Machine learning based vesselness measurement for coronary artery segmentation in cardiac CT volumes

Abstract

Automatic coronary centerline extraction and lumen segmentation facilitate the diagnosis of coronary artery disease (CAD), which is a leading cause of death in developed countries. Various coronary centerline extraction methods have been proposed and most of them are based on shortest path computation given one or two end points on the artery. The major variation of the shortest path based approaches is in the different vesselness measurements used for the path cost. An empirically designed measurement (e.g., the widely used Hessian vesselness) is by no means optimal in the use of image context information. In this paper, a machine learning based vesselness is proposed by exploiting the rich domain specific knowledge embedded in an expert-annotated dataset. For each voxel, we extract a set of geometric and image features. The probabilistic boosting tree (PBT) is then used to train a classifier, which assigns a high score to voxels inside the artery and a low score to those outside. The detection score can be treated as a vesselness measurement in the computation of the shortest path. Since the detection score measures the probability of a voxel to be inside the vessel lumen, it can also be used for the coronary lumen segmentation. To speed up the computation, we perform classification only for voxels around the heart surface, which is achieved by automatically segmenting the whole heart from the 3D volume in a preprocessing step. An efficient voxel-wise classification strategy is used to further improve the speed. Experiments demonstrate that the proposed learning based vesselness outperforms the conventional Hessian vesselness in both speed and accuracy. On average, it only takes approximately 2.3 seconds to process a large volume with a typical size of 512×512×200 voxels.

Robust and Fast Contrast Inflow Detection for 2D X-ray Fluoroscopy

“…On the other hand, because of preferable low dose radiation in clinics, false negatives on smaller vessels can hardly be avoided. It has been shown that [12] learning based vesselness measurement not only can achieve better performance than conventional filtering based approach [4] but also can be more computationally efficient. For this reason, a learning-based vesselness measurement is applied in our framework.…”

“…For this reason, a learning-based vesselness measurement is applied in our framework. While [12] tried to learn the vesselness in 3D CTA data, we train a learning-based classifier to measure the vesselness for a fluoroscopic image.…”

Automatic Heart Isolation in 3D CT Images

“…Here, we use a machine-learning based vesselness protection algorithm. As described in [14], the idea is to train a voxel classifier based on image context to tell the probability of the voxel being in a vessel. This algorithm is capable of quickly classifying a voxel to be vessel or not by applying a threshold to the returned vesselness probability.…”

Robust and Accurate Coronary Artery Centerline Extraction in CTA by Combining Model-Driven and Data-Driven Approaches

“…The generated ROI mesh is tight and it is then expanded a bit (5%) to provide a safety margin. The vessel-specific ROI mesh can be combined with the pericardium based coronary mask [10] to further constrain the search of major coronary centerlines. Fig.…”

“…(3) The first term is the cost for a single node, measuring how likely this point is at the center of the vessel. Here, a machine learning based vesselness [10] is used as the node cost. The second term is the total length of the path by summing the Euclidean distance between two neighboring points on the path.…”