June 2021
Volume 62, Issue 8
Open Access
ARVO Annual Meeting Abstract  |   June 2021
Surgical Phase Identification and Pupil Tracking in Phacoemulsification Cataract Surgery as a Foundation for Surgical Guidance
Author Affiliations & Notes
  • Rogerio Garcia Nespolo
    Bioengineering, University of Illinois at Chicago, Chicago, Illinois, United States
    Ophthalmology, University of Illinois at Chicago College of Medicine, Chicago, Illinois, United States
  • Emily Cole
    Ophthalmology, University of Illinois at Chicago College of Medicine, Chicago, Illinois, United States
  • Nita Valikodath
    Ophthalmology, University of Illinois at Chicago College of Medicine, Chicago, Illinois, United States
  • Cristian Luciano
    Bioengineering, University of Illinois at Chicago, Chicago, Illinois, United States
  • Yannek Leiderman
    Ophthalmology, University of Illinois at Chicago College of Medicine, Chicago, Illinois, United States
    Bioengineering, University of Illinois at Chicago, Chicago, Illinois, United States
  • Footnotes
    Commercial Relationships   Rogerio Garcia Nespolo, None; Emily Cole, None; Nita Valikodath, None; Cristian Luciano, None; Yannek Leiderman, None
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science June 2021, Vol.62, 2165. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Rogerio Garcia Nespolo, Emily Cole, Nita Valikodath, Cristian Luciano, Yannek Leiderman; Surgical Phase Identification and Pupil Tracking in Phacoemulsification Cataract Surgery as a Foundation for Surgical Guidance. Invest. Ophthalmol. Vis. Sci. 2021;62(8):2165.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : As a foundation to develop tools for surgical guidance, we sought to develop an agent capable of autonomously identifying the various steps and phases of phacoemulsification cataract surgery in real-time together with pupil segmentation, such that the generated output is capable of informing surgical decision making.

Methods : Heterogeneous videos were annotated by ophthalmic surgeons in order to achieve robustness of pupil detection, phase identification and tissue segmentation by the algorithm.
The application acquires video frames, in real time, from a surgical microscope-based video capture device. Then, a Region Based Convolutional Neural Network (R-CNN) performs the following functions for each analyzed frame (figure 1):
I) pupil location and area
ii) surgical phase identification according to the instruments in use

Results : 1. We evaluated the performance of the R-CNN via comparison with annotation of surgical videos performed by ophthalmic surgeons (Table 1). We achieved high values in accuracy, precision, and sensitivity across each of the four phases (idle, capsulorhexis, phacoemulsification and cortex removal), leading to F1-scores above 90%.
2. There was also strong correlation among the graders’ assessment of the size of the pupil with the pupil area detected by the algorithm, yielding precision, sensitivity, and intersection over union area (IoU) of 82.07%, 87.19%, and 95.14%, respectively.
3. The algorithm executed these tasks at an average processing speed of 82±20 frames per second (FPS), well above the output of of 60 FPS at which most contemporary surgical visualization systems display images.

Conclusions : It is important to state that no current machine learning solution combines phase identification with pupil tracking. We have developed a platform that provides the foundations for a real-time surgical guidance tool for phacoemulsification cataract surgery by using object detection and classification for surgical phase classification. Future machine learning-based tools can utilize these capabilities for the creation of novel surgical guidance tools and feedback mechanisms.

This is a 2021 ARVO Annual Meeting abstract.

 

Figure 1 - Pipeline for surgical phase recognition and pupil detection. The R-CNN retrieves the position of the pupil together with the classified phase.

Figure 1 - Pipeline for surgical phase recognition and pupil detection. The R-CNN retrieves the position of the pupil together with the classified phase.

 

Table 1 - Accuracy, precision and sensitivity for the phases being recognized during the video streaming of the test dataset.

Table 1 - Accuracy, precision and sensitivity for the phases being recognized during the video streaming of the test dataset.

×
×

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.

×