Channel Fitting Network for Retinal Lesion Segmentation from OCT Images.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Retinal lesions are a primary cause of age-related macular degeneration (AMD), significantly impacting the elderly population.
• Accurate detection and segmentation of retinal lesions are crucial for early AMD diagnosis and disease progression monitoring.
• Current segmentation models struggle with the diverse imaging characteristics of different AMD subtypes.

Purpose of the Study:

• To develop an enhanced deep neural network (DNN) model for precise retinal lesion segmentation.
• To improve the accuracy of retinal lesion segmentation by leveraging feature maps from a classification model.
• To address the challenge of segmenting retinal lesions with varied imaging features.

Main Methods:

• Proposed a novel channel fitting module to refine feature maps for the segmentation model.
• Utilized a classification model to guide the feature enhancement process, filtering irrelevant regions.
• Implemented five-fold cross-validation on a dataset of 2633 retinal images from 164 patients.

Main Results:

• The channel fitting module effectively enhanced feature maps, leading to more accurate segmentation.
• The proposed model demonstrated superior performance in segmenting retinal lesions compared to existing methods.
• Feature maps guided by the classification model proved more effective than those from segmentation models alone.

Conclusions:

• The proposed channel fitting module significantly improves the accuracy of retinal lesion segmentation.
• This approach offers a promising solution for the early diagnosis and monitoring of age-related macular degeneration.
• The integration of classification-guided features enhances the robustness of segmentation models for complex retinal pathologies.