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

Table 3 Prediction of post-radiation xerostomia based on salivary gland radiomics features

Authors, year	Dataset: sample size, cancer type	Time of xerostomia assessment, endpoint/scale	Imaging modality	VOI	Classifier / regression model(s)	Metric: maximum performance ^a
Sheikh et al. 2019 [110]	Train:216, HNSCC Test:50, HNSCC	3-month post-RT, CTCAE v4.0 ^b grade ≥ 2 vs. grade 0/1	Pre-treatment CT, T1-weighted MRI	Parotid and submandibular glands (bilateral)	Multivariable logistic regression	CV-AUC: 0.75 test-AUC: 0.70
Liu et al. 2019 [111]	Train:35, NPC Test:4, NPC	day of 10th and 30th RT, saliva amount (ml) over 5 min (a regression analysis)	CT at start and day of 10th RT fraction	Parotid glands (bilateral)	8 different regression models	CV-MSE: 0.9042 (10th fraction), 0.0569 (30th fraction) test-MSE: 0.0233 (30th fraction)
van Dijk et al. 2018 [112]	Train:68, HNSCC Test:25, HNSCC	12 moth post-RT, patient-rated moderate-to-severe xerostomia present vs. not present	Pre-treatment T1-weighted MRI	Parotid glands, (bilateral)	Multivariable logistic regression	n/a ^c
van Dijk et al. 2017 [113]	Total: 249, HNSCC	12 moth post-RT, EORTC QLQ-H, N35 questionnaire ^d moderate-to-severe xerostomia vs. not present	Pre-treatment contrast CT	Parotid and submandibular glands (bilateral)	Multivariable logistic regression	n/a ^c
van Dijk et al. 2018 [114]	Total: 161, HNSCC	12-month post-RT, EORTC QLQ-H questionnaire ^d moderate-to-severe xerostomia present vs. not present	Pre-treatment FDG PET	Contralateral parotid gland	Multivariable logistic regression	n/a ^c

^a The reported performance pertains to the maximum observed performance among the purest imaging feature-based models reported (i.e. the best model with fewest or no other predictors is reported)
^b Common Terminology Criteria for Adverse Events Version 4.0 [115]
^c “Pure” radiomics models were not built. Instead, the contribution of individual radiomics features to baseline models was investigated in terms of performance (gains)
^d European Organization for Research and Treatment of Cancer questionnaire module for quality of life assessments in head and neck cancer patients [116]
AUC Area under the receiver operating characteristics curve, CV Cross validation (of total set or training data set), HNSCC Head and neck SCC, MSE Mean squared error, NPC Nasopharyngeal carcinoma, RT Radiotherapy, Test Independent test data set, Total Only one data set used, Train Training data set

ISSN: 2059-7347