Predictive value of local control by 4'-[methyl-11C]-thiotymidine PET volume parameters in p16-negative oropharyngeal, hypopharyngeal, and supraglottic squamous cell carcinoma.

PURPOSE: We investigated the potential of baseline 4'-[methyl- 11 C]-thiothymidine ([ 11 C]4DST) PET for predicting loco-regional control of head and neck squamous cell carcinoma (HNSCC). METHODS: A retrospective analysis was performed using volumetric parameters, such as SUVmax, proliferative tumor volume (PTV), and total lesion proliferation (TLP), of pretreatment [ 11 C]4DST PET for 91 patients with HNSCC with primary lesions in the oral cavity, hypopharynx, supraglottis, and oropharynx, which included p16-negative patients. PTV and TLP were calculated for primary lesions and metastatic lymph nodes combined. We examined the association among the parameters and relapse-free survival and whether case selection focused on biological characteristics improved the accuracy of prognosis prediction. RESULTS: The area under the curves (AUCs) using PTV and TLP were high for the oropharyngeal/hypopharyngeal/supraglottis groups (0.91 and 0.87, respectively), whereas that of SUVmax was 0.66 ( P  < 0.01). On the other hand, the oral group had lower AUCs for PTV and TLP (0.72 and 0.77, respectively). When all cases were examined, the AUCs using PTV and TLP were 0.84 and 0.83, respectively. CONCLUSION: Baseline [ 11 C]4DST PET/CT volume-based parameters can provide important prognostic information with p16-negative oropharyngeal, hypopharyngeal, and supraglottic cancer patients.

Association of follow-up imaging frequency with temporal incidence and patterns of distant failure following (chemo) radiotherapy for HPV related oropharyngeal cancer.

OBJECTIVES: Emerging data supports radical intent therapy for oligometastatic (OM) relapsed human papilloma virus (HPV+) related oropharyngeal cancer (OPC). We assess the association of follow-up imaging frequency amongst HPV + OPC, with temporal and spatial patterns of distant relapse, to inform rationalisation of routine post-treatment imaging. MATERIALS AND METHODS: A retrospective single centre cohort study was carried out of consecutive HPV + OPC patients treated with radical intent (chemo)radiotherapy ((CT)RT) between 2011 and 2019. OM state was defined as ≤ 5 metastasis, none larger than 3 cm (OMs) or, if interval from last negative surveillance imaging > 6-months, then ≤ 10 metastasis, none larger than 5 cm, (OMp). Patients not meeting OMs / OMp criteria were deemed to have incurable diffuse metastatic disease (DMdiffuse). RESULTS: 793 HPV-OPC patients were identified with median follow-up 3.15years (range 0.2-8.9). 52 (6.6 %) patients had radiologically identified DM at first failure and were considered for analysis. The median time to recurrence was 15.1 months (range: 2.6-63 months). 87 % of distant metastasis (DM) occurred in the first two years after treatment. Twenty-seven (52 %) patients had OM (OMs or OMp) at time of failure, with 31 % having OMs. The median time from completion of treatment to diagnosis of DMdiffuse vs OM was 22.2 months (range: 2.6-63.1 months) vs 11.6 months (range: 3.5-32.5 months). The probability of being diagnosed with OM vs DMdiffuse increased with reducing interval from last negative surveillance scan to imaging identifying DM (≤6 months 88.9 %, 7-12 months 71.4 %, 13-24 months 35 %, > 24 months 22.2 %). CONCLUSION: We demonstrate that a reduced interval between last negative imaging and subsequent radiological diagnosis of DM is associated with increased likelihood of identification of OM disease. Consideration of increased frequency of surveillance imaging during the first two years of follow up is supported, particularly for patients at high risk of distant failure.

Longitudinal Pattern of Lymphedema and Fibrosis in Patients With Oral Cavity or Oropharyngeal Cancer: A Prospective Study.

PURPOSE: The study aimed to describe the prevalence, severity, and trajectory of internal lymphedema, external lymphedema, and fibrosis in patients with oral cavity or oropharyngeal (OCOP) cancer. METHODS AND MATERIALS: One hundred twenty patients with newly diagnosed OCOP cancer were enrolled in a prospective longitudinal study. Recruitment was conducted at a comprehensive medical center. Participants were assessed pretreatment; at end of treatment; and at 3, 6, 9, and 12 months post-cancer treatment. Validated clinician-reported measures and computed tomography were used to assess the study outcomes. RESULTS: Seventy-six patients who completed the 9- or 12-month assessments were included in this report. Examination of the external lymphedema and fibrosis trajectories revealed that the total severity score peaked between the end of treatment and 3 months posttreatment and then decreased gradually over time but did not return to baseline by 12 months posttreatment (P < .001). The longitudinal patterns of severity scores for patients treated with surgery only or with multimodality therapy were similar. Examination of the internal swelling trajectories revealed that all patients experienced a significant increase in sites with swelling immediately posttreatment. For patients treated with surgery only, swelling was minimal and returned to baseline by 9 to 12 months posttreatment. Patients receiving multimodal treatment experienced a gradual decrease in number of sites with swelling during the 12-month posttreatment period that remained significantly above baseline (P < .05). Computed tomography revealed different patterns of changes in prevertebral soft tissue and epiglottic thickness in the surgery-only and multimodality treatment groups during the 12-month posttreatment period. There were minimal changes in thickness in both regions in the surgery-only group. Patients with multimodal treatment had significant increases in thickness in both regions 3 months posttreatment that remained thicker at 12 months than at baseline (P < .001). CONCLUSIONS: Lymphedema and fibrosis are the common complications of OCOP cancer therapy. Routine assessment, monitoring, and timely treatment of lymphedema and fibrosis are critical.

Vision Transformer-Based Multilabel Survival Prediction for Oropharynx Cancer After Radiation Therapy.

PURPOSE: A reliable and comprehensive cancer prognosis model for oropharyngeal cancer (OPC) could better assist in personalizing treatment. In this work, we developed a vision transformer-based (ViT-based) multilabel model with multimodal input to learn complementary information from available pretreatment data and predict multiple associated endpoints for radiation therapy for patients with OPC. METHODS AND MATERIALS: A publicly available data set of 512 patients with OPC was used for both model training and evaluation. Planning computed tomography images, primary gross tumor volume masks, and 16 clinical variables representing patient demographics, diagnosis, and treatment were used as inputs. To extract deep image features with global attention, we used a ViT module. Clinical variables were concatenated with the learned image features and fed into fully connected layers to incorporate cross-modality features. To learn the mapping between the features and correlated survival outcomes, including overall survival, local failure-free survival, regional failure-free survival, and distant failure-free survival, we employed 4 multitask logistic regression layers. The proposed model was optimized by combining the multitask logistic regression negative-log likelihood losses of different prediction targets. RESULTS: We employed the C-index and area under the curve metrics to assess the performance of our model for time-to-event prediction and time-specific binary prediction, respectively. Our proposed model outperformed corresponding single-modality and single-label models on all prediction labels, achieving C-indices of 0.773, 0.765, 0.776, and 0.773 for overall survival, local failure-free survival, regional failure-free survival, and distant failure-free survival, respectively. The area under the curve values ranged between 0.799 and 0.844 for different tasks at different time points. Using the medians of predicted risks as the thresholds to identify high-risk and low-risk patient groups, we performed the log-rank test, the results of which showed significantly larger separations in different event-free survivals. CONCLUSION: We developed the first model capable of predicting multiple labels for OPC simultaneously. Our model demonstrated better prognostic ability for all the prediction targets compared with corresponding single-modality models and single-label models.

Cost-Effectiveness Analysis of PET-CT Surveillance After Treatment of Human Papillomavirus-Positive Oropharyngeal Cancer.

OBJECTIVE: To determine the cost-effectiveness of surveillance imaging with PET/CT scan among patients with human papillomavirus-positive oropharyngeal squamous cell carcinoma. STUDY DESIGN: Cost-effectiveness analysis. SETTING: Oncologic care centers in the United States with head and neck oncologic surgeons and physicians. METHODS: We compared the cost-effectiveness of 2 posttreatment surveillance strategies: clinical surveillance with the addition of PET/CT scan versus clinical surveillance alone in human papillomavirus-positive oropharyngeal squamous cell carcinoma patients. We constructed a Markov decision model which was analyzed from a third-party payer's perspective using 1-year Markov cycles and a 30-year time horizon. Values for transition probabilities, costs, health care utilities, and their studied ranges were derived from the literature. RESULTS: The incremental cost-effectiveness ratio for PET/CT with clinical surveillance versus clinical surveillance alone was $89,850 per quality-adjusted life year gained. Flexible fiberoptic scope exams during clinical surveillance would have to be over 51% sensitive or PET/CT scan cost would have to exceed $1678 for clinical surveillance alone to be more cost-effective. The willingness-to-pay threshold at which imaging surveillance was equally cost-effective to clinical surveillance was approximately $80,000/QALY. CONCLUSION: Despite lower recurrence rates of human papillomavirus-positive oropharyngeal cancer, a single PET/CT scan within 6 months after primary treatment remains a cost-effective tool for routine surveillance when its cost does not exceed $1678. The cost-effectiveness of this strategy is also dependent on the clinical surveillance sensitivity (flexible fiberoptic pharyngoscopy), and willingness-to-pay thresholds which vary by country.

Deep learning-based outcome prediction using PET/CT and automatically predicted probability maps of primary tumor in patients with oropharyngeal cancer.

BACKGROUND AND OBJECTIVE: Recently, deep learning (DL) algorithms showed to be promising in predicting outcomes such as distant metastasis-free survival (DMFS) and overall survival (OS) using pre-treatment imaging in head and neck cancer. Gross Tumor Volume of the primary tumor (GTVp) segmentation is used as an additional channel in the input to DL algorithms to improve model performance. However, the binary segmentation mask of the GTVp directs the focus of the network to the defined tumor region only and uniformly. DL models trained for tumor segmentation have also been used to generate predicted tumor probability maps (TPM) where each pixel value corresponds to the degree of certainty of that pixel to be classified as tumor. The aim of this study was to explore the effect of using TPM as an extra input channel of CT- and PET-based DL prediction models for oropharyngeal cancer (OPC) patients in terms of local control (LC), regional control (RC), DMFS and OS. METHODS: We included 399 OPC patients from our institute that were treated with definitive (chemo)radiation. For each patient, CT and PET scans and GTVp contours, used for radiotherapy treatment planning, were collected. We first trained a previously developed 2.5D DL framework for tumor probability prediction by 5-fold cross validation using 131 patients. Then, a 3D ResNet18 was trained for outcome prediction using the 3D TPM as one of the possible inputs. The endpoints were LC, RC, DMFS, and OS. We performed 3-fold cross validation on 168 patients for each endpoint using different combinations of image modalities as input. The final prediction in the test set (100) was obtained by averaging the predictions of the 3-fold models. The C-index was used to evaluate the discriminative performance of the models. RESULTS: The models trained replacing the GTVp contours with the TPM achieved the highest C-indexes for LC (0.74) and RC (0.60) prediction. For OS, using the TPM or the GTVp as additional image modality resulted in comparable C-indexes (0.72 and 0.74). CONCLUSIONS: Adding predicted TPMs instead of GTVp contours as an additional input channel for DL-based outcome prediction models improved model performance for LC and RC.

Radiomics analysis for the prediction of locoregional recurrence of locally advanced oropharyngeal cancer and hypopharyngeal cancer.

PURPOSE: By radiomic analysis of the postcontrast CT images, this study aimed to predict locoregional recurrence (LR) of locally advanced oropharyngeal cancer (OPC) and hypopharyngeal cancer (HPC). METHODS: A total of 192 patients with stage III-IV OPC or HPC from two independent cohort were randomly split into a training cohort with 153 cases and a testing cohort with 39 cases. Only primary tumor mass was manually segmented. Radiomic features were extracted using PyRadiomics, and then the support vector machine was used to build the radiomic model with fivefold cross-validation process in the training data set. For each case, a radiomics score was generated to indicate the probability of LR. RESULTS: There were 94 patients with LR assigned in the progression group and 98 patients without LR assigned in the stable group. There was no significant difference of TNM staging, treatment strategies and common risk factors between these two groups. For the training data set, the radiomics model to predict LR showed 83.7% accuracy and 0.832 (95% CI 0.72, 0.87) area under the ROC curve (AUC). For the test data set, the accuracy and AUC slightly declined to 79.5% and 0.770 (95% CI 0.64, 0.80), respectively. The sensitivity/specificity of training and test data set for LR prediction were 77.6%/89.6%, and 66.7%/90.5%, respectively. CONCLUSIONS: The image-based radiomic approach could provide a reliable LR prediction model in locally advanced OPC and HPC. Early identification of those prone to post-treatment recurrence would be helpful for appropriate adjustments to treatment strategies and post-treatment surveillance.

Interpretable radiomics method for predicting human papillomavirus status in oropharyngeal cancer using Bayesian networks.

OBJECTIVES: To develop a simple interpretable Bayesian Network (BN) to classify HPV status in patients with oropharyngeal cancer. METHODS: Two hundred forty-six patients, 216 of whom were HPV positive, were used in this study. We extracted 851 radiomics markers from patients' contrast-enhanced Computed Tomography (CT) images. Mens eX Machina (MXM) approach selected two most relevant predictors: sphericity and max2DDiameterRow. The area under the curve (AUC) demonstrated BN model performance in 30% of the data reserved for testing. A Support Vector Machine (SVM) based method was also implemented for comparison purposes. RESULTS: The Mens eX Machina (MXM) approach selected two most relevant predictors: sphericity and max2DDiameterRow. Areas under the Curves (AUC) were found 0.78 and 0.72 on the training and test data, respectively. When using support vector machine (SVM) and 25 features, the AUC was found 0.83 on the test data. CONCLUSIONS: The straightforward structure and power of interpretability of our BN model will help clinicians make treatment decisions and enable the non-invasive detection of HPV status from contrast-enhanced CT images. Higher accuracy can be obtained using more complex structures at the expense of lower interpretability. ADVANCES IN KNOWLEDGE: Radiomics is being studied lately as a simple imaging data based HPV status detection technique which can be an alternative to laboratory approaches. However, it generally lacks interpretability. This work demonstrated the feasibility of using Bayesian networks based radiomics for predicting HPV positivity in an interpretable way.

A comparison of loco-regional relapse pattern in HPV positive vs negative oropharyngeal cancers following radiotherapy; relation to pretreatment FDG-PET and radiotherapy target volumes.

BACKGROUND: Previous studies have shown that a large proportion of relapses in head-and neck squamous cell carcinoma (HNSCC) following radiotherapy (RT) occur in the pretreatment FDG-PET avid volume (GTV-PET). The aim of the current work was to see if this was valid also in an oropharynx squamous cell carcinoma (OPSCC) only population, and to compare the loco-regional relapse pattern between HPV positive and HPV negative patients. MATERIAL AND METHODS: Among 633 OPSCC patients treated between 2009 and 2017, 46 patients with known HPV (p16) status and isolated loco-regional relapse were included. Oncologists contoured relapse volumes (RV) on relapse scans (PET/CT, CT or MR), which were thereafter deformed to match the anatomy of the planning CTs. The point of origin (center of volume) of the deformed RVs were determined and analyzed in relation to the RT target volumes (GTV-PET, GTV and CTVs). The relapse pattern was compared between HPV positive and HPV negative patients using Fischer's exact test. RESULTS: Sixty RVs were contoured in the 46 patients. 55% (95% CI 44-67%) of relapses originated in GTV-PET, while the other RT volumes harbored 12% (5-20%) (GTV), 18% (9-28%) (high risk CTV) and 5% (0-11%) (low risk CTV) of relapses. Six relapses were found outside the RT target volumes. No significant difference in relapse pattern between HPV positive and HPV negative patients was found (p = .95). CONCLUSION: There were no signs of difference in loco-regional relapse pattern between HPV positive and HPV negative patients. In agreement with previous findings, GTV-PET was the most frequent RT target volume of relapse.

Evaluation treatment planning system for oropharyngeal cancer patient using machine learning.

Oropharyngeal cancer (OPC) comprises a group of various malignant tumours that grow in the throat, larynx, mouth, sinuses, and nose. THE RESEARCH AIMS: to investigate the performance of the OPC VMAT model by comparison to clinical plans in terms of dosimetric parameters and normal tissue complication probabilities. PURPOSE: Tune the model which at least matches the performance of clinical created photon treatment plans and analyse and find the most appropriate strategic plan scheme for OPC. METHODS AND MATERIALS: The machine learning (ML) plans are compared to the reference plans (clinical plans) based on dose constraints and target coverage. VMAT oropharynx ML model of Raystation development 11B version (non-clinical) was used. A model was trained by using different modalities. A different strategy of machine learning and clinical plans was performed for five patients. The dose Prescribed for OPC is 70 Gy, 2 Gy per fraction (2Gy/Fx). The PTV was derived for the primary tumour and secondary tumour, PTV+7000 cGy and PTV_5425 cGy volumetric modulated arc therapy (VMAT) were used with beams performing a full 360° rotation around the single isocenter. RESULTS: Organs at risk were observed that the volume of L-Eye in clinical plan (AF) for the case1 treatment planning could be successfully used ensuring efficiency and lower than MLVMAT and MLVMAT-org plans were 372 cGy, 697 cGy and 667 cGy respectively, while showed case2, case3, case4 and case5 are better to protect the critical organs in ML plan compare with a clinical plan. DHI for the PTV-7000 and PTV-5425 is between 1 and 1.34, While DCI for PTV-7000 and PTV-5425 is between 0.98 and 1.

Post-treatment 18F-fludeoxyglocuse-positron emission tomography in human papillomavirus-associated oropharyngeal cancer.

BACKGROUND: Human papillomavirus association has changed the landscape of treatment for oropharyngeal squamous cell carcinoma; it remains to be seen whether current post-treatment surveillance schedules are effective. OBJECTIVE: Evaluate whether post-treatment surveillance of oropharyngeal cancer through FDG-PET imaging is modified by human papillomavirus association. METHODS: A prospective cohort analysis of retrospective data was conducted for patients undergoing treatment of oropharyngeal cancer between 2016 and 2018. This study was conducted at a single large tertiary referral center in Brisbane, Australia. RESULTS: Two-hundred and twenty-four patients were recruited for the purposes of the study, 193 (86%) with HPV-associated disease. In this cohort FDG-PET had a sensitivity of 48.3%, specificity of 72.6%, positive predictive value of 23.7%, and negative predictive value of 88.8% in detecting disease recurrence. CONCLUSIONS: FDG-PET in HPV-associated oropharyngeal cancer has significantly lower positive predictive value when compared to non-HPV-associated oropharyngeal cancer. Caution should be used when interpreting positive post-treatment FDG-PET.

Dysphagia at 1 Year is Associated With Mean Dose to the Inferior Section of the Brain Stem.

PURPOSE: Dysphagia is a common toxicity after head and neck (HN) radiation therapy that negatively affects quality of life. We explored the relationship between radiation therapy dose to normal HN structures and dysphagia 1 year after treatment using image-based datamining (IBDM), a voxel-based analysis technique. METHODS AND MATERIALS: We used data from 104 patients with oropharyngeal cancer treated with definitive (chemo)radiation therapy. Swallow function was assessed pretreatment and 1 year posttreatment using 3 validated measures: MD Anderson Dysphagia Inventory (MDADI), performance status scale for normalcy of diet (PSS-HN), and water swallowing test (WST). For IBDM, we spatially normalized all patients' planning dose matrices to 3 reference anatomies. Regions where the dose was associated with dysphagia measures at 1 year were found by performing voxel-wise statistics and permutation testing. Clinical factors, treatment variables, and pretreatment measures were used in multivariable analysis to predict each dysphagia measure at 1 year. Clinical baseline models were found using backward stepwise selection. Improvement in model discrimination after adding the mean dose to the identified region was quantified using the Akaike information criterion. We also compared the prediction performance of the identified region with a well-established association: mean doses to the pharyngeal constrictor muscles. RESULTS: IBDM revealed highly significant associations between dose to distinct regions and the 3 outcomes. These regions overlapped around the inferior section of the brain stem. All clinical models were significantly improved by including mean dose to the overlap region (P ≤ .006). Including pharyngeal dosimetry significantly improved WST (P = .04) but not PSS-HN or MDADI (P ≥ .06). CONCLUSIONS: In this hypothesis-generating study, we found that mean dose to the inferior section of the brain stem is strongly associated with dysphagia 1 year posttreatment. The identified region includes the swallowing centers in the medulla oblongata, providing a possible mechanistic explanation. Further work including validation in an independent cohort is required.

Prognostic and predictive values of baseline and mid-treatment FDG-PET in oropharyngeal carcinoma treated with primary definitive (chemo)radiation and impact of HPV status: Review of current literature and emerging roles.

BACKGROUND AND PURPOSE: This study provides a review of the literature assessing whether semiquantitative PET parameters acquired at baseline and/or during definitive (chemo)radiotherapy ("prePET" and "iPET") can predict survival outcomes in patients with oropharyngeal squamous cell carcinoma (OPC), and the impact of human papilloma virus (HPV) status. MATERIAL AND METHODS: A literature search was carried out using PubMed and Embase between 2001 to 2021 in accordance with PRISMA. RESULTS: The analysis included 22 FDG-PET/CT studies [1-22], 19 pre-PET and 3 both pre-PET and iPET, The analysis involved 2646 patients, of which 1483 are HPV-positive (17 studies: 10 mixed and 7 HPV-positive only), 589 are HPV-negative, and 574 have unknown HPV status. Eighteen studies found significant correlations of survival outcomes with pre-PET parameters, most commonly primary or "Total" (combined primary and nodal) metabolic tumour volume and/or total lesional glycolysis. Two studies could not establish significant correlations and both employed SUVmax only. Two studies also could not establish significant correlations when taking into account of the HPV-positive population only. Because of the heterogeneity and lack of standardized methodology, no conclusions on optimal cut-off values can be drawn. Ten studies specifically evaluated HPV-positive patients: five showed positive correlation of pre-PET parameters and survival outcomes, but four of these studies did not include advanced T or N staging in multivariate analysis, and two studies only showed positive correlations after excluding high risk patients with smoking history or adverse CT features. Two studies found that prePET parameters predicted treatment outcomes only in HPV-negative but not HPV-positive patients. Two studies found that iPET parameters could predict outcomes in HPV-positive patients but not prePET parameters. CONCLUSION: The current literature supports high pre-treatment metabolic burden prior to definitive (chemo)radiotherapy can predict poor treatment outcomes for HPV-negative OPC patients. Evidence is conflicting and currently does not support correlation in HPV-positive patients.

Iterative Metal Artifact Reduction in Head and Neck CT Facilitates Tumor Visualization of Oral and Oropharyngeal Cancer Obscured by Artifacts From Dental Hardware.

RATIONALE AND OBJECTIVES: The purpose of this study was to evaluate the diagnostic utility of iterative metal artifact reduction (iMAR) in computed tomography (CT)-imaging of oral and oropharyngeal cancers when obscured by dental hardware artifacts and to determine the most appropriate iMAR settings for this purpose. MATERIALS AND METHODS: The study retrospectively enrolled 27 patients (8 female, 19 male; mean age 64±12.7years) with histologically confirmed oral or oropharyngeal cancer obscured by dental artifacts in contrast-enhanced CT. Raw CT data were reconstructed with ascending iMAR strengths (levels 1/2/3/4/5) and one reconstruction without iMAR (level 0). For subjective analysis, two blinded radiologists rated tumor visualization and artifact severity on a five-point Likert scale. For objective analysis, signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), and artifact index (AI) were determined. RESULTS: iMAR reconstructions improved the subjective image quality of tumor edge and contrast, and the objective parameters of tumor SNR and CNR, reaching their optimum at iMAR levels 4 and 5 (P<.001). AI decreased with iMAR reconstructions reaching its minimum at iMAR level 5 (P<.001). Tumor detection rates increased 2.4-fold with iMAR 5, 2.1-fold with iMAR 4, and 1.9-fold with iMAR 3 compared to reconstructions without iMAR. Disadvantages such as algorithm-induced artifacts increased significantly with higher iMAR strengths (P<.05), reaching a maximum with iMAR 5. CONCLUSION: iMAR significantly improves CT imaging of oral and oropharyngeal cancers, as confirmed by both subjective and objective measures, with best results at highest iMAR strengths.

CT-based deep multi-label learning prediction model for outcome in patients with oropharyngeal squamous cell carcinoma.

BACKGROUND: Personalized treatment is increasingly required for oropharyngeal squamous cell carcinoma (OPSCC) patients due to emerging new cancer subtypes and treatment options. Outcome prediction model can help identify low or high-risk patients who may be suitable to receive de-escalation or intensified treatment approaches. PURPOSE: To develop a deep learning (DL)-based model for predicting multiple and associated efficacy endpoints in OPSCC patients based on computed tomography (CT). METHODS: Two patient cohorts were used in this study: a development cohort consisting of 524 OPSCC patients (70% for training and 30% for independent testing) and an external test cohort of 396 patients. Pre-treatment CT-scans with the gross primary tumor volume contours (GTVt) and clinical parameters were available to predict endpoints, including 2-year local control (LC), regional control (RC), locoregional control (LRC), distant metastasis-free survival (DMFS), disease-specific survival (DSS), overall survival (OS), and disease-free survival (DFS). We proposed DL outcome prediction models with the multi-label learning (MLL) strategy that integrates the associations of different endpoints based on clinical factors and CT-scans. RESULTS: The multi-label learning models outperformed the models that were developed based on a single endpoint for all endpoints especially with high AUCs ≥ 0.80 for 2-year RC, DMFS, DSS, OS, and DFS in the internal independent test set and for all endpoints except 2-year LRC in the external test set. Furthermore, with the models developed, patients could be stratified into high and low-risk groups that were significantly different for all endpoints in the internal test set and for all endpoints except DMFS in the external test set. CONCLUSION: MLL models demonstrated better discriminative ability for all 2-year efficacy endpoints than single outcome models in the internal test and for all endpoints except LRC in the external set.

Machine learning-based automatic proton therapy planning: Impact of post-processing and dose-mimicking in plan robustness.

PURPOSE: Automated treatment planning strategies are being widely implemented in clinical routines to reduce inter-planner variability, speed up the optimization process, and improve plan quality. This study aims to evaluate the feasibility and quality of intensity-modulated proton therapy (IMPT) plans generated with four different knowledge-based planning (KBP) pipelines fully integrated into a commercial treatment planning system (TPS). MATERIALS/METHODS: A data set containing 60 oropharyngeal cancer patients was split into 11 folds, each containing 47 patients for training, five patients for validation, and five patients for testing. A dose prediction model was trained on each of the folds, resulting in a total of 11 models. Three patients were left out in order to assess if the differences introduced between models were significant. From voxel-based dose predictions, we analyze the two steps that follow the dose prediction: post-processing of the predicted dose and dose mimicking (DM). We focused on the effect of post-processing (PP) or no post-processing (NPP) combined with two different DM algorithms for optimization: the one available in the commercial TPS RayStation (RSM) and a simpler isodose-based mimicking (IBM). Using 55 test patients (five test patients for each model), we evaluated the quality and robustness of the plans generated by the four proposed KBP pipelines (PP-RSM, PP-IBM, NPP-RSM, NPP-IBM). After robust evaluation, dose-volume histogram (DVH) metrics in nominal and worst-case scenarios were compared to those of the manually generated plans. RESULTS: Nominal doses from the four KBP pipelines showed promising results achieving comparable target coverage and improved dose to organs at risk (OARs) compared to the manual plans. However, too optimistic post-processing applied to the dose prediction (i.e. important decrease of the dose to the organs) compromised the robustness of the plans. Even though RSM seemed to partially compensate for the lack of robustness in the PP plans, still 65% of the patients did not achieve the expected robustness levels. NPP-RSM plans seemed to achieve the best trade-off between robustness and OAR sparing. DISCUSSION/CONCLUSIONS: PP and DM strategies are crucial steps to generate acceptable robust and deliverable IMPT plans from ML-predicted doses. Before the clinical implementation of any KBP pipeline, the PP and DM parameters predefined by the commercial TPS need to be modified accordingly with a comprehensive feedback loop in which the robustness of the final dose calculations is evaluated. With the right choice of PP and DM parameters, KBP strategies have the potential to generate IMPT plans within clinically acceptable levels comparable to plans manually generated by dosimetrists.

Multifactorial estimation of clinical outcome in HPV-associated oropharyngeal squamous cell carcinoma via automated image analysis of routine diagnostic H&E slides and neural network modelling.

OBJECTIVE: Routine haematoxylin and eosin (H&E) photomicrographs from human papillomavirus-associated oropharyngeal squamous cell carcinomas (HPV + OpSCC) contain a wealth of prognostic information. In this study, we developed a high content image analysis (HCIA) workflow to quantify features of H&E images from HPV + OpSCC patients to identify prognostic features and predict patient outcomes. METHODS: First, we have developed an open-source HCIA tool for single-cell segmentation and classification of H&E images. Subsequently, we have used our HCIA tool to analyse a set of 889 images from diagnostic H&E slides in a retrospective cohort of HPV + OpSCC patients with favourable (FO, n = 60) or unfavourable (UO, n = 30) outcomes. We have identified and measured 31 prognostic features which were quantified in each sample and used to train a neural network (NN) model to predict patient outcomes. RESULTS: Univariate and multivariate statistical analyses revealed significant differences between FO and UO patients in 31 and 17 variables, respectively (P < 0.05). At the single-image level, the NN model had an overall accuracy of 72.5% and 71.2% in recognising FO and UO patients when applied to test or validation sets, respectively. When considering 10 images per patient, the accuracy of the NN model increased to 86.7% in the test set. CONCLUSION: Our open-source H&E analysis workflow and predictive models confirm previously reported prognostic features and identifies novel factors which predict HPV + OpSCC outcomes with promising accuracy. Our work supports the use of machine learning in digital pathology to exploit clinically relevant features in routine diagnostic pathology without additional biomarkers.

Incidence and predictive factors of retropharyngeal lymph node metastases in patients with oropharyngeal cancer undergoing multimodality treatment planning imaging.

BACKGROUND: We investigated the incidence and predictive factors of retropharyngeal lymph node (RPLN) metastases in patients with oropharyngeal cancer (OPC) undergoing multimodality treatment planning imaging before radiotherapy. METHODS: Consecutive patients with OPC treated with curative-intent radiotherapy from 2017 to 2019 were retrospectively analyzed. Treatment planning comprised contrast-enhanced computed tomography (CT), magnetic resonance imaging (MRI), and fluorodeoxyglucose-positron emission tomography (FDG-PET) unless contraindicated. RESULTS: Of 300 patients, 66 (22%) had radiological evidence of RPLN involvement on planning images, compared to 17 (6%) on diagnostic CT alone. On multivariate analysis, RPLN involvement was statistically (p < 0.05) associated with tonsil, soft palate, and posterior pharyngeal wall primaries, and with disease extension to the soft palate or vallecula. CONCLUSIONS: Multimodality treatment planning imaging reveals a high rate of RPLN metastases from OPC compared to diagnostic CT alone. Patients with tonsil, soft palate, or posterior pharyngeal wall primaries or disease extending to the soft palate or vallecula appear at higher risk.

Screening for extranodal extension in HPV-associated oropharyngeal carcinoma: evaluation of a CT-based deep learning algorithm in patient data from a multicentre, randomised de-escalation trial.

BACKGROUND: Pretreatment identification of pathological extranodal extension (ENE) would guide therapy de-escalation strategies for in human papillomavirus (HPV)-associated oropharyngeal carcinoma but is diagnostically challenging. ECOG-ACRIN Cancer Research Group E3311 was a multicentre trial wherein patients with HPV-associated oropharyngeal carcinoma were treated surgically and assigned to a pathological risk-based adjuvant strategy of observation, radiation, or concurrent chemoradiation. Despite protocol exclusion of patients with overt radiographic ENE, more than 30% had pathological ENE and required postoperative chemoradiation. We aimed to evaluate a CT-based deep learning algorithm for prediction of ENE in E3311, a diagnostically challenging cohort wherein algorithm use would be impactful in guiding decision-making. METHODS: For this retrospective evaluation of deep learning algorithm performance, we obtained pretreatment CTs and corresponding surgical pathology reports from the multicentre, randomised de-escalation trial E3311. All enrolled patients on E3311 required pretreatment and diagnostic head and neck imaging; patients with radiographically overt ENE were excluded per study protocol. The lymph node with largest short-axis diameter and up to two additional nodes were segmented on each scan and annotated for ENE per pathology reports. Deep learning algorithm performance for ENE prediction was compared with four board-certified head and neck radiologists. The primary endpoint was the area under the curve (AUC) of the receiver operating characteristic. FINDINGS: From 178 collected scans, 313 nodes were annotated: 71 (23%) with ENE in general, 39 (13%) with ENE larger than 1 mm ENE. The deep learning algorithm AUC for ENE classification was 0·86 (95% CI 0·82-0·90), outperforming all readers (p<0·0001 for each). Among radiologists, there was high variability in specificity (43-86%) and sensitivity (45-96%) with poor inter-reader agreement (κ 0·32). Matching the algorithm specificity to that of the reader with highest AUC (R2, false positive rate 22%) yielded improved sensitivity to 75% (+ 13%). Setting the algorithm false positive rate to 30% yielded 90% sensitivity. The algorithm showed improved performance compared with radiologists for ENE larger than 1 mm (p<0·0001) and in nodes with short-axis diameter 1 cm or larger. INTERPRETATION: The deep learning algorithm outperformed experts in predicting pathological ENE on a challenging cohort of patients with HPV-associated oropharyngeal carcinoma from a randomised clinical trial. Deep learning algorithms should be evaluated prospectively as a treatment selection tool. FUNDING: ECOG-ACRIN Cancer Research Group and the National Cancer Institute of the US National Institutes of Health.