Preoperative prediction of diffuse glioma type and grade in adults: a gadolinium-free MRI-based decision tree

Abstract

Objectives: To develop a gadolinium-free MRI-based diagnosis prediction decision tree (DPDT) for adult-type diffuse gliomas and to assess the added value of gadolinium-based contrast agent (GBCA) enhanced images.

Materials and methods: This study included preoperative grade 2-4 adult-type diffuse gliomas (World Health Organization 2021) scanned between 2010 and 2021. The DPDT, incorporating eleven GBCA-free MRI features, was developed using 18% of the dataset based on consensus readings. Diagnosis predictions involved grade (grade 2 vs. grade 3/4) and molecular status (isocitrate dehydrogenase (IDH) and 1p/19q). GBCA-free diagnosis was predicted using DPDT, while GBCA-enhanced diagnosis included post-contrast images. The accuracy of these predictions was assessed by three raters with varying experience levels in neuroradiology using the test dataset. Agreement analyses were applied to evaluate the prediction performance/reproducibility.

Results: The test dataset included 303 patients (age (SD): 56.7 (14.2) years, female/male: 114/189, low-grade/high-grade: 54/249, IDH-mutant/wildtype: 82/221, 1p/19q-codeleted/intact: 34/269). Per-rater GBCA-free predictions achieved ≥ 0.85 (95%-CI: 0.80-0.88) accuracy for grade and ≥ 0.75 (95%-CI: 0.70-0.80) for molecular status, while GBCA-enhanced predictions reached ≥ 0.87 (95%-CI: 0.82-0.90) and ≥ 0.77 (95%-CI: 0.71-0.81), respectively. No accuracy difference was observed between GBCA-free and GBCA-enhanced predictions. Group inter-rater agreement was moderate for GBCA-free (0.56 (95%-CI: 0.46-0.66)) and substantial for GBCA-enhanced grade prediction (0.68 (95%-CI: 0.58-0.78), p = 0.008), while substantial for both GBCA-free (0.75 (95%-CI: 0.69-0.80) and GBCA-enhanced (0.77 (95%-CI: 0.71-0.82), p = 0.51) molecular status predictions.

Conclusion: The proposed GBCA-free diagnosis prediction decision tree performed well, with GBCA-enhanced images adding little to the preoperative diagnostic accuracy of adult-type diffuse gliomas.

Key points: Question Given health and environmental concerns, is there a gadolinium-free imaging protocol to preoperatively evaluate gliomas comparable to the gadolinium-enhanced standard practice? Findings The proposed gadolinium-free diagnosis prediction decision tree for adult-type diffuse gliomas performed well, and gadolinium-enhanced MRI demonstrated only limited improvement in diagnostic accuracy. Clinical relevance Even inexperienced raters effectively classified adult-type diffuse gliomas using the gadolinium-free diagnosis prediction decision tree, which, until further validation, can be used alongside gadolinium-enhanced images to respect standard practice, despite this study showing that gadolinium-enhanced images hardly improved diagnostic accuracy.

Keywords: Brain neoplasms; Gadolinium; Glioma; Isocitrate dehydrogenase; Magnetic resonance imaging.

Fig. 1

Diagnosis prediction decision tree (DPDT) based on GBCA-free MRI sequences. Flow chart describes DPDT for adult-type diffuse gliomas encompassing seven VASARI (necrosis, diffusion, hemorrhage, non-enhancing tumor margin, calvarial remodeling, cysts, proportion of edema) and four non-VASARI (T2-FLAIR mismatch sign, T2 signal homogeneity, calcification, midline shift) imaging features guiding the histomolecular diagnosis decision. GBCA, gadolinium-based contrast agent; VASARI, Visually AcceSAble Rembrandt Images

Fig. 2

Patient enrollment diagram. The flow chart depicts the patients included and excluded in this study. IDH, isocitrate dehydrogenase

Fig. 3

Per-rater prediction performance of histomolecular diagnosis of adult-type diffuse gliomas using GBCA-free vs. GBCA-enhanced scans. Color bar charts show the prediction performance, including accuracy, sensitivity, and specificity levels in predicting tumor grade, IDH mutation, and 1p/19q-codeletion status using GBCA-free and GBCA-enhanced scans per rater (rater 1 = green bar, rater 2 = orange bar, rater 3 = blue bar). Comparison of GBCA-free and GBCA-enhanced predictions revealed no significant differences (p > 0.05) except for rater 1’s sensitivity in histopathological grade prediction (red star, p = 0.006). GBCA, gadolinium-based contrast agent

Fig. 4

Group inter-rater agreement in histomolecular diagnosis prediction of adult-type diffuse gliomas using GBCA-free vs. GBCA-enhanced scans. The color box plot shows inter-rater agreement in predicting tumor grade (low-grade: grade 2 vs. high-grade: grade 3/4) and molecular status (astrocytoma, IDH-mutant vs. oligodendroglioma, IDH-mutant and 1p/19q-codeleted vs. glioblastoma, IDH-wildtype) among all raters. Green bars depict the results based on the evaluation of GBCA-free scans (only pre-contrast sequences: pre-contrast T1-weighted, T2-weighted, FLAIR, DWI/ADC, and SWI (if present)), and red bars show the results of the evaluation using GBCA-enhanced scans (pre- and post-contrast sequences: pre-contrast T1-weighted, T2-weighted, FLAIR, DWI/ADC, SWI (if present) + post-contrast T1-weighted). Comparison of agreements between GBCA-free and GBCA-enhanced predictions was significant for tumor grade (red star, p = 0.008), while it was insignificant for molecular marker (p > 0.05). Note: The interpretation of agreement values was as follows: 0.01–0.20, slight; 0.21–0.40, fair; 0.41–0.60, moderate; 0.61–0.80, substantial; and 0.81–0.99, almost perfect. GBCA, gadolinium-based contrast agent; IDH, isocitrate dehydrogenase

Fig. 5

Intra-rater inter-group agreement in histomolecular diagnosis prediction of adult-type diffuse gliomas. The color box plot describes intra-rater inter-group agreement comparing GBCA-free and GBCA-enhanced histomolecular diagnosis predictions of each rater (rater 1 = green bar, rater 2 = orange bar, rater 3 = blue bar). Diagnosis predictions include tumor grade (low-grade: grade 2 vs. high-grade: grade 3/4) and molecular status (astrocytoma, IDH-mutant vs. oligodendroglioma, IDH-mutant and 1p/19q-codeleted vs. glioblastoma, IDH-wildtype) evaluations. Red triangles indicate prevalence-adjusted and bias-adjusted kappa (PABAK) values that compensate for the possible influence of dataset diagnosis imbalances. Comparison of agreements among all raters revealed significant differences for both tumor grade and molecular marker (p < 0.001). Note: The interpretation of agreement values was as follows: 0.01–0.20, slight; 0.21–0.40, fair; 0.41–0.60, moderate; 0.61–0.80, substantial; and 0.81–0.99, almost perfect. GBCA, gadolinium-based contrast agent; IDH, isocitrate dehydrogenase

Fig. 6

Group inter-rater agreement in evaluating imaging features included in the Diagnosis Prediction Decision tree (DPDT) for adult-type diffuse gliomas. The color box plot shows inter-rater agreement in evaluating single DPDT imaging features using GBCA-free or GBCA-enhanced scans among all raters. Green bars depict the results based on the assessment of GBCA-free scans (only pre-contrast sequences: pre-contrast T1-weighted, T2-weighted, FLAIR, DWI/ADC, and SWI (if present)), and red bars show the results of the evaluation using GBCA-enhanced scans (pre- and post-contrast sequences: pre-contrast T1-weighted, T2-weighted, FLAIR, DWI/ADC, SWI (if present) + post-contrast T1-weighted). Comparison of agreements between GBCA-free and GBCA-enhanced assessments was insignificant (p > 0.05) except for hemorrhage and midline shift (red stars, p = 0.02 and p = 0.04, respectively). Note: The interpretation of agreement values was as follows: 0.01–0.20, slight; 0.21–0.40, fair; 0.41–0.60, moderate; 0.61–0.80, substantial; and 0.81–0.99, almost perfect. GBCA, gadolinium-based contrast agent