Author (Year) | complications | Sample size | treatment | model/ algorithm | AUC(CI) | Prognostic factors & Feature Variables | Significant contributions and findings |
---|---|---|---|---|---|---|---|
Hubert S. Gabryś et al. [16] | Xerostomia | 153 | IMRT | LR-L1 LR-L2 LR-EN kNN SVM ET GTB | Early Stage (0–6 months): LR-L1 AUC Validation: 0.56 LR-L2 AUC Validation: 0.46 LR-EN AUC Validation: 0.54 kNN AUC Validation: 0.65 SVM AUC Validation: 0.57 ET AUC Validation: 0.44 GTB AUC Validation: 0.55 Late Stage (6–15 months): LR-L1 AUC Validation: 0.63 LR-L2 AUC Validation: 0.60 LR-EN AUC Validation: 0.56 kNN AUC Validation: 0.62 SVM AUC Validation: 0.52 ET AUC Validation: 0.55 GTB AUC Validation: 0.65 Long-term (15-24 months): LR-L1 AUC Validation: 0.86 LR-L2 AUC Validation: 0.86 LR-EN AUC Validation: 0.83 kNN AUC Validation: 0.74 SVM AUC Validation: 0.79 ET AUC Validation: 0.88 GTB AUC Validation: 0.77 Longitudinal Long-term (15–24 months): LR-L1 AUC Validation: 0.52 LR-L2 AUC Validation: 0.39 LR-EN AUC Validation: 0.52 kNN AUC Validation: 0.58 SVM AUC Validation: 0.57 ET AUC Validation: 0.51 GTB AUC Validation: 0.63 | Demographics: Age, Gender, Salivary Gland Shape, Volume, Sphericity, Eccentricity Volume Dose Histogram: Mean, Distribution, Skewness Spatial Dose Gradient: Gradient x, Gradient y, Gradient z Spatial Dose Distribution: η200, η020, η002 Spatial Dose Correlation: η110, η101, η011 Spatial Dose Skewness: η300, η030, η003 Spatial Dose Co-skewness: η012, η021, η120, η102, η210, η201 | 1. The integration of organ and dose shape descriptors has a positive impact on predicting xerostomia 2. The prediction of xerostomia is dependent on patient-specific and non-dosimetric factors, emphasizing the importance of personalized data for risk assessment 3. These insights offer detailed machine learning methodologies that are valuable for future radiomics and dosiomics in the establishment of NTCP (Normal Tissue Complication Probability) models |
Tsair-Fwu Lee et al. ( 2014) | Xerostomia | 206 | IMRT | LASSO & Logistic Regression | XER3m (LASSO-Suboptimal) Model: Number of factors is 3 AUC is 0.84 XER3m (LASSO-Optimal) Model: Number of factors is 8 AUC is 0.86 XER3m (Likelihood) Model: Number of factors is 9 AUC is 0.85 XER12m (LASSO-Suboptimal) Model: Number of factors is 5 AUC is 0.84 XER12m (LASSO-Optimal) Model: Number of factors is 9 AUC is 0.87 XER12m (Likelihood) Model: Number of factors is 11 AUC is 0.86 | XER3m Related Factors: Dmean-c, Dmean-i, Age, Economic Status, T Stage, AJCC Stage, Smoking, Education Level, Chemotherapy (C/T), Node Classification, Baseline Xerostomia, SIB or SQM, Gender, Family History, Marital Status XER12m Related Factors: Dmean-i, Dmean-c, Smoking, T Stage, Baseline Xerostomia, Alcohol Issues, Family History, Node Classification, Gender, Age, Economic Status, Chemotherapy (C/T), AJCC Stage, Marital Status, SIB or SQM | 1. Utilizing the Least Absolute Shrinkage and Selection Operator (LASSO) to construct a multivariate logistic regression model effectively predicts the incidence of moderate to severe xerostomia in head and neck cancer patients undergoing Intensity-Modulated Radiation Therapy (IMRT) 2. Through LASSO, eight prognostic factors were identified for the 3-month time point: Dmean-c, Dmean-i, age, financial status, T-stage, AJCC stage, smoking, and education. For the 12-month time point, nine prognostic factors were identified: Dmean-i, education, Dmean-c, smoking, T-stage, baseline xerostomia, alcohol consumption, family medical history, and lymph node classification 3. During the process of selecting the optimal number of prognostic factors via LASSO, fine-tuning was performed using the Hosmer–Lemeshow test and AUC. For the 3-month time point, three optimal prognostic factors were selected: Dmean-c, Dmean-i, and age. For the 12-month time point, five optimal prognostic factors were selected: Dmean-i, education, Dmean-c, smoking, and T-stage 4. The overall performance of the NTCP model at both time points, as indicated by scaled Brier scores, Omnibus, and Nagelkerke R2 metrics, met certain standards and aligned with expected values 5. The multivariate NTCP model using LASSO was confirmed to be effective for predicting xerostomia in patients evaluated post-IMRT |
Tsair-Fwu Lee et al. ( 2014) | Xerostomia | 152 (HNSCC) 84 (NPC) | 3D-CRT IMRT | LASSO & Logistic Regression | XER HNSCC-3 m Model: Number of Factors = 3 AUC = 0.88 (Range: 0.86–0.91) XER HNSCC-12 m Model: Number of Factors = 3 AUC = 0.98 (Range: 0.97–0.98) XER NPC-3 m Model: Number of Factors = 4 AUC = 0.87 (Range: 0.83–0.90) XER NPC-12 m Model: Number of Factors = 3 AUC = 0.96 (Range: 0.95–0.97) | Dmean-c Dmean-i Age Economic Status T-Stage Education Level | The multivariate Normal Tissue Complication Probability (NTCP) model developed using the Least Absolute Shrinkage and Selection Operator (LASSO) effectively predicts the incidence of moderate to severe xerostomia in patients with Head and Neck Squamous Cell Carcinoma (HNSCC) and Nasopharyngeal Carcinoma (NPC) undergoing Intensity-Modulated Radiation Therapy (IMRT) Through LASSO, higher AUC performance was retained while selecting the fewest predictive factors, resulting in the establishment of four predictive models In all models, the average dose to the contralateral and ipsilateral salivary glands was chosen as the most important predictive factor. Other selected clinical and socio-economic factors include age, financial status, T-stage, and educational level The multivariate logistic regression model using LASSO techniques can improve the prediction of the incidence of xerostomia in HNSCC and NPC patients The predictive model developed for HNSCC cannot be directly applied to the NPC population undergoing IMRT and vice versa, necessitating validation |
Lisanne V. van Dijk et al. ( 2016) | Xerostomia | 249 | 3D-CRT IMRT VMAT | LASSO & Logistic Regression | XER12m Model without IBM Discrimination: AUC = 0.75 ( 0.69–0.81) XER12m Model with IBM Discrimination: AUC = 0.77 ( 0.71–0.82) XER12m Model without IBM Validation: AUC boot = 0.74 XER12m Model with IBM Validation: AUC boot = 0.76 | CT Image Biomarkers (IBMs) Short Run Emphasis (SRE): An image biomarker (IBM) that measures the heterogeneity of the parotid gland tissue Additional Parameters: Mean Contra-lateral Parotid Gland Dose: The average radiation dose received by the contra-lateral parotid gland during treatment Maximum CT Intensity of the Submandibular Gland: The highest computed tomography (CT) intensity value recorded for the submandibular gland Mean Dose to Submandibular Glands: The average radiation dose received by the submandibular glands during treatment | Existing models for predicting late-stage patient assessment of moderate to severe xerostomia (XER12m) and oral mucosal hypersecretion (STIC12m) after radiation therapy are primarily based on dose-volume parameters and baseline xerostomia (XERbase) or oral mucosal hypersecretion (STICbase) scores. However, the aim of the study is to improve these predictions by using patient-specific features based on CT image biomarkers (IBM) The research team prospectively collected planning CT scans and patient assessment outcome measurements for 249 head and neck cancer patients undergoing definitive radiation therapy (with or without systemic therapy) These potential image biomarkers (IBM) represent the geometric features, CT intensity, and textural characteristics of the salivary glands and submandibular glands Lasso regularization was used to create multivariate logistic regression models, and internal validation was performed through bootstrapping By adding the image biomarker "Short Run Emphasis" (SRE), which quantifies the heterogeneity of salivary gland tissue, to the average contralateral salivary gland dose and baseline xerostomia model, significant improvements were made in predicting xerostomia at 12 months For predicting oral mucosal hypersecretion at 12 months, researchers selected the maximum CT intensity of the submandibular gland as another image biomarker, in addition to baseline hypersecretion and the average dose to the submandibular gland By introducing image biomarkers representing the heterogeneity and density of the salivary glands, researchers improved predictions for xerostomia and oral mucosal hypersecretion at 12 months Providing image biomarkers can further guide the patient-specific response of healthy tissue to radiation doses in research |
Stefano Ursino et al ( 2021) | Dysphagia | 38 | RT IMRT | LRC SVC RFC | Predicting Dysphagia at 6 months: SVC: AUC = 0.82 LRC: AUC = 0.80 RFC: AUC = 0.83 Predicting Dysphagia at 12 months: SVC: AUC = 0.85 LRC: AUC = 0.82 RFC: AUC = 0.94 | Dose-Volume Histogram (DVH) features of the throat (SWOARs) Dose of Swallowing Risk Organs (SWOARs) Baseline and Post-Radiation 6 and 12 Months Penetration-Aspiration Score (P/A-VF) | Researchers developed a predictive model for Radiation-Induced Dysphagia (RID) based on Videofluoroscopy (VF) by incorporating Dose-Volume Histogram (DVH) parameters of Swallowing Risk Organs at Risk (SWOARs) into machine learning analysis The RID predictive model was developed using the dose of nine swallowing risk organs and the Penetration-Aspiration Score (P/A) from VF data at 6 and 12 months post-treatment Seventy-two dose features were extracted for each patient from the DVH and were analyzed using Linear Support Vector Classification (SVC), Logistic Regression Classification (LRC), and Random Forest Classification (RFC) Among 38 patients, the DVH features of SWOARs showed relevance at both 6 months (SVC's AUC 0.82; LRC's AUC 0.80; RFC's AUC 0.83) and 12 months (SVC's AUC 0.85; LRC's AUC 0.82; RFC's AUC 0.94) At 6 months, the SWOARs with the highest relevance and their corresponding features included the base of the tongue (V65 and Dmean), superior and middle constrictor muscles (V45, V55, V65, Dmp, Dmean, Dmax, and Dmin), and salivary glands (Dmean and Dmp). At 12 months, the features with the highest relevance included middle and inferior constrictor muscles (V55, Dmin, and Dmean; and V55, V65, Dmin, and Dmax), glottis (V55 and Dmax), laryngeal muscles (Dmax), and cervical esophagus (Dmax) A RID predictive model was trained and cross-validated, demonstrating high discriminative ability at both 6 and 12 months post-radiation therapy |
Jamie A. Dean et al. ( 2018) | Dysphagia | 263 | 3D-CRT IMRT | PLR SVC RFC | 6 months following RT: PLRstandard: AUC = 0.82 ± 0.04 SVCstandard: AUC = 0.82 ± 0.04 RFCstandard: AUC = 0.78 ± 0.05 PLRspatial: AUC = 0.75 ± 0.08 SVCspatial: AUC = 0.74 ± 0.08 RFCspatial: AUC = 0.75 ± 0.05 | PM receiving > 1 Gy/fraction | Researchers have proposed a model capable of predicting the severity of acute dysphagia in individual patients, which can be used to guide clinical decisions The goal of the study is to establish a model incorporating spatial dose metrics that can offer guidelines for radiation therapy planning, aiming to reduce the incidence of severe swallowing difficulties The researchers used radiation therapy doses to the pharyngeal mucosa (PM), including dose-volume and spatial dose metrics, along with clinical data, to develop a model for severe acute dysphagia Penalized Logistic Regression (PLR), Support Vector Classification (SVC), and Random Forest Classification (RFC) models were generated and internally (173 patients) and externally (90 patients) validated It was determined that the volume of the pharyngeal mucosa receiving moderate and high doses (greater than 1 Gy/fraction) is most correlated with severe acute dysphagia. In radiation therapy planning, these volumes should be minimized as much as possible to reduce the occurrence of severe acute dysphagia The performance of the Penalized Logistic Regression model using dose-volume metrics (PLR_standard) was comparable to more complex models and demonstrated excellent discriminative ability in external validation (Area Under the Curve, AUC = 0.82) |
Jamie A. Dean et al. ( 2016) | Mucositis | 351 | RT (Not Specifically Stated) | PLR SVC RFC | PLRstandard: AUC = 0.72 ± 0.09 SVCstandard: AUC = 0.72 ± 0.09 RFCstandard: AUC = 0.71 ± 0.09 PLRspatial: AUC = 0.72 ± 0.09 SVCspatial: AUC = 0.71 ± 0.09 RFCspatial: AUC = 0.70 ± 0.09 | Volumes of oral cavity receiving intermed—high dose | The aim of this study is to generate a predictive model for severe acute oral mucositis using spatial dose metrics and machine learning, which can guide clinical decision-making and inform treatment planning Researchers used radiation therapy dosages (dose-volume and spatial dose metrics) and clinical data to generate predictive models. They compared the performance of penalized logistic regression, support vector classification, and random forest classification models The performance of the standard dose-volume-based model was not significantly different from models that included spatial information. The discriminative ability was similar across all models, but the standard random forest classification model had the best calibration The average AUC and calibration slope for this model were 0.71 (SD = 0.09) and 3.9 (SD = 2.2), respectively The volume of the oral cavity receiving moderate and high doses is correlated with severe oral mucositis Reducing the volume of the oral cavity receiving moderate and high doses may potentially reduce the incidence of oral mucositis |
Ivo Beetz et al. (2012) | Xerostomia | 178 | IMRT | M-LR | XER6m Model AUC = 0.68 (0.60–0.76) | Moderate to severe dry mouth (XER M6) and sticky saliva (STIC M6) were assessed at 6 months before and after treatment using the EORTC QLQ-H&N35 questionnaire (For all questions, including those related to dry mouth and sticky saliva, a 4-point Likert scale was used.) The main predictive factors for dry mouth are the average dose to the contralateral salivary gland and baseline dry mouth The main predictive factors for sticky saliva are the average dose to the contralateral submandibular gland, the sublingual gland, and the minor salivary glands of the soft palate | This is a multi-center prospective study aimed at developing a multivariate logistic regression model The purpose of the study is to predict the risk of xerostomia and sticky saliva in patients with head and neck cancer 6 months after receiving IMRT. The study covers 178 patients with head and neck cancer. The results show that 51.6% of patients experienced xerostomia after treatment; 35.6% of patients reported issues with sticky saliva The main predictive factors for xerostomia are the average dose to the contralateral salivary gland and baseline xerostomia The main predictive factors for sticky saliva are the average dose to the contralateral submandibular gland, sublingual gland, and minor salivary glands in the soft palate The model proposed in this study can serve as a reference for optimizing future IMRT treatments Moderate to severe xerostomia (XER M6) and sticky saliva (STIC M6) were assessed using the EORTC QLQ-H&N35 questionnaire before and 6 months after treatment For all questions, including those related to xerostomia and sticky saliva, a 4-point Likert scale was used |
Ivo Beetz et al. [24] | Xerostomia | 165 | IMRT 3D-CRT | M-LR | XER6m Model AUC = 0.82 (0.76–0.89) | Moderate to severe dry mouth (XER M6) and sticky saliva (STIC M6) were assessed at 6 months before and after treatment using the EORTC QLQ-H&N35 questionnaire (For all questions, including those related to dry mouth and sticky saliva, a 4-point Likert scale was used.) | Dose distributions in minor salivary glands during 3D-CRT have limited impact on patient-rated salivary dysfunction symptoms Beyond the parotid and submandibular glands, only the sublingual glands showed a significant association with sticky saliva Reliable risk estimation needs other factors like age and baseline subjective scores Including these selected factors in predictive models enhances model performance significantly over just using dose volume histogram parameters |
Kuo Men et al. [19] | Xerostomia | 784 | IMRT | 3D rCNN | XER12m Model: AUC = 0.84 (0.74–0.91) No contour—AUC = 0.82 (0.72–0.90) No CT- AUC = 0.78 (0.67–0.88) | A subset of 40 images from the RTOG 0522 clinical trial had their features automatically extracted through deep learning | A toxicity prediction model using 3D rCNN was developed and evaluated The model extracted low- and high-level spatial features from CT planning images, radiation therapy dose distributions, and contours with 3D filters The proposed model showed promising results in predicting xerostomia Future studies focusing on more accurate definitions of xerostomia-associated regions can enhance the model's performance |
Benjamin S Rosen et al. [26] | Xerostomia | 105 | VMAT | PLR | Prediction of XER12m for ≥ 1 grade xerostomia using Dose/Clinical model (DVH/Clinical): AUC = 0.709 (95% CI, 0.603–0.815) Prediction of XER12m with added Radiomics model (DVH/Clinical + Radiomics): AUC = 0.719 (95% CI, 0.603–0.830) Prediction of XER12m for ≥ 2 grade xerostomia using Dose/Clinical model (DVH/Clinical): AUC = 0.692 (95% CI, 0.615–0.770) Prediction of XER12m with added contralateral salivary gland changes slightly improved predictive performance (DVH/Clinical + Radiomics): AUC = 0.776 (95% CI, 0.643–0.912) | CBCT Image Features Patient Demographics Follow-up and Clinical Outcomes | 1. A methodology has been introduced for using on-board CBCT to measure treatment-related PG changes during HNC radiotherapy 2. Early treatment CBCT measurements of PG density changes were linked to long-term xerostomia 3. These CBCT-measured changes offer better predictions than PG dose alone 4. The CBCT analysis can be conducted with minimal additional cost, making it a viable option for an adaptive radiotherapy platform |
Khadija Sheikh et al [27] | Xerostomia | 266 | IMRT VMAT TomoTherapy | LASSO + Generalized linear models (multiple LR) | XER3m: DVH-AUC = 0.63 (0.51–0.81) CT-AUC = 0.57 (0.45–0.71) MR-AUC = 0.66 (0.54–0.82) CT + MR-AUC = 0.70 (0.57–0.82) DVH + CT-AUC = 0.56 (0.40–0.68) DVH + CT + MR-AUC = 0.60 (0.50–0.73) Clinical + CT + MR-AUC = 0.73 (0.62–0.86) Clinical + DVH + CT + MR-AUC = 0.68 (0.52–0.80) | IBMs (Image Biomarkers) CT and MR Imaging Dose-Volume Histogram (DVH) Parameters | 1. Baseline image features from both parotid and submandibular glands can potentially serve as clinical surrogates for baseline function 2. Features from the submandibular glands might offer insights into unstimulated salivary function, enhancing predictions of post-RT xerostomia susceptibility 3. While combining all data showed a trend towards better prediction, further research is needed to ascertain the advantages of merging imaging modalities for xerostomia prediction 4. Prediction models based on these features can deepen our comprehension of radiation-induced xerostomia and aid in tailoring radiation treatment plans to reduce toxicity |