Introduction

An improved understanding of infarct evolution following reperfusion therapy may be an opportunity to therapeutically mitigate secondary injury processes associated with blood flow restoration. Measurement of infarct volume by magnetic resonance imaging (MRI) has been used as a secondary endpoint in many randomized control trials [1, 2]; however, segmentation of infarct volumes, both pre-reperfusion and postreperfusion therapy, has been problematic as there is a wide range of relative and absolute diffusion-weighted imaging (DWI) and apparent diffusion coefficient (ADC) thresholds that have been used for this purpose [3,4,5,6,7,8,9,10,11,12,13,14,15,16].

A major reason for the variations in DWI thresholds used to segment the infarct core is related to the proposed definitions of infarct core and the heterogeneity of study designs; therefore, standardization of MRI sequences and threshold-based techniques applied for measuring stroke lesion volumes are overdue. Future clinical therapeutic trials and observational acute stroke studies using MR DWI will require standardization of MRI sequences and threshold-based techniques applied for measuring stroke lesion volumes and to permit comparison between studies.

In this work, we examined the stroke lesion volume and its evolution evaluated by MR imaging during the immediate post-treatment phase (<5 h) and at 24 h. Specifically, we determined whether different ADC and DWI thresholds affect the measurement of lesion volume at two time points after reperfusion: <5 h and 24 h. In addition, we quantified the differences in volumetric change between the two imaging time points measured by the prespecified absolute ADC and relative DWI intensity thresholds and quantified the mean differences by estimating the bias and the limits with which differences will lie with 95% confidence. We hypothesize that the change in stroke lesion volumes between two time points will not depend on the absolute ADC and relative DWI thresholds applied, and therefore provides a standardized metric for future stroke trials.

Material and Methods

Patient Details

This was a prospective single-center study including a total of 33 acute ischemic stroke patients. Reperfusion treatment, either tissue plasminogen activator (tPA), mechanical thrombectomy or both, was administered before MR imaging. Written informed consent was obtained from either the patient or the surrogate. The Calgary Health Regional Ethics Board (CHREB) institutional ethical committee approved the study.

Inclusion criteria were: (i) acute ischemic stroke last seen normal within 12 h as per the ESCAPE trial inclusion criteria [17], (ii) occlusion detected by computed tomography angiography (CTA) with hemispheric stroke involving the middle cerebral artery, posterior cerebral artery, or anterior cerebral artery territory, (iii) age ≥18 years and (iv) MR imaging performed immediately following acute treatment <5 h (baseline), and at 24 h (follow-up) after treatment.

MRI Acquisition

All imaging was performed on a 3.0 T clinical MR scanner (Discovery 750; General Electric Healthcare, Waukesha, WI) using a 12-channel receive-only head neck and spine (HNS) coil. Subjects were positioned supine with head first position into the head coil. Diffusion-tensor imaging (DTI) was performed using a spin echo – echo planar imaging (SE-EPI) sequence (repetition time/echo time [TR/TE] = 9000/94.3 ms, field of view [FOV] = 250–350 mm, flip angle = 90o, number of signals averaged [NSA] = 1, EPI factor = 128, acquisition matrix = 144 × 144 and slice thickness = 5 mm without gap) with diffusion sensitizing gradients (b = 1000 s/mm2) applied in 15 directions and one non-diffusion-weighted (b = 0 s/mm2) volume.

Reperfusion Assessment

Reperfusion status was determined using the thrombolysis in cerebral infarction (TICI) score based on Intra-arterial Therapy (IAT) on digital subtraction angiography (DSA; [18]). Where DSA was not available, the TICI was assessed using the modified arterial occlusion lesion (mAOL) approach from multi-phase CTA (mCTA) data [17].

Image Processing

The DTI data were motion-corrected by aligning all DW volumes to the b = 0 s/mm2 volume using the ANTONIA software [19]. The pre-processed data were subsequently used to compute an average DWI and the ADC map using the Stejskal-Tanner equation [20]. Lesion volumes were semi-automatically segmented at baseline and follow-up using commonly used ADC thresholds of 550 × 10−6 mm2/s, 600 × 10−6 mm2/s, 630 × 10−6 mm2/s and 650 × 10−6 mm2/s using the ANTONIA software [7, 9,10,11,12, 20]. Lesion segmentation using relative DWI intensity thresholds was performed using MeVisLab software [21]. Therefore, the mean (μ) and standard deviation (σ) of the DWI values in normal contralateral brain tissue was obtained from the average DWI datasets using manually defined regions of interest (ROI) approximately mirroring the visible ischemic region. Different relative thresholds (t) were calculated as t = (μ + k*σ), with k = 0.85, 1.28, 1.64, 1.96, and 2.7 [7, 13,14,15,16]. The DWI thresholds were used to segment the lesion using a volume growing approach. The absolute change in lesion volume (infarct growth) between baseline and follow-up was calculated for each ADC and DWI threshold by subtracting the baseline volume from the corresponding follow-up volume. This method for measuring DWI volume is highly correlated to other methodologies, such as voxel-based analysis [22]. Any acquisitions containing hemorrhage were not included in the analysis (n = 4 for <5 h, n = 5 for 24 h).

Statistical Analysis

Calculation of the distributions of all variables was performed using descriptive statistics, e.g. means, medians, standard deviations, frequency distributions, interquartile range (IQR) and estimated proportions. Non-parametric Wilcoxon test was used for comparisons between baseline and follow-up absolute lesion volumes using different ADC and DWI thresholds. Kruskal-Wallis H (K-independent sample test) was performed for comparisons of the absolute change in lesion volume (infarct growth) from the corresponding ADC and DWI thresholds. We quantified the differences in volumetric change acquired using the prespecified ADC and DWI thresholds by estimating the bias, which is the mean difference of the volumetric change, and the limits within which most differences will lie at 95% confidence intervals. We estimated these limits from the mean standard deviation of the differences. For example, for the thresholds to agree in terms of these volumetric measurements, we wanted them to be the same for high and low threshold values used for lesion measurement. This can be checked from the Bland-Altman Plot [23]. A p-value of ≤0.05 was considered significant. All statistical analyses were performed using Statistical Package for the Social Sciences (SPSS) software (version 19.0; IBM, Armonk, NY).

Results

Subject demographics are summarized in Table 1 and one patient was excluded due to Digital Imaging and Communications in Medicine (DICOM) image corruption. The median absolute lesion volumes for baseline and follow-up were significantly different for each ADC threshold 550 × 10−6 mm2/s, 600 × 10−6 mm2/s, 630 × 10−6 mm2/s and 650 × 10−6 mm2/s (p < 0.05; Table 2). Similarly, the median absolute lesion volumes for baseline and follow-up were significantly different for each DWI threshold k = 0.85, 1.28, 1.64, 1.96, and 2.7 (p < 0.05; Table 3). Fig. 1 shows the representative examples for both diffusion sequences. For both ADC and DWI sequences, there was a significant difference in lesion volume between each time point (p < 0.05). Fig. 2 shows the Bland-Altman plots for the absolute change in lesion volumes between baseline and follow-up for the absolute ADC and relative DWI thresholds. Although the bias of agreement was similar when the absolute ADC and relative DWI thresholds were similar, both the bias and limits of agreement increased as the difference of the compared ADC and DWI threshold increases. For example, when comparing lesion growth as determined using the ADC threshold of 550 × 10−6 mm2/s with 600 × 10−6 mm2/s, the mean difference between the measurements is 2.5 ml but the limits of agreement, i. e. 95% CI around the mean difference, can be as high as 12.4 ml. When the difference in thresholds applied increases, for instance, when comparing absolute volume change at thresholds 550 × 10−6 mm2/s and 650 × 10−6 mm2/s the mean difference (the bias agreement) is 7.2 ml and the 95% CI around that mean difference is as high as 29.5 ml. Similar results were obtained for DWI absolute volume change (Fig. 2ii)

Table 1 Clinical details of all the patients (n = 33)
Full size table
Table 2 Median (IQR) baseline and follow-up lesion volumes (ml), and the median absolute change in lesion volumes (ml) from baseline to follow-up for specific ADC thresholds
Full size table
Table 3 Median (IQR) baseline and follow-up lesion volumes (ml), and the median absolute change in lesion volume (ml) from baseline to follow-up for specific DWI thresholds
Full size table
Fig. 1
figure 1

Following ischemia reperfusion, the absolute stroke injury volume is dependent on Apparent diffusion coefficient (ADC) and relative Diffusion weighted imaging (DWI) thresholds. In this example, a 45-year-old female (Thrombolysis in Cerebral Infarction [TICI] 2b reperfusion) presenting with a severe stroke at baseline with National Institute of Health Stroke Scale (NIHSS) score of 20 demonstrated M1 occlusion on Computed Tomography Angiogram (CTA). Intravenous-tissue plasminogen activator (tPA) was administered at 3.35 h following stroke symptoms and MR imaging was performed at 4.08 h post-stroke. Reperfusion was confirmed by digital subtraction angiogram (DSA) at this time point. a–e The absolute lesion volume determined by applying (i); f–j absolute ADC lesion volume at baseline and at follow-up for ADC thresholds of 550 × 10−6 mm2/s to 650 × 10−6 mm2/s and (ii) a–f; absolute DWI lesion volume at baseline and g–i at follow-up using thresholds of K = 0.85 to k = 2.7. The ADC and relative DWI thresholds demonstrates considerable variation depending on the threshold used

Full size image
Fig. 2
figure 2

Shows the Bland-Altman plots comparing (i) apparent diffusion coefficient (ADC) thresholds: a absolute change in lesion volume (ml) for 550 and 600 × 10−6 mm2/s, b absolute change in lesion volume for 550 and 650 × 10−6 mm2/s. (ii) Relative diffusion weighted imaging (DWI) thresholds: c absolute change in lesion volume (ml) for k = 0.85 and k = 1.28, d absolute change in lesion volume for k = 0.85 and k = 2.7. In all plots, the Y‑axis depicts the difference between the thresholds while the X‑axis shows the mean of both threshold values. The bias of agreement is represented by the gap between the X‑axis (bold black line) and the corresponding zero line (dashed line). The shaded area represents the confidence intervals around the mean difference. The red line represents the ± 95% (2 SD) limits of agreement

Full size image

Discussion

In this study, the absolute lesion volumes differed when different ADC absolute and DW intensity thresholds were applied at both 5‑h and 24-h time points. For instance, the median absolute lesion volume at 24 h for ADC 550 × 10−6 mm2/s versus 650 × 10−6 mm2/s was 16.9 ml versus 4.8 ml, almost a 4-fold difference. Without standardization of the measurement of MR diffusion lesion volumes using absolute ADC thresholds and DWI intensity thresholds, the infarct volumes significantly vary, which will have consequences for interpretation of stroke volume data, comparison of such data between studies, and affect sample size calculations for future clinical trials. Also, contrary to our hypothesis, the absolute lesion volume changes (lesion growth over 24 h after acute revascularization treatment) for both ADC and DWI sequences tended to be threshold dependent, which may have clinical significance. The limits of agreement shown in the Bland-Altman analysis increase as the difference between comparative thresholds increase, and the mean bias of agreement also changes when the threshold comparisons are more disparate. The large difference in volume change between specific thresholds at the 95% CI (limits of agreement) support the conclusion from these data that any two threshold methods are not comparable and that one threshold method cannot reliably replace another threshold method. These observations have significant importance in terms of measuring stroke lesion growth reliably and reproducibly as a suitable outcome measure in MR imaging stroke clinical trials by applying standardized post-imaging analysis methods.

Also of relevance are the limitations of automated thresholding techniques for volumetric analysis of the ischemic lesion. We observed several cases where the ADC and DWI thresholds either failed to segment the lesion completely or underestimated the lesion size compared to visual assessment by experts. Generally, following cerebral ischemia there is restricted diffusivity of water in the extracellular space related to cytotoxic edema, and the diffusivity of water is a time-dependent phenomenon [24]; there is an initial drop in ADC and concomitant increase in DWI [25]. The decline in ADC starts to normalize after 7–10 days, after which the ADC starts to increase supporting the presence of vasogenic edema [25]; however, in some cases, we observed an increase in ADC and a decrease in DWI signal intensity, contradicting the conventional DWI signature of ischemia. This is reflective of the acute increase in water in the extracellular space indicating severe vasogenic edema in the acute phase of ischemia reperfusion [26, 27].

In this study we demonstrated that absolute lesion volumes measured and change in lesion volumes over 24 h after recanalization treatment is dependent on the absolute ADC and relative DWI thresholds used. Standardization of stroke lesion segmentation, either by automated or manual techniques, will be required based on these results. Attempts to salvage ischemic brain tissue with thrombectomy do not always produce a significant clinical benefit in a substantial proportion of stroke patients that receive the therapy. One consideration in the design of future therapeutic stroke trials is the utilization of surrogates of disease progression, such as the temporal change of DWI volume. Measuring an acute change in DWI volume calculated from serially acquired MRI could significantly enrich clinical trial design by reducing sample sizes required to show a biological effect of a potential neuroprotective agent compared with traditional clinical outcome measures or infarct volume measured on MRI at a single time point, e.g. 30-day fluid-attenuated inversion recovery (FLAIR) MRI [28]. We anticipate that our results will support the need for greater standardization of image segmentation methods for measuring acute stroke lesions. This will be imperative to our understanding of reperfusion related secondary lesion growth that will need to be explored in future phase II clinical therapeutic stroke trials, with the aim of improving stroke outcome by attenuating secondary injury processes.