Looking through the imaging perspective: the importance of imaging necrosis in glioma diagnosis and prognostic prediction – single centre experience

Patients with a primary diagnosis of glioma (June 2013–May 2021) were retrospectively included. Inclusion and exclusion criteria are presented in Supplementary Figure 1. Clinical information of patients was retrieved from the electronic medical records, and follow-up information was obtained through clinical interviews. Follow-up survival data were available until May 31, 2021. Overall survival (OS) was calculated from the initial surgery date to the date of death, or the date of the last follow-up visit if the patient was alive or lost to follow-up.

This retrospective analysis was in accordance with the ethical standards of the institutional and national research committee and was approved by the ethics committee of our institution ([2021]209). The requirement for written informed consent was waived due to the retrospective nature of this study.

Participants underwent conventional (T1/T2-weighted images [T1WI/T2WI], T2-weighted fluidattenuated inversion recovery [T2WI-FLAIR] and sagittal view of contrast-enhanced three-dimensional T1 MPRAGE images) and DCE-MRI imaging using a 3.0T MR system (Magnetom Verio, Siemens Medical Solutions, Erlangen, Germany) with a 64-channel head-neck coil. The parameter details of the conventional MRI and the DCE-MRI were elaborated in Supplementary Appendix 1.

All DCE-MRI data were transferred to the post-processing workstation (detailed in Supplementary Appendix 2). Pharmacokinetic parameters, including the transfer constant (ktrans), extravascular extracellular volume fraction (ve), rate constant (kep = ktrans/ve), and initial area under the curve in the first 60 s (iauc), were automatically generated. Regions of interest (ROIs) were selected across three consecutive maximum tumor parenchyma slices. At each slice, one ROI was put in tumor parenchyma (hereafter termed “tumor”), according to T2WI-FLAIR, and enhanced T1WI, avoiding necrosis, cystic, and vessel areas. Another two approximate 2-mm-diameter ROIs were put in tumor peripheral zones (hereafter termed “edema”, within a 1-cm distance from the outer enhancing tumor margin) and contralateral normal-appearing brain tissues (hereafter termed “control”) (Figure 1). The mean values of each DCE-MRI metric was recorded.

FIGURE 1.

As mentioned in the introduction, examples of imaging necrosis, defined as a region within the tumor that does not enhance or shows markedly diminished enhancement, high signal intensity on T2WI, low signal intensity on T1WI, and an irregular border, are shown in Figure 2 and Supplementary Figure 2. Two experienced radiologists reviewed all conventional MRIs. Then they determined whether there was Im_necrosis by consensus. One of these two experienced radiologists and a third radiologist repeatedly assessed 68 cases after the initial assessment to assess the inter-observer agreement. The assessed images were randomized within each type of pathology, and the observers were blinded to the clinical and pathological information and thoroughly acquainted with the criteria.

FIGURE 2.

Pa_necrosis was defined according to pathological reports provided by the Pathology department of our hospital, if available. The status of 1p19q codeletion, EGFR amplification, chr7 gain/10 loss (+7/−10), and CDKN2A/B homozygous deletion were determined by fluorescence in situ hybridization with a specific probe. IDH mutation was determined by high-throughput sequencing, including IDH1 and IDH2 mutations. The pathological diagnosis and grading of gliomas were reassigned according to the 2021 WHO CNS classification (Supplementary Figure 3).^4,14,15

Statistical Analysis Data were analyzed using IBM SPSS Statistics 26 software, the SPSSAU data scientific analysis platform (https://spssau.com/), and the R programming language (version 4.1.2, The R Foundation for Statistical Computing). Normally distributed continuous variables were compared using unpaired t-tests, whereas non-parametric tests were used for non-normally distributed variables. Descriptive data are expressed as mean ± SD, except where otherwise stated. Unpaired t-tests, non-parametric tests, and chi-squared tests were used to compare differences between parameters. Receiver Operating Characteristics (ROC) curves were used to evaluate diagnostic efficacy. Simple kappa was calculated to assess the consistency of different diagnoses and inter-observer agreement. Kaplan–Meier survival analysis was used to analyze survival data. Hazard ratios (HR) were estimated according to the Cox proportional hazard method. A two-sided p value < 0.05 was considered significant. Detailed statistical methods are shown in Supplementary Appendix 3.

We initially identified 150 eligible patients (median age = 46 years, range 21–79 years), and 104 (69.33%) were male (Table 1). All the diagnoses assigned to the patients according to the latest integrated histomolecular classification criterion were presented in Supplementary Figure 3 and Supplementary Table 1. Pa_necrosis was identified in 70/76 of highgrade gliomas (HGGs, CNS WHO grade 4) and 3/43 of low-grade gliomas (LGGs, CNS WHO grade 2 and 3) which were oligodendrogliomas, IDHmutant and 1p/19q-deleted, while Im_necrosis was identified in 70/76 of HGGs and 12/43 of LGGs.

There was 1/77 HGG without enhancement but with positive status of EGFR amplification, thus diagnosed as glioblastomas, IDH-wildtype, while there were 32/45 LGGs with enhancement diagnosed as oligodendrogliomas, IDH-mutant and 1p/19q-deleted (n = 15) and astrocytoma, IDH-mutant (n = 17). And most HGGs were manifested as ring enhancement, and most LGGs had patchy and punctate enhancement. All the clinical information was presented in Table 1 and Supplementary Appendix 4.

In this study, the following four groups were determined: Im+Pa_necrosis group (representing patients with both Im_necrosis and Pa_necrosis, n = 74), no_necrosis group (representing patients without Im_necrosis nor Pa_necrosis, n = 28), Only Im_necrosis group (representing patients with Im_necrosis but without Pa_necrosis, n = 7), and Only Pa_necrosis group (representing patients with Pa_necrosis but without Im_necrosis, n = 4) groups. Detailed clinical, imaging and paychological information of Only Im_necrosis group and Only Pa_necrosis group were shown in Table 2. We found strong agreement between Im_necrosis and Pa_necrosis (Kappa = 0.767, p < 0.001, 95%CI: 0.637–0.897).

Besides, there was strong inter-observer agreement in identifying imaging necrosis (Kappa = 0.668, p < 0.001, 95%CI: 0.489–0.846). And the spotlike, dotted, long-strip, long tubular, and fissural enhancements (Figure 3) which were easily misdiagnosed as imaging necrosis should be avoided.

FIGURE 3.

Most HGGs (85.37%) were found to have Im_necrosis, while most LGGs (83.78%) were without Im_necrosis. There were 4/30 WHO grade 2 patients with Im_necrosis. Of those, two were diagnosed as oligodendrogliomas, IDH-mutant and 1p/19q-deleted, and two as astrocytomas, IDH-mutant.

Significant differences in the presence of Im_necrosis with a large effect size were found between HGGs and LGGs and among different grades of gliomas (Table 1, p < 0.001). Cochran–Armitage tests showed an upward trend in Im_necrosis from lower to higher grades of gliomas (p < 0.001). Multiple comparisons with Bonferroni correction showed that the difference between WHO grades (any two grades) and Im_necrosis was significant (all p < 0.01).

There were significant correlations between the expression of other molecular markers such as IDH, 1p19q, and CDKN2A/B and the presence of Im_necrosis. According to Table 1, the proportion of IDH-wildtype, 1p19q-non-codeletion, or CDKN2A/B-positive cases with Im_necrosis was significantly higher than that of cases without Im_necrosis with a medium effect size, respectively (75.82% vs. 32.08%, 85.94% vs. 51.11%, 20% vs. 0, respectively). However, no significant correlation was found between Im_necrosis and EGFR amplification or +7/−10 cytogenetic signature (p > 0.05) (Table 1).

One-hundred and thirty patients were included in the final survival analysis. Compared with gliomas with Im_necrosis, patients without Im_necrosis had a significantly longer survival time (p < 0.001, Figure 4A). By reference to the gliomas with Pa_necrosis, patients without Pa_necrosis had a significantly longer survival time as well (p < 0.001, Figure 4B).

FIGURE 4.

The differences among the OS of Im+Pa_necrosis, no_necrosis, Only Im_necrosis, and Only Pa_necrosis groups were statistically significant (p < 0.01, Figure 4C). Further, after Bonferroni correction, there were significant differences between Im+Pa_necrosis and no_necrosis groups (p < 0.001). According to Figure 4C, the OS of the Only Pa_necrosis group (n = 2) was shorter than the OS of no_necrosis group (n = 28) and Only Im_necrosis group (n = 7). Between the two survival curves of no_necrosis and the Only Im_necrosis groups, there were marked crossovers, but within a certain period (time spanning about from 5-month to 40-month postoperatively), the OS of the Only Im_necrosis group was shorter than the OS of no_necrosis group.

Further, when added significant variables such as age, IDH, 1p19q, and Im_necrosis into the multivariate Cox proportional hazards regression analyses, only Imnecrosis (HR = 2.113, 95% CI: 1.015–4.402, p = 0.046) was significant and independently related to the patients’outcome, indicating that Im_necrosis is an independent and unfavourable prognostic factor.

Since pathology is the golden standard for necrosis diagnosis, we analyzed the associations with Pa_necrosis and DCE-MRI metrics. Most DCE-MRI metrics demonstrated a significant difference in identifying gliomas with Pa_necrosis with a very large effect size (Table 3). Kep was significantly higher for gliomas with Pa_necrosis than those without Pa_necrosis, while other DCE-MRI metrics showed the opposite trend. ROCs analysis showed that the Tumor-ve-Mean displayed the best diagnostic performance with the largest AUC of 0.891 (95%CI: 0.788–0.995, p < 0.0001), and the optimal cut-off point was 0.17 with a sensitivity of 96% and specificity of 83.3%.

Similarly, we performed the analysis regarding Im_necrosis (Table 3), and the Tumor-ve-Mean displayed the best diagnostic performance as well, with the most significant AUC of 0.929 (95%CI: 0.872–0.986, p < 0.0001) and the optimal cut-off point was 0.17 with a sensitivity of 89.2% and specificity of 91.9%.