Skip to main content

Table 2 Prediction of locoregional recurrence, treatment response, and survival

From: Applications of radiomics in precision diagnosis, prognostication and treatment planning of head and neck squamous cell carcinomas

Author, yearDataset: sample size, HNSCC typeImaging modalityTherapyOutcome, model, analysis type(Endpoint:) variables: metric: maximum performance a
Aerts et al.
2014 [18]
Test: 231, HNSCC
(HNSCC cohort used for validation only, training on 422 lung cancer primaries)
Pre-treatment contrast CTRT + chemotherapy or RT + chemotherapy + surgery or RT aloneOS,
multivariable Cox regression,
regression
Radiom:
test-CI 1:0.69
test-CI 2: 0.69
Radiom+Clin:
test-CI 1: 0.70
test-CI 2: 0.69
Clin:
test-CI 1: 0.69
test-CI 2: 0.66
Bogowicz et al.
2017 [78]
Train: 128, HNSCC
Test: 50, HNSCC
3-month post treatment FDG-PETRT + cisplatin / cetuximab or RT + cisplatin + cetuximabLC,
multivariable Cox regression,
regression
Radiom:
CV-CI: 0.74–0.76
test-CI: 0.71–0.73
(study evaluated reproducibility of identical features using two different software - performance range is reported)
Bogowicz et al.
2017 [25]
Train: 121, HNSCC
Test: 51, HNSCC
Pre-treatment FDG-PET, contrast CTRT + cisplatin / cetuximab or RT + cisplatin + cetuximabLC,
3 different regression methods,
regression
Radiom:
CV-CI: 0.77
test-CI: 0.73
Bogowicz et al.
2017 [50]
Train: 93, HNSCC
Test: 56, HNSCC
Pre-treatment contrast CTRT + cisplatin / cetuximabLC,
multivariable Cox regression,
regression
Radiom:
train-CI: 0.75
test-CI: 0.78
Clin:
train-CI: 0.79
test-CI: 0.73
Radiom + Clin:
train-CI: 0.80
test-CI: 0.76
Bogowicz et al.
2019 [79]
Train: 77, HNSCC
Test: 51, HNSCC
Pre-treatment contrast CTRT + cisplatin / cetuximab or RT + cisplatin + cetuximabLC and LRC,
multivariable Cox regression,
regression
LC:
Radiom:
CV-CI: 0.81
test-CI: 0.70
LRC:
Radiom:
CV-CI: 0.75
test-CI: 0.67
(addition of lymph node to primary tumor radiomics features was investigated –the best performance reported)
Bogowicz et al.
2019 [80]
Total: 45, HNSCCPre-treatment perfusion CTIMRT + cisplatin / cetuximab or IMRT + cisplatin + cetuximabLC,
multivariable Cox regression,
regression
Radiom:
CV-CI: 0.79
Clin:
CV-CI: 0.66
Cheng et al.
2013 [81]
Total: 70, OPSCCPre-treatment FDG-PETRT + platinum-based chemotherapy / cetuximab or RT aloneDSS, OS and PFS,
multivariate Cox regression,
regression
DSS:
Radiom:
HR: 0.28 (p = 0.001)
OS:
Radiom:
HR: 0.46 (p = 0.017)
PFS:
Radiom:
HR: 0.32 (p = 0.001)
(Only 1 radiomics feature was tested in multivariate Cox regression, along with clinicopathological and FDG-PET variables)
Cheng et al.
2015 [82]
Total: 88, OPSCCPre-treatment FDG-PETRT + chemotherapy / biotherapy or RT alonePFS and DSS,
multivariate Cox regression,
regression
PFS:
Radiom:
HR: 4.38 (p = 0.002)
DSS:
Radiom:
HR: 4.24 (p = 0.005)
(single radiomics features were tested in multivariate Cox regression, along with clinicopathological and FDG-PET variables)
Cozzi et al.
2019 [83]
Train: 70, HNSCC
Test: 40, HNSCC
Pre-treatment contrast CTRT + chemotherapy or RT + chemotherapy + induction-chemotherapyOS, PFS and LC,
multivariable Cox regression, regression
OS:
Radiom:
train-CI: 0.88
test-CI: 0.90
PFS:
Radiom:
train-CI: 0.72
test-CI: 0.80
LC:
Radiom:
train-CI: 0.72
test-CI: 0.80
Feliciani et al. 2018 [84]Total: 90, HNSCCPre-treatment FDG-PETIMRT + platinum-based chemotherapy w/ or w/o adjuvant / neoadjuvant chemotherapyPFS and OS,
multivariate Cox regression,
regression
PFS:
Radiom + Clin:
CV-CI: 0.76
Clin:
CV-CI: 0.65
OS:
Radiom + Clin:
CV-CI: 0.76
Clin:
CV-CI: 0.73
Folkert et al. 2017 [85]Train: 174, OPSCC
Test: 65, OPSCC
Pre-treatment FDG-PETRT + platinum-based chemotherapy / cetuximab / multidrug regimensACM, LF and DM,
multivariable logistic regression,
classification
ACM:
Radiom + Clin:
CV-AUC: 0.65
test-AUC: 0.60
LF:
Radiom + Clin:
CV-AUC: 0.73
test-AUC: 0.68
DM:
Radiom + Clin:
CV-AUC: 0.66
test-AUC: 0.65
Ger et al.
2019 [35]
Train 1: 377, HNSCC (CT)
Train 2: 345, HNSCC (PET)
Test 1: 349, HNSCC (CT)
Test 2: 341, HNSCC (PET)
Pre-treatment contrast CT, pre-treatment FDG-PET (separately analyzed)Not reported (definitive RT as part of treatment was inclusion criterion)OS,
multivariable Cox regression,
regression
Radiom:
test-AUC 1: 0.72 (CT)
test-AUC 2: 0.59 (PET)
Clin:
test-AUC 1: 0.73 (CT)
(AUC was calculated at 3 years post treatment, with patients with risk prediction > median assigned to the high-risk group)
Kuno et al. 2017 [86]Total: 62, HNSCCPre-treatment contrast CTIMRT + chemotherapy w/ or w/o induction chemotherapy or IMRT aloneLF,
multivariate Cox regression,
regression
Radiom:
HR: 3.75–8.61
(8 features were significant after adjusting for clinical variables; the HR range is reported above)
Leger et al. 2017 [30]Train: 213, HNSCC
Test: 80, HNSCC
Pre-treatment non-contrast CTRT + chemotherapyLRC and OS,
11 different ML algorithms,
regression
LRC:
Radiom:
test-CI: 0.71
OS:
Radiom:
test-CI: 0.64
Leijenaar et al. 2015 [87]Test: 542, OPSCC
(validation of radiomics signature by Aerts et al. [18])
Pre-treatment contrast CTIMRT + chemotherapy or IMRT aloneOS,
multivariable Cox regression,
regression
Radiom:
test-CI: 0.63
Liu et al.
2016 [28]
Train: 42, NPC
Test: 11, NPC
Pre-treatment T2, contrast-enhanced T1 MRI, diffusion weighted MRIRT + cisplatinTherapy response (complete/partial response vs. stable / progressive disease),
k-nearest neighbors, neural network,
classification
Radiom:
CV-acc: 0.95
CV-sens: 0.97
CV-spec: 091
test-acc: 0.91
test-sens: 0.88
test-spec: 1
Lv et al.
2019 [88]
Total: 296, HNSCC
(various partitions in train/test were evaluated)
Pre-treatment non-contrast CT, FDG-PETRT + chemotherapy or RT aloneRFS, MFS and OS,
multivariate Cox regression,
regression
RFS:
Radiom:
mean test-CI: 0.61
Radiom + Clin:
mean test-CI: 0.60
Clin:
mean test-CI: 0.58
MFS:
Radiom:
mean test-CI: 0.70
Radiom + Clin:
mean test-CI: 0.71
Clin:
mean test-CI: 0.61
OS:
Radiom:
mean test-CI: 0.62
Radiom + Clin:
mean test-CI: 0.65
Clin:
mean test-CI: 0.62
(the mean was calculated across all test partitions)
Lv et al.
2019 [24]
Train: 85, NPC
Test: 43, NPC
Pre-treatment CT, FDG-PETIMRT + cisplatin or IMRT alonePFS,
multivariate Cox regression,
regression
Radiom:
train-CI: 0.76
test-CI: 0.62
Radiom + Clin:
train-CI: 0.75
test-CI: 0.75
Clin:
train-CI: 0.71
test-CI: 0.75
M.D. Anderson C.C.H.a.N.Q.I.W.G. 2018 [31]Train: 255, OPSCC
Tune: 165, OPSCC
Test: 45, OPSCC
Pre-treatment contrast CTOne or combinations of: IMRT / chemotherapy / induction chemotherapy / neck dissectionLC,
multivariate Cox regression,
regression
Overall performance evaluation of Cox models not reported
Mo et al.
2019 [89]
Train: 80, HYSCC
Test: 33, HYSCC
Pre-treatment non-contrast CT and contrast-CTLaryngeal-preservation treatments (RT, chemotherapy, induction-chemotherapy, neck dissection)PFS,
multivariable Cox regression,
regression
Radiom:
train-CI: 0.79
test-CI: 0.76
Radiom + Clin:
train-CI: 0.80
test-CI: 0.76
Clin:
train-CI: 0.63
test-CI: 0.54
Ou et al.
2017 [90]
Total: 120, HNSCCPre-treatment CTCRT / IMRT + cisplatin / cetuximabOS and PFS,
multivariable Cox regression,
regression
OS:
Radiom:
HR: 0.3 (p = 0.02)
PFS:
Radiom:
HR: 0.3 (p = 0.01)
Ouyang et al. 2017 [91]Train: 70, NPC
Test: 30, NPC
Pre-treatment T2, contrast-enhanced T1 MRINot reportedPFS,
multivariable Cox regression,
regression
Radiom:
train-HR: 5.14 (p < 0.001)
test-HR: 7.28 (p = 0.015)
Parmar et al. 2015 [29]Train: 101, HNSCC
Test: 95, HNSCC
Pre-treatment contrast CTRT + chemotherapy or RT + chemotherapy + surgery or RT aloneOS,
12 different ML classifiers,
classification
Radiom:
test-AUC: 0.79
Parmar et al. 2015 [54]Train: 136, HNSCC
Test: 95, HNSCC
Pre-treatment contrast CTRT + chemotherapy or RT + chemotherapy + surgery or RT aloneOS,
multivariable Cox regression,
regression
Radiom:
test-CI: 0.63
Ulrich et al. 2019 [92]Total: 30, OPSCC and LSCCPre-treatment 18F-fluorothymidine PETRT + platinum-based chemotherapyPFS,
univariate Cox regression,
regression
Radiom:
HR: 4.10 (p = 0.001)
Vallieres et al. 2017 [23]Train: 194, HNSCC
Test: 106, HNSCC
Pre-treatment FDG-PET, non-contrast CTRT + platinum-based chemotherapy / cetuximab or RT aloneLR, DM and OS,
logistic regression, random forests,
classification
(regression analysis was performed for a subset of models; see publication)
LR:
Radiom:
test-AUC: 0.64
Radiom + Clin:
test-AUC: 0.69
DM:
Radiom:
test-AUC: 0.86
Radiom + Clin:
test-AUC: 0.86
OS:
Radiom:
test-AUC: 0.62
Radiom + Clin:
test-AUC: 0.74
Wang et al. 2018 [27]Total: 120, NPCPre-treatment T2, contrast-enhanced T1 MRIInduction-chemotherapy (cisplatin + 5-fluorouracil + docetaxel)Early response to induction chemotherapy,
“Rad-score”,
classification
Radiom:
train-AUC: 0.82
internally bootstrap-validated train-AUC: 0.82
Zhai et al. 2019 [93]Train: 240, HNSCC
Test: 204, HNSCC
Pre-treatment contrast CTRT + chemotherapy / cetuximab or RT aloneLC, RC, MFS and DFS,
multivariate Cox regression,
regression
LC:
Radiom:
train-CI: 0.62
test-CI: 0.62
Radiom + Clin:
train-CI: 0.66
test-CI: 0.64
Clin:
train-CI: 0.64
test-CI: 0.62
RC:
Radiom:
train-CI: 0.78
test-CI: 0.80
Radiom + Clin:
train-CI: 0.78
test-CI: 0.80
Clin:
train-CI: 0.74
test-CI: 0.76
MFS:
Radiom:
train-CI: 0.73
test-CI: 0.68
Radiom + Clin:
train-CI: 0.72
test-CI: 0.71
Clin:
train-CI: 0.71
test-CI: 0.70
DFS:
Radiom:
train-CI: 0.66
test-CI: 0.65
Radiom + Clin:
train-CI: 0.69
test-CI: 0.70
Clin:
train-CI: 0.66
test-CI: 0.66
Zhang et al. 2017 [94]Train: 80, NPC
Test: 33, NPC
Pre-treatment T2, contrast-enhanced T1 MRINot reportedPFS,
“Rad-score”,
classification
Radiom:
train-AUC: 0.89
test-AUC: 0.82
Zhang et al. 2017 [95]Train: 70, NPC
Test: 40, NPC
Pre-treatment T2, contrast-enhanced T1 MRINot reportedLF and DF,
9 different ML classifiers,
classification
LF and DF:
Radiom:
test-AUC: 0.85
Zhang et al. 2017 [96]Train: 88, NPC
Test: 30, NPC
Pre-treatment T2, contrast-enhanced T1 MRINot reportedPFS,
univariate / multivariable Cox regression,
regression
Radiom:
train-CI: 0.76
test-CI: 0.74
Clin:
train-CI: 0.65
test-CI: 0.63
Radiom + Clin:
train-CI: 0.78
test-CI: 0.72
  1. a The reported performance pertains to the maximum observed performance among all models of each respective category (i.e. we are reporting the highest achieved performance, in case different radiomics features / models / signatures or clinical predictors / models were tested). For radiomics-based models, the performance of the purest imaging feature-based model is reported. (i.e. the model with fewest or no other predictors)
  2. acc Accuracy, ACM All-cause mortality, AUC Area under the receiver operating characteristics curve, CI Concordance index, Clin Non-radiomic predictor(s) or model(s) (“clinical”), CRT Conformal radiotherapy, DF/DM Distant failure/metastasis, DFS Disease-free survival, DSS Disease-specific survival, HNSCC Head and neck SCC, HR Hazard ratio, HYSCC Hypopharyngeal SCC, IMRT Intensity-modulated radiotherapy, LC/LF Local tumor control/failure, LR Locoregional recurrence, LRC Locoregional control, LSCC Laryngeal SCC, MFS Metastasis-free survival, NPC Nasopharyngeal carcinoma, OPSCC Oropharyngeal SCC, OS Overall survival, PFS Progression-free survival, Radiom Radiomics model, radiomic feature(s) or feature combinations (“signature”, “Rad-score”), RC Regional control, RFS Recurrence-free survival, RT Radiotherapy, sens Sensitivity, spec Specificity, test Independent test dataset, total Only one dataset used, train Training dataset, tune Validation set used for hyperparameter tuning