Lisanne Vania van Dijk

Introduction:Traditional NTCP models for predicting radiation-induced pneumonitis (RP) in lung cancer patients often rely on simplified dose characteristics, such as the mean lung dose, neglecting the spatial heterogeneity of dose distributions. This study introduces a deep learning (DL)-based NTCP model that integrates spatial dose information with clinical and anatomical data to improve RP prediction in loco-regionally advanced-stage lung cancer patients undergoing radiotherapy.Methods:A single-institute, retrospective cohort of 1,056 patients treated between 2013 and 2024 was used. RP was defined as CTCAE grade ≥2 within 6 or 12 months from the start of radiotherapy. The dataset was randomly split into a development and independent test set, stratified by RP incidence and immunotherapy status. A ResNet18-based DL model was trained using five-fold cross-validation. Inputs included planning CT, 3D dose distribution, lung minus gross tumour volume (Lungs-GTV) segmentation, and clinical variables (age, gender, smoking history, pulmonary comorbidities, immunotherapy, chemotherapy, surgical interventions, dose per fraction). Model performance was compared to a recent published RP NTCP model. To improve model interpretability, attentions maps generated using Grad-Cam++ highlighted regions of interest.Results:The DL-based NTCP model achieved enhanced performance on the independent test set (AUC=0.79, 95%CI[0.72-0.85]) compared to the reference model (AUC=0.76, 95%CI[0.70-0.83]) for predicting RP within 12 months. Attention maps mainly highlighted low-to-mid dose regions in the ipsilateral lung for higher RP prediction risk.Conclusion:Incorporating spatial dose information via a ResNet18-based DL model enhances RP prediction over traditional NTCP models.

Background and Purpose: Due to superior soft tissue contrast, MRI may provide more prognostic information than CT/PET for outcome prediction. This study aims to compare the prognostic value of MRI with CT and PET in deep learning models for local control, regional control and overall survival in oropharyngeal cancer patients compared to a clinical benchmark model.Materials and Methods: A dataset comprising 266 oropharyngeal cancer patients was assembled. Each patient’s data includes pretreatment axial T1 and a coronal T2 MRI scan, CT and PET scans, gross tumor volume of the primary tumor, clinical parameters and information on local control, regional control and overall survival. Various 2D and 3D convolutional neural networks were trained using images of contoured gross tumor volume with and without a margin for outcome prediction.Results: The 2D models using T2 images within a bounding box region determined by the gross tumor volume achieved concordance index of 0.88 and 0.75 for local control and overall survival prediction, respectively. Additionally, MRI-based models achieved higher concordance indexes than CT- or PET-based models for local control prediction. In comparison to a clinical benchmark model, the T2-based 2D model showed improved local control prediction (concordance index: 0.88 vs. 0.80), and combining the T2 prediction with clinical model improved overall survival prediction (concordance index 0.81 vs. 0.78). Furthermore, the clinical + MRI models showed enhanced performance compared to a clinical routine survival risk stratification system.Conclusions: MRI-based deep learning models can improve prediction of local control and overall survival in oropharyngeal cancer.

Purpose/Objective(s) Osteoradionecrosis of the jaw (ORN) is a severe complication of head and neck cancer (HNC) radiotherapy (RT), significantly impacting patient quality of life. Current dose assessment relies on whole-mandible dosimetry, limiting personalized, tooth-specific region risk evaluation. The lack of standardized methods to document and analyze dose distributions to tooth-bearing regions further hampers dental and maxillofacial decision-making. This study develops and evaluates RADMAP, a (semi-)automated tool for segmenting tooth-based jaw regions, enabling patient- and tooth-specific radiation dose mapping to improve dental dose reporting and ORN risk assessment. Methods and Materials A total of 736 tooth locations from 23 HNC patients treated with definitive RT were analyzed, including 11 who developed ORN. The RADMAP tool applies an angular ray-based algorithm to automatically segment the mandible and maxilla into 32 tooth-specific jaw regions, with optional manual refinement (semi-automated), outputting radiation dose mapping in both tabular and odontogram formats. Mean dose values from manually contoured tooth roots were compared with RADMAP-segmented alveolar regions (fully and semi-automated). Interobserver agreement among six users was evaluated. Whole-mandible, manual tooth, and tooth-based jaw segment (alveolar and basal) doses were compared between regions with and without ORN development. Results Mean dose correlation between RADMAP-segmented alveolar jaw regions and manually segmented tooth roots was high for both fully and semi-automated approaches (R² = 0.98, pConclusion RADMAP enables accurate, tooth-specific dose mapping of the mandible and maxilla, and potential improved ORN risk differentiation beyond whole-mandible dosimetry. Tooth-based dose reporting can personalize treatment, enhance multidisciplinary communication, and support prevention of radiation-related orodental sequelae.

Early identification of patients at risk of immune effector cell–associated hematotoxicity (ICAHT) is essential to minimize nonrelapse mortality. The CAR-HEMATOTOX (HT) score is an implemented risk-stratification tool for ICAHT, infections, and survival in patients with relapsed/refractory large B-cell lymphoma (R/R LBCL) receiving chimeric antigen receptor T-cell therapy (CART). Although validated in its defining study, the HT score was developed in a small cohort, necessitating independent external validation. This study externally validates the HT score in a real-world population-based cohort of adults with R/R LBCL receiving CART. The HT score, based on absolute neutrophil count (ANC), hemoglobin, platelets, C-reactive protein, and ferritin, was calculated before lymphodepleting chemotherapy. Of 245 consecutive patients, 171 (70%) had an HT score of ≥2 (HThigh). The initial end point, clinically significant neutropenia (ANC of <500/μL for ≥14 days), occurred in 21% of patients. The binary HT score was associated with clinically significant neutropenia (odds ratio [OR], 2.94; 95% confidence interval [CI], 1.27-6.80; P = .012) with a good predictive performance (area under the curve = 0.73). Similar results were achieved for early and late ICAHT grade ≥3 (OR, 2.92; 95% CI, 1.19-7.14; P = .019; and OR, 2.42; 95% CI, 1.31-4.47; P = .005). A trend toward an association with severe infections was observed (OR, 2.02; 95% CI, 0.91-4.48; P = .085). HThigh patients had a lower progression-free and overall survival (hazard ratio [HR], 1.84; 95% CI, 1.15-2.93; P = .011; and HR, 2.83; 95% CI, 1.64-4.87; P < .001, respectively). The HT score identified CART-treated patients with R/R LBCL at risk of clinically significant neutropenia, poor survival outcomes, and potentially severe infections.

Accurate delineation of orodental structures on computed tomography (CT) is critical for image-guided assessments of radiation-associated bone injury. This dataset comprises curated CT imaging and expert-defined segmentation masks for 60 patients with head and neck cancer treated with radiotherapy (RT), including delineations of mandibular and maxillary sub-volumes and individual teeth. Segmentation guidelines were informed by anatomical differences across sub-regions and aligned with the ClinRad osteoradionecrosis (ORN) staging system. The dataset includes converted NIfTI files of simulation CT images, RT dose distributions, and delineated structures. All segmentations were performed manually using a standardized protocol in a commercial treatment planning system and converted to research-ready formats using open-source tools. This dataset may facilitate the development and validation of automated segmentation tools, dose mapping applications, and image-based ORN detection pipelines in head and neck cancer survivors.

Osteoradionecrosis of the jaw (ORNJ) is a radiation-induced late toxicity that can dramatically decrease patients’ quality of life. Recent increases in survival rates of head and neck cancers associated with human papillomavirus (HPV) infection have resulted in a higher frequency of radiation-induced toxicities, particularly ORNJ. Recent work with Normal Tissue Complication Probability (NTCP) models and a Weibull Accelerated Failure Time (WAFT) model have further developed our understanding of ORNJ clinical/dosimetric risk factors and longitudinal features, respectively. In this data descriptor, 1129 head and neck cancer (HNC) patients received curative intent radiotherapy (RT) at MD Anderson Cancer Center and were follow-up with clinical and radiological assessments at 3-6, 12, 18, 24 months, and then annually following the conclusion of RT for development of ORNJ. This data, in addition to the patients’ demographic, supplementary clinical, and dosimetric information was recorded in a comma-separated value file embedded within this data descriptor. This large, longitudinal dataset is a significant resource for further systematic analysis of post-RT normal tissue outcomes in HNC.

Background & purpose Late radiation-associated dysphagia after head and neck cancer (HNC) significantly impacts patient’s health and quality of life. Conventional normal tissue complication probability (NTCP) models use discrete dose parameters to predict toxicity risk but fail to fully capture the complexity of this side effect. Deep learning (DL) offers potential improvements by incorporating 3D dose data for all anatomical structures involved in swallowing. This study aims to enhance dysphagia prediction with 3D DL NTCP models compared to conventional NTCP models. Materials & methods A multi-institutional cohort of 1484 HNC patients was used to train and validate a 3D DL model (Residual Network) incorporating 3D dose distributions, organ-at-risk segmentations, and CT scans, with or without patient- or treatment-related data. Predictions of grade ≥2 dysphagia (CTCAEv4) at six months post-treatment were evaluated using area under the curve (AUC) and calibration curves. Results were compared to a conventional NTCP model based on pre-treatment dysphagia, tumour location, and mean dose to swallowing organs. Attention maps highlighting regions of interest for individual patients were assessed. Results DL models outperformed the conventional NTCP model in both the independent test set (AUC=0.80-0.84 versus 0.76) and external test set (AUC=0.73-0.74 versus 0.63) in AUC and calibration. Attention maps showed a focus on the oral cavity and superior pharyngeal constrictor muscle. Conclusion DL NTCP models performed better than the conventional NTCP model, suggesting the benefit of using 3D-input over the conventional discrete dose parameters. Attention maps highlighted relevant regions linked to dysphagia, supporting the utility of DL for improved predictions.