Detection of Polyps in Colonoscopic Videos Using Saliency Map-Based Modified Particle Filter

In this article, an automatic polyp detection system for endoscopic video frames is proposed. Manual inspection of each frame for polyp localization in the colonoscopic video has many adversaries. This work proposes a real-time tracking framework for polyp region segmentation in hugely acquired colonoscopic video frames. In our work, the polyp region in the frame is roughly detected by a saliency map at first, followed by a modified tracking mechanism for localization. The work suggests the use of a visual saliency map as the measurement model for tracking. The saliency map is composed of four probability maps generated by incorporating the characteristics associated with the polyps. The elliptical shape of the polyps is used by the particles for final refinement using an active contour (AC) model. The tracking efficiency and the segmentation score achieved using the proposed method suggest that our method can be used for polyp detection and localization. The proposed method achieves an average dice score of 66.06% in the CVC clinic Database. Our method can be employed in both online as well as off-line endoscopic video sequences. A GUI is also designed using the proposed method as an automatic polyp detection system.

Automated summary

This article introduces an automatic polyp detection system for endoscopic video frames which utilizes saliency maps and tracking mechanisms for localization. The method achieves high tracking efficiency and segmentation scores, proving to be effective for polyp detection and localization, with promising results in the CVC clinic Database.