Improved cGAN based linear lesion segmentation in high myopia ICGA images

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Myopia prevalence is rising globally, with linear lesions like lacquer cracks and myopic stretch lines being key indicators in high myopia.
• Indocyanine green angiography (ICGA) is crucial for visualizing these retinal lesions.

Purpose of the Study:

• To develop an automated method for segmenting linear lesions in ICGA images for quantitative high myopia analysis.
• To improve the accuracy and efficiency of diagnosing high myopia.

Main Methods:

• An improved conditional generative adversarial network (cGAN) was proposed, utilizing a partial densely connected network as the generator.
• Feature reuse and time-saving network design were incorporated.
• Dice loss and weighted binary cross-entropy loss were implemented to address data imbalance.

Main Results:

• The proposed cGAN-based method demonstrated superior performance in segmenting linear lesions compared to existing networks.
• The enhanced network effectively handles data imbalance issues common in medical image segmentation.

Conclusions:

• The developed AI method offers a promising tool for accurate and quantitative analysis of high myopia.
• This technology can aid ophthalmologists in diagnosing and managing high myopia more effectively.