Modified DeeplabV3+ with multi-level context attention mechanism for colonoscopy polyp segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Gastroenterology

Background:

• Colon cancer is a leading cause of death worldwide, necessitating improved diagnostic methods.
• Manual analysis of colonoscopy images is time-consuming and prone to error.
• Automated computer-aided techniques are crucial for efficient and reliable diagnosis of gastrointestinal disorders.

Purpose of the Study:

• To develop an automated framework for diagnosing colonoscopy diseases using medical image analysis.
• To improve the accuracy and efficiency of polyp detection and segmentation in colonoscopy images.
• To create a model capable of automatically distinguishing polyps from other image features.

Main Methods:

• A modified DeeplabV3+ model was developed for polyp segmentation in colonoscopy images.
• The model's encoder utilized a pre-trained dilated convolutional residual network for enhanced feature extraction.
• The modified model's performance was evaluated against state-of-the-art segmentation methods on public datasets.

Main Results:

• The modified DeeplabV3+ model achieved high segmentation accuracy, with Dice similarity coefficients of 0.97 on the CVC-Clinic DB dataset and 0.95 on the Kvasir-SEG dataset.
• The proposed model demonstrated improved segmentation efficiency and effectiveness in both software and hardware implementations.
• Minor modifications to the DeeplabV3+ architecture led to significant performance gains.

Conclusions:

• The developed framework offers a robust solution for automated polyp segmentation in colonoscopy.
• The enhanced DeeplabV3+ model can assist endoscopists in reducing the risk of polyps progressing to cancer.
• This computer-aided approach shows promise for widespread clinical adoption in diagnosing gastrointestinal illnesses.