ROP lesion segmentation via sequence coding and block balancing.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Retinopathy of prematurity (ROP) is a significant cause of vision loss in premature infants.
• Accurate detection and segmentation of ROP lesions are critical for diagnosis and treatment.
• Subtle and small ROP lesions present diagnostic challenges for both human experts and automated systems.

Purpose of the Study:

• To develop and evaluate a novel deep learning model for improved segmentation of retinopathy of prematurity lesions.
• To address the challenges posed by subtle and small ROP lesions in automated diagnostic systems.

Main Methods:

• Introduction of the Sequence encoding and Block balancing-based Segmentation Network (SeBSNet).
• Integration of domain knowledge coding, sequence coding learning (SCL), and block-weighted balancing (BWB) techniques.
• Utilizing SeBSNet for the segmentation of retinopathy of prematurity lesions.

Main Results:

• SeBSNet achieved superior performance in ROP lesion segmentation compared to state-of-the-art methods.
• Achieved average ROC_AUC of 98.84%, PR_AUC of 71.90%, and Dice score of 66.88%.
• Incorporating SeBSNet techniques into ROP classification networks significantly enhanced classification performance.

Conclusions:

• SeBSNet offers a robust and effective solution for the automated segmentation of ROP lesions.
• The proposed model holds promise for improving the diagnosis and management of retinopathy of prematurity.
• The developed techniques can enhance both segmentation and classification tasks in ROP analysis.