Efficient Vessel Feature Detection for Endoscopic Image Analysis

Abstract

Distinctive feature detection is an essential task in computer-assisted minimally invasive surgery (MIS). For special conditions in an MIS imaging environment, such as specular reflections and texture homogeneous areas, the feature points extracted by general feature point detectors are less distinctive and repeatable in MIS images. We observe that abundant blood vessels are available on tissue surfaces and can be extracted as a new set of image features. In this paper, two types of blood vessel features are proposed for endoscopic images: branching points and branching segments. Two novel methods, ridgeness-based circle test and ridgeness-based branching segment detection are presented to extract branching points and branching segments, respectively. Extensive in vivo experiments were conducted to evaluate the performance of the proposed methods and compare them with the state-of-the-art methods. The numerical results verify that, in MIS images, the blood vessel features can produce a large number of points.More importantly, those points are more robust and repeatable than the other types of feature points. In addition, due to the difference in feature types, vessel features can be combined with other general features, which makes them new tools for MIS image analysis. These proposed methods are efficient and the code and datasets are made available to the public.

Video‐based 3D reconstruction, laparoscope localization and deformation recovery for abdominal minimally invasive surgery: a survey

“…Bartoli provided a uterus dataset, which contains tissue deformation caused by instrument interactions. In an image dataset was collected for evaluation of the repeatability of feature detectors. The dataset in contains hundreds of images sampled from in vivo videos taken during colon surgeries; the images in this dataset were taken at different viewpoints, and the ground truth homography mappings are available.…”

“…In an image dataset was collected for evaluation of the repeatability of feature detectors. The dataset in contains hundreds of images sampled from in vivo videos taken during colon surgeries; the images in this dataset were taken at different viewpoints, and the ground truth homography mappings are available. Puerto‐Souza and Mariottini provided the hierarchical multi‐affine (HMA) feature matching toolbox for MIS images, which contains 100 image pairs representing various surgical scenes, such as instrument occlusion, fast camera motion and organ deformation.…”

“…In the third category, feature detectors exploit the second derivatives of raw images to detect interest points defined by blobs and ridges. Most methods in this category are based on analysis of the Hessian matrix . In the Hessian affine detector , the determinants of the Hessian matrices were calculated for all pixels, and the local maxima were selected as feature points.…”

“…The speeded‐up robust features (SURF) feature detector replaced the Gaussian filters with box filters to obtain a faster speed . It has been reported that general feature point detectors do not perform well in MIS images . Lin et al .…”

Computer-Aided Endoscopic Diagnosis Without Human-Specific Labeling

“…The commonly used classification models are support vector machine (SVM) [17], [18], neural network [19], fuzzy logic principles [20], clustering-based methods [21], and filter-based methods [22]. For supportive systems, they usually provide some support to guide and make a diagnosis easier for clinicians, such as enhancing image quality [23], detecting informative frames [24], pose detection for endoscopy [25], WCE color video segmentation [26], and feature detection [27].…”