Outcome prediction models incorporating clinical variables for Head and Neck Squamous cell Carcinoma: A systematic review of methodological conduct and risk of bias.

Multiple outcome prediction models have been developed for Head and Neck Squamous Cell Carcinoma (HNSCC). This systematic review aimed to identify HNSCC outcome prediction model studies, assess their methodological quality and identify those with potential utility for clinical practice. Inclusion criteria were mucosal HNSCC prognostic prediction model studies (development or validation) incorporating clinically available variables accessible at time of treatment decision making and predicting tumour-related outcomes. Eligible publications were identified from PubMed and Embase. Methodological quality and risk of bias were assessed using the checklist for critical appraisal and data extraction for systematic reviews of prediction modelling studies (CHARMS) and prediction model risk of bias assessment tool (PROBAST). Eligible publications were categorised by study type for reporting. 64 eligible publications were identified; 55 reported model development, 37 external validations, with 28 reporting both. CHARMS checklist items relating to participants, predictors, outcomes, handling of missing data, and some model development and evaluation procedures were generally well-reported. Less well-reported were measures accounting for model overfitting and model performance measures, especially model calibration. Full model information was poorly reported (3/55 model developments), specifically model intercept, baseline survival or full model code. Most publications (54/55 model developments, 28/37 external validations) were found to have high risk of bias, predominantly due to methodological issues in the PROBAST analysis domain. The identified methodological issues may affect prediction model accuracy in heterogeneous populations. Independent external validation studies in the local population and demonstration of clinical impact are essential for the clinical implementation of outcome prediction models.

Standardising Breast Radiotherapy Structure Naming Conventions: A Machine Learning Approach.

In progressing the use of big data in health systems, standardised nomenclature is required to enable data pooling and analyses. In many radiotherapy planning systems and their data archives, target volumes (TV) and organ-at-risk (OAR) structure nomenclature has not been standardised. Machine learning (ML) has been utilised to standardise volumes nomenclature in retrospective datasets. However, only subsets of the structures have been targeted. Within this paper, we proposed a new approach for standardising all the structures nomenclature by using multi-modal artificial neural networks. A cohort consisting of 1613 breast cancer patients treated with radiotherapy was identified from Liverpool & Macarthur Cancer Therapy Centres, NSW, Australia. Four types of volume characteristics were generated to represent each target and OAR volume: textual features, geometric features, dosimetry features, and imaging data. Five datasets were created from the original cohort, the first four represented different subsets of volumes and the last one represented the whole list of volumes. For each dataset, 15 sets of combinations of features were generated to investigate the effect of using different characteristics on the standardisation performance. The best model reported 99.416% classification accuracy over the hold-out sample when used to standardise all the nomenclatures in a breast cancer radiotherapy plan into 21 classes. Our results showed that ML based automation methods can be used for standardising naming conventions in a radiotherapy plan taking into consideration the inclusion of multiple modalities to better represent each volume.

Larynx cancer survival model developed through open-source federated learning.

INTRODUCTION: Federated learning has the potential to perfrom analysis on decentralised data; however, there are some obstacles to survival analyses as there is a risk of data leakage. This study demonstrates how to perform a stratified Cox regression survival analysis specifically designed to avoid data leakage using federated learning on larynx cancer patients from centres in three different countries. METHODS: Data were obtained from 1821 larynx cancer patients treated with radiotherapy in three centres. Tumour volume was available for all 786 of the included patients. Parameter selection among eleven clinical and radiotherapy parameters were performed using best subset selection and cross-validation through the federated learning system, AusCAT. After parameter selection, ß regression coefficients were estimated using bootstrap. Calibration plots were generated at 2 and 5-years survival, and inner and outer risk groups' Kaplan-Meier curves were compared to the Cox model prediction. RESULTS: The best performing Cox model included log(GTV), performance status, age, smoking, haemoglobin and N-classification; however, the simplest model with similar statistical prediction power included log(GTV) and performance status only. The Harrell C-indices for the simplest model were for Odense, Christie and Liverpool 0.75[0.71-0.78], 0.65[0.59-0.71], and 0.69[0.59-0.77], respectively. The values are slightly higher for the full model with C-index 0.77[0.74-0.80], 0.67[0.62-0.73] and 0.71[0.61-0.80], respectively. Smoking during treatment has the same hazard as a ten-years older nonsmoking patient. CONCLUSION: Without any patient-specific data leaving the hospitals, a stratified Cox regression model based on data from centres in three countries was developed without data leakage risks. The overall survival model is primarily driven by tumour volume and performance status.

An investigation of the conformity, feasibility, and expected clinical benefits of multiparametric MRI-guided dose painting radiotherapy in glioblastoma.

Background: New technologies developed to improve survival outcomes for glioblastoma (GBM) continue to have limited success. Recently, image-guided dose painting (DP) radiotherapy has emerged as a promising strategy to increase local control rates. In this study, we evaluate the practical application of a multiparametric MRI model of glioma infiltration for DP radiotherapy in GBM by measuring its conformity, feasibility, and expected clinical benefits against standard of care treatment. Methods: Maps of tumor probability were generated from perfusion/diffusion MRI data from 17 GBM patients via a previously developed model of GBM infiltration. Prescriptions for DP were linearly derived from tumor probability maps and used to develop dose optimized treatment plans. Conformity of DP plans to dose prescriptions was measured via a quality factor. Feasibility of DP plans was evaluated by dose metrics to target volumes and critical brain structures. Expected clinical benefit of DP plans was assessed by tumor control probability. The DP plans were compared to standard radiotherapy plans. Results: The conformity of the DP plans was >90%. Compared to the standard plans, DP (1) did not affect dose delivered to organs at risk; (2) increased mean and maximum dose and improved minimum dose coverage for the target volumes; (3) reduced minimum dose within the radiotherapy treatment margins; (4) improved local tumor control probability within the target volumes for all patients. Conclusions: A multiparametric MRI model of GBM infiltration can enable conformal, feasible, and potentially beneficial dose painting radiotherapy plans.

Delineation uncertainties of tumour volumes on MRI of head and neck cancer patients.

Background: During the last decade, radiotherapy using MR Linac has gone from research to clinical implementation for different cancer locations. For head and neck cancer (HNC), target delineation based only on MR images is not yet standard, and the utilisation of MRI instead of PET/CT in radiotherapy planning is not well established. We aimed to analyse the inter-observer variation (IOV) in delineating GTV (gross tumour volume) on MR images only for patients with HNC. Material/methods: 32 HNC patients from two independent departments were included. Four clinical oncologists from Denmark and four radiation oncologists from Australia had independently contoured primary tumour GTVs (GTV-T) and nodal GTVs (GTV-N) on T2-weighted MR images obtained at the time of treatment planning. Observers were provided with sets of images, delineation guidelines and patient synopsis. Simultaneous truth and performance level estimation (STAPLE) reference volumes were generated for each structure using all observer contours. The IOV was assessed using the DICE Similarity Coefficient (DSC) and mean absolute surface distance (MASD). Results: 32 GTV-Ts and 68 GTV-Ns were contoured per observer. The median MASD for GTV-Ts and GTV-Ns across all patients was 0.17 cm (range 0.08-0.39 cm) and 0.07 cm (range 0.04-0.33 cm), respectively. Median DSC relative to a STAPLE volume for GTV-Ts and GTV-Ns across all patients were 0.73 and 0.76, respectively. A significant correlation was seen between median DSCs and median volumes of GTV-Ts (Spearman correlation coefficient 0.76, p < 0.001) and of GTV-Ns (Spearman correlation coefficient 0.55, p < 0.001). Conclusion: Contouring GTVs in patients with HNC on MRI showed that the median IOV for GTV-T and GTV-N was below 2 mm, based on observes from two separate radiation departments. However, there are still specific regions in tumours that are difficult to resolve as either malignant tissue or oedema that potentially could be improved by further training in MR-only delineation.

Open-source distributed learning validation for a larynx cancer survival model following radiotherapy.

INTRODUCTION: Prediction models are useful to design personalised treatment. However, safe and effective implementation relies on external validation. Retrospective data are available in many institutions, but sharing between institutions can be challenging due to patient data sensitivity and governance or legal barriers. This study validates a larynx cancer survival model performed using distributed learning without any sensitive data leaving the institution. METHODS: Open-source distributed learning software based on a stratified Cox proportional hazard model was developed and used to validate the Egelmeer et al. MAASTRO survival model across two hospitals in two countries. The validation optimised a single scaling parameter multiplied by the original predicted prognostic index. All analyses and figures were based on the distributed system, ensuring no information leakage from the individual centres. All applied software is provided as freeware to facilitate distributed learning in other institutions. RESULTS: 1745 patients received radiotherapy for larynx cancer in the two centres from Jan 2005 to Dec 2018. Limiting to a maximum of one missing value in the parameters of the survival model reduced the cohort to 1095 patients. The Harrell C-index was 0.74 (CI95%, 0.71-0.76) and 0.70 (0.66-0.75) for the two centres. However, the model needed a scaling update. In addition, it was found that survival predictions of patients undergoing hypofractionation were less precise. CONCLUSION: Open-source distributed learning software was able to validate, and suggest a minor update to the original survival model without central access to patient sensitive information. Even without the update, the original MAASTRO survival model of Egelmeer et al. performed reasonably well, providing similar results in this validation as in its original validation.

Repeatability of radiotherapy dose-painting prescriptions derived from a multiparametric magnetic resonance imaging model of glioblastoma infiltration.

Background and purpose: Glioblastoma (GBM) patients have a dismal prognosis. Tumours typically recur within months of surgical resection and post-operative chemoradiation. Multiparametric magnetic resonance imaging (mpMRI) biomarkers promise to improve GBM outcomes by identifying likely regions of infiltrative tumour in tumour probability (TP) maps. These regions could be treated with escalated dose via dose-painting radiotherapy to achieve higher rates of tumour control. Crucial to the technical validation of dose-painting using imaging biomarkers is the repeatability of the derived dose prescriptions. Here, we quantify repeatability of dose-painting prescriptions derived from mpMRI. Materials and methods: TP maps were calculated with a clinically validated model that linearly combined apparent diffusion coefficient (ADC) and relative cerebral blood volume (rBV) or ADC and relative cerebral blood flow (rBF) data. Maps were developed for 11 GBM patients who received two mpMRI scans separated by a short interval prior to chemoradiation treatment. A linear dose mapping function was applied to obtain dose-painting prescription (DP) maps for each session. Voxel-wise and group-wise repeatability metrics were calculated for parametric, TP and DP maps within radiotherapy margins. Results: DP maps derived from mpMRI were repeatable between imaging sessions (ICC > 0.85). ADC maps showed higher repeatability than rBV and rBF maps (Wilcoxon test, p = 0.001). TP maps obtained from the combination of ADC and rBF were the most stable (median ICC: 0.89). Conclusions: Dose-painting prescriptions derived from a mpMRI model of tumour infiltration have a good level of repeatability and can be used to generate reliable dose-painting plans for GBM patients.

Novel methodology to quantify dehydration in head and neck cancer radiotherapy using DIXON MRI.

INTRODUCTION: Head and neck cancer (HNC) patients are at risk of weight change, due to inadequate nutrition intake or dehydration, when receiving radiotherapy (RT). This study aimed to develop methodology to measure water content changes on magnetic resonance imaging (MRI) scans of the head and neck region over the course of RT. METHODS: Retrospective datasets of 54 patients were analysed. Eligible patients had been treated for HNC with cisplatin chemoradiation (CRT) or RT alone and underwent a minimum of 2 MRI scans from weeks 0, 3 and 6 of their treatment. Anatomical regions consisting of ≥90% water, on T2-weighted DIXON MRI sequences, were contoured. Water volume changes of all patients were evaluated, within an anatomically standardised external volume, by comparing the absolute water fraction volume (cc) (VEx90WF) and relative water fraction volume (%) (RelVEx90WF) at weeks 0 and 6 of RT. RESULTS: There was a statistically significant difference between the RelVEx90WF at weeks 0 and 6 (P = 0.005). However, no statistically significant difference was identified between weeks 0 and 6 VEx90WF (P = 0.064). There were no statistically significant differences identified between patients who received CRT versus RT alone. CONCLUSION: This study developed a novel method for measuring changes in water fraction volumes over time, using T2-weighted DIXON MRIs. The methodology created in this study requires further validation through phantom imaging, with known fat and water values.

Automated post-operative brain tumour segmentation: A deep learning model based on transfer learning from pre-operative images.

Automated brain tumour segmentation from post-operative images is a clinically relevant yet challenging problem. In this study, an automated method for segmenting brain tumour into its subregions has been developed. The dataset consists of multimodal post-operative brain scans (T1 MRI, post-Gadolinium T1 MRI, and T2-FLAIR images) of 15 patients who were treated with post-operative radiation therapy, along with manual annotations of their tumour subregions. A 3D densely-connected U-net was developed for segmentation of brain tumour regions and extensive experiments were conducted to enhance model accuracy. A model was initially developed using the publicly available BraTS dataset consisting of pre-operative brain scans. This model achieved Dice Scores of 0.90, 0.83 and 0.78 for predicting whole tumour, tumour core, and enhancing tumour subregions when tested on BraTS20 blind validation dataset. The acquired knowledge from BraTS was then transferred to the local dataset. For augmentation purpose, the local dataset was registered to a dataset of MRI brain scans of healthy subjects. To improve the robustness of the model and enhance its accuracy, ensemble learning was used to combine the outputs of all the trained models. Even though the size of the dataset is very small, the final model can segment brain tumours with a high Dice Score of 0.83, 0.77 and 0.60 for whole tumour, tumour core and enhancing core respectively.