Three-dimensional continuous max flow optimization-based serous retinal detachment segmentation in SD-OCT for central...

Area of Science:

• Ophthalmology
• Medical Imaging
• Computational Biology

Background:

• Serous retinal detachment is crucial for diagnosing central serous chorioretinopathy (CSC).
• Accurate segmentation of neurosensory retinal detachment (NRD) and pigment epithelial detachment (PED) is essential for clinical evaluation.
• Current methods may lack automation and precision in assessing these detachments.

Purpose of the Study:

• To develop an automatic, three-dimensional segmentation method for detecting NRD and PED in spectral domain optical coherence tomography (SD-OCT) images.
• To provide a quantitative tool for evaluating serous retinal detachment in CSC patients.
• To enhance the clinical assessment of central serous chorioretinopathy.

Main Methods:

• Utilized random forest classification to construct a probability map based on structural texture, intensity, and layer thickness.
• Employed a continuous max flow optimization algorithm for segmenting fluid regions associated with retinal detachment.
• Trained and validated the method on 37 SD-OCT volumes from CSC cases.

Main Results:

• Achieved high accuracy for NRD segmentation: 92.1% true positive volume fraction (TPVF), 0.53% false positive volume fraction (FPVF), 94.7% positive predictive value (PPV), and 93.3% dice similarity coefficient (DSC).
• Demonstrated strong performance for PED segmentation: 92.5% TPVF, 0.14% FPVF, 80.9% PPV, and 84.6% DSC.
• The method proved to be an effective automatic tool for assessing serous retinal detachment.

Conclusions:

• The proposed 3D automatic segmentation method accurately detects NRD and PED in SD-OCT images.
• This tool has significant potential to improve the clinical evaluation and management of central serous chorioretinopathy.
• The high performance metrics suggest clinical applicability for automated retinal detachment assessment.