Front. Oncol. Frontiers in Oncology Front. Oncol. 2234-943X Frontiers Media S.A. 10.3389/fonc.2023.1217461 Oncology Erratum Erratum: Imaging biomarkers of glioblastoma treatment response: a systematic review and meta-analysis of recent machine learning studies Frontiers Production Office * Frontiers Media SA, Lausanne, Switzerland

Approved by: Frontiers Editorial Office, Frontiers Media SA, Switzerland

*Correspondence: Frontiers Production Office, production.office@frontiersin.org

24 05 2023 2023 13 1217461 05 05 2023 05 05 2023 Copyright © 2023 Frontiers Production Office 2023 Frontiers Production Office

This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

 An Erratum on Imaging biomarkers of glioblastoma treatment response: a systematic review and meta-analysis of recent machine learning studies by Booth TC, Grzeda M, Chelliah A, Roman A, Al Busaidi A, Dragos C, Shuaib H, Luis A, Mirchandani A, Alparslan B, Mansoor N, Lavrador J, Vergani F, Ashkan K, Modat M and Ourselin S (2022) Front. Oncol. 12:799662. doi: 10.3389/fonc.2022.799662 glioblastoma machine learning monitoring biomarkers meta-analysis artificial intelligence treatment response deep learning glioma section-in-acceptance Cancer Imaging and Image-directed Interventions

Due to a production error, there was an error in the published Table 1 . The 4th row of the table started at the second column instead of the first column, causing the contents of the last column to move to the next row, resulting in a formatting error. The corrected Table 1 appears below.

Studies using machine learning in the development of glioblastoma monitoring biomarkers.

Author Target condition Reference standard Dataset(s) Available demographic information Methodology Features selected Test set performance
a Kim J.Y. et al. (34) Early true progression or Early pseudoprogression Mixture of histopathology and imaging follow up Training = 61Testing = 34 T 1 C, FLAIR, DWI, DSC Training =age mean ± SD (range)58 ± 11 (34–83)male 38 (62%)Testing =age mean ± SD62 ± 12 male 25 (74%)Data from Korea Retrospective2 centers: 1 train & 1 external test set.LASSO feature selection with 10-fold CVLinear generalized model First-order,volume/shape, Second-order (texture), wavelet.ADC & CBV parameters included. Recall 0.71Specificity 0.90Precision 0.83BA 0.81F1 0.77AUC 0.85 (CI 0.71 – 0.99)
Kim J.Y. et al. (35) Early true progression or Early pseudoprogression Mixture of histopathology and imaging follow up Training = 59Testing = 24 T 1 C, FLAIR, DTI, DSC Training =age mean ± SD61 ± 11male 37 (63%)Testing =age mean ± SD59 ± 12male 9 (38%)Data from Korea Retrospective1 centerLASSO feature selection with 10-fold CVLinear generalized model First-order,Second-order(texture), wavelet.FA & CBV parameters included. Recall 0.80Specificity 0.63Precision 0.36BA 0.72F1 0.50AUC 0.67 (0.40 – 0.94)
Bacchi S. et al. (36) True progression or PTRE (HGG) Histopathology for progression and imaging follow up for pseudoprogression Training = 44Testing = 11 T 1 C, FLAIR, DWI Combined =age mean ± SD56 ± 10male 26 (47%)Data from Australia Retrospective1 center3D CNN & 5-fold CV CNN.FLAIR & DWIparameters Recall 1.00Specificity 0.60Precision 0.75BA 0.80F1 0.86AUC 0.80
Elshafeey N. et al. (37) True progression or b PTRE Histopathology Training = 98Testing = 7DSC, DCE Training =age mean ± SD50 ± 13male 14 (58%)No testing demographic informationData from USA Retrospective3 centersmRMR feature selection. 1 test.1) decision tree algorithm C5.02) SVMincluding LOO and 10-fold CV Ktrans & CBV parameters Insufficient published data to determine diagnostic performance(CV training results available recall 0.91; specificity 0.88)
Verma G. et al. (38) True progression or Pseudoprogression Mixture of histopathology and imaging follow up Training = 273D-EPSI Training =age mean ± SD64 ± 10male 14 (52%)Data from USA Retrospective1 centerMultivariate logistic regressionLOOCV Cho/NAA & Cho/Cr No test set(CV training results available recall 0.94; specificity 0.87)
Ismail M. et al. (39) True progression or Pseudoprogression Mixture of histopathology and imaging follow up Training = 59Testing = 46 T 1 C, T 2/FLAIR Training =age mean(range) 61 (26–74)male 39 (66%)Testing =age mean (range) 56 (25–76)male 30 (65%)Data from USA Retrospective2 centers: 1 train & 1 external test set.SVM & 4-fold CV Global & curvature shape Recall 1.00Specificity 0.67Precision 0.88BA 0.83F1 0.94
a Bani-Sadr A. et al. (40) True progression or Pseudoprogression Mixture of histopathology and imaging follow up Training = 52Testing = 24 T 1 C, FLAIRMGMT promoter status Combined =age mean ± SD58 ± 11male 45 (59%)Data from France Retrospective1 centerRandom Forest. Second-order features+/-MGMT promoter status Recall 0.94 (0.71 - 1.00)Specificity 0.38 (0.09 - 0.76)Precision 0.36BA 0.66F1 0.84AUC 0.77& non-MRI:Recall 0.80 (0.56 - 0.94)Specificity 0.75 (0.19 - 0.99)Precision 0.86BA 0.74F1 0.83AUC 0.85
Gao X.Y. et al. (41) True progression or PTRE (HGG) Mixture of histopathology and imaging follow up Training = 34Testing = 15(per lesion) T 1 C, FLAIR Combined =age mean ± SD51 ± 11male 14 (36%)(per patient)Data from China Retrospective2 centersSVM & 5-fold CV T 1 C, FLAIRsubtraction map parameters Recall 1.00Specificity 0.90Precision 0.83BA 0.95F1 0.91AUC 0.94 (0.78 – 1.00)
Jang B-S. et al. (42) True progression or Pseudoprogression Mixture of histopathology and imaging follow up Training = 59 Testing = 19 T 1 C & clinical features & IDH/MGMTpromoter status Training =age median (range)56 (22–77)male 41 (70%)Testing =age mean ± SD53 (28–75)male 10 (53%)Data from Korea Retrospective2 centers1 train & 1 external test set.CNN LSTM & 10-fold CV(compared to Random Forest) CNN T 1 C parameters+/-Age; Gender; MGMT status; IDH mutation; radiotherapy dose and fractions; follow-up interval Recall 0.64Specificity 0.50Precision 0.64BA 0.57F1 0.63AUC 0.69 & non-MRI:Recall 0.72Specificity 0.75Precision 0.80BA 0.74F1 0.76AUC 0.83
Li M. et al. (43) True progression or b PTRE Imaging follow up Training = 84DTI No demographic informationData from USA Retrospective.1 centerDC-AL GAN CNNwith SVM including 5 and 10 and 20-fold CV(compared to DCGAN, VGG, ResNet, and DenseNet) CNN. DTI No test set(CV training results only available: Recall 0.98Specificity 0.88AUC 0.95)
Akbari H. et al. (44) True progression or Pseudoprogression Histopathology Training = 40Testing = 23Testing = 20 T 1 C, T 2/FLAIR, DTI, DSC, DCE Combinedinternal =age mean (range)57 (33–82)male 38 (60%)No external demographic informationData from USA Retrospective2 centers. 1 train & test. 1 external test set.imagenet_vgg_f CNN SVM & LOOCV First-order, second-order (texture).CBV, PH, TR, T 1 C, T 2/FLAIRparameters included. Recall 0.70Specificity 0.80Precision 0.78BA 0.75F1 0.74AUC 0.80
Li X. et al. (45) Early True progression or early pseudoprogression (HGG) Mixture of histopathology and imaging follow up Training = 362 T 1 C, T 2, multi-voxel & single-voxel 1H-MRS, ASL Training = age mean (range) 50 (19–70)male 218 (60%)Data from China RetrospectiveGabor dictionary and sparse representation classifier (SRC) Sparse representations No test set(CV training results only available:Recall 0.97Specificity 0.83)
Manning P et al. (46) True progression or pseudoprogression Mixture of histopathology and imaging follow up Training = 32DSC, ASL Training = age mean ± SD56 ± 13male 22 (69%)Data from USA Retrospective1 centerLinear discriminant analysis & LOOCV CBF and CBV parameters included. No test set(CV training results only available:Recall 0.92Specificity 0.86 AUC 0.95)
Park J.E. et al., 2020 (47) Early True progression or early pseudoprogression Mixture of histopathology and imaging follow up Training = 53Testing = 33 T 1 C Training = age mean ± SD56 ± 11male 31 (59%)Testing = age mean ± SD62 ± 12male 25 (76%)Data from Korea Retrospective2 centers. 1 train & test. 1 external test set.Random Forest feature selection with 10-fold CV (Automated segmentation) First-order, volume/shape, Second-order (texture), wavelet parameters included. Recall 0.61Specificity 0.47Precision 0.58BA 0.54F1 0.59AUC 0.65 (0.46 – 0.84)
Lee J. et al. (48) True progression or b PTRE (HGG) Histopathology Training = 43 T 1, T 1 C, T 2, FLAIR, (subtractions: T 1 C - T 1, T 2- FLAIR) ADC parameters. Training =age mean ± SD (range)52 ± 13 (16–74)male 24 (56%)Data from USA Retrospective1 centerCNN-LSTM.3-fold CV CNN-LSTM parameters. No test set(CV training results only available:AUC 0.81 (0.72 - 0.88))
Kebir S. et al. (49) True progression or b PTRE Imaging follow up Training = 30Testing = 14O-(2[18F]-fluoroethyl)-L-tyrosine (FET) Combined = age mean ± SD (range)57 ± 11 (34-79)male 34 (77%)Data from Germany Retrospective1 centerLinear discriminant analysis.3-fold CV TBRmean TBRmax TTPmin parameters. Recall 1.00Specificity 0.80Precision 0.90BA 0.92F1 0.95AUC 0.93 (0.78 - 1.00)
Cluceru J. et al. (50) Early True progression or early pseudoprogression (HGG) Histopathology Training = 139DSC, MRSI, DWI, DTI Training = age median (range)52 (21–84)Male 83 (60%)Data from USAEthnicity:White 112 (80%)American Indian 1 (1%)Asian 6 (4%(Pacific Islander 2 (1%)Other 18 (13%) Retrospective1 centerMultivariate logistic regression.5-fold CV Cho, Cho/Cr, Cho/NAA & CBV parameters. No test set(CV training results only available:Recall 0.65 (0.33 - 0.96);Specificity 0.62 (0.21 - 1.00)AUC 0.69 (0.51 - 0.87))
Jang B.S. et al. (51) True progression or b PTRE Mixture of histopathology and imaging follow up (including PET) (i) (trained model = 78)testing = 104(ii) all training = 182 T 1 C & clinical, molecular, timings, radiotherapy data Testing = age median (range)55 (25-76)male 59 (67%)Data from Korea Retrospective(i) 6 centers1 external test set.CNN LSTM(ii) 7 centers1 training setCNN LSTM & 10-fold CV CNN T 1 C parameters and Age; Gender; MGMT status; IDH mutation; radiotherapy dose and fractions; follow-up interval (i) Insufficient published data to determine diagnostic performance(ii) No test set(CV training results available AUPRC 0.87)

Within publication some data appears mathematically discrepant.

Within publication discrepant or unclear information (e.g. interval after radiotherapy).

Unless otherwise stated, glioblastoma alone was analyzed.

PTRE, post-treatment related effects; HGG, high-grade glioma.

MRI sequences: T1 C, postcontrast T1-weighted; T2, T2-weighted; FLAIR, fluid-attenuated inversion recovery; DSC, dynamic susceptibility-weighted; DCE, dynamic contrast-enhanced; DWI, diffusion-weighted imaging; DTI, diffusor tensor imaging; ASL, arterial spin labelling; MRI parameters: ADC, apparent diffusion coefficient; FA, fractional anisotropy; TR, trace (DTI); CBV, cerebral blood volume; PH, peak height; Ktrans, volume transfer constant.

Magnetic resonance spectroscopy: 1H-MRS, 1H-magnetic resonance spectroscopy; 3D-EPSI, 3D echo planar spectroscopic imaging.

1H-MRS parameters: Cr, creatine; Cho, choline; NAA, N-acetyl aspartate.

Nuclear medicine: TBR, tumor-to-brain ratio; TTP, time-to-peak.

Molecular markers: MGMT, O6-methylguanine-DNA methyltransferase; IDH, isocitrate dehydrogenase.

Machine learning methodology: CV, cross validation; LOOCV, leave-one-out cross validation; SVM, support vector machine; CNN, convolutional neural network; LASSO, least absolute shrinkage and selection operator; LSTM, long short-term memory; mRMR, minimum redundancy and maximum relevance; VGG, Visual Geometry Group (algorithm); DCGAN, deep convolutional generative adversarial network; DC-AL GAN, DCGAN with AlexNet.

Statistical measures: CI, confidence intervals; BA, balanced accuracy; AUC, area under the receiver operator characteristic curve; AUPRC, area under the precision-recall curve.

The publisher apologizes for this error. The original version of this article has been updated.