July 2019
Volume 60, Issue 9
Free
ARVO Annual Meeting Abstract  |   July 2019
Machine Learning for the automated interpretation of Optical Coherence Tomography Angiography for Age-related Macular Degeneration
Author Affiliations & Notes
  • Konstantinos Balaskas
    Moorfields Eye Hospital, London, London, ENGLAND, United Kingdom
  • Abdullah Alfahaid
    Moorfields Eye Hospital, London, London, ENGLAND, United Kingdom
  • Hagar Khalid
    Moorfields Eye Hospital, London, London, ENGLAND, United Kingdom
  • Panagiotis Sergouniotis
    Manchester Royal Eye Hospital, United Kingdom
  • NIKOLAS PONTIKOS
    Moorfields Eye Hospital, London, London, ENGLAND, United Kingdom
  • Pearse Andrew Keane
    Moorfields Eye Hospital, London, London, ENGLAND, United Kingdom
  • Footnotes
    Commercial Relationships   Konstantinos Balaskas, Heidelberg (R), TopCon (R); Abdullah Alfahaid, None; Hagar Khalid, None; Panagiotis Sergouniotis, None; NIKOLAS PONTIKOS, None; Pearse Keane, Allergan (R), Bayer (R), Haag-Streitt (R), Heidelberg (R), Novartis (R), TopCon (R), Zeiss (R)
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science July 2019, Vol.60, 3095. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Konstantinos Balaskas, Abdullah Alfahaid, Hagar Khalid, Panagiotis Sergouniotis, NIKOLAS PONTIKOS, Pearse Andrew Keane; Machine Learning for the automated interpretation of Optical Coherence Tomography Angiography for Age-related Macular Degeneration. Invest. Ophthalmol. Vis. Sci. 2019;60(9):3095.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : This report summarises the findings of experiments conducted on Moorfields and Manchester datasets using our new Hybrid Machine Learning Algorithm named “Hybrid Machine Learning Approach Using LBP Descriptor and PCA for Age-Related Macular Degeneration (AMD) Classification in Optical Coherence Tomography Angiography (OCTA) Images”. Two classifiers were tested; K-Nearest Neighbour (KNN) and Support Vector Machine (SVM). The aim of the report is to demonstate the perfomance of the algorithm in distinguidhing between normal and abnormal OCTAs and those with Dry as opposed to Wet AMD.

Methods : In our approach, before passing the feature vectors (LBP values) generated from the OCTA images for classification, the Principal Component Analysis (PCA) is applied. The PCA is a statistical procedure that converts a high number of possibly correlated feature vectors into a lower number of uncorrelated feature vectors. Therefore, PCA is more likely to increase the chance of having multiple correlated features and
redundant features. Two classifiers were tested namely K-Nearest Neighbour (KNN) and Support Vector Machine (SVM).

Results : Our two algorithms were evaluated based on their ability to distinguish between the various classes. The Manchester dataset involves 23 Healthy and 23 Wet AMD eyes, while the Moorfields includes 166 Wet AMD
and 79 Dry AMD eyes. The classification was performed on all classes available as a binary classification problem, namely Healthy vs Wet AMD and Wet AMD vs Dry AMD. We used the Receiver Operating Curve to test algorithm performance. The following results were obtained:
Healthy vs Wet AMD: Choriocapillaris 100% ± 0 Outer 99% ± 1 Deep 98% ± 3 Superficial 95% ± 5 All layers: 96% ± 2
Wet AMD vs Dry AMD: Choriocapillaris 83% ± 2 Outer 85% ± 3 Deep 80% ± 4 Superficial 75% ± 3 All layers 78% ± 4

Conclusions : In conclusion, the reported Machine Learning algorithm shows good performance and resilience in performing classification tasks on OCTA images, specifically to distinguish between normal and Wet AMD scans and well as between scans from Dry and Wet AMD cases. Further iterations of the algorith will furhter enhance its performance.

This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.

×
×

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.

×