An Automatic Image Processing Method Based on Artificial Intelligence for Locating the Key Boundary Points in the...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate measurement of central serous chorioretinopathy (CSCR) lesion diameter is crucial for assessing disease severity and treatment effectiveness.
• Current manual measurement methods using optical coherence tomography (OCT) B-scan images are unreliable, inefficient, and subjective.

Purpose of the Study:

• To develop an automated artificial intelligence (AI) based image processing framework for precise measurement of CSCR lesion areas.
• To improve the accuracy and efficiency of CSCR lesion diameter measurement by overcoming the limitations of manual methods.

Main Methods:

• A joint AI framework combining an initial location module (ILM) with multitask learning and a single agent reinforcement learning module (SARLM).
• ILM provides preliminary boundary point locations, while SARLM refines accuracy using a Markov decision process.
• The framework integrates ILM's initial guidance with SARLM's exploration for enhanced generalization and accuracy.

Main Results:

• The AI-based joint framework demonstrated effectiveness in rapidly and accurately measuring CSCR lesion diameters.
• The combination of multitask learning and reinforcement learning improved location accuracy and efficiency.
• The method alleviates time-consuming issues associated with SARLM by enabling agents to work in localized regions.

Conclusions:

• The proposed AI framework offers a valuable tool for accurate and efficient CSCR lesion diameter measurement.
• This automated approach has significant clinical application value for managing CSCR.
• The integration of ILM and SARLM represents an innovative solution for objective and repeatable CSCR assessment.