advertisement

Topcon

Editors Selection IGR 24-1

Artificial intelligence: Detecting glaucoma progression with AI

Sasan Moghimi

Comment by Sasan Moghimi on:

106413 Deep Learning-Assisted Detection of Glaucoma Progression in Spectral-Domain OCT, Mariottoni EB; Mariottoni EB; Datta S; Shigueoka LS et al., Ophthalmology. Glaucoma, 2023; 6: 228-238


Find related abstracts


With the introduction of spectral-domain optical coherence tomography (SDOCT), objective and reproducible structural measurements became feasible. However, there is still no consensus on how to determine the presence of glaucoma progression. Trend-based analysis may be insensitive to small, localized changes and event-based algorithms do not take into account the time during which the changes have occurred. In cross-sectional studies artificial intelligence algorithms, such as deep-learning models has been shown to be promising for detection of glaucoma and even sometimes superior to the evaluation of the experts.1-3 However, majority of the prior studies exploring usage of artificial intelligence model for detection of glaucoma progression have primarily focused on visual field.4

In this longitudinal study, Mariottoni and colleagues have develop and validate a deeplearning model for detection of glaucoma progression using SDOCT measurements of retinal nerve fiber layer (RNFL) thickness from 816 eyes of416 individuals of Duke Glaucoma Registry. First, the presence of glaucoma progression was defined by the assessment of two glaucoma specialists using overview of RNFL thickness profiles. DL convolutional neural network was trained to assess SD-OCT RNFL thickness measurements (768 measurements at equally spaced points around the optic nerve) of two visits (a baseline and a follow-up visit) along with time between visits to predict the probability of glaucoma progression.

The DL model significantly outperformed trend-based analyses with an AUC of 0.938, a sensitivity of 87.3% and a specificity of 86.4%. Trend-based analysis using global RNFL thickness showed a sensitivity of only 46.1% and specificity of 92.6%. Interestingly, likelihood ratios for the DL model were associated with large changes in the probability of progression in approximately 74% of the SDOCT tests, indicating that in the vast majority of SD-OCT tests the DL model would provide useful information to clarify the presence of progression.

The main concern of the study is the subjective reference standard they adopted for glaucoma progression, which only utilized the overview of RNFL profile assessed by graders

The main concern of the study is the subjective reference standard they adopted for glaucoma progression, which only utilized the overview of RNFL profile assessed by graders. This approach is prone to variability, and validation with visual field testing, as well as a comparison with other methods, would provide a better perspective on the performance of these progression algorithms. Additionally, as with other DL algorithms, the 'black box' nature of the internal features used in the predictions is not entirely clear. However, the use of innovative visualizations in the study addressed these concerns and showed that the model was able to pinpoint the likely locations of regions of progression using heatmaps. Generalizability of the model across other population and other devices should also kept in our mind. It is an important aspect to consider when assessing the applicability and real-world utility of these models.5

The current report is the first study to show an application of deep-learning models to assess progression with SDOCT. The model agreed well with expert judgments and outperformed conventional trend-based analyses of change, while also providing indication of the likely locations of change. While the proposed method validate a model to assess event-based analysis of progression, future studies needed to consider rate of change and predict trend-based progression and also the generalizability of artificial models for glaucoma progression detection.

References

  1. Asaoka R, Murata H, Hirasawa K, et al. Using Deep Learning and Transfer Learning to Accurately Diagnose Early-Onset Glaucoma From Macular Optical Coherence Tomography Images. Am J Ophthalmol. 2019;198:136-145.
  2. Kamalipour A, Moghimi S, Khosravi P, et al. Deep Learning Estimation of 10-2 Visual Field Map Based on Circumpapillary Retinal Nerve Fiber Layer Thickness Measurements. Am J Ophthalmol. 2023;246:163-173.
  3. Phene S, Dunn RC, Hammel N, et al. Deep Learning and Glaucoma Specialists: The Relative Importance of Optic Disc Features to Predict Glaucoma Referral in Fundus Photographs. Ophthalmology. 2019;126(12):1627-1639.
  4. Wang M, Shen LQ, Pasquale LR, et al. An Artificial Intelligence Approach to Detect Visual Field Progression in Glaucoma Based on Spatial Pattern Analysis. Invest Ophthalmol Vis Sci. 2019;60(1):365-375.
  5. Fan R, Alipour K, Bowd C, et al. Detecting Glaucoma from Fundus Photographs Using Deep Learning without Convolutions: Transformer for Improved Generalization. Ophthalmology Science. 2023;3(1):100233.


Comments

The comment section on the IGR website is restricted to WGA#One members only. Please log-in through your WGA#One account to continue.

Log-in through WGA#One

Issue 24-1

Change Issue


advertisement

Oculus