An automated ASPECTS method with atlas-based segmentation

Highlights

• An automated ASPECTS method with atlas-based segmentation is proposed.

• The performance of different parameters and thresholds are evaluated and compared.

• Brain density shift is better for quantitative analysis of early ischemic changes.

• The accuracy of proposed method achieves 80% on the test set.

Abstract

Background and Purpose

As a simple and reliable systematic method to evaluate the early ischemic changes in the blood supply region of the middle cerebral artery of patients with ischemic stroke, the Alberta Stroke Program Early CT score (ASPECTS) can be used for rapid semi-quantitative evaluation of ischemic lesions, which is helpful to select potential candidates for intravenous and intra-arterial therapies, determine the thrombolytic effect and long-term prognosis. This method mainly relies on doctors’ visual observation. However, due to different levels of doctor's experience, the poor inter-reader agreement may result in errors in the final ASPECTS. The purpose of this work was to propose an automated semi-quantitative method for the diagnosis of acute ischemic stroke based on non-contrast computed tomography (NCCT), to provide a reference for doctors in the diagnosis and evaluation.

Methods

NCCT data from a total of 90 patients were included for auto-ASPECTS training and testing. After preprocessing CT images, the regions of interest (ROI) for ASPECTS were labeled using atlas-based segmentation. The mean difference, mean ratio and brain density shifts (BDS) of the corresponding regions of the contralateral brain were used as the standard for quantitative analysis. The auto-ASPECTS method was developed and validated to predict early ischemic changes whose performance was evaluated by the agreement (accuracy) of predictions and consensus scores of two observers.

Results

A comparison was made among the results on mean difference, mean ratio, BDS and the combination of multiple parameters as the standard. The result of using BDS alone was relatively better than the result of using any other parameter alone or any combination of multiple parameters, and accuracy in the test set was 0.80. In the test set, accuracy with using different BDS thresholds increased by 6.67% compared with using the consistent BDS threshold. After dichotomy of auto-ASPECTS and consensus scores with the threshold of 7, the agreement of them was 83.3% and there was no significant difference between the two distributions (p = 0.344) in McNemar test.

Conclusions

The proposed auto-ASPECTS method for NCCT images can provide useful information for early diagnosis and evaluation of patients with acute ischemic stroke (AIS).

Introduction

Stroke is a leading cause of mortality and disability worldwide, with a global lifetime risk of approximately 25% [1]. Stroke is the disease with the highest mortality and has 270 million high-risk groups and over 12 million sufferers in China. There are about 5.5 million new stroke cases and more than 2 million deaths in China every year [2], [3], [4]. Its recurrence rate in three years is more than 20% with only 2% timely treatment rate, and direct medical expenses are over 7 billion dollars per year, which has become an important reason for poverty in China [5,6]. Although the imaging technology in the field of medical devices has been constantly improved, the image post-processing still relies on tedious manual operation and marking. What's worse, sometimes the clinician has to spend a lot of time off work completing the daily scan of the image files.

The Alberta Stroke Program Early CT score (ASPECTS) [7] proposed by BARBER et al. in 2000 is a simple and reliable systematic method to evaluate the early ischemic changes in the blood supply region of the middle cerebral artery in AIS patients. It can be used for rapid semi-quantitative evaluation of ischemic lesions, selecting potential candidates for intravenous and intra-arterial therapies [8] and predicting the thrombolytic effect and long-term prognosis [7,9]. The ASPECTS method divides CT images of the brain into ten regions, on the basis of the importance of middle cerebral artery blood flow. Each of the 10 regions has the same weight, accounting for 1 point, and the total score is 10 points. The number of regions with early ischemic changes is subtracted from the total score, and the value obtained is used as the score result for ischemic core judgement and treatment selection.

However, there are several factors such as the time window between CT scanning and symptom onset [10], viewing window width and level settings [11], experienced doctors [12] influencing the reliability of the visual early ischemic changes (EICs) detection on NCCT. Even among experienced doctors, differences may exist in identifying and quantifying the early radiological signs of acute ischemic stroke on NCCT [13]. ASPECTS may be not reliable when utilized as a criterion for treatment decisions [14]. Therefore, an objective and automated method is needed to assist doctors in identifying and quantifying early ischemic brain damage. From the number of papers about ASPECTS that are retrieved from PubMed for each year in 2000 - 2019 in Fig. 1, it can be seen that the importance of ASPECTS in clinical applications is increasing.

Changes in brain tissue density and cortical swelling can be analyzed by CT values as a sign of EICs [15]. At present, some relevant studies have used CT densitometry to determine EICs in acute ischemic stroke. Bendszus et al. [16] proposed a post-processing method to detect acute MCA territory infarcts by subtracting the density histogram of the entire cerebral hemisphere in the CT images, and the results showed that the detection rate of acute infarcts increased significantly by human observers. ASPECTS regions were not used in the study, which limits its applicability in clinical practice. V. Puetz proposed a modified definition of EIC in the ASPECTS methodology. Focal swelling regions without associated hypoattenuation on NCCT do not contribute to ASPECTS [15].

On the basis of using CT densitometry to determine EICs in acute ischemic stroke, Stoel et al. [17] described an automated atlas-based ASPECTS method. They proposed an automatic method calculating the ASPECTS by using an atlas-based segmentation method to obtain the regions of interest in ASPECTS and quantifying tiny EICs based on the brain density shifts (BDS), which referred to shifts corresponding to the minimum overlap or maximum cross-correlation of the density histograms of the regions in both hemispheres, between the corresponding regions of contralateral brain. However, the results of using other statistical features of CT images and using different BDS thresholds for each region were not studied in this paper.

Therefore, we proposed an atlas-based segmentation method to quantify EICs and evaluated the performance of the method with brain density shift (BDS), mean difference and mean ratio or different groups of them as the parameters to determine EICs. And the accuracy of our method achieved 80% on the test set for all ASPECTS regions. What's more, we found different thresholds affected the performance of the proposed method and did several comparative experiments to find the better parameters and thresholds. In our performance evaluation experiment, a training set with 30 samples was used to validate the effectiveness of the automatic ASPECTS method and select the best template and parameters and a test set with 60 samples was used to verify the agreement between the scores calculated by our automatic ASPECTS method and the consensus scores. And after dichotomy of ASPECTS scores with the threshold of 7, the agreement percentage (0.83) of the results of our method in the test set are comparable with the agreement percentage of two observers (0.83) [17]. In addition, the kappa analysis showed similar result (κ = 0.665) with a commercially available software (RAPID ASPECTS) (κ = 0.7) [18].