Deep learning segmentation of hyperautofluorescent fleck lesions in Stargardt disease

Abstract

Stargardt disease is one of the most common forms of inherited retinal disease and leads to permanent vision loss. A diagnostic feature of the disease is retinal flecks, which appear hyperautofluorescent in fundus autofluorescence (FAF) imaging. The size and number of these flecks increase with disease progression. Manual segmentation of flecks allows monitoring of disease, but is time-consuming. Herein, we have developed and validated a deep learning approach for segmenting these Stargardt flecks (1750 training and 100 validation FAF patches from 37 eyes with Stargardt disease). Testing was done in 10 separate Stargardt FAF images and we observed a good overall agreement between manual and deep learning in both fleck count and fleck area. Longitudinal data were available in both eyes from 6 patients (average total follow-up time 4.2 years), with both manual and deep learning segmentation performed on all (n = 82) images. Both methods detected a similar upward trend in fleck number and area over time. In conclusion, we demonstrated the feasibility of utilizing deep learning to segment and quantify FAF lesions, laying the foundation for future studies using fleck parameters as a trial endpoint.

Next-Generation Sequencing Applications for Inherited Retinal Diseases

“…For example, ML has been utilised for the detection of fleck lesions in fundus autofluorescence imaging, a characteristic feature of Stargardt disease. The ML approach could accurately identify and quantify fleck lesions, a potential outcome measure for future clinical trials [170]. Similarly, optic disc photography has been used to systematically detect abnormalities of the optic disc [171].…”

Stargardt disease and progress in therapeutic strategies

“…However, automated segmentation of the diseased RPE-Bruch's membrane is unreliable and requires manual adjustment using an "adaptive" approach or automated deep learning segmentation (20,21). Flecks are best visualized on FAF and their growth and life cycle may serve as a biomarker for disease progression where one study has developed deep learning segmentation of flecks in STGD1 (22). FAF imaging can also illustrate early macular RPE atrophy.…”

Stargardt disease: Multimodal imaging: A review

“… 76 Regions affected by flecks have reduced function 77 but this imaging feature alone was not predictive of future sensitivity loss 78 . Nevertheless, changes in hyperautofluorescent lesion area have been proposed as a potential clinical trial endpoint using an automated segmentation algorithm to delineate the lesion boundaries 79,80 . Another key FAF feature is the mapping of regions with decreased autofluorescence (DAF).…”

“… 78 Nevertheless, changes in hyperautofluorescent lesion area have been proposed as a potential clinical trial endpoint using an automated segmentation algorithm to delineate the lesion boundaries. 79 , 80 Another key FAF feature is the mapping of regions with decreased autofluorescence (DAF). Varied levels of darkness, as compared to the absent FAF signal at blood vessels and the optic nerve head, have been used to classify these lesions into definitely and questionably DAF (DDAF, QDAF respectively).…”

“… 76 To date, time and ambiguity of manual boundary segmentation have impeded the widespread adoption of this technique in clinical trials. The development of a deep learning‐based segmentation algorithm by Charng et al 80 may improve the accuracy and efficiency of measuring hyperautofluorescent fleck area.…”