Automated Classification and Segmentation in Colorectal Images Based on Self-Paced Transfer Network

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Gastroenterology

Background:

• Manual diagnosis of colorectal diseases is time-consuming and labor-intensive.
• Accurate segmentation and classification of colorectal polyps are crucial for timely surgical resection.
• Challenges exist due to unbalanced labels and difficult colorectal image data.

Purpose of the Study:

• To develop an automated method for colorectal disease classification and segmentation.
• To improve diagnostic accuracy and efficiency for colonoscopists.
• To assist in identifying polyps requiring surgical intervention.

Main Methods:

• A self-paced transfer VGG network (STVGG) was proposed for classification, utilizing ImageNet pretraining and self-paced learning.
• Features from the trained STVGG model were shared with a U-Net segmentation network to avoid redundant learning.
• The method was trained and evaluated on 3061 colorectal images.

Main Results:

• The proposed method achieved 96% classification accuracy.
• Demonstrated superior segmentation performance compared to other methods.
• Accurately segmented polyps from surrounding tissues.

Conclusions:

• The developed deep learning approach significantly enhances colorectal disease classification and segmentation.
• The method shows potential for assisting colonoscopists in early polyp identification and guiding resection decisions.
• Automated analysis can improve the efficiency and accuracy of colorectal cancer screening.