Explainable prediction model for the human papillomavirus status in patients with oropharyngeal squamous cell carcinoma using CNN on CT images.

Several studies have emphasised how positive and negative human papillomavirus (HPV+ and HPV-, respectively) oropharyngeal squamous cell carcinoma (OPSCC) has distinct molecular profiles, tumor characteristics, and disease outcomes. Different radiomics-based prediction models have been proposed, by also using innovative techniques such as Convolutional Neural Networks (CNNs). Although some of these models reached encouraging predictive performances, there evidence explaining the role of radiomic features in achieving a specific outcome is scarce. In this paper, we propose some preliminary results related to an explainable CNN-based model to predict HPV status in OPSCC patients. We extracted the Gross Tumor Volume (GTV) of pre-treatment CT images related to 499 patients (356 HPV+ and 143 HPV-) included into the OPC-Radiomics public dataset to train an end-to-end Inception-V3 CNN architecture. We also collected a multicentric dataset consisting of 92 patients (43 HPV+ , 49 HPV-), which was employed as an independent test set. Finally, we applied Gradient-weighted Class Activation Mapping (Grad-CAM) technique to highlight the most informative areas with respect to the predicted outcome. The proposed model reached an AUC value of 73.50% on the independent test. As a result of the Grad-CAM algorithm, the most informative areas related to the correctly classified HPV+ patients were located into the intratumoral area. Conversely, the most important areas referred to the tumor edges. Finally, since the proposed model provided additional information with respect to the accuracy of the classification given by the visualization of the areas of greatest interest for predictive purposes for each case examined, it could contribute to increase confidence in using computer-based predictive models in the actual clinical practice.

The observational clinical registry (cohort design) of the European Reference Network on Rare Adult Solid Cancers: The protocol for the rare head and neck cancers.

INTRODUCTION: Care for head and neck cancers is complex in particular for the rare ones. Knowledge is limited and histological heterogeneity adds complexity to the rarity. There is a wide consensus that to support clinical research on rare cancer, clinical registries should be developed within networks specializing in rare cancers. In the EU, a unique opportunity is provided by the European Reference Networks (ERN). The ERN EURACAN is dedicated to rare adults solid cancers, here we present the protocol of the EURACAN registry on rare head and neck cancers (ClinicalTrials.gov Identifier: NCT05483374). STUDY DESIGN: Registry-based cohort study including only people with rare head and neck cancers. OBJECTIVES: to help describe the natural history of rare head and neck cancers;to evaluate factors that influence prognosis;to assess treatment effectiveness;to measure indicators of quality of care. METHODS: Settings and participants It is an hospital based registry established in hospitals with expertise in head and neck cancers. Only adult patients with epithelial tumours of nasopharynx; nasal cavity and paranasal sinuses; salivary gland cancer in large and small salivary glands; and middle ear will be included in the registry. This registry won't select a sample of patients. Each patient in the facility who meets the above mentioned inclusion criteria will be followed prospectively and longitudinally with follow-up at cancer progression and / or cancer relapse or patient death. It is a secondary use of data which will be collected from the clinical records. The data collected for the registry will not entail further examinations or admissions to the facility and/or additional appointments to those normally provided for the patient follow-up. Variables Data will be collected on patient characteristics (eg. patient demographics, lifestyle, medical history, health status); exposure data (eg. disease, procedures, treatments of interest) and outcomes (e.g. survival, progression, progression-free survival, etc.). In addition, data on potential confounders (e.g. comorbidity; functional status etc.) will be also collected. Statistical methods The data analyses will include descriptive statistics showing patterns of patients' and cancers' variables and indicators describing the quality of care. Multivariable Cox's proportional hazards model and Hazard ratios (HR) for all-cause or cause specific mortality will be used to determine independent predictors of overall survival, recurrence etc. Variables to include in the multivariable regression model will be selected based on the results of univariable analysis. The role of confounding or effect modifiers will be evaluated using stratified analysis or sensitivity analysis. To assess treatment effectiveness, multivariable models with propensity score adjustment and progression-free survival will be performed. Adequate statistical (eg. marginal structural model) methods will be used if time-varying treatments/confounders and confounding by indication (selective prescribing) will be present. RESULTS: The registry initiated recruiting in May 2022. The estimated completion date is December 2030 upon agreement on the achievement of all the registry objectives. As of October 2022, the registry is recruiting. There will be a risk of limited representativeness due to the hospital-based nature of the registry and to the fact that hospital contributing to the registry are expert centres for these rare cancers. Clinical Follow-up could also be an issue but active search of the life status of the patients will be guaranteed.