Direct Estimation of Choroidal Thickness in Optical Coherence Tomography Images with Convolutional Neural Networks

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate choroidal thickness estimation is crucial for diagnosing eye diseases using computer-aided systems.
• Current segmentation-based methods for choroidal thickness estimation from OCT images are highly dependent on segmentation accuracy.

Purpose of the Study:

• To propose a novel, direct method for estimating choroidal thickness using CNNs, eliminating the need for explicit choroidal segmentation.
• To evaluate the performance of the proposed direct estimation method against traditional segmentation-based approaches.

Main Methods:

• A direct method utilizing convolutional neural networks (CNNs) was developed for choroidal thickness estimation.
• OCT B-scan images were processed by a trained CNN model to estimate thickness on cropped patches.
• The mean choroidal thickness was calculated by averaging the estimations from individual patches.

Main Results:

• The proposed direct CNN-based method demonstrated highly competitive results compared to established segmentation-based methods.
• The experimental evaluation on 150 OCT volumes confirmed the efficacy and robustness of the direct estimation technique.
• The findings suggest that direct choroidal thickness estimation is a promising alternative to segmentation-dependent methods.

Conclusions:

• Direct estimation of choroidal thickness using CNNs is a viable and promising approach for computer-aided eye disease diagnosis.
• This method overcomes the limitations of segmentation accuracy in traditional approaches.
• The CNN-based direct estimation shows potential for improving the efficiency and reliability of ophthalmic diagnostics.