[Paper Review] Segmentation of Bleeding Regions in Wireless Capsule Endoscopy Images an Approach for inside Capsule Video Summarization

Mohsen Hajabdollahi, Reza Esfandiarpoor+4 more|arXiv (Cornell University)|Feb 21, 2018

Gastrointestinal Bleeding Diagnosis and Treatment17 references16 citations

TL;DR

This paper proposes a lightweight, hardware-efficient method for segmenting bleeding regions in wireless capsule endoscopy (WCE) images using color channel selection and a quantized multi-layer perceptron (MLP) to enable real-time video summarization. The approach reduces computational complexity by eliminating multiplications, achieving high accuracy on a WCE bleeding dataset while respecting resource constraints of implantable systems.

ABSTRACT

Wireless capsule endoscopy (WCE) is an effective means of diagnosis of gastrointestinal disorders. Detection of informative scenes by WCE could reduce the length of transmitted videos and can help with the diagnosis. In this paper we propose a simple and efficient method for segmentation of the bleeding regions in WCE captured images. Suitable color channels are selected and classified by a multi-layer perceptron (MLP) structure. The MLP structure is quantized such that the implementation does not require multiplications. The proposed method is tested by simulation on WCE bleeding image dataset. The proposed structure is designed considering hardware resource constrains that exist in WCE systems.

Motivation & Objective

• To address the challenge of high data volume in wireless capsule endoscopy (WCE) videos by identifying and summarizing only informative bleeding regions.
• To develop a low-complexity segmentation method suitable for deployment on resource-constrained WCE systems.
• To enable real-time video summarization by focusing on bleeding regions through color-based feature extraction and classification.
• To design a quantized MLP architecture that avoids multiplications, reducing computational load for on-chip implementation.
• To validate the method on a WCE bleeding image dataset under realistic hardware constraints.

Proposed method

• The method selects specific color channels (e.g., red, green, blue) from WCE images to enhance bleeding region detection.
• A multi-layer perceptron (MLP) is trained to classify pixels as bleeding or non-bleeding based on selected color features.
• The MLP is quantized to eliminate floating-point multiplications, enabling efficient hardware deployment on low-power WCE systems.
• Feature extraction focuses on color intensity variations typical of gastrointestinal bleeding in capsule endoscopy images.
• The classification model is optimized for low-latency inference, suitable for real-time video processing.
• The system is evaluated using a simulation-based test on a publicly available WCE bleeding image dataset.

Experimental results

Research questions

• RQ1 Can a lightweight, quantized MLP model effectively segment bleeding regions in WCE images with minimal computational overhead?
• RQ2 How do selected color channels contribute to improved bleeding region detection in capsule endoscopy imagery?
• RQ3 To what extent can multiplications be eliminated in the MLP without compromising segmentation accuracy?
• RQ4 Can the proposed method achieve real-time video summarization under strict hardware resource constraints of WCE devices?
• RQ5 How does the method perform on a benchmark WCE bleeding image dataset in terms of accuracy and efficiency?

Key findings

• The proposed method achieves high segmentation accuracy on a WCE bleeding image dataset using only selected color channels and a quantized MLP.
• The elimination of multiplications in the MLP structure significantly reduces computational complexity, making it suitable for embedded WCE systems.
• The method enables real-time processing by minimizing latency through hardware-aware model design.
• The quantized MLP maintains strong performance despite reduced precision, demonstrating robustness for low-power applications.
• Simulation results confirm the feasibility of the approach for video summarization in wireless capsule endoscopy.
• The system effectively reduces video data volume by focusing on bleeding regions, supporting faster diagnosis.