Comparison of EPI DWI and STEAM DWI in Early Postoperative MRI Controls After Resection of Tumors of the Central Nervous System

The MRIs of neurosurgical patients with early postoperative MRI were retrospectively analyzed. The study was ethically approved by the institutional review board (No. 32/05/21, Ethics Committee of the University of Göttingen).

We included all patients with early postoperative MRI, who were scanned between 1 October 2019 and 30 September 2021. Inclusion criteria were (i) a time interval up to 72 h after neurosurgical brain operation of brain tumors, (ii) a standard MRI protocol, as described in the next paragraph, and (iii) STEAM DWI.

The database search in our Radiology Information System (medavis RIS 4, medavis GmbH, Karlsruhe, Germany) with the search terms: “STEAM” and “postoperative” found 50 patients, 7 of whom were excluded because the interval between surgery and MRI was more than 3 days. Of the patients 43 (21 female) were included with an MRI scan performed within 72 h after surgery. Mean age at the time of MRI was 49 ± 22 years (mean ± standard deviation), range 4–81 years. The average time between end of the operation and the MRI was 37 ± 20 h (mean ± standard deviation, median 41 h).

The pathologies comprised 25 glioblastoma multiforme World Health Organization (WHO) grade IV, 3 high-grade astrocytomas WHO grade III, 10 low-grade gliomas WHO grade II, and 5 pilocytic astrocytomas WHO grade I.

All MRIs were performed on the same 3‑Tesla MR scanner (MAGNETOM Prisma, Siemens, Munich, Germany). Our early postoperative MRI protocol included a high-resolution axial T2-weighted turbo spin echo (TSE) sequence with 2.5 mm slice thickness (echo time TE 108 ms, repetition time TR 3000 ms, in-plane resolution 0.5 × 0.5 mm), and a native and contrast-enhanced sagittal T1-weighted volume interpolated breathhold examination (VIBE) sequence with 1.0 mm slice thickness (TE 2 ms, TR 5 ms, in-plane resolution 1.0 × 1.0 mm). Additionally, an axial STEAM DWI sequence (TE 4.53 ms, TR 7.52 ms, 1 × 1 mm in-plane resolution, 3 mm slice thickness, voxel volume 3 mm³, 6 gradient directions, 144 s acquisition time, with DWI acquisition at b = 0 s/mm² and b = 1000 s/mm², apparent diffusion coefficient (ADC) map) was applied in tandem with a standard EPI DWI (Siemens 3scan_trace_p2, TE 67 ms, TR 3000 ms, 0.62 × 0.62 mm in-plane resolution, 5.2 mm slice thickness, voxel volume 2 mm³, 96 s acquisition time, with DWI acquisition at b = 0 s/mm² and b = 1000 s/mm², ADC map) in order to increase the detection rate for postoperative stroke in patients within 72 h after neurosurgery. For further details about the sequences, see Supplemental Table 1.

The STEAM DWI relies on a spin-echo diffusion module preceding a single-shot STEAM imaging technique with highly undersampled radial encodings. Iterative online image reconstruction is accomplished by nonlinear inversion with spatial regularization. The undersampling factor is 20, for further details see [3, 4].

The underlying technique of EPI DWI is conventionally a single-shot spin-echo echo planar imaging sequence with built-in diffusion gradients [7]. A 3-scan trace approach is used to account for the diffusion anisotropy. Image acquisitions are performed sequentially using 3 orthogonal diffusion gradient directions [7].

All patient data were pseudonymized and listed in a picture archiving and communication system (PACS) folder. Client software used was GE Centricity™ Universal Viewer (GE Healthcare, Chicago, IL, USA). Three neuroradiologists (rater 1, rater 2 and rater 3), with 2, 4 and 8 years of diagnostic MRI experience, respectively, independently evaluated the cases for the occurrence and severity of artifacts. The diagnostic accuracy regarding a residual tumor or a postoperative subacute infarction was assessed (i) with EPI DWI and without STEAM DWI and (ii) with STEAM DWI and without EPI DWI. Preoperative and postoperative T2-weighted and T1-weighted images with and without contrast agent were also available. A 4-week time interval was observed in order to minimize recall bias for the independent assessment of the two DWI sequences. The confidence in the imaging diagnosis of a residual tumor or a postoperative subacute infarction was indicated separately by each rater on a visual analog scale from 0 to 100%.

The T2-weighted TSE in combination with diffusion-weighted images with b = 1000 s/mm² of (i) EPI and (ii) STEAM of every patient were rated for the presence and severity of air artifacts (AAR air artifacts rating; 0 no artifacts, 1 small artifacts, 2 moderate artifacts, 3 severe artifacts). The rating was done for 3 regions (frontal subdural, ventricular system and resection cavity). Figure 1 demonstrates the rating scale. For further analysis, MRI data were dichotomized for the presence of distorting artifacts in the resection cavity (0–1 without, 2–3 with influencing artifacts).

The resection cavity was rated for the presence and severity of blood artifacts as well.

We defined the transversal T2-weighted TSE sequence 3 months after surgery as our gold standard for the determination of sensitivity and specificity. All gold standard diagnoses were discussed and determined jointly. We considered every T2-hyperintense lesion or defect, (despite tumor growth) with a short axis diameter of at least 2 mm as an infarct (T2-weighted TSE transversal as described above). Connected lesions were counted as one.

In addition to the visual determination of the infarction lesions and the indication whether they would suspect residual tumor, each radiologist had a subjective certainty of the diagnosis on a visual analog scale (0–10). An explicit image quality and sharpness rating of both sequences was not conducted.

Two blinded raters (rater 1 and rater 2) separately marked all infarct areas using T2-weighted TSE transversal and either EPI DWI (with b = 1000 s/mm² and ADC map) or STEAM DWI (with b = 1000 s/mm² and STEAM ADC map) with the segmentation tool of 3D Slicer (http://​3dslicer.​org, version 4.11 Linux). Areas with an increased signal on b = 1000 s/mm² imaging and a reduction of the corresponding ADC values were segmented in each slice, as demonstrated in Supplemental Fig. 1. All resection borders were counted (independent of thickness). Infarct volumes were calculated by 3D Slicer. The values of both raters were averaged for further analysis. An example of the segmentation tool is shown in Fig. 2.

The program Statistica, version 13 (TIBCO Software Inc., Palo Alto, CA, USA) was used for statistical analysis. P-values below 0.05 were defined as statistically significant. The detection rate, sensitivity and specificity of STEAM DWI to detect postoperative stroke as compared to EPI DWI was calculated for every individual rater as well as for all raters together. Shapiro-Wilk test [8] rejected the hypothesis of a normal distribution of the detection rates and measured volumes. Hence, the nonparametric Wilcoxon matched pairs [9] test was used for the comparison of observations and volume data.

Fleiss’ kappa [10] was calculated to determine the overall raw agreement between the three neuroradiologists.

Postoperative diffusion-restricted lesions (diameter > 3 mm) were detected in 43 of 43 patients. On average 2.2 ± 1.4 (mean ± standard deviation, rater 1: 2.1 ± 1.3, rater 2: 2.2 ± 1.4, rater 3: 2.3 ± 1.4) diffusion-restricted lesions were detected per patient using EPI DWI and 2.9 ± 1.3 (rater 1: 2.8 ± 1.2, rater 2: 2.9 ± 1.4, rater 3: 3.0 ± 1.4) lesions using STEAM DWI. An illustration of the frequency distribution is provided in Supplemental Fig. 2.

In patients with air artifacts (AAR > 1), the mean count of diffusion-restricted lesions was 2.0 ± 1.3 (EPI DWI), and 2.8 ± 1.2 (STEAM DWI).

In 26 patients a 3-month follow-up MRI was available (gold standard). For these patients, the overall detection rate (for infarct lesions) of EPI DWI was 74% (63/85) compared to 100% of STEAM DWI. Subacute infarct lesions were completely missed by EPI DWI in two patients (sensitivity: EPI DWI 92%, STEAM DWI 100%). Specificity was 100% for both sequences.

A postoperative irradiation following the Stupp protocol [11] was already started in 16 patients at the time of follow-up.

The mean diagnostic confidence (on visual analog scale 0–10, three neuroradiologists) for the detection of infarct lesions was 64 ± 28% for EPI DWI (significant difference, paired t‑test, p < 0.001), and 96 ± 17% for STEAM DWI and for residual tumors 89 ± 23% and 95 ± 14%, respectively (no significant difference, paired t‑test, p = 0.099).

Overall Fleiss’ kappa for the counting of infarct lesions (three raters) was 0.60 (moderate agreement) for EPI DWI and 0.57 (moderate agreement) for STEAM DWI. We noticed more counted lesions in STEAM than in EPI DWI for each rater. The interrater differences were not primarily based on different visual detections, but on the fact that two or three directly adjacent lesions were counted differently by the raters, either as one contiguous lesion or separately. Logically, Fleiss’ kappa for the difference of counted infarct lesions in EPI and STEAM DWI showed a better, substantial agreement of 0.70 (3 raters). As this quantitative assessment did not seem sufficient, we also carried out a volumetric analysis.

In 43 of 43 patients, measurable infarcts were detected by both DWI sequences. The mean infarct volume per patient was 5.8 ± 5.5 cm³ (mean ± standard deviation, median 4.2 cm³, minimum 0.1 cm³, maximum 25 cm³) using STEAM DWI with b = 1000 s/mm² and ADC and 5.1 ± 5.5 cm³ (median 4.0 cm³, minimum 0 mm³, maximum 25 cm³) using EPI DWI with b = 1000 s/mm² and ADC.

In cases of no severe air (n = 15) in the resection cavity, both DWI showed similar results (STEAM 4.2 ± 2.7 cm³ vs. EPI DWI 4.0 ± 2.4 cm³). For the 28 cases with air artifacts, an average volume of 6.6 ± 5.9 cm³ was detected using STEAM DWI, and 5.8 ± 5.8 cm³ using EPI DWI. Table 1 displays the measured volumes.

Wilcoxon matched pairs test showed significant (p < 0.02) differences between both (EPI and STEAM DWI) volumes in patients with severe air artifacts in the resection cavity, but not in patients without such artifacts (p > 0.5). A graphic illustration of the results is given in Fig. 3.

Intracranial air was noticed in every patient. EPI DWI presented with corresponding artifacts in every case. In more detail, air was seen as frontal pneumocephalus in 39 patients, in the resection cavity in 33 cases and in the ventricular system in 16 cases. An illustrative case showing the benefits of STEAM DWI is demonstrated in Fig. 4.

More specifically, the air in the resection cavity was in the form of a small microbubble (AAR 1) in 5 cases. In 28 patients (24 patients AAR 2, 4 patients AAR 3), EPI DWI was distorted by air artifacts in the resection cavity, especially in very early postoperative controls (< 24 h). Table 2 illustrates the distribution of air and blood artifacts.

The mean sum (per rater) of detected infarct lesions using EPI DWI was 89, while an average total of 120 lesions was identified via STEAM DWI, as detailed shown in Table 3. It is not surprising that more lesions were found even in the group without air artifacts, as there were of course susceptibility artifacts due to blood deposits in this group as well. Figure 5 shows an example of a resected high-grade glioma.

Blood artifacts were noticed in 17 patients using EPI DWI, but did not appear severe (9 cases moderate, 8 cases minor). Blood or air artifacts were not noticed using STEAM DWI. In 14 patients, neither severe air nor blood artifacts were noticed in the operation area using EPI DWI. In these cases, postoperative infarctions were equally detected with both sequences.

In general, the contribution of DWI alone to the quality of postoperative imaging is multifactorial, subjective and hard to measure. Of particular importance, however, is the advantage of being able to safely rule out infarction in symptomatic patients, as shown in Fig. 5.

In 3 of 25 patients with glioblastoma, we detected T2-hyperintense, contrast-enhanced tissue and correlating distorting susceptibility artifacts on EPI DWI. In these cases, STEAM DWI helped to reliably differentiate between a contrast-enhancing residual tumor and a subacute infarct (see Fig. 6 for an example).

In general, we expect an even higher benefit of STEAM DWI in early postoperative MRI for low-grade gliomas, which could not be finally shown in this study due to the small number of cases and only minor air inclusions in these patients.

In the cases with intraventricular air STEAM DWI showed better anatomic visualization of the adjacent brain areas and postoperative infarctions. In these cases it was possible to match infarct location and contrast enhancement to reliably reveal the residual tumor.

The only moderate interrater agreement is not surprising, as the detection and counting of ischemic changes after surgery is a complex task that requires three-dimensional imagination of contiguous regions and is difficult to objectify. Different voxel size, slice thickness and volume of the two DWIs may influence the measured volumes by partial volume effects [24]. If this were the case, the volumes of patients without air artifacts should also differ, but they do not.

A longitudinal follow-up of several months would increase the confidence of the diagnosis in the case of residual tumors. At the time of this evaluation, only 26 follow-up scans (gold standard) had been performed. As these cases already showed significant differences (EPI 74% vs. STEAM 100% sensitivity, both 100% specificity) and a clear superiority, we finished the analyses. An explicit image quality and sharpness rating of both sequences was not carried out because we suspected a strong correlation with diagnostic confidence.

Unfortunately, we could not test the sequence in intraoperative MRI, as this is not common practice in our hospital. Additional information can be expected in this set-up as well.

This study served as a pilot project to gain first experience with STEAM DWI in postoperative MRI. Further studies with larger sample size are required to corroborate these findings. Different classes of pathologies could also be evaluated, e.g., intracranial hematomas or aneurysms.