A Classification-Guided Segmentation Algorithm Based on Deep Learning for Epithelium Segmentation in...

Area of Science:

• Digital Pathology
• Computational Histopathology
• Medical Image Analysis

Background:

• Epithelium in radicular cysts (RCs) provides crucial diagnostic information for periapical lesions.
• Accurate segmentation of epithelium in whole slide images (WSIs) aids pathologists in diagnosis and treatment planning.
• Class imbalance due to fragmented epithelium poses challenges for automated segmentation methods.

Purpose of the Study:

• To develop an automated, accurate segmentation algorithm for epithelium in histopathological images of RCs.
• To address the class imbalance problem in automated segmentation of fragmented epithelium.
• To improve the efficiency and objectivity of pathological diagnosis in RCs.

Main Methods:

• Proposed a two-stage Classification-Guided Segmentation Algorithm (CGSA).
• The first stage uses a classification network for Region of Interest (ROI) localization.
• The second stage employs a segmentation network guided by the classification results to segment epithelium.

Main Results:

• CGSA demonstrated superior performance compared to a one-stage segmentation method.
• Achieved an overall Dice coefficient of 0.722 and Intersection over Union (IoU) of 0.593 for epithelium segmentation.
• Improved Dice's coefficient by 5.5% and IoU by 5.9% over a standard UNet-based one-stage method.

Conclusions:

• The CGSA framework effectively segments epithelium in RCs, overcoming class imbalance issues.
• This automated approach offers a valuable alternative to labor-intensive and subjective manual annotation.
• The proposed method enhances the speed and precision of histopathological diagnosis in radicular cyst analysis.