Medical Image Segmentation Terminology

Definition

An image processing technique that divides an image into a number of specific regions with unique properties. It is a key step before image analysis.

The so-called medical image segmentation is the process of dividing a medical image into a number of disjointed "connected" regions based on certain similarity features of the medical image (e.g., brightness, color, texture, area, shape, location, local statistical features, or spectral features, etc.), where the relevant features are shown in the same region. The relevant features in the same region show consistency or similarity, while in different regions show obvious differences, that is to say, there is some kind of discontinuity in the pixels on the boundary of the region.

A region, as a connected set of pixels and a basic segmentation unit in image segmentation, can be defined according to different connectivity: 4-connected region and 8-connected region. The connectivity of a region means that between any two pixels in a region, there exists a connectivity path consisting of pixels belonging exactly to this region. If the connectivity of a region is determined only on the basis of neighboring pixels in quadrature (top, bottom, left, right) or quadrature (top left, bottom left, top right, bottom right), it is called a 4-connected region, and if the connectivity of a region is determined on the basis of neighboring pixels in quadrature and quadrature at the same time it is called an 8-connected region.

1, medical image segmentation: medical image processing and analysis of complex and critical steps in the field, its purpose is to medical image with some special meaning of the part of the segmentation, and extract the relevant features for clinical diagnosis and treatment and pathology research to provide a reliable basis to assist doctors to make more accurate diagnosis.

2, medical image segmentation: medical image imaging has a variety of image modalities, such as MR, CT and so on. Each element in a two-dimensional image is called a pixel, and each element in a three-dimensional image is called a voxel. In some cases, a three-dimensional image can be represented as a series of two-dimensional slices for observation, with the advantage of low computational complexity and small memory requirements.

3, medical image segmentation: automatic segmentation of targets from medical images is a difficult task, because medical images have a high complexity and lack of simple linear features; in addition, the accuracy of the segmentation results are also affected by some of the volumetric effects, gray scale inhomogeneity, artifacts, the proximity of the gray scale between different soft tissues and other factors. For the correction techniques commonly employed, artifacts in MR and CT images can be classified as:

(1) artifacts that require appropriate filtering algorithms, such as noise artifacts, sensitivity artifacts, and artifacts in the presence of non-sharp edges; (2) artifacts that require appropriate image restoration algorithms, such as motion artifacts; and (3) artifacts that require specific algorithms, such as partial-volume and grayscale inhomogeneities. Image processing field Although many algorithms exist to deal with the above problems, medical image segmentation is still a complex and challenging problem. From the point of view of medical image processing process, grayscale-based and texture-based feature techniques are the conventional classification methods. In addition, the use of machine learning tools to optimize these image segmentation algorithms is the current technique that has received more attention．

4. Some of the commonly used methods for CT image segmentation are: threshold based, region based, deformation model based, fuzzy based and neural network based.

5, the impact of factors:

(1) noise: due to the imaging equipment, imaging principles and the individual's own differences, medical images generally contain a lot of noise. Since the noise is independent of the location and space constraints, thus the distribution of noise can be utilized to achieve noise reduction.

(2) Artifacts: Artifacts are generally generated during image alignment and 3D reconstruction (e.g., CT), and can only be less, not eliminated, in principle.Artifacts in CT imaging include: partial volume effects, bar artifacts, motion artifacts, beam-hardening artifacts, ringing artifacts, and metal artifacts. Due to the existence of these artifacts to the CT image segmentation brings a certain degree of difficulty, different tissue parts segmentation accuracy is not the same.