¹⁸F-fluorodeoxyglucose positron emission tomography/computed tomography (¹⁸F-FDG PET/CT) has become a cornerstone in the management of head and neck squamous cell carcinoma (HNSCC). While conventional imaging defines anatomical extent, PET provides complementary metabolic information crucial for staging, prognostication, treatment adaptation, and surveillance. This review aims to synthesize current evidence on the role of PET imaging in HNSCC, integrating recent data on quantitative biomarkers, novel radiotracers, and artificial intelligence (AI)–based analytical approaches. PET/CT demonstrates superior sensitivity for nodal and distant metastases compared with CT or MRI, particularly in advanced-stage or unknown-primary disease. Quantitative metrics such as SUVmax, metabolic tumor volume, and total lesion glycolysis provide independent prognostic value beyond TNM staging. In the post-treatment setting, ¹⁸F-FDG PET/CT reliably identifies complete metabolic response, reducing unnecessary neck dissections. Limitations include false positives early after therapy due to inflammation or osteoradionecrosis; dual-timepoint acquisition and PET/MRI may enhance specificity. New tracers targeting hypoxia (¹⁸F-FMISO), proliferation (¹⁸F-FLT), angiogenesis (⁶⁸Ga-RGD), and cancer-associated fibroblasts (⁶⁸Ga-FAPI) show promise for biological target delineation. AI methods, including radiomics and deep learning, have demonstrated potential for automated nodal detection and outcome prediction, though clinical validation remains limited. ¹⁸F-FDG PET/CT, complemented by emerging molecular tracers and AI-driven analysis, represents a pivotal tool for personalized management of HNSCC. By integrating anatomical, metabolic, and biological information, PET imaging supports refined staging, individualized radiotherapy planning, and improved patient outcomes. PET/CT improves post-treatment response assessment in HNSCC. High NPV supports PET/CT in surveillance strategies. PET/CT enhances the detection of regional and distant recurrences. Imaging timing post-CRT is key for diagnostic accuracy. PET/CT may guide personalized follow-up and salvage therapy.

Background: Cancer-related sarcopenia (CRS) is a significant complication of head and neck carcinoma (HNC), characterised by muscle degeneration and poor clinical outcomes. Although various dietary and therapeutic interventions have been explored, most of them remain empirical, and the molecular mechanisms underlying CRS are not yet fully understood. Methods: Transcription profiles of muscle fragments from 29 HNC patients and 8 control donors were analysed by bulk RNA sequencing (6/29 and 3/8) and/or RT-qPCR (29/29 and 5/8). In parallel, differentiating human myoblasts (AB1190) were subjected to indirect co-culture with two types of effector cells: HNC cells (FaDu) or control epithelial cells (NHEK). The contactless effects of effector cells on target myoblasts were investigated using cell imaging to assess muscular differentiation, RT-qPCR and Western blot to assess gene expression. Results: Bulk RNA sequencing identified 789 differentially expressed transcripts between HNC and control samples. Subsequent RT-qPCR analysis focused on IL32 and BIRC3 mRNAs (up-regulated in HNC samples) and ACE1 mRNA (down-regulated). Among male HNC patients, the IL32/ACE1 mRNA ratio was significantly elevated in CRS cases (p = 0.0001, effect size r = 0.57) and correlated with the severity of muscle atrophy (negative correlation with the Skeletal Muscle Index at a threshold of 10%: p = 0.093, r = -0.41). In contrast, no such trend was observed for the BIRC3/ACE1 ratio. Exposure of human myoblasts to HNC cells induced inhibition of myogenesis and strong up-regulation of IL32 mRNA and protein. In contrast, these effects were absent or much smaller under exposure to NHEK controls. Conclusions: IL32 is a potential biomarker for CRS in HNC patients. In addition, the HNC-myoblast co-cultivation model provides a promising in vitro system to study CRS mechanisms, potentially reducing the reliance on animal models.

Purpose/Objective Sarcomatoid carcinoma of head and neck is a rare type of squamous cell carcinoma (SCC) with sarcomatoid features associated with a poor prognosis.This study aimed to summarize the clinical characteristics, prognosis and treatment options for these patients. Material/Methods Patients diagnosed with a sarcomatoid carcinoma of the head and neck between December 2012 to November 2022 were selected from the pathology database of Gustave Roussy. The data (patients, tumor characteristics, treatment, outcomes, etc.) were collected retrospectively. Multivariable prognosis analysis of progression free survival (PFS), and overall survival (OS) were performed. Results A total of 57 cases of sarcomatoid carcinoma were included. There were 41 (72%) males and 16 (28%) females. The median age at diagnosis was 64 years. The oral cavity (32%) and the larynx (26%) were the most common tumor sites, followed by the oropharynx (23%), posterior wall of the pharynx and maxillary sinus (5%). 17 patients (30%) were no smokers.More than half of the patients (65%) presented with advanced-stage disease (T3, T4), 95% of them had no metastatic disease at diagnosis. A total of 37 patients (65%) had no previous history of head and neck cancers (de novo tumors), and 25 (44%) arised in irradiated areas.Among all, 28 (49%) underwent a surgical treatment, for 9 (32%) the resection was R1, 9 patients had perineural invasion, and 6 of them presented vascular embols. 32% had nodal metastasis.The other 51 % of patients received radiochemotherapy (n=15), palliative chemotherapy (n=10), or best supportive care (n=4).The follow-up was 51.8 months. The median [95%CI] OS and PFS were 29.7 months [13.8;49.2] and 9.5 months [7.5;22.8].The multivariable prognostic analysis showed that age, T stage, site (larynx and oral cavity), no history of head and neck cancer, and the presence of a tumor in irradiated site were significantly correlated with a poor OS. With the exception ofno history of head and neck cancer, all of the above negatively influenced the PFS Conclusion Sarcomatoid carcinoma is a different entity from the conventional SCC of head and neck. This type of tumor is very aggressive and has a poor prognosis. Tumors in irradiated site have a significant poor prognostic.Further clinical trials are needed to establish the best treatment strategy for these tumors.

PURPOSE:Patients with human papillomavirus (HPV)-positive oropharyngeal cancer (OPC) and advanced stage and/or significant smoking history are at higher risk of relapse. Induction immunotherapy before chemoradiation (CRT) may improve outcomes. This randomized phase II trial assessed the feasibility and safety of induction nivolumab before CRT in this high-risk population. METHODS:Eligible patients had HPV-positive OPC with either T4 and/or N2/N3 disease or a smoking history >10 pack-years. Patients were randomly assigned 1:2 to receive either standard CRT (70 Gy with cisplatin, control arm [CA], n = 20) or two infusions of nivolumab followed by CRT (experimental arm [EA], n = 41). The primary end point was the rate of patients who received full treatment in due time (FTDT), defined as (1) two nivolumab infusions on days 1 and 13-17, (2) CRT started between days 27-37 after the first nivolumab infusion, (3) no radiotherapy break ≥7 days, (4) >95% of theoretical/prescribed RT dose, and (5) cisplatin dose received ≥200 mg/m2. If two patients or less in the EA failed FTDT, the strategy would be considered feasible. Secondary end points included oncologic outcomes and toxicity. RESULTS:Between July 2019 and September 2021, 62 patients were randomly assigned. Median follow-up was 37.5 months. The primary end point was not met: four of 41 patients in EA received <200 mg/m2 cisplatin. Grade 4 to 5 acute adverse events occurred only in EA, in seven patients. The 2-year cumulative incidence (95% CI) of relapse was 7.3% (1.9 to 18.0) in EA versus 15.0% (3.6 to 34.0) in CA. CONCLUSION:Induction nivolumab before CRT did not meet the predefined feasibility threshold because of reduced cisplatin dosing after toxicity in 10% of patients. The relapse incidence was numerically lower in the EA but this finding is exploratory and requires confirmation.

Pheochromocytomas and paragangliomas (PPGLs) are rare neuroendocrine tumors originating from neural crest-derived chromaffin tissue, marked by clinical heterogeneity and substantial genetic underpinnings. With up to 70

BACKGROUND:Effective radiotherapy requires balancing target coverage with sparing of organs at risk. We developed a fully automated, Python-based pipeline to integrate white matter tracts (WMTs) into stereotactic brain radiotherapy (SBRT) planning. METHODS:The pipeline uses a probabilistic atlas of 52 WMTs and includes image pre-processing, affine and non-linear registration to a standardized space, and dose mapping into the atlas framework. Various interpolation techniques were assessed to optimize dose-mapping accuracy. Robustness was assessed through sensitivity analyses of four interpolation methods for affine and non-linear transformations, resulting in 16 combinations, and local perturbations of registration fields. Prospective validation was performed in nine patients with pre-treatment diffusion MRI (GE 3T, 64 directions, b = 1000, 2 mm isotropic, 4'30 acquisition), where atlas-based WMTs were compared with individual tractography using TractSeg and DeepWMA. RESULTS:Applied to 108 patients, the pipeline achieved precise spatial registration, with B-spline interpolation providing the most consistent dose mapping. Sensitivity analyses showed negligible effects of interpolation or perturbations on the mean and minimum doses to WMTs, while the maximum dose in tracts located adjacent to the planning target volume was more sensitive, with rare deviations up to ∼1 Gy. In prospective validation, median Dice coefficients across all evaluated WMTs were 0.49 (range 0.00-0.66) for TractSeg versus DeepWMA, 0.45 (0.02-0.68) for TractSeg versus the atlas, and 0.31 (0.00-0.55) for DeepWMA versus the atlas. Despite modest Dice values, as consistently reported in tractography benchmarks, the atlas-based pipeline performed within the same range of variability as that observed between two widely used tractography methods, supporting its validity for clinical application. CONCLUSIONS:This pipeline enables robust and reproducible integration of WMTs into SBRT planning, validated against tractography and applicable when diffusion MRI is unavailable. It provides a foundation for establishing tract-specific dose constraints, advancing neurocognitive-sparing SBRT.

INTRODUCTION:The past two decades have seen a paradigm shift in the surgical treatment of sinonasal malignancies, moving from traditional open approaches to minimally invasive endoscopic techniques. Technological advancements have enabled the use of the endoscopic endonasal transcribriform approach (EETA) for tumors involving the anterior skull base (ASB). This study evaluates the oncologic outcomes of EETA for locally advanced sinonasal malignancies with ASB involvement. METHODS:We conducted a retrospective cohort study at a tertiary cancer center, including patients treated with EETA for sinonasal malignancies from January 2012 to January 2024. Data on demographic characteristics, clinical presentations, treatment protocols, and survival outcomes were analyzed. Survival probabilities were assessed using Kaplan-Meier curves, and subgroups comparisons were made using the log rank test. RESULTS:Forty-nine patients were included, predominantly with primary tumors (91.8%) and advanced stages (cT3-T4: 89.8%). EETA achieved R0 resection in 82.1% of cases. All eligible patients (88.9%) underwent postoperative radiotherapy, with 97.5% beginning treatment 6-8 weeks after surgery. The mean follow-up duration was 43.2 months. The 5-year overall survival (OS) was 66.5%, disease-free survival (DFS) was 58.7%, and local relapse-free survival (LRFS) was 75%. Patients with olfactory neuroblastoma had significantly higher OS and DFS compared to intestinal-type adenocarcinoma. Margin status significantly impacted survival, with better outcomes observed in patients with clear margins. CONCLUSION:EETA offers a viable and effective alternative for treating locally advanced sinonasal malignancies with ASB involvement, achieving acceptable oncologic outcomes with reduced morbidity.

Background/Objectives: To evaluate and optimize the reconstruction parameters of images acquired with a photon-counting CT scanner to achieve a stable radiomics signal. Methods: Radiomics is a quantitative imaging biomarker correlated to survival in oncology patients. Implementing radiomics in clinical routine remains challenging due to the feature’s instability. Photon-counting CT scans use innovative technology directly converting photons into electrical signals resulting in higher-resolution images with reduced artifacts. This study used two organic phantoms: a natural wet sponge and a dry sausage. UHR images were acquired using a NAEOTOM Alpha photon-counting CT scan (Siemens) with a 0.4 mm slice thickness and 0.3 × 0.3 mm pixel size. Tube current and voltage were fixed at 112 mA and 120 KvP. A total of 24 reconstruction parameter sets were obtained by combining different values of kernel (Br), quantitative iterative reconstruction (QIR), spectral reconstruction (keV), and matrix size. Ten successive acquisitions were obtained on both phantoms. In total, 93 radiomic features were extracted on an ROI using the default parameters of Pyradiomic 3.0.1. Each feature’s stability was evaluated using the coefficient of variation (CV) within each parameter set. Results: Of the 24 reconstruction parameter sets, 5 were selected based on best image quality by seven radiologists and three radiology technologists. Radiomics features were considered stable on a set when CV was less than 15%. Feature stability was impacted by reconstruction parameters and the phantom used. The most stable combination included 90 and 65 stable features of the 93 tested on the sausage and sponge respectively. It was configured with Br36, QIR 4, 60 keV, and a 1024 × 1024 matrix size. Conclusions: Images obtained on photon-counting CT scans offer promising radiomic feature stability with optimal parameter configurations that could be applied in a clinical setting.

PURPOSE:The purpose of this study was to evaluate baseline tumor burden from liquid biopsy (LB) and computed tomography (CT) as prognostic biomarkers and whether their combination refines stratification in metastatic solid cancers. MATERIALS AND METHODS:This retrospective cohort study included 1065 patients. Eligible patients underwent LB and chest-abdomen-pelvis CT examination at baseline. Radiologists outlined lesions on the largest axial slice, and total tumor volume (TTV) was approximated in three dimensions. LB Tumor fraction (TF) ≥ 10 % was considered high. To assess combined prognostic power of LB and TTV, patients were divided into three groups according to LB (circulating tumor deoxyribonucleic acid [ctDNA] detectability and TF) and each group was further divided into two subgroups using TTV thresholds determined by Youden's index. Overall survival (OS) analyses were performed using Cox proportional hazard models and Kaplan-Meier curves. RESULTS:A total of 560 patients (290 women and 270 men; median age, 61 years) with 31,314 annotated lesions were selected. The median OS was 11.28 months, and the median TTV was 96.68 cm3. The LB groups included patients with undetectable ctDNA (n = 102), patients with detectable ctDNA and low TF (n = 251), and patients with high TF (n = 207). Integrating TTV thresholds (18.7 cm3, 44.9 cm3, and 159.94 cm3) to LB groups significantly stratified the population on OS. Patients with undetectable ctDNA and TTV ≥ 18.7 cm3 had significantly shorter OS (median OS, 24.8 months) than those with TTV < 18.7 cm3 (median OS, > 35 months). CONCLUSION:Combining baseline CT tumor burden with LB could be a valuable prognostic tool for stratifying patients with metastatic solid cancer.

Background Tumor fraction (TF) at liquid biopsy is a potential noninvasive marker for tumor burden, but validation is needed. Purpose To evaluate TF as a potential surrogate for tumor burden, assessed at contrast-enhanced CT across diverse metastatic cancers. Methods This retrospective monocentric study included patients with cancer and metastatic disease, with TF results and contemporaneous contrast-enhanced CT performed between January 2021 and January 2023. The total tumor volume (TTV), representing CT tumor burden, was calculated by adding all lesion volumes and was computed by using manually outlined annotations of each lesion on the largest surface of the axial slice. TF greater than 10% was considered high. A training-validation split was applied. Correlations between TF and TTV were assessed using regression models and Spearman correlation coefficients. Receiver operating characteristic curve analysis established the TTV cutoff. The metastatic site, histology type, and TTV were used to predict liquid biopsy contributory status. Results Among 1065 patients (median age, 62 years [IQR: 53, 70]; 537 female), 56 288 lesions were annotated, mostly in the lung (n = 20 334), lymph nodes (n = 11 651), and liver (n = 10 277). A total of 763 liquid biopsies were contributive, 254 were noncontributive, and 48 failed. The training and validation sets included 745 and 320 patients, respectively. TF helped predict TTV with the linear model (R2 = 0.17; ρ = 0.41; P < .001). The TTV and TF categories achieved an area under the receiver operating characteristic curve (AUC) of 0.74 (95% CI: 0.71, 0.78), with an optimal cutoff of 151 cm3 for TTV and a TF cutoff of 10%. The sensitivity was 57% (204 of 359) and the specificity was 80% (525 of 658). TTV helped predict contributory status, with an AUC of 0.71 (95% CI: 0.67, 0.76) and an optimal cutoff greater than 37 cm3. Liver lesion volumes were significantly associated with a contributory liquid biopsy in the validation cohort. Conclusion While correlated, TF at liquid biopsy did not accurately represent the TTV at CT. © RSNA, 2024 Supplemental material is available for this article. See also the editorial by Koh in this issue.

Introduction: The incidence of venous thromboembolism is estimated to be around 3% of cancer patients. However, a majority of incidental pulmonary embolism (iPE) can be overlooked by radiologists in asymptomatic patients, performing CT scans for disease surveillance, which may significantly impact the patient’s health and management. Routine imaging in oncology is usually reviewed with delayed hours after the acquisition of images. Nevertheless, the advent of AI in radiology could reduce the risk of the diagnostic delay of iPE by an optimal triage immediately at the acquisition console. This study aimed to determine the accuracy rate of an AI algorithm (CINA-iPE) in detecting iPE and the duration until the management of cancer patients in our center, in addition to describing the characteristics of patients with a confirmed pulmonary embolism (PE). Materials and Methods: This is a retrospective analysis of the role of Avicenna’s CE-certified and FDA-cleared CINA-iPE algorithm in oncology patients treated at Gustave Roussy Cancer Campus. The results obtained from the AI algorithm were compared with the attending radiologist’s report and were analyzed by both a radiology resident and a senior radiologist. In case of any discordant results, the reason for this discrepancy was further investigated. The duration between the exact time of the CT scan and analysis was assessed, as well as the duration from the result’s report and the start of active management. Results: Out of 3047 patients, 104 alerts were detected for iPE (prevalence of 1.3%), while 2942 had negative findings. In total, 36 of the 104 patients had confirmed PE, while 68 alerts were false positives. Only one patient reported as negative by the AI tool was deemed to have a PE by the radiologist. The sensitivity and specificity of the AI model were 97.3% and 97.74%, while the PPV and NPV were 34.62% and 99.97%, respectively. Most causes of FP were artifacts (22 cases, 32.3%) and lymph nodes (11 cases, 16.2%). Seven patients experienced delayed diagnosis, requiring them to return to the ER for treatment after being sent home following their scan. The remaining patients received prompt care immediately after their testing, with a mean delay time of 8.13 h. Conclusions: The addition of an AI system for the detection of unsuspected PEs on chest CT scans in routine oncology care demonstrated a promising efficacy in comparison to human performance. Despite a low prevalence, the sensitivity and specificity of the AI tool reached 97.3% and 97.7%, respectively, with detection of all the reported clinical PEs, except one single case. This study describes the potential synergy between AI and radiologists for an optimal diagnosis of iPE in routine clinical cancer care. Clinical relevance statement: In the oncology field, iPEs are common, with an increased risk of morbidity when missed with a delayed diagnosis. With the assistance of a reliable AI tool, the radiologist can focus on the challenging analysis of oncology results while dealing with urgent diagnosis such as PE by sending the patient straight to the ER (Emergency Room) for prompt treatment.

The parotid glands are the largest of the major salivary glands. They can harbour both benign and malignant tumours. Preoperative work-up relies on MR images and fine needle aspiration biopsy, but these diagnostic tools have low sensitivity and specificity, often leading to surgery for diagnostic purposes. The aim of this paper is (1) to develop a machine learning algorithm based on MR images characteristics to automatically classify parotid gland tumours and (2) compare its results with the diagnoses of junior and senior radiologists in order to evaluate its utility in routine practice. While automatic algorithms applied to parotid tumours classification have been developed in the past, we believe that our study is one of the first to leverage four different MRI sequences and propose a comparison with clinicians. In this study, we leverage data coming from a cohort of 134 patients treated for benign or malignant parotid tumours. Using radiomics extracted from the MR images of the gland, we train a random forest and a logistic regression to predict the corresponding histopathological subtypes. On the test set, the best results are given by the random forest: we obtain a 0.720 accuracy, a 0.860 specificity, and a 0.720 sensitivity over all histopathological subtypes, with an average AUC of 0.838. When considering the discrimination between benign and malignant tumours, the algorithm results in a 0.760 accuracy and a 0.769 AUC, both on test set. Moreover, the clinical experiment shows that our model helps to improve diagnostic abilities of junior radiologists as their sensitivity and accuracy raised by 6 % when using our proposed method. This algorithm may be useful for training of physicians. Radiomics with a machine learning algorithm may help improve discrimination between benign and malignant parotid tumours, decreasing the need for diagnostic surgery. Further studies are warranted to validate our algorithm for routine use.

Liquid biopsy is an emerging technology capable of detecting cancer, while CT-imaging is the gold standard. We aim to correlate imaging features from CT-scans to liquid biopsy ctDNA tumor fraction (TF) and blood Tumor Mutational Burden (bTMB). This retrospective multicentric study included 1017 patients with metastatic cancer who underwent a variety of treatments. Liquid biopsy and contemporaneous CT-scans were collected. TF and bTMB values were computed using FoundationOne Liquid CDX solution by Roche, with high tumor fraction being over 10%. Two expert radiologists outlined all cancerous lesions in the largest axial diameter. Total tumor volume (TTV) was defined as the sum of estimated lesion volumes. Training set consisted of data from several imaging centers, while the validation set was single-center. Continuous value distributions by category were tested using the Mann-Whitney U-test, and a threshold was set with ROC curve analysis using Youden's Index. This cutoff defined high and low groups, and a Χ2 test of independence was used to examine the correlation. Sensitivity and specificity were reported with imaging features taken as the gold standard. Subanalyses were run on patients with liver lesions. Overall, 55294 lesions were annotated, most commonly located in the lung (n=20074), liver (n=11297) and lymph nodes (n=10222). The train and validation sets included 599 and 418 patients. Preliminary analyses were performed on the train set. Patients with low TF had significantly less TTV than those with high TF (p<0.001). ROC analysis yielded an AUC of 0.63 with a threshold of 106 cm3. Correlation was significant between TTV and TF categories (Χ2 test: p<0.001). The sensitivity and specificity were 76% and 42%. For patients with liver lesions (n=253) , TTV threshold was 108 cm3 with an AUC of 0.66. Correlation was significant between TTV and TF categories (Χ2 test: p<0.001). The sensitivity and specificity were 82% and 42%. Analysis showed significant correlation between TTV and ctDNA TF. Sensitivity was higher for patients with liver lesions, suggesting that some organs shed tumoral DNA more than others. The study of tumor heterogeneity on imaging and its correlation to bTMB is in progress.

Background: Our aim was to report the long-term outcomes of mandibular reconstruction using CAD-CAMdesigned 3D-printed porous titanium implants in patients not amenable to a free vascularized fibula flap reconstruction. Methods: The implants were designed with ProPlan CMF (R) 2.2 software and manufactured with a Selective Laser Melting (SLM) "layer-by-layer" 3D-printing of pure porous titanium powder beds. Primary endpoints were implant exposure and implant removal calculated using Gray's tests. Secondary endpoints were predictive factors of implant exposure and implant removal, and rates of dental rehabilitation. Results: Thirty-six patients were operated between 2015 and 2017 and were included in this study. Reconstruction using a porous titanium 3D-printed implant was proposed due to medical contraindication for a fibula free flap (n = 13), due to the failure of a previous fibula free flap reconstruction (n = 7), or due to refusal of a fibula free flap reconstruction by the patient (n = 16). The medical indications for mandibular reconstruction were a primary tumor requiring mandibulectomy in nine patients, mandibular osteoradionecrosis requiring mandibulectomy in nineteen patients, and secondary reconstruction in eight patients. The 2-year rates of implant exposure and implant removal were 69.4% and 52.8%. Reconstruction of the symphysis was a high-risk exposure variable (OR 30; p = 0.0003). Only one patient underwent a successful dental rehabilitation. Conclusion: The use of a porous titanium 3D- implant for mandibular reconstruction in head and neck cancer patients resulted in high rates of implant exposure and of implant removal, notably when symphysis involvement. (c) 2022 Elsevier Masson SAS. All rights reserved.

The term lymphoma includes a wide variety of different clinical entities including diffuse large B-cell lymphomas (DLBCL). Skeletal muscle or intramuscular lymphomas represent less than 2% of B-cell Lymphoma, they are quite rare, even more in the orofacial area. We present the case of a painless growing mass of the right cheek mimicking a chronic oral cellulitis in a 34-year-old man. Magnetic resonance imaging (MRI) of the mandible revealed a well-defined 7x3cm mass around the core of the mandible that invades the buccal floor and the subcutaneous planes. A whole-body 18F-FDG PET/CT for the initial diagnosis revealed an intensely isolated hypermetabolic band corresponding to a voluminous tumoral permeation. The diagnosis of a skeletal muscle diffuse large B-cell lymphoma was established after an intraoral biopsy. It was treated with 4 chemotherapy cures and showed complete remission at one year of follow-up. This atypical form of lymphoma should be integrated into the differential diagnosis of soft tissue tumors in the oral cavity.

Objectives This study proposes and evaluates a deep learning method that predicts surrogate images for contrast-enhanced T1 from multiparametric magnetic resonance imaging (MRI) acquired using only a quarter of the standard 0.1 mmol/kg dose of gadolinium-based contrast agent. In particular, the predicted images are quantitatively evaluated in terms of lesion detection performance. Materials and Methods This monocentric retrospective study leveraged 200 multiparametric brain MRIs acquired between November 2019 and February 2020 at Gustave Roussy Cancer Campus (Villejuif, France). A total of 145 patients were included: 107 formed the training sample (55 ± 14 years, 58 women) and 38 the separate test sample (62 ± 12 years, 22 women). Patients had glioma, brain metastases, meningioma, or no enhancing lesion. T1, T2-FLAIR, diffusion-weighted imaging, low-dose, and standard-dose postcontrast T1 sequences were acquired. A deep network was trained to process the precontrast and low-dose sequences to predict “virtual” surrogate images for contrast-enhanced T1. Once trained, the deep learning method was evaluated on the test sample. The discrepancies between the predicted virtual images and the standard-dose MRIs were qualitatively and quantitatively evaluated using both automated voxel-wise metrics and a reader study, where 2 radiologists graded image qualities and marked all visible enhancing lesions. Results The automated analysis of the test brain MRIs computed a structural similarity index of 87.1% ± 4.8% between the predicted virtual sequences and the reference contrast-enhanced T1 MRIs, a peak signal-to-noise ratio of 31.6 ± 2.0 dB, and an area under the curve of 96.4% ± 3.1%. At Youden's operating point, the voxel-wise sensitivity (SE) and specificity were 96.4% and 94.8%, respectively. The reader study found that virtual images were preferred to standard-dose MRI in terms of image quality (P = 0.008). A total of 91 reference lesions were identified in the 38 test T1 sequences enhanced with full dose of contrast agent. On average across readers, the brain lesion SE of the virtual images was 83% for lesions larger than 10 mm (n = 42), and the associated false detection rate was 0.08 lesion/patient. The corresponding positive predictive value of detected lesions was 92%, and the F1 score was 88%. Lesion detection performance, however, dropped when smaller lesions were included: average SE was 67% for lesions larger than 5 mm (n = 74), and 56% with all lesions included regardless of their size. The false detection rate remained below 0.50 lesion/patient in all cases, and the positive predictive value remained above 73%. The composite F1 score was 63% at worst. Conclusions The proposed deep learning method for virtual contrast-enhanced T1 brain MRI prediction showed very high quantitative performance when evaluated with standard voxel-wise metrics. The reader study demonstrated that, for lesions larger than 10 mm, good detection performance could be maintained despite a 4-fold division in contrast agent usage, unveiling a promising avenue for reducing the gadolinium exposure of returning patients. Small lesions proved, however, difficult to handle for the deep network, showing that full-dose injections remain essential for accurate first-line diagnosis in neuro-oncology.

This article presents an overview of the graphical user interfaces (GUIs) developed at CEA/SERMA (Service d’Études des Réacteurs et de Mathématiques Appliquées) in Saclay, France, which have been used for over forty years by engineers and scientists to build geometries and meshes for general-purpose lattice transport calculations (neutrons and photons). Several applications make use of these calculations, from fuel assembly to full core design, criticality and safety, needing consistency check of the geometry and input properties before starting any lattice calculation. The software pattern design of the GUIs is briefly discussed, showing also the rationale behind the two interfaces for the construction of the geometries for simple fuel assemblies and complex motifs including the reflector (colorsets). The new GUI, ALAMOS, specifically developed for APOLLO3® with a Python Application Programming Interface (API), is here presented as the successor of Silène, which was the first GUI released in the 1990s to serve APOLLO2 calculations. The considerable experience gained by Silène over the years with plenty of various applications has provided a crucial support for the development of ALAMOS.

Objectives: Our objective was to develop a predictive model using a machine learning signature to identify patients at high risk of relapse or death after treatment for HPV-positive oropharyngeal carcinoma. Materials and methods: Pre-treatment variables of 450 patients with HPV-positive oropharyngeal carcinoma treated with a curative intent comprised clinical items, imaging parameters and histological findings. The events considered were progression or residual disease after treatment, the recurrent disease after a disease-free interval and death. The endpoints were the prediction of events and progression-free survival. After feature Z-score normalisation and selection, random forest classifier models were trained. The best models were evaluated on recall, the F-score, and the ROC AUC metric. The clinical relevance of the best prediction model was evaluated using Kaplan-Meier analysis with a log-rank test. Results: The best random forest model predicted the 5-year risk of relapse-free survival with a recall of 79.1%, an Fl -score of 81.08%, and an AUC of the ROC curve of 0.89. The models performed poorly for the prediction of specific events of progression only, recurrence only or death only. The clinical relevance of the model was validated with a 5-year relapse-free survival of high-risk patients versus low-risk patients of 23.5% and 80%, respectively (p < 0.0001). Conclusion: Patients with HPV-driven oropharyngeal carcinoma at high risk of relapse-free survival could be identified with a predictive machine learning model using patient data before treatment.

The need for developing new biomarkers is increasing with the emergence of many targeted therapies. In this study, we used artificial intelligence (AI) to develop a multimodal model (PULS-AI) predicting the survival of solid tumor patients treated with antiangiogenic treatments. Our retrospective, multicentric study included 616 patients with 7 different cancer types: renal cell carcinoma, colorectal carcinoma, hepatocellular carcinoma, gastrointestinal carcinoma, melanoma, breast cancer, and sarcoma. A set of 196 patients was left out of the study for validation. Clinical data including patient, treatment, and cancer metadata were collected at baseline for all patients, as well as computed tomography (CT) and ultrasound (US) images. Radiologists annotated all metastases on the CT images and the visible tumor lesion on the US images. AI models were used to extract relevant features from the regions of interest on CT and US images. In addition, handcrafted features related to the tumor burden were extracted from the annotations of all lesions on CT such as the number of lesions and the tumor burden volume per organ (lungs, liver, skull, bone, other). Finally, a Cox regression model was fitted to the set of imaging features and clinical features. The annotation process led to 1147 annotated US images with lesions delineation and 4564 reviewed CTs, of which 989 were selected and fully annotated with a total of 9516 annotated lesions.The developed model reaches an average concordance index of 0.71 (0.67-0.75, 95% CI). Using a risk threshold of 50%, PULS-AI model is able to significantly isolate (log-rank test P-value < 0.001) high-risk patients from low-risk patients (respective median OS of 12 and 32 months) with a hazard ratio of 3.52 (2.35-5.28, 95% CI). The results of this study show that AI algorithms are able to extract relevant information from radiology images and to aggregate data from multiple modalities to build powerful prognostic tools. Such tools may provide assistance to oncology clinicians in therapeutic decision-making. Citation Format: Kathryn Schutte, Fabien Brulport, Sana Harguem-Zayani, Jean-Baptiste Schiratti, Ridouane Ghermi, Paul Jehanno, Alexandre Jaeger, Talal Alamri, Raphael Naccache, Leila Haddag-Miliani, Teresa Orsi, Jean-Philippe Lamarque, Isaline Hoferer, Littisha Lawrance, Baya Benatsou, Imad Bousaid, Mickael Azoulay, Antoine Verdon, François Bidault, Corinne Balleyguier, Victor Aubert, Etienne Bendjebbar, Charles Maussion, Nicolas Loiseau, Benoit Schmauch, Meriem Sefta, Gilles Wainrib, Thomas Clozel, Samy Ammari, Nathalie Lassau. PULS-AI: A multimodal artificial intelligence model to predict survival of solid tumor patients treated with antiangiogenics [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 1924.