaro Olate-Perez, Rodrigo A Valenzuela, Cristhian A Urzua; Combination of infrared reflectance and OCT imaging for age-related macular degeneration screening via deep learning. Invest. Ophthalmol. Vis. Sci. 2023;64(8):326.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

Purpose : Age-related macular degeneration (AMD) is one of the leading causes of visual impairment and blindness. Although there are no treatments to stop the progression of AMD, studies have shown that correcting risk factors can reduce the progression rate, therefore, it is important to provide timely detection of early-stage AMD. Our purpose is to develop and evaluate deep learning (DL) methodologies for the screening of AMD, using optical coherence tomography (OCT) with or without infrared reflectance (IR) images.

Methods : Five different DL methodologies to automatically detect AMD were evaluated. 4,523 IR/OCT image pairs from our own dataset were used for model training and AMD detection, and 46,316 images from the public OCT2017 dataset were used for transfer learning. OCT images were used either alone or with IR images as input for model training. The method trains a multimodal DL architecture and subsequently applies several Explainable Artificial Intelligence (XAI) strategies. The output is twofold: first, the screening result is generated, while a visual justification is also presented by considering three XAI techniques (Grad-CAM, Guided backpropagation, and Guided Grad-CAM). The main performance metrics considered in this study were the area under the ROC curve (AUC), accuracy, and precision.

Results : The DL methodology that combines IR and OCT images achieved an AUC of 0.979 and an accuracy of 0.945 and showed a slightly superior performance than the methodology that used OCT images as a single source, which achieved an AUC of 0.972 and an accuracy of 0.935. Grad-CAM provides a coarse heatmap, highlighting the regions in which the retinal damage can be located. Guided Grad-CAM provides a fine-grained visualization of the different layers of the retina.

Conclusions : The novel artificial intelligence-based approach presented here is a promising tool for screening a massive number of images for AMD with a level of accuracy at least equal to that of ophthalmologists. The proposed visual justification based on Grad-CAM and Guided Grad-CAM provides a diagnosis tool and resembles the procedure followed by an ophthalmologist to reach a diagnosis.

This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.

 

Fig. 1. Diagram for multimodal AMD prediction via convolutional neural networks and XAI techniques.

Fig. 1. Diagram for multimodal AMD prediction via convolutional neural networks and XAI techniques.

 

Table N°1: Test results for each methodology and deep learning classifier in terms of accuracy and area under the ROC curve (AUC).

Table N°1: Test results for each methodology and deep learning classifier in terms of accuracy and area under the ROC curve (AUC).

×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×