A novel framework for automated volume quantification of subretinal fluid in central serous chorioretinopathy
Conventional quantification approaches of subretinal fluid (SRF) not only necessitate time-consuming frame-by-frame processing but also struggle with accurate fluid segmentation within the initial and terminal segments of linear optical coherence tomography (OCT) B-scan images. Therefore, a novel radial scanning OCT image-driven framework consisting of three modules is proposed for SRF volume quantification. The contour extraction module facilitates the extraction of SRF contour information from individual B-scan images through image cropping, cross-sectional annotation, and edge detection. The three-dimensional reconstruction module unifies the spatial relationships of radial OCT B-scan images through spatial alignment, achieving structured representation of cross-sectional contours via contour point serialization and accomplishing SRF reconstruction through subsequent spatial mapping. The volume quantification module develops an adjacency relationship construction technique to compensate for missing data between adjacent scanning frames and, through microelement subdivision and discrete integration approximation, enables volume quantification of SRF. Besides, an interpolation strategy is further designed to explore the feasibility of accurate SRF volume quantification by the framework under sparse scanning conditions. The quantification results of the proposed framework are in close agreement with the gold standard. The absolute error (AE) metric reaches a minimum of 0.0011 mm 3 , while the relative error (RE) metric ranges from 0.16 to 7.82%, with the majority concentrated around 6% and the lowest value recorded as low as 0.16%. Moreover, with the interpolation strategy’s sampling factor set to 2, the framework achieves a minimum mean absolute percentage error (MAPE) of 0.30%. These extensive experiments demonstrate the significant advantage and competitive strength of the proposed framework for SRF volume quantification, while also suggesting that the interpolation strategy possesses substantial potential to maintain the framework’s quantification accuracy even with a limited number of radial OCT B-scan images.
Authors
- 徐鉴
- Fen Zhou (ORCID: https://orcid.org/0000-0003-2357-4819)
- Sukun Tian (ORCID: https://orcid.org/0000-0001-8289-5968)
- 智鹏 颜
- Jianguo Xu
- Jianxin Shen
- Yongchi Liu
Institutions
- Stomatology Hospital (CN)
- Second Affiliated Hospital of Nanjing Medical University (CN)
- Nanjing Medical University (CN)
- Nanjing University of Aeronautics and Astronautics (CN)
Publication Details
- Journal
- Scientific Reports
- Published
- 2026-11-21
- DOI
- https://doi.org/10.1038/s41598-026-78874-8
- Primary Topic
- Retinal Imaging and Analysis
- Type
- article
- Field-Weighted Citation Impact
- 0.00