Multi-Dynamic-Multi-Echo-based MRI for the Pre-Surgical Determination of Sellar Tumor Consistency: a Quantitative Approach for Predicting Lesion Resectability

The protocol of this prospective study was approved by the local ethics commission (Medical University of Vienna, Vienna, Austria) and performed in accordance with the Helsinki Declaration of 1975. All patients provided written, informed consent prior to MRI scanning and agreed to the scientific use of the acquired imaging data (EC-Nr.: 1549/2019).

Between January 2020 and January 2022, pre-surgical MRI was performed in 72 patients with a tumor in the pituitary region at the Department of Neuroradiology of a tertiary care hospital. All lesions have been morphologically delineated as pituitary macroadenomas based on MRI evaluation. Pituitary adenoma was histologically confirmed in 61 of the 72 patients (Table 1). Histological data were not available in 2 patients. Microadenoma patients were excluded from this investigation. Study sample characteristics are shown in Fig. 1 and Table 1.

MRI was performed using the institutional routine imaging protocol for pre-neurosurgical planning (Table 2). Pre-contrast, MDME-based sequences (coronal plane) were added to the protocol. In 24 patients, post-contrast, MDME-based T1-weighted data were also available. The imaging data were acquired on a Siemens MAGNETOM Vida (3 T) MR system. Via two repeated acquisition phases [phase a, a slice-selective saturation pulse (flip angle: 120°) was applied to saturate one slice; phase b, a series of slice-selective refocusing pulses (flip angle: 180°) and a slice-selective excitation pulse (flip angle: 90°) were applied to generate spin echoes for another slice], the MDME sequence derived information about the physical MR properties of the tissue (T1-/T2-relaxation parameters and spin density) [13, 14, 21]. Data acquisition was performed at two different echo times (24 ms and 107 ms) following the 90° pulses, which were applied at four different saturation recovery times throughout the MDME scan cycle (i.e., eight spin echo acquisitions per scan cycle). Quantitative maps for the determination of lesion-specific properties were generated using the MDME post-processing software “SyMRI” (Synthetic MR AB, Linköping, Sweden; Version 11.2.9). Generation of parametric maps was performed within approximately ten seconds (per imaging data set), after transfer of MDME-based acquisitions from the MR scanner to a separate workstation for quantitative analysis. Contrast-enhanced, MDME-based T1-weighted data were generated automatically and retrieved on a Picture Archiving and Communication System (PACS) workstation for qualitative pre-surgical assessment.

Prior to the analysis, a critical visual review of the acquired MR data was performed by one neuroradiologist (W.M.) with 15 years of experience in neuroimaging. Highly artifact-degraded acquisitions were excluded. The lesion-specific properties [T1R (ms); T2R (ms); PD (%)] were measured by manually placing regions of interest (ROIs) at three different slices of the tumor on pre-contrast, “SyMRI”-generated parametric maps (Fig. 2). Units for T1R (spin-lattice-relaxation), T2R (spin-spin-relaxation), and PD (number of nuclei in the area being imaged) were adopted as provided by the “SyMRI”-based default software settings. ROI positioning was performed by two independent raters (M.S.Y. and L.L., each with one year of experience in pituitary MRI), who were blinded to intraoperative resectability assessments. In four cases, only two ROIs were drawn due to the limited availability of slices of the lesion. The average values of the quantitative properties were automatically computed, based on each single voxel value within the drawn ROI. Mean values, based on the three ROI measurements in each lesion, were calculated and used for further analysis (Supplementary Table 1).

The assessment of the intraoperative tumor extractability, via an endoscopic, transnasal, transsphenoidal approach, was assessed by the attending neurosurgeons (A.M. and S.W.) during surgery. In contrast to soft-constituted adenomas, tumors composed of fibrotic components are considered difficult to resect [5]. Therefore, the tumors were divided into two groups: “hard-to-remove by aspiration” (hRAsp) and “easy-to-remove by aspiration” (eRAsp). eRAsp lesions were freely removable by aspiration and only minimal use of curettage was required. Tumors that were difficult to aspirate, and which made curettage and successive mechanical debulking necessary, were assessed as hRAsp [5, 22].

Qualitative assessments were performed on contrast-enhanced, MDME-based, VIBE-based, and MPRAGE-based T1-weighted data (coronal plane) (Fig. 3) by two independent senior neuroradiologists [J.F. (15 years of experience in neuroimaging) and W.M.]. To evaluate the pituitary region and parasellar anatomy, the following criteria were used [23]: the Knosp grade (for left/right cavernous sinus invasion) [23, 24]; parasellar invasiveness; normal gland position; detectability of the optic chiasm and left/right oculomotor nerve. The scoring criteria are explained in Supplementary Table 2 and Supplementary Fig. 1. Supplementary Fig. 2 shows the differences between pre- and post-contrast MDME-based acquisitions.

The statistical analyses were proposed and performed by one biomedical statistician with 30 years of experience (M.W.) using SPSS Statistics for Windows (Version 28.0; IBM Armonk, NY).

A p-value of p ≤ 0.05 was considered statistically significant. An intra-class correlation coefficient (ICC) analysis was used to assess the overall agreement between the mean values determined by both raters. ICC values ≥ 0.75 were considered strong agreements [25]. In case of strong correlations, the data based on the determinations performed by rater 1 were used for further analysis. A Mann-Whitney-U-test was performed to detect differences in T1R, T2R, and PD metrics between both groups (hRAsp vs. eRAsp). Receiver Operating Characteristic (ROC) curve analyses were performed to evaluate the predictive power of quantitative MRI metrics regarding the pre-surgical discrimination between hRAsp and eRAsp lesions. The Youden indices were calculated to determine optimal cutoff values for each quantitative metric.

The Cohen’s Kappa coefficient (κ) was used to detect: a) agreements between the observer’s qualitative assessments (on the basis of contrast-enhanced, MDME-based, VIBE-based, and MPRAGE-based T1-weighted data) (i.e., inter-rater agreement) and: b) concordances between the observer’s qualitative assessments based on the three different, post-contrast sequences (MDME vs. VIBE; MDME vs. MPRAGE; and VIBE vs. MPRAGE) (i.e., inter-sequence agreement). κ analyses were interpreted as proposed by Landis and Koch [26].

One of the 72 subjects was excluded from this study, due to the lack of coronal MDME sequence acquisitions. Furthermore, six of the 72 patients were excluded from the quantitative analysis of lesion consistency since they did not undergo a neurosurgical resection (Fig. 1). For the quantitative analysis, a total of 65 participants (mean age at data acquisition, 54 years ± 15) (female/male: 33/32) were enrolled in this study (Table 1). Based on the intraoperative assessment of tumor resectability, 24/65 (37%) tumors were classified as hRAsp, while 41/65 (63%) lesions were determined to be eRAsp. We found no evidence of differences between both groups in terms of age (unpaired t‑test: mean age, hRAsp: 57 years ± 13; eRAsp: 53 years ± 16; p = 0.22) or sex distribution (Pearson χ² test: hRAsp, female/male: 11/13; eRAsp, female/male: 22/19; p = 0.54).

In 24 of the 72 patients, contrast-enhanced, MDME-based T1-weighted sequence acquisitions were also available. One of the 24 patients with contrast-enhanced acquisitions had to be excluded from the qualitative analysis, due to severe motion-related artifacts that were visible only on the contrast-enhanced, MDME-based and VIBE-based T1-weighted MR imaging data. Thus, for the qualitative sub-analysis of the pituitary region/parasellar anatomy, a total of 23 participants (mean age at data acquisition, 52 years ± 18) (female/male: 12/11) were included in this study (Table 1).

The ICC analysis revealed strong agreements between T1R [ICC: 0.99 (95% CI: 0.98–0.99), p < 0.001], T2R [ICC: 0.99 (95% CI: 0.98–0.99), p < 0.001], and the PD metrics [ICC: 0.85 (95% CI: 0.75–0.91, p < 0.001)] as determined by both raters.

Significant differences were found in T1R (hRAsp: 1221.0 ms ± 211.9; eRAsp: 1500.2 ms ± 496.4; p = 0.003) and T2R metrics (hRAsp: 88.8 ms ± 14.5; eRAsp: 137.2 ms ± 166.6; p = 0.03) between lesions classified as hRAsp vs. eRAsp. No evidence of differences was found in tumor resectability based on lesion-specific PD metrics (hRAsp: 85.6% ± 5.2; eRAsp: 87.7% ± 4.5; p = 0.11) (Fig. 4).

The ROC analysis revealed significant results for T1R, with an area under the curve (AUC) of 0.72 (95% CI: 0.60–0.85), p = 0.003; and T2R, with an AUC of 0.66 (95% CI: 0.53–0.79), p = 0.03. There was no evidence of significance for PD metrics, with an AUC of 0.62 (95% CI: 0.48–0.76), p = 0.11 (Table 3).

For T1R, the sensitivity/specificity to predict easy tumor extractability was 78%/58% (Youden index: 0.364) at a cutoff value of 1248 ms. For T2R, the sensitivity/specificity to predict easy tumor resectability was 39%/96% (Youden index: 0.349) at a cutoff value of 110 ms. For PD, the sensitivity/specificity to predict easy tumor extractability was 76%/54% (Youden index: 0.298) at a cutoff value of 85% (Fig. 5).

Inter-rater agreement between raters was almost perfect (κ > 0.8) for 6/7 (MDME/MPRAGE) and 5/7 (VIBE) evaluated aspects (Supplementary Table 2 and Supplementary Fig. 1). Agreement between raters was lowest (κ < 0.8) for the detectability of the left oculomotor nerve (MDME/VIBE/MPRAGE) and the optic chiasm (VIBE).

Inter-sequence agreement was almost perfect in 5/7 (MDME vs. MPRAGE) and 4/7 (MDME vs. VIBE/VIBE vs. MPRAGE) evaluated aspects. Agreement between sequences was lowest for the detectability of the left/right oculomotor nerve (MDME vs. VIBE/MDME vs. MPRAGE/VIBE vs. MPRAGE) and the optic chiasm (MDME vs. VIBE/VIBE vs. MPRAGE). Detailed information about inter-rater and inter-sequence agreements are given in Tables 4 and 5 and Supplementary Table 2.

Several limitations require consideration. We included all pituitary tumors initially classified as macroadenomas based on MR imaging. Therefore, based on histology, various subtypes of macroadenomas were included. Furthermore, there were also other solitary tumor entities, mimicking macroadenomas, that were included in this study. The sample size for both quantitative and qualitative analyses was relatively small, which mandates the need for further studies to confirm our findings. Furthermore, there was a considerable delay between intravenous contrast agent administration and MDME-based sequence acquisitions for the qualitative sub-analysis. However, the acquired data proved sufficient to reliably study the applicability of MDME-based imaging data in a clinical setting. This study did not provide information on the feasibility of MDME-based imaging for the assessment of pituitary microadenomas since these require different imaging acquisition strategies [32]. Nonetheless, the relaxometric evaluation of pituitary microadenomas is of great interest and requires further consideration in the future.