[Paper Review] Polyp Segmentation in Colonoscopy Images Using Fully Convolutional Network

Mojtaba Akbari, Majid Mohrekesh+5 more|arXiv (Cornell University)|Feb 1, 2018

Colorectal Cancer Screening and Detection135 citations

TL;DR

This paper proposes a CNN-based polyp segmentation method for colonoscopy images, featuring a novel training patch selection strategy and post-processing of the probability map to improve segmentation accuracy on the CVC-ColonDB dataset.

ABSTRACT

Colorectal cancer is a one of the highest causes of cancer-related death, especially in men. Polyps are one of the main causes of colorectal cancer and early diagnosis of polyps by colonoscopy could result in successful treatment. Diagnosis of polyps in colonoscopy videos is a challenging task due to variations in the size and shape of polyps. In this paper we proposed a polyp segmentation method based on convolutional neural network. Performance of the method is enhanced by two strategies. First, we perform a novel image patch selection method in the training phase of the network. Second, in the test phase, we perform an effective post processing on the probability map that is produced by the network. Evaluation of the proposed method using the CVC-ColonDB database shows that our proposed method achieves more accurate results in comparison with previous colonoscopy video-segmentation methods.

Motivation & Objective

• Motivate accurate polyp segmentation in colonoscopy to aid early detection of colorectal cancer.
• Develop a convolutional neural network approach tailored for polyp segmentation in colonoscopy images.
• Introduce a novel training patch selection method to improve learning.
• Apply post-processing on network probability maps to enhance final segmentation.
• Evaluate the method against prior colonoscopy video-segmentation techniques on a standard dataset.

Proposed method

• Use a convolutional neural network for polyp segmentation in colonoscopy images.
• Propose a novel image patch selection method during training to improve learning efficiency and performance.
• Apply effective post-processing to the network-produced probability map during testing to refine segmentation.
• Evaluate the approach on the CVC-ColonDB database and compare with previous methods.

Experimental results

Research questions

• RQ1 Can a CNN-based framework achieve higher segmentation accuracy for polyps in colonoscopy images than prior methods?
• RQ2 Does targeted image patch selection during training improve segmentation performance?
• RQ3 Does post-processing of the probability map at test time improve final segmentation results?

Key findings

• The proposed method achieves more accurate results compared with previous colonoscopy video-segmentation methods on CVC-ColonDB.
• Training with a novel image patch selection strategy enhances learning effectiveness.
• Post-processing of the network probability map during testing contributes to improved segmentation outcomes.