Automated OCT angiography image quality assessment using a deep learning algorithm

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Optical coherence tomography angiography (OCTA) is a crucial imaging technique in ophthalmology.
• Standardized assessment of OCTA image quality is essential for reliable diagnostic interpretation.
• Current manual image quality assessment can be time-consuming and subjective.

Purpose of the Study:

• To develop and validate a deep learning algorithm (DLA) for automated image quality assessment in OCTA.
• To expedite and standardize the evaluation of OCTA image quality.

Main Methods:

• A deep convolutional neural network (DCNN) was trained and validated on 160 en-face macular OCTA scans.
• Images were retrospectively classified as sufficient or insufficient quality based on artifact and segmentation scores.
• The DLA was tested on 40 untrained OCTA images to evaluate its classification performance.

Main Results:

• The DLA achieved high accuracy in classifying OCTA images, with 90% sensitivity, 90% specificity, and 90% overall accuracy.
• The algorithm demonstrated strong performance with a validation accuracy of 100% and a training accuracy of 97%.
• The DLA effectively discriminated between sufficient and insufficient image quality (p < 0.001).

Conclusions:

• Deep learning offers a promising, automated approach for distinguishing between sufficient and insufficient OCTA image quality.
• This DLA has the potential to contribute to the establishment of standardized image quality criteria for OCTA.
• Automated quality assessment can enhance the efficiency and reliability of OCTA interpretation.