The Impact of MRI Features and Observer Confidence on the Treatment Decision-Making for Patients with Untreated Glioma

A Comparative and Summative Study of Radiomics-based Overall Survival Prediction in Glioblastoma Patients

PURPOSE

This study aimed to assess different machine learning models based on radiomic features, Visually Accessible Rembrandt Images features and clinical characteristics in overall survival prediction of glioblastoma and to identify the reproducible features.

MATERIALS AND METHODS

Patients with preoperative magnetic resonance scans were allocated into 3 data sets. The Least Absolute Shrinkage and Selection Operator was used for feature selection. The prediction models were built by random survival forest (RSF) and Cox regression. C-index and integrated Brier scores were calculated to compare model performances.

RESULTS

Patients with cortical involvement had shorter survival times in the training set (P = 0.006). Random survival forest showed higher C-index than Cox, and the RSF model based on the radiomic features was the best one (testing set: C-index = 0.935 ± 0.023). Ten reproducible radiomic features were summarized.

CONCLUSIONS

The RSF model based on radiomic features had promising potential in predicting overall survival of glioblastoma. Ten reproducible features were identified.

Reproducible imaging-based prediction of molecular subtype and risk stratification of gliomas across different experience levels using a structured reporting system

OBJECTIVES

To determine reproducible MRI parameters predictive of molecular subtype and risk stratification in glioma and develop a structured reporting system.

METHODS

All study patients were initially diagnosed with glioma, 141 from the Cancer Genome Atlas and 131 from our tertiary institution, as training and validation sets, respectively. Images were analyzed by three neuroradiologists with 1-7 years of experience. MRI features including contrast enhancement pattern, necrosis, margin, edema, T2/FLAIR mismatch, internal cyst, and cerebral blood volume higher than normal cortex were reported using a structured reporting system. The pathology was stratified into five risk types: (1) oligodendroglioma, isocitrate dehydrogenase [IDH]-mutant, 1p19q co-deleted; (2) diffuse astrocytoma, IDH-mutant, grade II-III; (3) glioblastoma, IDH-mutant, grade IV; (4) diffuse astrocytoma, IDH-wild, grade II-III; and (5) glioblastoma, IDH-wild, grade IV. Significant predictors were selected using multivariate logistic regression, and diagnostic performance was tested using a validation set.

RESULTS

Reproducible imaging parameters exhibiting > 50% agreement across readers included the presence of necrosis, T2/FLAIR mismatch, internal cyst, and predominant contrast enhancement. In the validation set, prediction of risk type 5 exhibited the highest diagnostic performance with AUCs of 0.92 (reader 1) and 0.93 (reader 2) with predominant enhancement, followed by risk type 2 with AUCs of 0.95 and 0.95 with T2/FLAIR mismatch sign and no necrosis, and risk type 1 with AUCs of 0.84 and 0.83 with internal cyst or necrosis. Risk types 3 and 4 were difficult to visually predict.

CONCLUSIONS

Imaging parameters with high reproducibility enabling prediction of IDH-wild-type glioblastoma, IDH-mutant/1p19q co-deletion oligodendroglioma, and IDH-mutant diffuse astrocytoma were identified.

KEY POINTS

• Reproducible MRI parameters for determining molecular subtypes of glioma included the presence of necrosis, T2/FLAIR mismatch, internal cyst, and predominant contrast enhancement. • IDH-wild type glioblastoma, IDH-mutant/1p19q co-deletion oligodendroglioma, and IDH-mutant low-grade astrocytoma were identified using MRI parameters with high inter-reader reproducibility. • Identification of IDH-wild type low-grade glioma and IDH-mutant glioblastoma was difficult by visual analysis.