Study of correlations between CT properties of retrieved cerebral thrombi with treatment outcome of stroke patients

The risk for stroke is increasing yearly due to the aging population on a global level and the accumulation of risk factors.1 Stroke is the second leading cause of death and the main cause for decreased quality of life.2 Time is of the essence when it comes to treating the stroke, where thrombolytic treatment, usually with systemic administration of recombinant tissue plasmin activator (rt-PA), and mechanical thrombectomy are the main approaches; however new experimental treatments are also under consideration.3

The main cause of stroke is the acute cerebral artery occlusion, which leads to ischemic stroke, where the oxygen supply is cut off.4 The occlusion can either be thrombotic or embolic, originating from various locations, which results in various thrombi structures. The structure can also give us information on the age of the thrombi.5 On a microscopic level, thrombi are composed of different levels of red and white blood cells intertwined with fibrin meshwork.6,7 The structure can strongly influence the permeability which is important for the success of the thrombolytic treatment.8

The diagnosis of stroke is a multiple step process, which consists of clinical evaluation and diagnostic imaging. The clinical evaluation includes the assessment of the effect of the stroke, which can be estimated using the modified Rankin Scale (mRS). This is a seven-level stroke scale ranging from 0, which stands for no symptoms to 6, which stands for death.9 Another scale serving the same purpose, however, with more grades is National Institutes of Health (NIH) severity stroke scale (NIHss).10

In diagnostics of stroke, computed tomography (CT) is the main tool in selection of patients that are suitable candidates for stroke treatment: firstly, to exclude hemorrhage, to exclude large areas of hypodense brain tissue, suggesting irreversible ischemia and to exclude stroke mimics. Native CT is complemented by CT angiography (CTA), which provides high diagnostic value in the detection of occlusion in high degree of stenosis as well as CT perfusion (CTP), which provides high specificity in the detection of ischemia and infarcted brain tissue. However, the most accurate assessment for acute stroke involving the site of occlusion, infarction core and salvageable brain tissue is a combination of different CT procedures involving CTA and CTP. In addition, CT scanning is fast, it is widely accessible and its price per scan is relatively low. However, sensitivity of CT to soft tissues cannot match with that of MRI.11,12 While it has been proven that CT can be used to determine some characteristics of the thrombi14, the field of connecting CT characteristics to clinical and procedure parameters has yet to be investigated.

The aim of this study is to demonstrate that CT images provide more information than currently used in routine stroke diagnostics. More specifically, mainly geometrical and macroscopic parameters of CT images are used, while information on the HU values of the thrombus causing the ischemic stroke, are usually overlooked or considered not important. In this study we want to show that this information is relevant for assessment of the thrombus microscopic structure and that these parameters correlate with some clinical parameters and can be therefore used for better treatment planning.

Histological sections of selected representative retrieved cerebral thrombi are shown in Figure 1. Histological slices were stained by hematoxylineosin and then analyzed for RBC proportions using digital image processing of the thrombi images acquired by using optical microscopy of the histological sections. All retrieved thrombi had a distinctly heterogeneous, laminated (multilayer) structure. The structure involves the interweaving of compacted erythrocyte-rich (red) regions with thinner (pink) coatings containing a combination of complementary-linked platelets and a fibrin network. The laminations are often folded and twisted, which is probably due to blood flow turbulence in the environment where the thrombi were formed and their turning as they traveled along the vessel. Among the retrieved thrombi there were no homogeneous single-layered (red only) thrombi. Which can be explain by their high susceptibility to thrombolysis.

Figure 1

Figure 2 shows an example of a CT image of a stroke patient with a clearly visible MCA segment on both hemispheres of the brain. In the example the MCA segment on the right hemisphere is occluded by the cerebral thrombus, while the MCA segment on the opposite hemisphere is normal. Exact location of each thrombus was determined from its CTA image first and then this location was used to obtain HU values along the thrombus from the corresponding CT image. Yellow line on the image is drawn along the thrombus and the symmetrical position of the normal MCA segment. Graphs in Figure 2A correspond to HU intensity profiles along both yellow lines: along the cerebral thrombus in Figure 2B and along the symmetrical section of the normal MCA segment in Figure 2C. For each HU, intensity profile was then calculated by average HU value HU_avg and HU value variability HU_var. Thus, each patient was characterized by these two CT parameters that were obtained for the occluded (occl) and normal (nor) MCA segment so that it became possible to calculate their absolute (diff) and relative differences (diff [%]), which are equal to: HU_avg_diff = HU_avg_occl–HU_avg_nor, HU_avg_diff[%] = 100·(1-HU_avg_nor/HU_avg_occl) and HU_var_diff = HU_var_occl–HU_var_nor, HU_var_diff[%] = 100·(1-HU_var_nor/HU_var_occl). Table 1 shows CT as well as histological (RBC proportion), clinical and procedure parameters of patients and their retrieved thrombi included in the study.

Figure 2

Correlation between different pairs of data groups from Table 1 were tested by the Pearson correlation coefficient and the univariate linear regression analysis. Results of this analysis for the pairs with the highest correlation is shown by linear regression graphs in Figure 3 and their corresponding correlation and linear regression parameters are formulated in Table 2. The linear regression statistical analysis showed significant differences (p < 0.05) only in three cases, i.e., for the group pars: outcome mRS vs. average HU in the occluded MCA, the number of passes with the thrombectomy device vs. the average HU in the occluded MCA, and the difference between the initial and the outcome mRS vs. the average HU in the occluded MCA. The next three pairs, also included in Table 2 and Figure 3, had 0.05 < p < 0.1 which can be considered as borderline significant. These pairs are: thrombus RBC proportion vs. the difference between the average HU in the occluded and normal MCA, initial mRS vs. the average HU in the occluded MCA and the difference between the initial and the outcome mRS vs. difference between average HU in the occluded and normal MCA. From the graphs in Figure 3A,E it can be inferred that the average HU in the occluded MCA is in positive correlation with both the initial and the outcome mRS. As the regression coefficient (slope k) is higher for the outcome mRS than the initial mRS, the correlation between difference of the initial and the outcome mRS (mRS_diff = mRS_start-mRS_end) and average HU in the occluded MCA is negative (Figure 3C). The number of passes of the thrombectomy device is also in positive correlation with the average HU in the occluded MCA (Figure 3b). The remaining two of the more significant correlations involve differentiation between the average HU in the occluded and the normal MCA which is in positive correlation with the RBC proportion (Figure 3D); and the negative correlation with the difference of mRSs (Figure 3F). From the results in Table 2, it can also be seen that the data group pairs with higher coefficients of determination R² in the linear regression analysis also have higher Pearson coefficients of correlation. Therefore, R² can also be considered a measure to the extent of correlation.

Figure 3

CT was already employed to study properties of artificially made blood clots. In a study by Kirchhof et al. it was concluded that CT has a potential to differentiate between clots with high fibrinogen content and those with low content, but is unable to distinguish between pure plasma clots and low hematocrit clots.16 In a more recent study by Brinjikji et al. single-energy and dual-energy CT were compared in their power to differentiate among different types of artificial blood clots.17 They confirmed superiority of dual-energy CT in differentiation between high and low RBC content blood clots. However, studies where CT properties of blood clots in vivo are studied are relatively scarce.18In our study on correlations between CT properties of MCA thrombi and thrombectomy procedure parameters, treatment outcome, and histology the highest correlation was found between average HU in the occluded MCA (HU_avg_occl) and the stroke outcome assessed with mRS (mRS_end). According to our findings, higher average HU in the occluded MCA suggests higher mRS. This is because the most important factor among the factors that contribute to the assessment of mRS (such as extent of penumbra tissue, collateral vasculature, time form onset of the occlusion to the intervention, etc.) is the morphology of the thrombus, especially its compaction. Due to their retraction, thrombi with higher compaction, which are often older with low serum content19,20 have higher average HU values, while fresh thrombi with more serum spaces21, have lower overall HU.

The correlation between the number of passes with the thrombectomy device and average HU of the thrombi can also be associated with the thrombi organization. More organized (hard, fibrin-rich) thrombi can attach strongly with the arterial wall, which may result in the more difficult retrieval of such thrombus during the interventional procedure.22 This suggest, that if higher average HU values in the MCA occlusion are found on the CT scan, it is getting more likely that more than one pass will be needed to successfully retrieve the thrombus during the interventional procedure.

When treating patients with stroke, time is essential.23 The longer time taken till reestablishment of the normal cerebral blood flow, the less oxygen is present in the brain, which may result in brain cell and nerve connection loss. This contributes to more serious symptoms and a higher mRS classification over time.23 Outcome after mechanical thrombectomy can be more dependent on the capacity of collateral circulation than on the time from stroke onset.24 However, time still plays an important role in saving tissue at risk (penumbra). In addition, the longer time taken till resolving the occlusion, the thrombus can retract further and therefore becomes more compact with a low serum content. Clot retraction is an important contributing factor for the correlation that was found between the initial mRS (mRS_start before the treatment) and the average HU of the thrombus (HU_avg_occl). This factor may also explain the correlation between the average HU of the thrombus (HU_avg_occl) and the mRS classification difference between the initial and the outcome scores (mRS_diff). As already pointed out, a poorer treatment outcome can be expected in patients with poor collateral status.25 However, time since the formation of thrombus also impacts its properties, especially its retraction, which increases with the time since thrombus formation.26 The correlation between the average HU in the occluded MCA and the difference between the initial and the outcome mRS is negative (Figure 3C). This is because patients with a high average HU in the occluded MCA have usually higher initial mRS classification and their outcome mRS, on average is only slightly lower or even equal to the initial mRS due to the poor treatment results; e.g. patients with the initial mRS of 5 have often the outcome mRS of 4 or 5 so that the mRS difference (mRS_diff) is 0 or 1, which corresponds to a slim or no treatment improvement. In patients with low average HU in occluded MCA, the initial mRS classification is lower and therefore, their outcome mRS is even much lower, e.g., patients with the initial mRS of 1-3 can recover to mRS of 0 or 1, so that the mRS difference is in the range 0-3. The treatment outcome in these patients is better due to the short time frame from thrombus formation to intervention which is associated with earlier reperfusion and smaller brain tissue damage.27

In addition to average HU in the occluded MCA segment (HU_avg_occl), its difference to the corresponding value in normal MCA segment (HU_avg_ diff) was measured as well. The average HU difference value has in principle advantage over the single-sided MCA values (HU_avg_occl and HU_ avg_nor) as it is insensitive to possible offsets in HU values in CT images. The offsets cancel out in the subtraction of average HU values between the occluded and the normal MCA segments. Possible HU value offset can have an origin in image processing where consecutive three NCE CT image slices in the section with both MCA segments were stacked together. Therefore, the measured HU can in principle contain unwanted contributions from arterial wall and other tissues surrounding the vessel. In addition, the reference HU, that is measured in non-occluded symmetrical MCA segment, can be elevated due to possible arterial wall calcification, hematocrit level.

Positive correlation between RBC proportion (RBC [%]) and the difference between average HU in the occluded and normal MCA (HU_avg_diff) can be explained by the presence of iron in RBCs.28 The thrombi with a higher proportion of RBC have a higher iron content and therefore absorb more X-rays so that an increased HU values are detected in the CT images of these thrombi. In addition, thrombi with a higher proportion of RBC are also more compact, which further contributes to greater X-ray absorption. Another reason for increased HU values in the MCA region are also calcifications in the vessel wall, which cause higher X-ray absorption. If that is the case MCA may appear hyperdense also without thrombus in it.

Pretreatment thrombus evaluation can be of great clinical importance. Failure of thrombolytic treatment can be expected in fibrin rich thrombi. Thrombus composition can also have an impact on interventional treatment planning (mechanical thrombectomy). Optimal technique (stent retriever vs. aspiration) or device type selection can potentially be chosen on the basis of preoperative imaging data.

Major drawback of this study is limited number of cases (n = 25) and the use of sequential NCE CT. CT image resolution was insufficient to avoid the partial-volume effect in measurement HU profiles along the thrombi so that the HU profiles can be contaminated with HU values of the surrounding tissues. Specifically, in this study CT image analysis included stacking of three consecutive CT slices could result in inclusion of other brain tissues than the thrombus or normal MCA vessel because of the inferior bend of the MCA that can be anatomically present. Additional problem with the HU values of the vessel wall is that they can vary also due to possible atherosclerotic soft calcifications and these calcifications are often asymmetric, so that using the HU profile of the symmetric non-occluded MCA segment as the reference becomes questionable. If the spiral CT would be used instead of sequential CT, the CT images could also be processed in other (rotated) planes and the stacked slices could then be positioned and oriented ideally parallel to both MCA segments without compromising the quality of the observed vessel. This would make the HU measurements as well as the results more precise.

Routinely acquired CT images of stroke patients provide also information on HU values of thrombi that is usually ignored. In this study a relation between the HU value of a thrombus and its composition was confirmed. As more compact thrombus may represent a bigger problem for the interventional procedure a priori assessment of thrombus compaction is very important for good intervention planning. The present study also provides foundations for further studies where the accuracy of study could be improved by increasing the number of samples (patients and thrombi) and having more accurate spiral CT images of patients with stroke that would allow improve HU value determination in the targeted region.