Three-dimensional diabetic macular edema thickness maps based on fluid segmentation and fovea detection using deep...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic macular edema (DME) diagnosis traditionally relies on central retinal thickness (CRT).
• CRT may not fully capture the complex morphometry and fluid distribution in DME.
• Accurate quantification is crucial for effective patient management and follow-up.

Purpose of the Study:

• To develop and apply a novel 3D morphometry method for DME diagnosis.
• To go beyond the limitations of central retinal thickness (CRT) in quantifying DME.
• To utilize deep convolution neural networks (DCNNs) for enhanced DME analysis and patient follow-up.

Main Methods:

• Collected Optical Coherence Tomography (OCT) scans from 229 eyes of 160 patients.
• Utilized DCNNs (U-Net, sASPP, HRNetV2-W48, HRNetV2-W48+OCR) for segmentation of cystoid macular edema (CME), subretinal fluid (SRF), and fovea detection.
• Generated 3D retinal thickness maps for CME, SRF, and overall retina, divided by the Early Treatment Diabetic Retinopathy Study (ETDRS) grid.

Main Results:

• Achieved high dice similarity coefficients (DSC) for fluid segmentation: 0.78 (CME) and 0.82 (SRF).
• Demonstrated accurate fovea detection with an average deviation of 145.7±117.8 µm.
• 3D thickness maps identified center-involved DME and fluid location (above/below fovea) missed by CRT measurements.

Conclusions:

• 3D macular edema thickness maps provide more intuitive morphometry and detailed statistics than CRT.
• This advanced method supports more accurate diagnoses of DME.
• The 3D maps enhance the follow-up care for patients with DME.