Feature analysis

Chapter 4

Atherosclerosis plaque recognition

PCI with the stent strut medication as a progressive treatment indeed can relieve or cure the symptom of patients who catch CAD. Recognizing and measuring the location, angle, categories of the atherosclerosis plaque is the elementary and significant procedure before applying PCI, which provides beneficial information to specialists during the following treatment step, such as vessel inner examination and stent implantation. However, the manual qualitative analysis job of classification for the vessel lesion tissue and tracing of plaque components is time-consuming to doctors because a single-time IVOCT data set of a patient usually contains hundreds of in-vivovessel images with respect to one segment vascular.

Identifying and classifying rapidly and accurately atherosclerosis plaques and obtaining its relative measurement data instead of manual work are challenge tasks. As such, automated methods for IVOCT tissue characterization is necessary for the cardiovascular research and clinical practice. In this chapter, I focus on the need for developing an automated tissue identification method, and we will discuss the tissue feature definition and extraction of the vessel lesion.

radial and circumferential directions, respectively. As such, an obvious phenomenon can be observed in IVOCT images. First, the vessel tissue (including the lesion plaques) appears significantly different due to the existence of the tissue attenuation coefficientµ presenting difference. Different types of vessel tissue can be displayed clearly in the OCT image with the optical coherence technique. Secondly, the ability of light penetration gradually declines as the depth increases in the radial direction. Only the superficial layer of the IVOCT image can be clearly observed, while the most rest of the region manifest dark area containing useless information. Third, mixed plaques increase the difficulty of lesion plaques recognition.

Tissue stratification would occur in the radial direction, that is the different tissues would appear in one radial direction. Except for the center high-light region of the IVOCT image, the rest surrounding areas of the vessel imaging present poor signal and low contrast. The superficial tissue near the lumen boundary maybe consist of different types of tissue but its distribution is continuous in a limited region. According to the human vessel image acquisition procedure described in Sec. 2.1, the catheter scans the vessel with continuous angles to form a vessel OCT image. That is the tissue imaging presents continuity in the circumferential dimension and the contrast among different tissues is obvious. Figure 4.1 displays an example of the similarity and contrast comparison of the intensity along the radial and circumferential direction by using different colors in the superficial layer of the vessel wall. Based on the statistical tissue thickness data indicated in Tab. 4.1, we used 8 pixels as the basic depth for each layer and 24 pixels as the unit step length to partition regions in the circumferential direction to define an interesting region to compare the variance of intensity and texture of adjacent areas. Histologically, the intensity characteristic of tissues on the same layer having a fixed thickness presents homogeneous and coherent. Additionally, tissue stratification would occur in the radial direction, that is the different tissues would appear in one radial direction. That is the intensity distribution has a gradient property in the radial direction and certain continuous in the circumferential direction because of the heterogeneous of the vessel lesion tissue. Notably, Except for the center high-light region of the IVOCT image, the rest surrounding areas of the vessel imaging present poor signal and low contrast, which is worthless to the lesion plaque research.

Furthermore, several researches[13, 28, 43, 93] have proved that lipid plaques associate with a high attenuation coefficient or appear in signal-poor regions, the fibrous plaque has a low attenuation coefficient or is in signal-rich regions, and the calcified plaque presents low attenuation coefficient but with sharp borders. Soest et al.[86] investigated that the attenuation coefficient(µ_t)of the healthy vessel wall and fibrous plaque were 2–5mm⁻¹, calcified plaque µt ≈6±1.0mm⁻¹, lipid tissue µt⩾10mm⁻¹. Observing the Tab. 4.1, the thickness of the healthy vessel wall and the calcified plaque are in a limited range while the lipid plaque

4.1 Feature analysis 69

Fig. 4.1 (A) is the original IVOCT image, and (B) presents the characterization of the intensity similarity and variation in the superficial tissue nearby the lumen boundary.

occurs in a dynamic range due to its diffuse attribute. The fibrous plaque is classed into three categories based on its thickness measurement. Several studies have investigated to utilize the intensity profile of A-line for the tissue or lesion plaque classification analysis. The intensity change in the A-line profile indirectly or directly reflects the absorption and scattering attributes of different tissues. However, analyzing the vessel lesion plaque identification only with the intensity profile can not effectively and completely solve the tasks of atherosclerosis plaque identification, classification and quantitative measurement. Some papers combined the A-line profile with light attenuation for the feature analysis of the vessel tissue, such as Ughi et al.[83] used a rectangular window to iteratively fit the OCT A-lines for differentk values. At everyk, all the possible fits were calculated, and the best fitting curve was selected as the indicator µ_t of the corresponding tissue. Although the single A-line as the direct and important analysis object to extract significant features for vessel tissue quantitative measurement, note that a single A-line only contains 1-D information to present the vessel tissue appearance with OCT technology, e.g. intensity profile or attenuation degree, can’t provide extra information for the lesion plaque identification and classification. Additionally, according to the A-line characteristic analysis of per type of the vessel tissue and quantitative thickness measurement researches of different tissues (illustrated in Tab. 4.1) in the earlier literatures, the proximity luminal boundary region (PLBR) of a vessel contains more useful morphology information about the vessel tissue (health vessel wall or lesion plaques). The relationship between adjacent A-lines will not be indicated and the local region con-texture information of the vessel tissue will also be lost simultaneously if only considering using methods based on the single A-line along the radius direction for the tissue classification analysis. Undoubtedly, the information of the 2-D region containing multiple adjacent

A-lines would give out more benefits for tissue identification and recognition. Obviously, the texture and intensity relativity in a set of adjoining A-lines can be easily detected with the existing graph image analysis method. Meanwhile, extra features could be discovered and analyzed to help the lesion plaque examination in clinical research.

Table 4.1 Appearance of tissue thickness measurement in the previous studies Tissue type Thickness measurement

Healthy vessel wall <0.3mm[86]

Fibrous plaque

Three categories[88]:

<0.065mm;

0.065–0.15mm;

>0.15mm Lipid plaque 3.9±2.1mm[28]

Calcified plaque <0.7mm[91]