Active-Supervised Model for Intestinal Ulcers Segmentation Using Fuzzy Labeling.

Area of Science:

• Medical imaging analysis
• Artificial intelligence in gastroenterology
• Computational pathology

Background:

• Inflammatory bowel disease (IBD) diagnosis relies on colonoscopy, but current image analysis is subjective and inefficient.
• Accurate intestinal ulcer segmentation is crucial for IBD diagnosis but faces challenges like noisy labels and dataset variability.
• Existing methods struggle with the complexity and subjectivity of manual image scoring, impacting diagnostic accuracy.

Purpose of the Study:

• To develop advanced machine learning techniques for precise intestinal ulcer segmentation in IBD diagnosis.
• To address the challenges of labeling noise and performance variability in medical image segmentation models.
• To enhance the accuracy and efficiency of IBD diagnosis through multi-category ulcer segmentation.

Main Methods:

• Proposed an active ulcer segmentation algorithm utilizing fuzzy labeling and collaborative training.
• Implemented a segmentation model that leverages pixel-wise confidence from fuzzy labels for robustness against noisy annotations.
• Introduced a data adaptation strategy with active learning to select informative samples, improving model adaptability across diverse datasets.

Main Results:

• The collaborative training model demonstrated enhanced robustness to noisy labels through network cooperation.
• The active learning strategy significantly improved model performance and adaptability across different public and hospital datasets.
• Experimental validation confirmed the effectiveness of the proposed methods in handling segmentation challenges.

Conclusions:

• The developed machine learning approach offers a more precise and efficient method for IBD diagnosis.
• Addressing noisy labels and dataset variability is key to improving automated medical image analysis for IBD.
• This research paves the way for improved diagnostic tools in gastroenterology, aiding in early and accurate IBD detection.