Physics and Imaging in Radiation Oncology最新文献

{"title":"Multicentre prospective risk analysis of a fully automated radiotherapy workflow","authors":"Geert De Kerf ,&nbsp;Ana Barragán-Montero ,&nbsp;Charlotte L. Brouwer ,&nbsp;Pietro Pisciotta ,&nbsp;Marie-Claude Biston ,&nbsp;Marco Fusella ,&nbsp;Geoffroy Herbin ,&nbsp;Esther Kneepkens ,&nbsp;Livia Marrazzo ,&nbsp;Joshua Mason ,&nbsp;Camila Panduro Nielsen ,&nbsp;Koen Snijders ,&nbsp;Stephanie Tanadini-Lang ,&nbsp;Aude Vaandering ,&nbsp;Tomas M. Janssen","doi":"10.1016/j.phro.2025.100765","DOIUrl":"10.1016/j.phro.2025.100765","url":null,"abstract":"<div><h3>Background and Purpose</h3><div>Fully automated workflows (FAWs) for radiotherapy treatment preparation are feasible, but remain underutilized in clinical settings. A multicentre prospective risk analysis was conducted to support centres in managing FAW-related risks and to identify workflow steps needing improvement.</div></div><div><h3>Material and Methods</h3><div>Eight European radiotherapy centres performed a failure mode and effect analysis (FMEA) on a hypothetical FAW, with a manual review step at the end. Centres assessed occurrence, severity and detectability of provided, or newly added, failure modes to obtain a risk score. Quantitative analysis was performed on curated data, while qualitative analysis summarized free text comments.</div></div><div><h3>Results</h3><div>Manual review and auto-segmentation were identified as the highest-risk steps and the highest scoring failure modes were associated with inadequate manual review (high detectability and severity score), incorrect (i.e. outside of intended use) application of the FAW (high severity score) and protocol violations during patient preparation (high occurrence score). The qualitative analysis highlighted amongst others the risk of deviation from protocol and the difficulty for manual review to recognize automation errors. The risk associated with the technical parts of the workflow was considered low.</div></div><div><h3>Conclusions</h3><div>The FMEA analysis highlighted that points where people interact with the FAW were considered higher risk than lack of trust in the FAW itself. Major concerns were the ability of people to correctly judge output in case of low generalizability and increasing skill degradation. Consequently, educational programs and interpretative tools are essential prerequisites for widespread clinical application of FAWs.</div></div>","PeriodicalId":36850,"journal":{"name":"Physics and Imaging in Radiation Oncology","volume":"34 ","pages":"Article 100765"},"PeriodicalIF":3.4,"publicationDate":"2025-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143791665","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Detection of the failed-tolerance causes of electronic-portal-imaging-device-based in vivo dosimetry using machine learning for volumetric-modulated arc therapy: A feasibility study","authors":"Nipon Saiyo ,&nbsp;Hironori Kojima ,&nbsp;Kimiya Noto ,&nbsp;Naoki Isomura ,&nbsp;Kosuke Tsukamoto ,&nbsp;Shotaro Yamaguchi ,&nbsp;Yuto Segawa ,&nbsp;Junya Kohigashi ,&nbsp;Akihiro Takemura","doi":"10.1016/j.phro.2025.100785","DOIUrl":"10.1016/j.phro.2025.100785","url":null,"abstract":"<div><h3>Background and Purpose</h3><div>When electronic-portal-imaging-device (EPID)-based <em>in vivo</em> dosimetry (IVD) identifies dose tolerance failures, the cause of the failures should be evaluated. This study aimed to develop a machine-learning (ML) model to classify the cause of EPID-based IVD failures in volumetric-modulated arc therapy (VMAT) treatment.</div></div><div><h3>Materials and Methods</h3><div>Twenty-three prostate VMAT plans were used to recalculate the dose distribution in homogeneous phantom images as no-error (NE) plans. Errors in the randomized multileaf collimator (RMLC) position, monitor unit (MU) variation, lateral position, pitch rotation, and roll rotation were simulated. The IVD results of the NE plans and introduced errors were obtained using EPIgray software. Support vector machines (SVMs) were used to develop ML models for each error. The accuracy percentage, F1-score, and area under the receiver operating characteristic (ROC) curve (AUC) were used to evaluate models’ performances. The models were verified using five additional plans with an Alderson Rando phantom.</div></div><div><h3>Results</h3><div>The models obtained accuracies of over 90% and F1-scores of 0.9 for the RMLC position and MU variation. For lateral position, pitch rotation, and roll rotation errors, the accuracies were 66.1%, 65.2%, and 66.8%, and the F1-scores were 0.66, 0.65, and 0.67, respectively. The AUCs for all the errors were over 0.7. Additionally, the model verification results consistently classified EPIgray data for all the error types.</div></div><div><h3>Conclusion</h3><div>The developed ML models classified the causes of the failed tolerance of the EPID-based IVD.</div></div>","PeriodicalId":36850,"journal":{"name":"Physics and Imaging in Radiation Oncology","volume":"34 ","pages":"Article 100785"},"PeriodicalIF":3.4,"publicationDate":"2025-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144089485","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Breakthroughs in modern brachytherapy are transforming cancer care","authors":"Georgina Fröhlich,&nbsp;Alfonso Gomez-Iturriaga,&nbsp;Gemma Eminowicz,&nbsp;Christoph Bert,&nbsp;Åsa Carlsson Tedgren,&nbsp;Alexandra J. Stewart,&nbsp;Bernd Wisgrill,&nbsp;Ludvig P. Muren","doi":"10.1016/j.phro.2025.100783","DOIUrl":"10.1016/j.phro.2025.100783","url":null,"abstract":"","PeriodicalId":36850,"journal":{"name":"Physics and Imaging in Radiation Oncology","volume":"34 ","pages":"Article 100783"},"PeriodicalIF":3.4,"publicationDate":"2025-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144134588","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Enhancing patient-specific deep learning based segmentation for abdominal magnetic resonance imaging-guided radiation therapy: A framework conditioned on prior segmentation","authors":"Francesca De Benetti ,&nbsp;Nikolaos Delopoulos ,&nbsp;Claus Belka ,&nbsp;Stefanie Corradini ,&nbsp;Nassir Navab ,&nbsp;Thomas Wendler ,&nbsp;Shadi Albarqouni ,&nbsp;Guillaume Landry ,&nbsp;Christopher Kurz","doi":"10.1016/j.phro.2025.100766","DOIUrl":"10.1016/j.phro.2025.100766","url":null,"abstract":"<div><h3>Background and purpose:</h3><div>Conventionally, the contours annotated during magnetic resonance-guided radiation therapy (MRgRT) planning are manually corrected during the RT fractions, which is a time-consuming task. Deep learning-based segmentation can be helpful, but the available patient-specific approaches require training at least one model per patient, which is computationally expensive. In this work, we introduced a novel framework that integrates fraction MR volumes and planning segmentation maps to generate robust fraction MR segmentations without the need for patient-specific retraining.</div></div><div><h3>Materials and methods:</h3><div>The dataset included 69 patients (222 fraction MRs in total) treated with MRgRT for abdominal cancers with a 0.35 T MR-Linac, and annotations for eight clinically relevant abdominal structures (aorta, bowel, duodenum, left kidney, right kidney, liver, spinal canal and stomach). In the framework, we implemented two alternative models capable of generating patient-specific segmentations using the planning segmentation as prior information. The first one is a 3D UNet with dual-channel input (i.e. fraction MR and planning segmentation map) and the second one is a modified 3D UNet with double encoder for the same two inputs.</div></div><div><h3>Results:</h3><div>On average, the two models with prior anatomical information outperformed the conventional population-based 3D UNet with an increase in Dice similarity coefficient <span><math><mrow><mo>&gt;</mo><mn>4</mn><mspace></mspace><mtext>%</mtext></mrow></math></span>. In particular, the dual-channel input 3D UNet outperformed the one with double encoder, especially when the alignment between the two input channels is satisfactory.</div></div><div><h3>Conclusion:</h3><div>The proposed workflow was able to generate accurate patient-specific segmentations while avoiding training one model per patient and allowing for a seamless integration into clinical practice.</div></div>","PeriodicalId":36850,"journal":{"name":"Physics and Imaging in Radiation Oncology","volume":"34 ","pages":"Article 100766"},"PeriodicalIF":3.4,"publicationDate":"2025-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143918321","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Reconstructed dose and geometric coverage for tight margins using intrafraction re-planning on an integrated magnetic resonance imaging and linear accelerator system for prostate cancer patients","authors":"Ingeborg van den Berg ,&nbsp;Cornel Zachiu ,&nbsp;Eline N. de Groot-van Breugel ,&nbsp;Thomas Willigenburg ,&nbsp;Gijsbert H. Bol ,&nbsp;Jan J.W. Lagendijk ,&nbsp;Bas W. Raaymakers ,&nbsp;Harm H.E. van Melick ,&nbsp;Cornelis A.T. van den Berg ,&nbsp;Jochem R.N. van der Voort van Zyp ,&nbsp;Johannes C.J. de Boer","doi":"10.1016/j.phro.2025.100776","DOIUrl":"10.1016/j.phro.2025.100776","url":null,"abstract":"<div><h3>Background and purpose</h3><div>A sub-fractionation workflow enables a substantial reduction in planning target volume (PTV) margin in prostate cancer (PCa) patients by reducing systematic motion during magnetic resonance (MR)-guided radiotherapy. This study assessed geometric and reconstructed dose outcomes in patients treated with a tight-margin sub-fractionation workflow on a combined linear accelerator with a 1.5 T MRI scanner (MR-Linac).</div></div><div><h3>Materials and methods</h3><div>We evaluated the sub-fractionation workflow with tight margins (2–3 mm) on 128 PCa patients who completed treatment with 5 × 7.25 Gy (36.25 Gy total dose). A traffic light protocol was applied based on residual motions to detect patients with unexpectedly large motions. When ’red’ traffic light criteria were met, plans with larger margins (5 mm isotropic) were adopted for subsequent fractions. Intra- and inter-fraction dose accumulation was performed via an in-house developed deformable image registration algorithm.</div></div><div><h3>Results</h3><div>A total of 89 % (114/128) of patients completed treatment with the initial tight margins. The mean 3D intrafraction shifts were 1.0 mm (SD: 0.6 mm) in the group with the tight margins and 1.9 mm (SD: 1.5 mm) in the patient group who switched to large margins. The median accumulated D99% was 34.9 Gy (interquartile range: 34.0–35.3 Gy) for patients with prostate shifts who switched to larger margins. In 57 % (8/14) of these patients, the accumulated D99% was above the threshold of 34.4 Gy.</div></div><div><h3>Conclusions</h3><div>Tight margins of 2–3 mm can be safely applied for at least 95 % (122/128) of the PCa patients undergoing a sub-fractionation workflow on a 1.5 T MR-linac.</div></div>","PeriodicalId":36850,"journal":{"name":"Physics and Imaging in Radiation Oncology","volume":"34 ","pages":"Article 100776"},"PeriodicalIF":3.4,"publicationDate":"2025-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144070054","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Uncertainties in outcome modelling in radiation oncology","authors":"Lukas Dünger ,&nbsp;Emily Mäusel ,&nbsp;Alex Zwanenburg ,&nbsp;Steffen Löck","doi":"10.1016/j.phro.2025.100774","DOIUrl":"10.1016/j.phro.2025.100774","url":null,"abstract":"<div><div>Outcome models predicting e.g. survival, tumour control or radiation-induced toxicities play an important role in the field of radiation oncology. These models aim to support the clinical decision making and pave the way towards personalised treatment. Both validity and reliability of their output are required to facilitate clinical integration. However, models are influenced by uncertainties, arising from data used for model development and model parameters, among others. Therefore, quantifying model uncertainties and addressing their causes promotes the creation of models that are sufficiently reliable for clinical use. This topical review aims to summarise different types and possible sources of uncertainties, presents uncertainty quantification methods applicable to various modelling approaches, and highlights central challenges that need to be addressed in the future.</div></div>","PeriodicalId":36850,"journal":{"name":"Physics and Imaging in Radiation Oncology","volume":"34 ","pages":"Article 100774"},"PeriodicalIF":3.4,"publicationDate":"2025-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144070698","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Clinical implementation of patient-specific quality assurance for synthetic computed tomography","authors":"Francesca Nella ,&nbsp;Stephanie Tanadini-Lang,&nbsp;Riccardo Dal Bello","doi":"10.1016/j.phro.2025.100764","DOIUrl":"10.1016/j.phro.2025.100764","url":null,"abstract":"<div><h3>Background and purpose</h3><div>In a magnetic resonance (MR) only planning workflow, MR image is the sole dataset acquired. In order to calculate the dose deposition, a synthetic CT (sCT) is generated to substitute the planning computed tomography (CT). This study aimed to establish acceptance criteria for the clinical implementation of patient-specific quality assurance (PSQA) for sCT.</div></div><div><h3>Materials and methods</h3><div>A retrospective study was conducted on 60. 30 patients underwent a CT scan in treatment position and an MR in diagnostic position. 30 patients had both CT and MR images acquired in treatment position. For the latter group, a sCT for dose calculation was generated and compared against three PSQA methods: recalculation on (A) water override of the body, (B) tissue classes with bulk density overrides and (C) planning CT. The relative dose differences (ΔD [%]) between the sCT and the PSQA methos were evaluated.</div></div><div><h3>Results</h3><div>ΔD for PTV Dmean for method (A) were within 3% for pelvis and 4% for brain cohorts, with standard deviations below 1%. Methods (B) and (C) remained within 2% and 1%, respectively, with deviations up to 1%.</div></div><div><h3>Conclusion</h3><div>The present study proposes a robust PSQA method for MR-only planning. Method (A) is a valuable tool for identifying potential large outliers for Dmean deviations (&gt; 5 %) and it is proposed as the routine PSQA. Method (B) can be used for pelvis cases to improve detection to the 2 % level if method (A) fails. If both (A) and (B) fail, method (C) can be used as a fall-back.</div></div>","PeriodicalId":36850,"journal":{"name":"Physics and Imaging in Radiation Oncology","volume":"34 ","pages":"Article 100764"},"PeriodicalIF":3.4,"publicationDate":"2025-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143791666","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Development and external multicentric validation of a deep learning-based clinical target volume segmentation model for whole-breast radiotherapy","authors":"Maria Giulia Ubeira-Gabellini ,&nbsp;Gabriele Palazzo ,&nbsp;Martina Mori ,&nbsp;Alessia Tudda ,&nbsp;Luciano Rivetti ,&nbsp;Elisabetta Cagni ,&nbsp;Roberta Castriconi ,&nbsp;Valeria Landoni ,&nbsp;Eugenia Moretti ,&nbsp;Aldo Mazzilli ,&nbsp;Caterina Oliviero ,&nbsp;Lorenzo Placidi ,&nbsp;Giulia Rambaldi Guidasci ,&nbsp;Cecilia Riani ,&nbsp;Andrei Fodor ,&nbsp;Nadia Gisella Di Muzio ,&nbsp;Robert Jeraj ,&nbsp;Antonella del Vecchio ,&nbsp;Claudio Fiorino","doi":"10.1016/j.phro.2025.100749","DOIUrl":"10.1016/j.phro.2025.100749","url":null,"abstract":"<div><h3>Background and purpose:</h3><div>In order to optimize the radiotherapy treatment and minimize toxicities, organs-at-risk (OARs) and clinical target volume (CTV) must be segmented. Deep Learning (DL) techniques show significant potential for performing this task effectively. The availability of a large single-institute data sample, combined with additional numerous multi-centric data, makes it possible to develop and validate a reliable CTV segmentation model.</div></div><div><h3>Materials and methods:</h3><div>Planning CT data of 1822 patients were available (861 from a single center for training and 961 from 8 centers for validation). A preprocessing step, aimed at standardizing all the images, followed by a 3D-Unet capable of segmenting both right and left CTVs was implemented. The metrics used to evaluate the performance were the Dice similarity coefficient (DSC), the Hausdorff distance (HD), and its 95th percentile variant (HD_95) and the Average Surface Distance (ASD).</div></div><div><h3>Results:</h3><div>The segmentation model achieved high performance on the validation set (DSC: 0.90; HD: 20.5 mm; HD_95: 10.0 mm; ASD: 2.1 mm; epoch 298). Furthermore, the model predicted smoother contours than the clinical ones along the cranial–caudal axis in both directions. When applied to internal and external data the same metrics demonstrated an overall agreement and model transferability for all but one (Inst 9) center.</div></div><div><h3>Conclusion:</h3><div>. A 3D-Unet for CTV segmentation trained on a large single institute cohort consisting of planning CTs and manual segmentations was built and externally validated, reaching high performance.</div></div>","PeriodicalId":36850,"journal":{"name":"Physics and Imaging in Radiation Oncology","volume":"34 ","pages":"Article 100749"},"PeriodicalIF":3.4,"publicationDate":"2025-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143767951","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Dose assessment in moving targets and organs at risk during carbon ion therapy for pancreatic cancer with respiratory gating","authors":"Christina Stengl ,&nbsp;Jeppe B. Christensen ,&nbsp;Iván D. Muñoz ,&nbsp;Alexander Neuholz ,&nbsp;Stephan Brons ,&nbsp;Eduardo G. Yukihara ,&nbsp;Jakob Liermann ,&nbsp;Oliver Jäkel ,&nbsp;José Vedelago","doi":"10.1016/j.phro.2025.100775","DOIUrl":"10.1016/j.phro.2025.100775","url":null,"abstract":"<div><h3>Background and purpose</h3><div>Carbon ion radiotherapy (CIRT) has demonstrated promising treatment outcomes for pancreatic cancer. However, breathing-induced organ motion can compromise the efficacy of the treatment, leading to under- or over-dosage within the target and organs at risk (OARs). In this work, the dose during CIRT was simultaneously measured at the target and OARs using an anthropomorphic phantom to evaluate the effectiveness of respiratory gating for compensating breathing motion.</div></div><div><h3>Materials and methods</h3><div>The <u>P</u>ancreas <u>P</u>hantom for <u>I</u>on b<u>e</u>am <u>T</u>herapy (PPIeT) was irradiated with carbon ions. The phantom features a pancreas with a virtual tumour and OARs including a duodenum, kidneys, a spine and a spinal cord. Breathing-induced organ motion was imitated with amplitudes of 0 mm (control), 5 mm, 10 mm and 20 mm while irradiating with and without gating. Dose measurements were performed using an ionisation chamber and passive detectors.</div></div><div><h3>Results</h3><div>The prescribed uniform dose of 1.37 Gy in the virtual tumour was experimentally validated for the control. Breathing-induced motion of 20 mm led to a 75 % dose coverage at the target improving to 91 % with gating. For the OARs, the mean dose varied according to the organ, with gating showing no significant differences.</div></div><div><h3>Conclusions</h3><div>Accurate CIRT dosimetry with variable breathing-induced motions can be conducted with PPIeT for a pancreatic tumour and the OARs. Gating mitigated the effects of breathing-induced motion in the tumour.</div></div>","PeriodicalId":36850,"journal":{"name":"Physics and Imaging in Radiation Oncology","volume":"34 ","pages":"Article 100775"},"PeriodicalIF":3.4,"publicationDate":"2025-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143941762","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Predicting the hypoxic volume of head and neck tumors from fluorodeoxyglucose positron emission tomography images using artificial intelligence","authors":"Wei Zhao ,&nbsp;Milan Grkovski ,&nbsp;Heiko Schoder ,&nbsp;Aditya P. Apte ,&nbsp;John Humm ,&nbsp;Nancy Y. Lee ,&nbsp;Joseph O. Deasy ,&nbsp;Harini Veeraraghavan","doi":"10.1016/j.phro.2025.100769","DOIUrl":"10.1016/j.phro.2025.100769","url":null,"abstract":"<div><h3>Background and purpose</h3><div>Tumor hypoxia is linked to lower local control rates and increased distant disease progression during head and neck (HN) radiotherapy. ¹⁸F-fluoromisonidazole (¹⁸F-FMISO) positron emission tomography (PET) imaging measured hypoxia can aid dose selection for HN patients, but its availability is limited. Hence, we tested the hypothesis that an artificial intelligence (AI) model could synthesize ¹⁸F-FMISO-like images from routinely acquired ¹⁸F-fluorodeoxyglucose (¹⁸F-FDG) PET images in order to predict primary tumor or metastatic lymph node hypoxic volumes.</div></div><div><h3>Materials and methods</h3><div>One hundred and thirty-four (training = 84, validation = 13, testing = 21, additional testing = 16) HN carcinoma patients, treated with chemoradiotherapy between 2011 and 2018 and scanned at treatment baseline with ¹⁸F-FDG PET/computed tomography (CT) and ¹⁸F-FMISO dynamic PET/CT, were analyzed. A pix2pix-architecture-based generative adversarial network was trained to yield 2D voxel-wise FMISO hypoxia images of target-to-blood ratios (TBRs) directly from the ¹⁸F-FDG PET/CT image slices. The hypoxic volume was defined consistent with clinical procedure as the malignant volume with TBR values above 1.2. The AI model hypoxia predictions were compared against scaled ¹⁸F-FDG PET values.</div></div><div><h3>Results</h3><div>The AI model hypoxic volume predictions were well-correlated with ¹⁸F-FMISO hypoxic volumes on the held-out test subjects (Pearson correlation testing R = 0.96, additional testing R = 0.91, p &lt; 0.001). Predictions from globally scaled ¹⁸F-FDG PET images also produced a significantly correlated but worse prediction.</div></div><div><h3>Conclusion</h3><div>Voxel-wise prediction of hypoxia for HN cancers from a 2D deep learning model using FDG-PET images as inputs was shown to be feasible. Testing on larger institutional and multi-institutional cohorts is required to establish generalizability.</div></div>","PeriodicalId":36850,"journal":{"name":"Physics and Imaging in Radiation Oncology","volume":"34 ","pages":"Article 100769"},"PeriodicalIF":3.4,"publicationDate":"2025-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143928940","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}