Imaging Biomarkers of Glioblastoma Treatment Response: A Systematic Review and Meta-Analysis of Recent Machine Learning Studies

Abstract

Objective: Monitoring biomarkers using machine learning (ML) may determine glioblastoma treatment response. We systematically reviewed quality and performance accuracy of recently published studies.

Methods: Following Preferred Reporting Items for Systematic Reviews and Meta-Analysis: Diagnostic Test Accuracy, we extracted articles from MEDLINE, EMBASE and Cochrane Register between 09/2018-01/2021. Included study participants were adults with glioblastoma having undergone standard treatment (maximal resection, radiotherapy with concomitant and adjuvant temozolomide), and follow-up imaging to determine treatment response status (specifically, distinguishing progression/recurrence from progression/recurrence mimics, the target condition). Using Quality Assessment of Diagnostic Accuracy Studies Two/Checklist for Artificial Intelligence in Medical Imaging, we assessed bias risk and applicability concerns. We determined test set performance accuracy (sensitivity, specificity, precision, F1-score, balanced accuracy). We used a bivariate random-effect model to determine pooled sensitivity, specificity, area-under the receiver operator characteristic curve (ROC-AUC). Pooled measures of balanced accuracy, positive/negative likelihood ratios (PLR/NLR) and diagnostic odds ratio (DOR) were calculated. PROSPERO registered (CRD42021261965).

Results: Eighteen studies were included (1335/384 patients for training/testing respectively). Small patient numbers, high bias risk, applicability concerns (particularly confounding in reference standard and patient selection) and low level of evidence, allow limited conclusions from studies. Ten studies (10/18, 56%) included in meta-analysis gave 0.769 (0.649-0.858) sensitivity [pooled (95% CI)]; 0.648 (0.749-0.532) specificity; 0.706 (0.623-0.779) balanced accuracy; 2.220 (1.560-3.140) PLR; 0.366 (0.213-0.572) NLR; 6.670 (2.800-13.500) DOR; 0.765 ROC-AUC.

Conclusion: ML models using MRI features to distinguish between progression and mimics appear to demonstrate good diagnostic performance. However, study quality and design require improvement.

Keywords: artificial intelligence; deep learning; glioblastoma; glioma; machine learning; meta-analysis; monitoring biomarkers; treatment response.

Figure 1

Longitudinal series of MRI images in two patients (A, B) with glioblastoma, IDH-wildtype. All images are axial T₁-weighted after contrast administration. Images (Aa–Ad) demonstrate tumor progression. (Aa) Pre-operative MRI of a glioblastoma in the occipital lobe. (Ab) Post-operative MRI five days after resection; there is no contrast enhancement therefore no identifiable residual tumor. (Ac) The patient underwent a standard care regimen of radiotherapy and temozolomide. A new enhancing lesion at the inferior margin of the post-operative cavity was identified on MRI at three months after radiotherapy completion. (Ad) The enhancing lesion continued to increase in size three months later and was confirmed to represent tumor recurrence after repeat surgery. Images (Ba–Bd) demonstrate pseudoprogression. (Ba) Pre-operative MRI of a glioblastoma in the insula lobe. (Bb) Post-operative MRI at 24 hours after surgery; post-operative blood products are present but there is no contrast enhancement therefore no identifiable residual tumor. (Bc) The patient underwent a standard care regimen of radiotherapy and temozolomide. A new rim-enhancing lesion was present on MRI at five months after radiotherapy completion. (Bd) Follow-up MRI at monthly intervals showed a gradual reduction in the size of the rim-enhancing lesion without any change in the standard care regimen of radiotherapy and temozolomide or corticosteroid use. The image shown here is the MRI four months later.

Figure 2

Flow diagram of search strategy.

Figure 3

Forest plots showing sensitivity and specificity.

Figure 4

Summary Receiver Operator Characteristic Curve (SROC) of diagnostic performance accuracy. The summary point estimate and surrounding 95% confidence region is shown.