Diagnostic performance of clinical properties and conventional magnetic resonance imaging for determining the IDH1 mutation status in glioblastoma: a retrospective study

Participants and data collection

A total of 221 consecutive patients were screened in this retrospective study; all were admitted to our department for treatment and were pathologically diagnosed with gliomas between January 1, 2016 and December 31, 2017. The Medical Ethic Committee of Chinese PLA General Hospital had approved this study. The study inclusion criteria were as follows: (1) patients who were admitted to the hospital with complete clinical information, including age, gender, and imaging data; (2) patients who underwent routine preoperative T1- and T2-weighted imaging, as well as contrast-enhanced T1(T1+C)—and T2-fluid attenuated inversion recovery (FLAIR) imaging; (3) patients who had a pathological diagnosis of GBM after surgical resection, with a clear molecular pathological classification and grading according to the 2016 WHO standard; (4) and patients who had a first-time diagnosis without a history of surgery, radiotherapy, or chemotherapy. The study exclusion criteria were as follows: (1) patients who had incomplete or low quality preoperative MRI data before surgery; and (2) patients whose images were poorly registered during image processing.

Of the 221 patients assessed for study inclusion, 59 were pathologically diagnosed with GBM, nine of these were further excluded in accordance with the above exclusion criteria; seven because of incomplete and/or low quality MR images, one because of confusing immunohistochemical findings, and one because of poor image registration during image processing. Eventually 50 patients including 35 men and 15 women (average age, 48.8 years) were enrolled. According to the presence or absence of the IDH1 mutation, the patients were divided into IDH1-mutant and IDH1-wild type groups for further analyses.

Image acquisition

Anatomical imaging sequences included T1-weighted, T2-weighted, T2-FLAIR, and T1+C images, all of which were acquired using a 1.5-T high-field-strength superconducting magnet in an iMRI system diagnostic room (Espree, Siemens, Germany) (Chen et al., 2012). Examination parameters for each were as follows: for T1, repetition time/echo time (TR/TE) 2,020/4.38 ms; layer thickness, one mm; field of view (FOV), 250 × 250 mm; sequence duration, 5 min 30 s; image resolution, 1 × 1 × 1 mm; for T2, TR/TE, 4,000/97 ms; layer thickness, three mm; FOV, 230 × 230 mm; sequence duration, 2 min; image resolution, 0.7 × 0.7 × 4 mm; and for T2-FLAIR, TR/TE, 9,000/84 ms; layer thickness, five mm; sequence duration, 4 min; image resolution, 0.45 × 0.45 × 5.5; For T1 plus; C imaging, the parameters were identical to those used for T1, with gadopentetate dimeglumine as the contrast agent.

Imaging registration, tumor segmentation, and volume calculation

Patient MRI data were imported into 3DSlicer software (www.slicer.org) in the DICOM format. Sequence resampling was performed using the Resample Scalar Volume module in order to ensure equivalent layer thickness and voxel numbers by referencing the T2-FLAIR sequence (the FLAIR sequence displays tumors well and is thus helpful for tumor segmentation) and setting each voxel to a fixed volume of 1 × 1 × 5.235 mm. The BRAINS module of the 3DSlicer software was used for linear registration of rigid bodies between sequences. Further details regarding the procedure can be found on the following website: https://www.slicer.org/wiki/Documentation/Nightly/Registration/RegistrationLibrary.

Whole tumors were segmented in the Editor module according to the FLAIR images. Segmented tumor images were referred to as Label-1, representative of the whole tumor. Contrast-enhanced regions of the tumor were segmented using the T1+C sequence and referred to as Label-2. Using the software’s Add Scalar Volume module, the whole tumor (Label-1) was superimposed onto the tumor-core region (Label-2) to obtain a new image. This was referred to as Label-3, and showed both the tumor-edema region and the tumor-core region. The volumes of the tumor-edema region and the tumor-core regions were separately calculated using the data statistic module Label Statistic in 3DSlicer software. Figure 1 depicts this procedure.

Imaging information evaluation

Basic patient data, including age, sex, and the pathological diagnosis, were collected. Two deputy chief physicians of neuroradiology, who were blinded to all patient’s pathological background recorded and analyzed the location and laterality of all lesions, and determined the presense of lesion enhancement and cystic necrosis. Basic patient information, imaging data, and tumor volume statistics were integrated. Disagreements or controversies regarding the imaging data were resolved though communication and consultation until a consensus was reached.

Pathological examination

All pathological specimens were obtained by surgical resection and diagnosed by neuropathologists in the Department of Pathology. Diagnostic criteria were based on the 2016 World Health Organization classification of tumors of the central nervous system (Louis et al., 2016). The pathological specimens were subjected to routine staining and immunohistochemical analyses, and the IDH1 mutation status was confirmed by IDH1 R132 antibody (internal clone H09; Dianova, Hamburg, Germany).

Statistical methods

Data processing and analysis was performed using the EmpowerStats software (www.empowerstats.com, X&Y Solutions, Inc., Boston, MA, USA). Graphs were generated using GraphPad Prism 5 (Version 5.01, GraphPad, La Jolla, CA, USA). Differences between the IDH-mutant and IDH-wild type groups were statistically evaluated, with continuous variables subjected to the Kruskal–Wallis rank sum test and categorical variables subjected to the chi-square test. If a count variable had a theoretical value of <10, Fisher’s exact probability test was used. For extremely complex analyses, such as the analysis of predictive equations in the diagnostic model, the advanced diagnostic module of the EmpowerStats software was employed. In cases of statistical differences, receiver operating characteristic (ROC) curves were generated, and the sensitivity, specificity, positive predictive value, and negative predictive value were calculated. All tests were two-sided hypothesis tests, with the test level set to α = 0.05.

Participant demographics

Demographic information and characteristics for the IDH1-mutant and IDH1-wild type groups were separately recorded and analyzed (Table 1). There was a significant difference in age, the tumor location, and the overall tumor volume between the two groups. There were no significant differences between the two groups with regard to sex, tumor laterality, the presence of lesion enhancement, the presence of cystic change, or the tumor-edema and tumor-core volume. Moreover, there was a significant difference in terms of tumor distribution in the cerebral lobes: IDH1-mutant tumors were mainly located in the temporal (50%) and frontal (40%) lobes, whereas IDH1-wild type tumors were most often located in the temporal lobe (47.5%).

Relationship between patient age and IDH1 mutation status

There was a statistical difference in age between the two groups: the IDH1-mutant patients were younger than the IDH1-wild type patients (34.8 ± 10.8 vs 52.3 ± 11.6 years) (Fig. 2A). Curve fitting analysis of the patient age compared with the IDH1 mutation status revealed that the probability of an IDH1-wild type status in a GBM patient gradually increased with an increase in age (Fig. 2B).

Relationship between tumor volume and IDH1 mutation status

There was no significant difference in the tumor-edema and tumor-core volumes between the two genotype groups. However, there was a significant difference in the overall tumor volume between the two groups: IDH1-mutant patients exhibited larger overall tumor volumes than did IDH1-wild type patients (32.51 ± 10.63 vs 25.11 ± 10.44 cm³; Fig. 3A). Curve fitting analysis of the tumor size compared with the IDH1 mutation status revealed that the probability of an IDH1-mutant status in a GBM patient gradually increased with an increase in the tumor volume (Fig. 3B).

ROC curve analyses

Receiver operating characteristic curve analyses were performed for age and the tumor volume, the two parameters that significantly differed between the two groups (Fig. 4). When the age threshold was set to 42.5 years, the sensitivity and specificity of our model were 80% and 82%, respectively, and the AUC was 0.87. When the tumor volume threshold was set to 33.6 cm³, the sensitivity and specificity were 70% and 80%, respectively, and the AUC was 0.80 (Table 2). The following logistic regression model was used to establish a diagnostic prediction model that incorporated both age and the tumor volume for joint diagnosis: logit(IDH1) = 2.16674 − 0.13007 × Age + 0.07703 × Tumor volume. When the threshold was set to −0.89, the sensitivity and specificity were 70% and 93%, respectively, and the AUC was 0.88 (Table 2).