Sharing a whole-/total-body [18F]FDG-PET/CT dataset with CT-derived segmentations: an ENHANCE.PET initiative

Sharing a whole-/total-body [18F]FDG-PET/CT dataset with CT-derived segmentations: an ENHANCE.PET initiative | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Data Note Sharing a whole-/total-body [18F]FDG-PET/CT dataset with CT-derived segmentations: an ENHANCE.PET initiative Daria Ferrara, Manuel Pires, Sebastian Gutschmayer, Josef Yu, and 26 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7169062/v2 This work is licensed under a CC BY 4.0 License Status: Posted Version 2 posted You are reading this latest preprint version Show more versions Abstract We present a large whole-body and total-body curated dataset of dual-modality 2-deoxy-2-[18F]fluoro-D-glucose (FDG)-Positron Emission Tomography/Computed Tomography (PET/CT) studies, consisting of 1,597 PET/CT images and the corresponding CT-derived segmentations of 130 target regions. This multi-center dataset includes images from individuals without overt disease and patients with different pathologies (lung cancer, lymphoma, and melanoma). Target regions were first automatically segmented from CT images using an in-house software, and subsequently verified and corrected by physicians-in-training. In total, the segmented regions encompass 130 volumes, including abdominal organs, muscles, bones, cardiac subregions, vessels, adipose tissue, and skeletal muscle around the third lumbar vertebra. PET/CT images and corresponding CT-derived segmentations are provided in anonymized NIfTI format. The dataset can be used for deep learning training, validation, or multi-modality image analysis and thus fills an important gap in available resources to advance the use of PET/CT data in clinical management. Medical Physics [18F]FDG-PET/CT images anatomical segmentations open-sourcing Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Background & Summary In recent years, the field of biomedical engineering and medical physics has witnessed an increase in the complexity of data 1 , driven by rapid advancements in imaging technologies. Traditional data analysis methods have become increasingly inadequate to analyse these large data sets to meet the demand for greater diagnostic precision, and the shift toward personalized treatment strategies 2 , 3 . To support personalised medicine, more efficient, automated approaches capable of processing and interpreting large-scale datasets are needed. Artificial intelligence (AI) and machine learning (ML) have emerged as powerful tools in this context, offering the ability to identify complex patterns and insights that may not be apparent through conventional methods. However, the effectiveness of AI, particularly of deep learning, is heavily dependent on the availability of large, high-quality and heterogeneous datasets, thus requiring extensive training on vast amounts of data to achieve generalizability and robustness 4 . Combined positron emission tomography (PET) and computed tomography (CT) integrates both anatomical and functional imaging capabilities, making it indispensable for diagnosing, staging, and monitoring diseases, such as oncological disorders 5 – 7 . Despite the clinical importance of PET/CT datasets, open sourcing of imaging data is hindered by strict regulations, and analyses are often conducted in-house 8 on limited data. At present, very few nuclear medicine datasets with annotated lesions are publicly available : only 1,014 PET/CT lung cancer, lymphoma, melanoma, and healthy control cases from the AutoPET challenge 9 , and another 845 head and neck cancer cases through the HECKTOR challenge 10 . In contrast, Ma et al 11 identified over one million open-source non-nuclear medicine datasets, most of which originating from radiology and not segmented, including more than 350,000 from CT scans alone. This restricts the development and validation of computational methods for functional imaging, such as image and tumor segmentation, volumetric analysis (e.g., for body composition assessment 12 ), and radiomics. Recent advancements in PET/CT technology, particularly the shift from single-organ imaging 13 to total-body PET/CT systems 14 , 15 , allow for simultaneous imaging of multiple organs, fueling multi-organ analyses 16 and the exploration of systemic metabolic abnormalities 17 , 18 . However, the development of reliable AI methods for automated analysis of these complex datasets requires access to comprehensive open-source resources, including both images and high-quality segmentations of anatomical structures, which are critical for applications such as diagnosis, treatment planning 19 , volumetric analysis, and patient -specific dosimetry 20 , 21 . In the field of CT imaging alone, few open-source datasets include corresponding anatomical segmentations. Rister et al. 22 presented a dataset of 140 abdominal, neck-to-pelvis, and whole-body CT images from patients with liver cancer, segmented into six organ regions. The WORD dataset 23 comprises 170 abdominal CT images, primarily from prostate, cervical, or rectal cancer cases, along with the segmentations of 16 abdominal organs. These studies, however, are limited in the number of available CT images and segmented regions, and they do not extensively cover different pathologies. More recently, Koitka et al. introduced the Sparsely Annotated Region and Organ Segmentation (SAROS) dataset 24 , which consists of 900 abdominal, thoracic, or whole-body CT images from various pathologies. This work focused on 13 semantic body regions and six body parts, including annotations for every fifth image slice. Similar scope and scale were achieved in the AbdomenCT-1k study 25 , which focused on the liver, kidneys, spleen, and pancreas segmentations, and the comprehensive TotalSegmentator dataset 26 , with CT images of the abdomen, pelvis, or thorax segmented into a total of 104 regions of interest. However, these datasets are limited in scope, often focusing on specific body regions rather than total-body imaging. It is understood that CT images alone are sufficient for many applications, such as volumetric analysis for body composition 12 or the delineation of organs at risk in radiotherapy treatment planning 27 . In other applications, however, the functional information from PET imaging is essential as it provides complementary insights into disease mechanisms that CT alone cannot offer. For example, in pathological settings, [18F]FDG-uptake can help track disease progression by detecting systemic changes in metabolism, such as those seen in patients with infections 28 , 29 , chronic inflammation 30 , metabolic syndrome 31 or cancer-associated cachexia 32 – 34 . In studies involving healthy cohorts, longitudinal [18F]FDG PET/CT imaging allows for monitoring metabolic activity in participants and how it changes with aging or other factors 35 – 37 . Also, a more complete understanding of normal physiological metabolism would help identify deviations that may signal early stages of disease 18 . While the aforementioned AutoPET 9 and HECKTOR 10 challenges provide large PET/CT datasets, they focus on segmentations of pathological tissues but ignore healthy anatomical regions. In the present study, we address the limited availability of open source PET/CT images with segmented tissues as part of our ENHANCE.PET 38 initiative, which aims to facilitate the sharing of open-source tools and datasets to support research within the PET community. We curated a large [18F]FDG PET/CT dataset with anatomical segmentations fully verified by human readers. This dataset includes 1,597 whole-body and total-body PET/CT scans, along with corresponding CT-derived segmentations of 130 non-pathological tissues per scan. The initial segmentations were generated using our in-house tool, MOOSE 39 , for automatic CT segmentation and were manually verified and corrected using 3D Slicer 40 , a software platform for image analysis. The data include contributions from the LuCaPET consortium (grant number ERAPerMed_324, “ Clinical decision support for predicting cachexia in cancer patients using hybrid PET/CT imaging ”) and from the AutoPET Challenge 9 , whose images and lesion segmentations were already available as open-source on The Cancer Imaging Archive 41 . Focused mainly on the oncological cases of lung cancer, melanoma, and lymphoma (Fig. 1 ), the ENHANCE.PET 1.6k dataset also includes participants without known disease. The dataset is provided in anonymized NIfTI format to ensure patient privacy, along with demographic details and CT and PET acquisition parameters as non-imaging metadata. Compared to other publicly available datasets, ENHANCE.PET 1.6k uniquely focuses on the segmentations of organ volumes while avoiding pathological tissues (e.g., tumors, Fig. 2 ). This dataset is particularly suited for training models aimed at automatic image segmentation and identification of healthy tissues. We believe that the open availability of this dataset will advance the differential understanding of healthy and pathological tissues in computational medicine. This comprehensive resource is now available to facilitate future research and advanced data analysis in whole-body PET/CT imaging. We anticipate that applications such as developing and validating deep learning algorithms for automated data analysis and studies on disease-related systemic abnormalities will greatly benefit from this high-quality data collection. Methods Data collection The ENHANCE.PET 1.6k dataset was acquired in accordance with the guidelines set forth in the Declaration of Helsinki . Images were acquired between 1999 and 2022 from various institutions and studies, summarized in Fig. 3 : the open-source dataset AutoPET 9 , the University Hospital Leipzig in Germany ( IRB : 259/18-ek) and the Azienda Ospedaliero Universitaria Careggi in Italy ( IRB : 21306_oss) as part of the LuCaPET consortium. Participant demographics across the different clinical conditions are summarized in Table 1 . Table 1 Demographics and clinical details of the participants included in the study. Partner University Clinical Condition # Sex Age [years] Weight [kg] Height [cm] Azienda Ospedaliero Universitaria Careggi, IT Lung Cancer 199 72F / 127M 71 ± 10 72 ± 15 168 ± 9 University Hospital Leipzig, DE Lung Cancer 384 110F / 274M 65 ± 11 76 ± 16 172 ± 9 Open-source dataset AutoPET 9 Lung Cancer Lymphoma Melanoma Negative Findings 168 145 188 513 65F / 103M 69F / 76M 77F / 111M 233F / 280M 60 ± 15 57 ± 18 60 ± 16 60 ± 16 80 ± 18 79 ± 19 79 ± 20 79 ± 18 173 ± 10 173 ± 11 172 ± 9 171 ± 11 Imaging Protocols Three different PET/CT systems were used for image acquisition at the participating medical centres: Siemens Biograph mCT (N = 1398), Philips Gemini TF (N = 179), and GE Healthcare Discovery MI (N = 20). At all three sites, diagnostic CT scans were acquired with X-ray tube voltages between 100 kVp and 140 kVp, and CT data were reconstructed with a slice thickness between 1 mm and 5 mm. Details on the CT reconstruction parameters are provided in Table 2 . Table 2 Summary of CT systems and reconstruction parameters of the ENHANCE.PET 1.6k dataset. Partner University Azienda Ospedaliero Universitaria Careggi, IT N = 199 University Hospital Leipzig, DE N = 384 Open-source dataset AutoPET 9 N = 1014 PET/CT System Manufacturer GE Medical System (19) Philips (180) Siemens Siemens System Model Discovery MI (19) Gemini TF TOF 16 (180) Biograph 16 (207) Sensation 16 (177) Biograph 128 (900) SOMATOM Definition (114) kVp 120 (196) 140 (3) 120 100 (2) 120 (938) 140 (74) Filter Type N/A NONE FLAT Convolutional Kernel N/A STANDARD (19) B10f (325) B31f (56) B40f (3) I30f (31) I31f (471) B30f (24) B31f (488) Axial Pixel Size (mm) 0.98–1.37 0.98 0.69–0.98 Slice Thickness (mm) 3.75 (19) 5 (180) 2 (36) 3 (347) 1 (52) 2 (141) 3 (821) Focal Spot Size (mm) N/A 0.7 (330) 1.2 (54) 1.2 Participants were asked to fast for 6 hours before the examinations and were scanned in the supine position, with arms up. Each subject underwent a static PET acquisition following an intravenous injection of [18F]FDG (314 ± 48 MBq). Uptake times varied across the three sites, with an average of (68 ± 21) minutes post-injection. PET images were reconstructed with attenuation and scatter corrections applied using the corresponding CT data. Details on the CT and PET acquisition parameters are reported for each participant as non-imaging parameters in the available spreadsheet files. The download link is provided in the Data Records section. Segmentations and Data Processing PET/CT images were retrieved in anonymized DICOM format from the participants and centralized at the Medical University of Vienna. The metadata were used to extract relevant information about the CT and PET acquisition protocols as well as essential demographic details of the participants. For subsequent analysis and segmentation, all data were converted to NIfTI format using the dcm2niix DICOM to NIfTI converter 42 . To ensure that participants could not be visually identified from their CT images 43 – 45 , both the PET and CT images from the Azienda Ospedaliero Universitaria Careggi and the University Hospital Leipzig were edited: in the PET images, voxels between the upper part of the skull segmentation and the bottom of the brain, within a cylinder of 16 voxels in the z-direction, were set to zero (Fig. 4 B). Similarly, the corresponding CT region was set to -1000 Hounsfield Units to simulate air. The images from the AutoPET challenge were left unedited as per their version available online. To maximize efficiency and accelerate the workflow, the processing of the entire ENHANCE.PET 1.6k dataset was done serially: automatic segmentation of the CT images, manual refinement of the derived labels, and retraining of the original segmentation models, according to the following scheme. Our in-house developed software, MOOSE 39 , was first used for the automatic segmentation of 384 lung cancer images from the University Hospital Leipzig. The resulting segmentations were manually refined by 10 medical students using the 3D Slicer image analysis software 40 . For each dataset, a student was randomly assigned to verify and correct the segmentations, addressing possible systematic errors such as inaccuracies at anatomical borders of target regions, mislabelling between left and right regions, or misclassification of regions with similar intensities on CT. A second student was then tasked with reviewing the first student's work and correcting any remaining mistakes. Once both students agreed on the final version of the dataset, it was reviewed by a radiology resident and a nuclear medicine resident. The PET images overlapped with the corresponding CT were used to exclude tumor volumes from the segmentation of various abdominal organs. These preliminary segmentations were divided into seven different classes, as shown in Fig. 5 : organs, cardiac, muscles, ribs, peripheral bones, vertebrae, and body composition around the L3 vertebra area. These masks were then used to retrain our models using nnU-Net 46 . Details on the retraining process are provided in the Technical Validation section. The newly trained models were subsequently used for the segmentation of the second dataset, originating from Azienda Ospedaliero Universitaria Careggi, and again underwent manual refinement and quality control described above. The same workflow of automatic segmentation, manual refinement, and model retraining was then applied to the remaining images from the AutoPET 9 open-source dataset. In the case of the AutoPET data, since the corresponding lesion segmentations were available online, they were taken as ground truth and directly subtracted from the organ segmentations without the need for manual correction. To maximize data variability and ensure robust performance across different scanning conditions, we included in our training dataset 86 additional total-body CT images (30F/56M, 53 ± 15 years, 87 ± 19 kg, 172 ± 9 cm) from the University of California Davis, California (USA), acquired on a United Imaging uEXPLORER PET/CT system. The segmentations were generated using the same workflow as for the other datasets. Their inclusion helped account for different acquisition protocols (arms-down positioning) and provided broader representation of physiological variations, as they covered various inflammatory and pathological conditions, such as head and neck cancer (N = 5), arthritis (N = 43), genito-urinary cancer (N = 9), and healthy controls (N = 29). While these cases cannot currently be shared due to privacy restrictions, we are working to make them publicly available in the future. At each stage of retraining, the size of the training data increased, improving model performance. This iterative process allowed for efficient verification and faster corrections by the medical students without compromising precision, especially in regions where systematic errors had been identified and were hindering the manual correction process. Segmentation of the fingers and hand bones showed the most significant improvement as the training dataset grew (Fig. 6 ). In the first round of training, several instances of left-right misclassification were identified, especially when the hands were crossed over the abdomen or above the head. However, this issue progressively improved with each retraining step. Another improvement achieved through more extensive training data was the automatic inclusion of the quadratus lumborum muscle in the "skeletal muscle" label for body composition, which had previously been missing and required manual correction (Fig. 6 ). Data Records The PET/CT images and corresponding segmentations from Azienda Ospedaliero Universitaria Careggi, University Hospital Leipzig, and the AutoPET 9 dataset are stored on Amazon Web Services (AWS, https://aws.amazon.com/ ) and can be downloaded either directly following MOOSE 39 installation via command line, as described in the Code Availability section, or at the following link: https://enhance-pet.s3.eu-central-1.amazonaws.com/enhance-pet-1_6k/enhance-pet-1_6k.zip The imaging data are stored in three separate folders containing the CT images, PET images, and ground truth segmentations, respectively. Within the segmentations folder, there are seven subfolders corresponding to different segmentation classes: "Body-Composition," "Cardiac," "Muscles," "Organs," "Peripheral Bones," "Ribs," and "Vertebrae." Each folder contains NIfTI data files, named sequentially from 0001.nii.gz to 1597.nii.gz. The directory structure of the ENHANCE.PET 1.6k dataset is shown in Fig. 7 . A JSON file containing the complete list of segmentations and their corresponding intensities within the multi-class files is available for download at the following link: https://enhance-pet.s3.eu-central-1.amazonaws.com/enhance-pet-1_6k/labels.json In addition to the imaging data, non-imaging information is provided in two spreadsheet files. The CT-details.xlsx file contains details on the CT acquisition parameters (e.g., PET/CT system manufacturer and model, kVp, filter type, convolutional kernel, axial pixel size, slice thickness, and focal spot size) for each participant. The PT-details.xlsx file provides the corresponding demographic information (e.g., clinical indication, sex, age, weight, height) as well as PET acquisition parameters (e.g., injected activity, acquisition date and time, radioactivity injection details, image units, slope, intercept, system model and manufacturer). Technical Validation We used the ENHANCE.PET 1.6k dataset with the additional 86-image contribution from the University of California, Davis to develop a deep learning-based method for the automatic segmentation of CT scans: 80% of the images and corresponding labels were sampled from the total with a stratified sampling, thus ensuring that the original proportion of data by medical facility and clinical condition remained unchanged. The 1346 selected imaging data served as a training dataset for multiple segmentation models targeting different anatomical regions, including bones of the limbs and skull, thoracic cage bones, vertebrae and sacrum, major abdominal organs, lower back muscles, cardiac tissues, and body composition around the L3 vertebra. A detailed list of regions segmented by each model is shown in Fig. 5 . Prior to training, all images and labels were resampled with SimpleITK ( https://simpleitk.org/about.html ) from the original resolution to a voxel spacing of 1.5 x 1.5 x 1.5 mm using B-spline interpolation. This provided a resolution high enough to segment fine structures on the CT images, while considering the computational burden of training. The models were trained using nnU-Net 46 , a state-of-the-art self-configuring framework based on the U-Net architecture for semantic segmentation. The training process was performed over 2000 epochs. To assess the model performance, we tested all segmentation models on the remaining 20% of the ENHANCE.PET 1.6k. Segmentation accuracy was evaluated using the Dice Similarity Coefficient (DSC) to quantify the overlap between predicted segmentations and the reference labels and with the Average Symmetric Surface Distance (ASSD) 47 to estimate the average distance between surface voxels of the reference labels and the automated segmentation. The averaged results for the generated models are shown in Fig. 8 , and the metrics for each label are reported in Table 4 . Table 4 Mean Dices and Average Symmetric Surface Distance (ASSD) per segmented volume between the reference labels of the test dataset (N = 337, 20% of the total ENHANCE.PET 1.6k) and the labels resulting from the AI prediction. Left and right regions were merged for this analysis. Model Regions Mean Dice ± St Dev Mean ASSD ± St Dev [mm] Organs Adrenal Glands Bladder Brain Gallbladder Kidney Liver Lung Lower Lobe Lung Middle Lobe Lung Upper Lobe Pancreas Spleen Stomach Thyroid 0.82 ± 0.12 0.90 ± 0.18 0.96 ± 0.13 0.88 ± 0.15 0.96 ± 0.05 0.98 ± 0.03 0.95 ± 0.12 0.94 ± 0.09 0.97 ± 0.05 0.90 ± 0.08 0.96 ± 0.08 0.95 ± 0.06 0.87 ± 0.19 0.5 ± 0.5 1.0 ± 2.0 0.6 ± 0.4 0.8 ± 1.7 0.5 ± 0.7 0.6 ± 0.5 0.6 ± 1.4 0.8 ± 1.8 0.9 ± 0.6 0.7 ± 0.7 0.7 ± 0.8 0.7 ± 1.0 0.6 ± 0.8 Cardiac Myocardium Atrium Ventricle Aorta Iliac Artery Iliac Vena Inferior Vena Cava Portal Splenic Vein Pulmonary Artery 0.92 ± 0.05 0.96 ± 0.03 0.96 ± 0.03 0.95 ± 0.02 0.90 ± 0.09 0.92 ± 0.07 0.92 ± 0.06 0.82 ± 0.18 0.94 ± 0.05 0.5 ± 0.4 0.5 ± 0.4 0.5 ± 0.4 0.5 ± 0.2 0.6 ± 1.9 0.8 ± 1.8 0.6 ± 0.5 0.9 ± 1.8 0.8 ± 0.6 Muscles Autochthon Gluteus Maximus Gluteus Medius Gluteus Minimus Iliopsoas 0.98 ± 0.01 0.98 ± 0.01 0.98 ± 0.01 0.96 ± 0.03 0.97 ± 0.02 0.3 ± 0.2 0.3 ± 0.2 0.3 ± 0.2 0.3 ± 0.2 0.3 ± 0.2 Ribs Rib 1 Rib 2 Rib 3 Rib 4 Rib 5 Rib 6 Rib 7 Rib 8 Rib 9 Rib 10 Rib 11 Rib 12 Sternum 0.86 ± 0.12 0.88 ± 0.12 0.88 ± 0.11 0.90 ± 0.10 0.90 ± 0.09 0.91 ± 0.09 0.91 ± 0.09 0.91 ± 0.09 0.90 ± 0.10 0.90 ± 0.11 0.89 ± 0.14 0.86 ± 0.18 0.94 ± 0.03 0.5 ± 0.6 0.4 ± 0.8 0.4 ± 0.6 0.4 ± 0.9 0.5 ± 2.0 0.5 ± 1.8 0.6 ± 2.4 0.6 ± 2.3 0.7 ± 1.9 0.8 ± 2.7 0.9 ± 3.2 0.9 ± 3.7 0.5 ± 1.1 Vertebrae Vertebra C1 Vertebra C2 Vertebra C3 Vertebra C4 Vertebra C5 Vertebra C6 Vertebra C7 Vertebra T1 Vertebra T2 Vertebra T3 Vertebra T4 Vertebra T5 Vertebra T6 Vertebra T7 Vertebra T8 Vertebra T9 Vertebra T10 Vertebra T11 Vertebra T12 Vertebra L1 Vertebra L2 Vertebra L3 Vertebra L4 Vertebra L5 Hip Sacrum 0.89 ± 0.10 0.92 ± 0.09 0.91 ± 0.09 0.90 ± 0.09 0.90 ± 0.11 0.90 ± 0.09 0.91 ± 0.09 0.93 ± 0.09 0.93 ± 0.10 0.92 ± 0.10 0.93 ± 0.09 0.93 ± 0.07 0.93 ± 0.08 0.93 ± 0.09 0.94 ± 0.08 0.94 ± 0.08 0.94 ± 0.10 0.94 ± 0.11 0.94 ± 0.12 0.93 ± 0.13 0.93 ± 0.14 0.93 ± 0.15 0.93 ± 0.16 0.90 ± 0.20 0.97 ± 0.08 0.96 ± 0.08 0.5 ± 0.7 0.4 ± 0.4 0.4 ± 0.4 0.4 ± 0.5 0.4 ± 0.4 0.4 ± 0.4 0.4 ± 0.6 0.4 ± 0.7 0.4 ± 0.9 0.4 ± 0.9 0.4 ± 0.9 0.4 ± 0.5 0.4 ± 0.6 0.4 ± 1.0 0.4 ± 0.9 0.5 ± 1.0 0.5 ± 1.2 0.5 ± 1.4 0.6 ± 1.7 0.7 ± 2.2 0.7 ± 2.4 0.7 ± 2.6 0.8 ± 2.6 0.6 ± 2.0 0.3 ± 0.2 0.3 ± 0.2 Peripheral Bones Carpal Clavicle Femur Fibula Fingers Humerus Metacarpal Metatarsal Patella Radius Scapula Skull Tarsal Tibia Toes Ulna 0.76 ± 0.33 0.91 ± 0.16 0.96 ± 0.15 0.93 ± 0.19 0.55 ± 0.39 0.94 ± 0.15 0.71 ± 0.33 0.92 ± 0.17 0.92 ± 0.19 0.66 ± 0.42 0.91 ± 0.16 0.90 ± 0.16 0.95 ± 0.17 0.95 ± 0.16 0.90 ± 0.15 0.68 ± 0.40 0.8 ± 0.6 0.5 ± 2.1 0.3 ± 0.6 0.3 ± 0.5 10.0 ± 37.5 0.4 ± 1.0 5.2 ± 22.7 0.8 ± 4.1 0.5 ± 1.0 0.5 ± 0.8 0.3 ± 0.4 0.5 ± 0.6 1.4 ± 4.5 0.3 ± 0.3 0.6 ± 1.3 0.5 ± 0.6 Body Composition Skeletal Muscle Subcutaneous Fat Visceral Fat 0.85 ± 0.08 0.87 ± 0.08 0.87 ± 0.09 2.0 ± 1.5 2.4 ± 1.8 1.5 ± 1.6 All models achieved high accuracy, with mean DSC values exceeding 0.85 across most regions and mean ASSD values below 3 mm in all regions. The "Muscles" model achieved the highest overlap and the lowest prediction error, with an average DSC of 0.97 ± 0.02 and an ASSD of 0.3 ± 0.2 mm (Fig. 8 ). Cardiac, organ, and vertebrae models also achieved high average DSC values, exceeding 0.90. The peripheral bones model had the lowest performance, with an average DSC of 0.86 ± 0.27 and the highest variation in ASSD, at 0.8 ± 7.2 mm. The lower performance in these regions is likely due to their small size and thin anatomical structures: the digits of the hand had the most significant negative impact on model performance, with some cases of left/right misclassification identified (especially when patients underwent imaging with their hands crossed over the abdomen), resulting in an average DSC of 0.55 ± 0.39 and an ASSD of 10 ± 37 mm (Table 4 ). Similarly, the segmentation of the metacarpals yielded a DSC of 0.71 ± 0.33 and an ASSD of 5 ± 23 mm. Other regions with lower overlap included the portal and splenic veins (DSC: 0.82 ± 0.18) and the adrenal glands (DSC: 0.82 ± 0.12), most likely due to their low contrast resolution in CT imaging of the test dataset, which makes delineation more challenging. The ribs and body composition models also showed higher variation in ASSD, at 0.6 ± 2.1 mm and 2.0 ± 1.7 mm, respectively. The ENHANCE.PET 1.6k dataset proved to be satisfactory for training models for automated CT image segmentation. This dataset has the potential to contribute significantly to further advancements in deep learning-based algorithms, including attempts to improve segmentation models performance or the addition of new volumes of interest not covered in the present study. The dual availability of both CT and PET images, together with the inclusion of segmentations for multiple anatomical regions, makes the ENHANCE.PET 1.6k dataset particularly valuable for research focused on diseases that affect multiple organs or systems, such as metabolic disorders or systemic inflammatory diseases 48 , 49 , or for studies on normal glucose metabolism in healthy tissues. As a limitation, since the segmentations were derived from the CT images, the alignment with the corresponding PET images may be compromised in cases of significant patient motion, which was not systematically assessed in this study. We hope that this open-source dataset will accelerate developments in medical imaging, ultimately contributing to the advancement of personalized medicine and more effective clinical decision-making. Usage Notes The contributions of the Azienda Ospedaliero-Universitaria Careggi and the University Hospital Leipzig to the ENHANCE.PET 1.6k dataset are licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0). Data from the AutoPET 9 Challenge are licensed under the Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0). All imaging data are presented in NIfTI format, ensuring participants’ privacy while allowing for easy use in further analysis. This format can be opened with most visualization software, including 3D Slicer ( https://www.slicer.org/ ) and ITK-SNAP ( http://www.itksnap.org/pmwiki/pmwiki.php ). DICOM to NIfTI conversion was performed using dcm2nii 42 , and all image processing was conducted using Python. Declarations Competing interests The authors declare that there are no conflicts of interest related to this project. L.K.S.S and T.B. are co-founders of Zenta GmbH. R.D.B has received research support from Lilly and from United Imaging Healthcare during the course of this study. UC Davis has a revenue-sharing agreement with United Imaging Healthcare. Author contributions D.F., M.P., S.G., T.B., L.K.S.S.: data analysis and processing, project conceptualization, manuscript writing, development and maintenance of the software MOOSE 39 . Acknowledgements This research was funded in whole or in part by the Austrian Science Fund (FWF) ( 10.55776/I5902 ), under the ERA-NET Cofund scheme of the Horizon 2020 Research and Innovation Framework Programme of the European Commission Research Directorate-General, Grant Agreement No. 779282, which includes national funding for the partners by the FWF ( 10.55776/I5902 ), Innovation Fund Denmark, Regione Toscana, and Saxon State Ministry for Science, Culture and Tourism (Germany). For open access purposes, the author has applied a CC BY public copyright license to any author accepted manuscript version arising from this submission. Sebastian Gutschmayer worked in parts under a research agreement between Siemens Healthineers and the Medical University Vienna. Armin Frille was supported by the postdoctoral fellowship ‘MetaRot program’ (clinician scientist program) from the Federal Ministry of Education and Research (BMBF), Germany (FKZ 01EO1501, IFB Adiposity Diseases), a research grant from the ‘Mitteldeutsche Gesellschaft für Pneumologie (MDGP) e.V.’ (2018-MDGP-PA-002), a junior research grant from the Medical Faculty, University of Leipzig (934100-012), and a graduate fellowship from the ‘Novartis Foundation’. UC Davis data were acquired using support from the grants: National Psoriasis Foundation 19-4200:NPF, NIH R01AR076088, NIH R01AR085314, NIH K12CA138464, NIH P30CA093373, NIH R01CA206187. Code Availability The segmentation software MOOSE and the presented open-source dataset are part of the ENHANCE.PET ( https://enhance.pet/ ) initiative for facilitating data sharing and collaboration within the PET community. In case of downloading and usage of the dataset, please cite the ENHANCE.PET initiative and website. MOOSE code is open-source and available online with extensive documentation, and can be accessed on GitHub at https://github.com/ENHANCE-PET/MOOSE . Following MOOSE installation within a Python environment, the dataset can be downloaded via the command line: References Zhang S, Metaxas D (2016) Large-Scale medical image analytics: Recent methodologies, applications and Future directions. Med Image Anal 33:98–101 Pinto-Coelho L (2023) How artificial intelligence is shaping medical imaging technology: A survey of innovations and applications. Bioeng (Basel) 10 Khalifa M, Albadawy M (2024) AI in diagnostic imaging: Revolutionising accuracy and efficiency. Comput Methods Programs Biomed Update 5:100146 Shen C et al (2020) An introduction to deep learning in medical physics: advantages, potential, and challenges. Phys Med Biol 65:05TR01 Beyer T et al (2000) A combined PET/CT scanner for clinical oncology. J Nucl Med 41:1369–1379 Townsend DW (2008) Dual-modality imaging: combining anatomy and function. J Nucl Med 49:938–955 Wechalekar K, Sharma B, Cook G (2005) PET/CT in oncology—a major advance. Clin Radiol 60:1143–1155 Larson DB, Magnus DC, Lungren MP, Shah NH, Langlotz CP (2020) Ethics of using and sharing clinical imaging data for artificial intelligence: A proposed framework. Radiology 295:675–682 Gatidis S et al (2022) A whole-body FDG-PET/CT Dataset with manually annotated Tumor Lesions. Sci Data 9:601 Oreiller V et al (2022) Head and neck tumor segmentation in PET/CT: The HECKTOR challenge. Med Image Anal 77:102336 Ma J et al (2024) Segment anything in medical images. Nat Commun 15:654 Winder C et al (2024) Automated extraction of body composition metrics from abdominal CT or MR imaging: A scoping review. Eur J Radiol 181:111764 Jones T, Townsend D (2017) History and future technical innovation in positron emission tomography. J Med Imaging (Bellingham) 4:011013 Badawi RD et al (2019) First human imaging studies with the EXPLORER total-body PET scanner. J Nucl Med 60:299–303 Prenosil GA et al (2022) Performance Characteristics of the Biograph Vision Quadra PET/CT System with a Long Axial Field of View Using the NEMA NU 2-2018 Standard. J Nucl Med 63:476–484 Shiyam Sundar LK, Hacker M, Beyer T, Whole-Body PET (2022) Imaging: A Catalyst for Whole-Person Research? J Nucl Med 64:197–199 Sun T et al (2022) Identifying the individual metabolic abnormities from a systemic perspective using whole-body PET imaging. Eur J Nucl Med Mol Imaging 49:2994–3004 Gutschmayer S, Muzik O, Hacker M, Ferrara D, Zuehlsdorff S, Newiger H, Beyer T, Sundar LS (2022) Towards holistic assessment of human physiology: Fully-automated construction of tracer independent total-body PET/CT normative database using diffeomorphisms [abstract]. in Annual Congress of the European Association of Nuclear Medicine October 15–19 , Spain vol. 49 225–226 (Springer Science and Business Media LLC, 2022) van Ginneken B, Schaefer-Prokop CM, Prokop M (2011) Computer-aided diagnosis: how to move from the laboratory to the clinic. Radiology 261:719–732 Bednarz B (2023) Theranostics and patient-specific dosimetry. Semin Radiat Oncol 33:317–326 Poel R et al (2021) The predictive value of segmentation metrics on dosimetry in organs at risk of the brain. Med Image Anal 73:102161 Rister B, Yi D, Shivakumar K, Nobashi T, Rubin DL (2020) CT-ORG, a new dataset for multiple organ segmentation in computed tomography. Sci Data 7:381 Luo X et al (2022) A large scale dataset, benchmark and clinical applicable study for abdominal organ segmentation from CT image. Med Image Anal 82:102642 Koitka S, SAROS et al (2024) A dataset for whole-body region and organ segmentation in CT imaging. Sci Data 11:483 Ma J et al (2020) AbdomenCT-1K: Is abdominal organ segmentation A solved problem? arXiv [cs.CV] Wasserthal J et al (2023) TotalSegmentator: Robust Segmentation of 104 Anatomic Structures in CT Images. Radiology: Artif Intell 5:e230024 Chen X et al (2021) A deep learning-based auto-segmentation system for organs-at-risk on whole-body computed tomography images for radiation therapy. Radiother Oncol 160:175–184 Hess S, Hansson SH, Pedersen KT, Basu S, Høilund-Carlsen P (2014) F. FDG-PET/CT in infectious and inflammatory diseases. PET Clin 9:497–519 Yao J et al (2022) The imaging quantification of multiple organs by dynamic 18F-FDG PET/CT in discharged COVID-19 patients: A prospective pilot study. Int J Med Sci 19:1539–1547 Patel NH et al (2021) Heightened splenic and bone marrow uptake of 18F-FDG PET/CT is associated with systemic inflammation and subclinical atherosclerosis by CCTA in psoriasis: An observational study. Atherosclerosis 339:20–26 Bang J-I et al (2020) Blood pool activity on F-18 FDG PET/CT as a possible imaging biomarker of metabolic syndrome. Sci Rep 10:17367 Jiang Y et al (2022) Abnormal [18F]FDG uptake in liver and adipose tissue: a potential imaging biomarker for cancer-associated cachexia. Eur Radiol. 10.1007/s00330-022-09226-z Ferrara D et al (2024) Detection of cancer-associated cachexia in lung cancer patients using whole-body [18F]FDG-PET/CT imaging: A multi-centre study. J Cachexia Sarcopenia Muscle. 10.1002/jcsm.13571 Yu J et al (2024) Systemic metabolic and volumetric assessment via whole-body [18F]FDG-PET/CT: Pancreas size predicts cachexia in head and neck squamous cell carcinoma. Cancers (Basel) 16:3352 Pak K et al (2024) Glucose metabolism and radiodensity of abdominal adipose tissue: A 5-year longitudinal study in a large PET cohort. Clin Endocrinol (Oxf) 101:623–630 Ishii S et al (2024) Investigation of factors affecting CT attenuation and glucose metabolism of bone marrow as seen on PET/CT scans. Am J Nucl Med Mol Imaging 14:22–30 Sharma P, Chatterjee P, Alvarado LA, Dwivedi AK (2023) Standardized uptake value of normal organs on routine clinical [18F]FDG PET/CT: impact of tumor metabolism and patient-related factors. Nucl Med Rev Cent East Eur 26:1–10 Sundar LKS et al (2023) The ENHANCE-PET Framework: An initiative to engage the imaging community in Advancements in Total-Body PET Analysis. EANM’23 Abstract Book Congress Sep 9–13, 2023. Eur J Nucl Med Mol Imaging 50:1–898 Sundar LKS et al (2022) Fully Automated, Semantic Segmentation of Whole-Body 18F-FDG PET/CT Images Based on Data-Centric Artificial Intelligence. J Nucl Med 63:1941–1948 Fedorov A et al (2012) 3D Slicer as an image computing platform for the Quantitative Imaging Network. Magn Reson Imaging 30:1323–1341 Clark K et al (2013) The Cancer Imaging Archive (TCIA): maintaining and operating a public information repository. J Digit Imaging 26:1045–1057 Li X, Morgan PS, Ashburner J, Smith J, Rorden C (2016) The first step for neuroimaging data analysis: DICOM to NIfTI conversion. J Neurosci Methods 264:47–56 Selfridge AR et al (2023) Facial anonymization and privacy concerns in total-body PET/CT. J Nucl Med 64:1304–1309 Schwarz CG et al (2019) Identification of anonymous MRI research participants with face-recognition software. N Engl J Med 381:1684–1686 Clunie DA et al (2023) Report of the medical image DE-identification (MIDI) task group - best practices and recommendations. ArXiv Isensee F, Jaeger PF, Kohl SAA, Petersen J, Maier-Hein KH (2021) nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation. Nat Methods 18:203–211 Yeghiazaryan V, Voiculescu I (2018) Family of boundary overlap metrics for the evaluation of medical image segmentation. J Med Imaging (Bellingham) 5:015006 Katal S, Eibschutz LS, Saboury B, Gholamrezanezhad A, Alavi A (2022) Advantages and applications of total-body PET scanning. Diagnostics (Basel) 12:426 Nakamoto R et al (2019) Diffusely Decreased Liver Uptake on FDG PET and Cancer-Associated Cachexia With Reduced Survival. Clin Nucl Med 44:634–642 Additional Declarations The authors declare no competing interests. Cite Share Download PDF Status: Posted Version 2 posted You are reading this latest preprint version Show more versions Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7169062","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Data Note","associatedPublications":[],"authors":[{"id":489620031,"identity":"2459c0cd-4ba8-40cb-b958-7d08f09c840c","order_by":0,"name":"Daria Ferrara","email":"","orcid":"https://orcid.org/0000-0002-8711-8081","institution":"QIMP Team, Medical University of Vienna, Vienna, Austria","correspondingAuthor":false,"prefix":"","firstName":"Daria","middleName":"","lastName":"Ferrara","suffix":""},{"id":489620032,"identity":"c63b467a-e863-4ffb-9dc3-81fd1b229e51","order_by":1,"name":"Manuel Pires","email":"","orcid":"","institution":"QIMP Team, Medical University of Vienna, Vienna, Austria","correspondingAuthor":false,"prefix":"","firstName":"Manuel","middleName":"","lastName":"Pires","suffix":""},{"id":489620033,"identity":"3980209e-727d-4181-be26-b5566f948c1d","order_by":2,"name":"Sebastian Gutschmayer","email":"","orcid":"","institution":"QIMP Team, Medical University of Vienna, Vienna, Austria","correspondingAuthor":false,"prefix":"","firstName":"Sebastian","middleName":"","lastName":"Gutschmayer","suffix":""},{"id":489620034,"identity":"6168b0bf-f9cf-49d4-a78d-0fcbfc19333d","order_by":3,"name":"Josef Yu","email":"","orcid":"","institution":"QIMP Team, Medical University of Vienna, Vienna, Austria. Division of Nuclear Medicine, Medical University of Vienna, Department of Biomedical Imaging and Image-guided Therapy, Vienna, Austria","correspondingAuthor":false,"prefix":"","firstName":"Josef","middleName":"","lastName":"Yu","suffix":""},{"id":490931328,"identity":"ded4dc55-7958-4764-b724-81a7f8ae16b0","order_by":4,"name":"Yasser G. Abdelhafez","email":"","orcid":"","institution":"Department of Radiology, University of California Davis, Sacramento, California, USA","correspondingAuthor":false,"prefix":"","firstName":"Yasser","middleName":"G.","lastName":"Abdelhafez","suffix":""},{"id":490931329,"identity":"acfe689f-4bf4-409e-b8fd-7f0877d56b83","order_by":5,"name":"Elisabetta Abenavoli","email":"","orcid":"","institution":"Division of Nuclear Medicine, Azienda Ospedaliero Universitaria Careggi, Florence, Italy","correspondingAuthor":false,"prefix":"","firstName":"Elisabetta","middleName":"","lastName":"Abenavoli","suffix":""},{"id":490931330,"identity":"1ff921ac-f622-4762-9fa8-37cf4b217318","order_by":6,"name":"Ramsey D. Badawi","email":"","orcid":"","institution":"Department of Radiology, University of California Davis, Sacramento, California, USA","correspondingAuthor":false,"prefix":"","firstName":"Ramsey","middleName":"D.","lastName":"Badawi","suffix":""},{"id":490931331,"identity":"6ebe921b-b528-4475-a917-a2c981701b7f","order_by":7,"name":"Abhijit J. Chaudhari","email":"","orcid":"","institution":"Department of Radiology, University of California Davis, Sacramento, California, USA","correspondingAuthor":false,"prefix":"","firstName":"Abhijit","middleName":"J.","lastName":"Chaudhari","suffix":""},{"id":490931332,"identity":"e62ac969-1677-429a-9871-98f04541ce47","order_by":8,"name":"Moon S. Chen, Jr.","email":"","orcid":"","institution":"Comprehensive Cancer Center, University of California Davis, Sacramento, California, USA","correspondingAuthor":false,"prefix":"","firstName":"Moon","middleName":"S.","lastName":"Chen","suffix":"Jr."},{"id":490931333,"identity":"39057962-e747-4af4-a904-6e695d74551c","order_by":9,"name":"Simon R. Cherry","email":"","orcid":"","institution":"Department of Radiology, University of California Davis, Sacramento, California, USA","correspondingAuthor":false,"prefix":"","firstName":"Simon","middleName":"R.","lastName":"Cherry","suffix":""},{"id":490931334,"identity":"fca374d7-7085-4e69-a665-6205bd6c6bee","order_by":10,"name":"Armin Frille","email":"","orcid":"","institution":"Department of Respiratory Medicine, University Hospital Leipzig, Leipzig, Germany","correspondingAuthor":false,"prefix":"","firstName":"Armin","middleName":"","lastName":"Frille","suffix":""},{"id":490931670,"identity":"d7235942-165b-4ec9-bd14-183320faa280","order_by":11,"name":"Barbara K. Geist","email":"","orcid":"","institution":"Division of Nuclear Medicine, Medical University of Vienna, Department of Biomedical Imaging and Image-guided Therapy, Vienna, Austria","correspondingAuthor":false,"prefix":"","firstName":"Barbara","middleName":"K.","lastName":"Geist","suffix":""},{"id":490931671,"identity":"032a7d5b-f887-41c9-b6ba-16e59be34db1","order_by":12,"name":"Stefan Gruenert","email":"","orcid":"","institution":"Division of Nuclear Medicine, Medical University of Vienna, Department of Biomedical Imaging and Image-guided Therapy, Vienna, Austria","correspondingAuthor":false,"prefix":"","firstName":"Stefan","middleName":"","lastName":"Gruenert","suffix":""},{"id":490931672,"identity":"9f93ed2b-ec93-4e6c-8111-33c9008ea60f","order_by":13,"name":"Marcus Hacker","email":"","orcid":"","institution":"Division of Nuclear Medicine, Medical University of Vienna, Department of Biomedical Imaging and Image-guided Therapy, Vienna, Austria","correspondingAuthor":false,"prefix":"","firstName":"Marcus","middleName":"","lastName":"Hacker","suffix":""},{"id":490931673,"identity":"5ce0d043-6a7e-4245-b783-55a57137ed2a","order_by":14,"name":"Swen Hesse","email":"","orcid":"","institution":"Department of Nuclear Medicine, University Hospital Leipzig, Leipzig, Germany","correspondingAuthor":false,"prefix":"","firstName":"Swen","middleName":"","lastName":"Hesse","suffix":""},{"id":490931674,"identity":"55ca7e26-1b4e-4695-b224-2bb432e527b7","order_by":15,"name":"Teresa Kerkhoff","email":"","orcid":"","institution":"Department of Respiratory Medicine, University Hospital Leipzig, Leipzig, Germany","correspondingAuthor":false,"prefix":"","firstName":"Teresa","middleName":"","lastName":"Kerkhoff","suffix":""},{"id":490931675,"identity":"21143a0a-e9f9-47de-b38d-3f1e4fb5dcb1","order_by":16,"name":"Pia Linder","email":"","orcid":"","institution":"Division of Nuclear Medicine, University of Tübingen, Tübingen, Germany","correspondingAuthor":false,"prefix":"","firstName":"Pia","middleName":"","lastName":"Linder","suffix":""},{"id":490931676,"identity":"47db2f59-73cb-43c9-a5d5-8c4c273bb8c1","order_by":17,"name":"Johanna Pappisch","email":"","orcid":"","institution":"Department of Respiratory Medicine, University Hospital Leipzig, Leipzig, Germany","correspondingAuthor":false,"prefix":"","firstName":"Johanna","middleName":"","lastName":"Pappisch","suffix":""},{"id":490932296,"identity":"49fc3716-123f-4875-920d-c1d0e8784bbb","order_by":18,"name":"Smilla Pusitz","email":"","orcid":"","institution":"Division of Nuclear Medicine, Medical University of Vienna, Department of Biomedical Imaging and Image-guided Therapy, Vienna, Austria","correspondingAuthor":false,"prefix":"","firstName":"Smilla","middleName":"","lastName":"Pusitz","suffix":""},{"id":490932297,"identity":"ceb4880a-39e2-41b3-abf6-2b79e191104a","order_by":19,"name":"Osama A. Raslan","email":"","orcid":"","institution":"Department of Radiology, University of California Davis, Sacramento, California, USA","correspondingAuthor":false,"prefix":"","firstName":"Osama","middleName":"A.","lastName":"Raslan","suffix":""},{"id":490932298,"identity":"51cb1df1-e5d6-4f5c-b069-179012044c58","order_by":20,"name":"Ivo Rausch","email":"","orcid":"","institution":"QIMP Team, Medical University of Vienna, Vienna, Austria","correspondingAuthor":false,"prefix":"","firstName":"Ivo","middleName":"","lastName":"Rausch","suffix":""},{"id":490932299,"identity":"d40ff2cf-bde1-4a9d-82b0-eafb67a154e1","order_by":21,"name":"Siba P. Raychaudhuri","email":"","orcid":"","institution":"Department of Medicine and Dermatology, UC Davis School of Medicine, Sacramento, California, USA","correspondingAuthor":false,"prefix":"","firstName":"Siba","middleName":"P.","lastName":"Raychaudhuri","suffix":""},{"id":490932300,"identity":"6a9b15f6-85a8-4d75-9cc3-cb837d04b06e","order_by":22,"name":"Osama Sabri","email":"","orcid":"","institution":"Department of Nuclear Medicine, University Hospital Leipzig, Leipzig, Germany","correspondingAuthor":false,"prefix":"","firstName":"Osama","middleName":"","lastName":"Sabri","suffix":""},{"id":490932301,"identity":"8ad52117-1e3c-41a6-88f1-f0060ea16d4b","order_by":23,"name":"Fabian Schmidt","email":"","orcid":"","institution":"Division of Nuclear Medicine, University of Tübingen, Tübingen, Germany. Werner Siemens Imaging Center, Department of Preclinical Imaging and Radiopharmacy, University of Tübingen, Germany","correspondingAuthor":false,"prefix":"","firstName":"Fabian","middleName":"","lastName":"Schmidt","suffix":""},{"id":490932302,"identity":"e02c0de5-9077-4b37-8de3-e3aee9312fb9","order_by":24,"name":"Roberto Sciagrà","email":"","orcid":"","institution":"Division of Nuclear Medicine, Azienda Ospedaliero Universitaria Careggi, Florence, Italy","correspondingAuthor":false,"prefix":"","firstName":"Roberto","middleName":"","lastName":"Sciagrà","suffix":""},{"id":490932303,"identity":"d5120e28-8be6-4f0d-9f6d-25fb22db7cb9","order_by":25,"name":"Benjamin Spencer","email":"","orcid":"","institution":"Department of Radiology, University of California Davis, Sacramento, California, USA","correspondingAuthor":false,"prefix":"","firstName":"Benjamin","middleName":"","lastName":"Spencer","suffix":""},{"id":490932304,"identity":"a8889b80-356b-4377-ab36-e60a9f05d2ff","order_by":26,"name":"Guobao Wang","email":"","orcid":"","institution":"Department of Radiology, University of California Davis, Sacramento, California, USA","correspondingAuthor":false,"prefix":"","firstName":"Guobao","middleName":"","lastName":"Wang","suffix":""},{"id":490932305,"identity":"068d2bd5-5b01-4eb1-95ac-c1433ffcddcc","order_by":27,"name":"Hubert Wirtz","email":"","orcid":"","institution":"Department of Respiratory Medicine, University Hospital Leipzig, Leipzig, Germany","correspondingAuthor":false,"prefix":"","firstName":"Hubert","middleName":"","lastName":"Wirtz","suffix":""},{"id":490932306,"identity":"6a3097e0-2042-4d44-be02-a0ca83c988f3","order_by":28,"name":"Thomas Beyer","email":"","orcid":"","institution":"QIMP Team, Medical University of Vienna, Vienna, Austria","correspondingAuthor":false,"prefix":"","firstName":"Thomas","middleName":"","lastName":"Beyer","suffix":""},{"id":490932307,"identity":"2abf3381-c681-475b-85f3-62031af1c4f8","order_by":29,"name":"Lalith Kumar Shiyam Sundar","email":"data:image/png;base64,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","orcid":"","institution":"QIMP Team, Medical University of Vienna, Vienna, Austria. DIGIT-X Lab, Department of Radiology, LMU Munich, Germany","correspondingAuthor":true,"prefix":"","firstName":"Lalith","middleName":"Kumar Shiyam","lastName":"Sundar","suffix":""}],"badges":[],"createdAt":"2025-07-20 10:28:20","currentVersionCode":2,"declarations":{"humanSubjects":true,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":true,"humanSubjectConsent":true,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-7169062/v2","doiUrl":"https://doi.org/10.21203/rs.3.rs-7169062/v2","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":88365869,"identity":"f2df5103-6277-4191-926f-23c2669e93e4","added_by":"auto","created_at":"2025-08-05 17:26:00","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":70275,"visible":true,"origin":"","legend":"\u003cp\u003eComparison of clinical indications of cases included in the ENHANCE.PET 1.6k dataset and other open-source CT images and CT-derived segmentations (TotalSegmentator\u003csup\u003e26\u003c/sup\u003e, SAROS\u003csup\u003e24\u003c/sup\u003e and WORD\u003csup\u003e23\u003c/sup\u003e). The clinical indications for cases within the TotalSegmentator dataset were derived from the non-imaging parameters provided as a CSV file (\u003ca href=\"https://zenodo.org/records/8367088\"\u003ehttps://zenodo.org/records/8367088\u003c/a\u003e).\u003c/p\u003e","description":"","filename":"Figure1.png","url":"https://assets-eu.researchsquare.com/files/rs-7169062/v2/483f1a3963fc551fd17ea0ce.png"},{"id":88366079,"identity":"f5132807-690a-49cd-ad43-2caa4fd25a64","added_by":"auto","created_at":"2025-08-05 17:34:01","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":738269,"visible":true,"origin":"","legend":"\u003cp\u003eExample of liver segmentation from the “CTLiver” sample data of the 3D Slicer\u003csup\u003e40\u003c/sup\u003e software. Segmentations were performed using MOOSE\u003csup\u003e39\u003c/sup\u003e, our in-house tool for automatic CT segmentation trained with the ENHANCE.PET 1.6k dataset, as well as TotalSegmentator\u003csup\u003e26\u003c/sup\u003e. Both models accurately segmented the liver volume. However, MOOSE excluded the large liver lesion from the segmentation. In contrast, the output from TotalSegmentator included both healthy and pathological tissue. The ability of MOOSE to differentiate small non-/malignant tissue in low-contrast CT images remains to be studied.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7169062/v2/276d8b9cc24167b198550edd.png"},{"id":88366078,"identity":"de2dc2a6-98c1-4c49-961b-87279b643ad2","added_by":"auto","created_at":"2025-08-05 17:34:01","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":91342,"visible":true,"origin":"","legend":"\u003cp\u003eGeographic distributions and clinical indications of the ENHANCE.PET 1.6k dataset. Red dots on the map represent the three clinical facilities of University of Tübingen, Germany; University Hospital Leipzig, Germany; and Azienda Ospedaliero Universitaria Careggi, Italy.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7169062/v2/695b165a1ef9f32d2526000b.png"},{"id":88365872,"identity":"cf3adf02-fde8-4a2b-ac16-ad0c3a5fce6e","added_by":"auto","created_at":"2025-08-05 17:26:01","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":323126,"visible":true,"origin":"","legend":"\u003cp\u003eCoronal and sagittal views of (A) an original CT image from the AutoPET Challenge\u003csup\u003e9\u003c/sup\u003e and (B) an anonymized CT image, defaced, from the Azienda Ospedaliero Universitaria Careggi, Italy. Defacing of the PET/CT images was performed as an additional measure for complete anonymization of patients prior to data open-sourcing.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7169062/v2/d8566ca0f1d5b35de62271c4.png"},{"id":88365878,"identity":"267f6e79-fa5e-43a4-9b82-b6c9fb7d3663","added_by":"auto","created_at":"2025-08-05 17:26:01","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":2103312,"visible":true,"origin":"","legend":"\u003cp\u003eComplete list of segmented target regions per dataset. L = left; R = right.\u003c/p\u003e","description":"","filename":"Figure5.png","url":"https://assets-eu.researchsquare.com/files/rs-7169062/v2/e14ed9d5243162da5342883c.png"},{"id":88365880,"identity":"6d90494b-9178-4e34-8ad5-bfd99cdbbe59","added_by":"auto","created_at":"2025-08-05 17:26:01","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":85319,"visible":true,"origin":"","legend":"\u003cp\u003ePerformance of the CT-segmentation models on selected target regions as a function of the training dataset size. Performance was evaluated on 20% of the ENHANCE.PET 1.6k dataset (N=337), as described in the Technical Validation section. As the training dataset size increased, DICE scores for segmentation improved, primarily due to the correction of systematic segmentation errors, such as left/right misclassification in the hands and the inclusion of missing regions in the \"skeletal muscle\" segmentation. Similarly, the average symmetric surface distance (ASSD) decreased as the dataset size grew.\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-7169062/v2/701662dbfcbf53d26a3b7fad.png"},{"id":88365875,"identity":"f4819047-5734-42f5-ae1f-0465359862c7","added_by":"auto","created_at":"2025-08-05 17:26:01","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":184109,"visible":true,"origin":"","legend":"\u003cp\u003eFolder structure of the ENHANCE.PET 1.6k dataset. Each image and segmentation are provided in anonymized NIfTI format and named with ascending unique IDs from 0001.nii.gz to 1597.nii.gz. For each participant, the CT image, PET image, and segmentations of cardiac subregions and vessels, muscles, organs, peripheral bones, ribs, vertebrae, and body composition (including skeletal muscle and adipose tissue) around the L3 vertebra region are provided.\u003c/p\u003e","description":"","filename":"floatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-7169062/v2/b66169bc5500b370c45ea8ff.png"},{"id":88366081,"identity":"e00f69f3-ac96-472c-b750-1561a9cb6024","added_by":"auto","created_at":"2025-08-05 17:34:01","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":376621,"visible":true,"origin":"","legend":"\u003cp\u003eMean Dice scores and Average Symmetric Surface Distance (ASSD) per available segmentation model between reference labels of the test dataset (N=337, 20% of the total ENHANCE.PET 1.6k) and the labels resulting from MOOSE\u003csup\u003e39\u003c/sup\u003e prediction.\u003c/p\u003e","description":"","filename":"floatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-7169062/v2/e4cbebcff3ae004ff5b90dad.png"},{"id":88367044,"identity":"e8e7e9ab-5f0d-488e-bf4a-6c52c28448ff","added_by":"auto","created_at":"2025-08-05 17:50:03","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":5087537,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7169062/v2/52376a20-9f8d-4afd-ae8a-37b28c9ec24a.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"Sharing a whole-/total-body [18F]FDG-PET/CT dataset with CT-derived segmentations: an ENHANCE.PET initiative","fulltext":[{"header":"Background \u0026 Summary","content":"\u003cp\u003eIn recent years, the field of biomedical engineering and medical physics has witnessed an increase in the complexity of data\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e, driven by rapid advancements in imaging technologies. Traditional data analysis methods have become increasingly inadequate to analyse these large data sets to meet the demand for greater diagnostic precision, and the shift toward personalized treatment strategies\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e,\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e. To support personalised medicine, more efficient, automated approaches capable of processing and interpreting large-scale datasets are needed. Artificial intelligence (AI) and machine learning (ML) have emerged as powerful tools in this context, offering the ability to identify complex patterns and insights that may not be apparent through conventional methods. However, the effectiveness of AI, particularly of deep learning, is heavily dependent on the availability of large, high-quality and heterogeneous datasets, thus requiring extensive training on vast amounts of data to achieve generalizability and robustness\u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eCombined positron emission tomography (PET) and computed tomography (CT) integrates both anatomical and functional imaging capabilities, making it indispensable for diagnosing, staging, and monitoring diseases, such as oncological disorders\u003csup\u003e\u003cspan additionalcitationids=\"CR6\" citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e–\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e. Despite the clinical importance of PET/CT datasets, open sourcing of imaging data is hindered by strict regulations, and analyses are often conducted in-house\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e on limited data. At present, very few nuclear medicine datasets with annotated lesions are \u003cem class=\"Highlight ht71194251-f7a6-4c2d-a145-3d9f25b46662\" highlight=\"true\" htmatch=\"publicly available\" htloopnumber=\"264975710\" style=\"font-style: inherit;\"\u003epublicly available\u003c/em\u003e: only 1,014 PET/CT lung cancer, lymphoma, melanoma, and healthy control \u003cem class=\"Highlight ht71194251-f7a6-4c2d-a145-3d9f25b46662\" highlight=\"true\" htmatch=\"case*\" htloopnumber=\"264975710\" style=\"font-style: inherit;\"\u003ecases\u003c/em\u003e from the AutoPET challenge\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e, and another 845 head and neck cancer \u003cem class=\"Highlight ht71194251-f7a6-4c2d-a145-3d9f25b46662\" highlight=\"true\" htmatch=\"case*\" htloopnumber=\"264975710\" style=\"font-style: inherit;\"\u003ecases\u003c/em\u003e through the HECKTOR challenge\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e. In contrast, Ma et al\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e identified over one million open-source non-nuclear medicine datasets, most of which originating from radiology and not segmented, including more than 350,000 from CT scans alone. This restricts the development and \u003cem class=\"Highlight ht4fc55b9d-f515-4fa8-9e5d-6731d62f45ba\" highlight=\"true\" htmatch=\"validat*\" htloopnumber=\"264975710\" style=\"font-style: inherit;\"\u003evalidation\u003c/em\u003e of computational methods for functional imaging, such as image and tumor segmentation, volumetric analysis (e.g., for body composition assessment\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e), and radiomics.\u003c/p\u003e\u003cp\u003eRecent advancements in PET/CT technology, particularly the shift from single-organ imaging\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e to total-body PET/CT systems\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e,\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e, allow for simultaneous imaging of multiple organs, fueling multi-organ analyses\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e and the exploration of systemic metabolic abnormalities\u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e,\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e. However, the development of reliable AI methods for automated analysis of these complex datasets requires access to comprehensive open-source resources, including both images and high-quality segmentations of anatomical structures, which are critical for applications such as diagnosis, treatment planning\u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e, volumetric analysis, and \u003cem class=\"Highlight ht71194251-f7a6-4c2d-a145-3d9f25b46662\" highlight=\"true\" htmatch=\"patient\" htloopnumber=\"264975710\" style=\"font-style: inherit;\"\u003epatient\u003c/em\u003e-specific dosimetry\u003csup\u003e\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e,\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eIn the field of CT imaging alone, few open-source datasets include corresponding anatomical segmentations. Rister et al.\u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e presented a dataset of 140 abdominal, neck-to-pelvis, and whole-body CT images from patients with liver cancer, segmented into six organ regions. The WORD dataset\u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e comprises 170 abdominal CT images, primarily from prostate, cervical, or rectal cancer \u003cem class=\"Highlight ht71194251-f7a6-4c2d-a145-3d9f25b46662\" highlight=\"true\" htmatch=\"case*\" htloopnumber=\"264975710\" style=\"font-style: inherit;\"\u003ecases,\u003c/em\u003e along with the segmentations of 16 abdominal organs. These studies, however, are limited in the number of available CT images and segmented regions, and they do not extensively cover different pathologies. More recently, Koitka et al. introduced the Sparsely Annotated Region and Organ Segmentation (SAROS) dataset\u003csup\u003e\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e, which consists of 900 abdominal, thoracic, or whole-body CT images from various pathologies. This work focused on 13 semantic body regions and six body parts, including annotations for every fifth image slice. Similar scope and scale were achieved in the AbdomenCT-1k study\u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u003c/sup\u003e, which focused on the liver, kidneys, spleen, and pancreas segmentations, and the comprehensive TotalSegmentator dataset\u003csup\u003e\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e, with CT images of the abdomen, pelvis, or thorax segmented into a total of 104 regions of interest.\u003c/p\u003e\u003cp\u003eHowever, these datasets are limited in scope, often focusing on specific body regions rather than total-body imaging. It is understood that CT images alone are sufficient for many applications, such as volumetric analysis for body composition\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e or the delineation of organs at risk in radiotherapy treatment planning\u003csup\u003e\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e. In other applications, however, the functional information from PET imaging is essential as it provides complementary insights into disease mechanisms that CT alone cannot offer. For example, in pathological settings, [18F]FDG-uptake can help track disease progression by detecting systemic changes in metabolism, such as those seen in patients with infections\u003csup\u003e\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e,\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e, chronic inflammation\u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e, metabolic syndrome\u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e or cancer-associated cachexia\u003csup\u003e\u003cspan additionalcitationids=\"CR33\" citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e–\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e\u003c/sup\u003e. In studies involving healthy cohorts, longitudinal [18F]FDG PET/CT imaging allows for monitoring metabolic activity in participants and how it changes with aging or other factors\u003csup\u003e\u003cspan additionalcitationids=\"CR36\" citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e–\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e\u003c/sup\u003e. Also, a more complete understanding of normal physiological metabolism would help identify deviations that may signal early stages of disease\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e. While the aforementioned AutoPET\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e and HECKTOR\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e challenges provide large PET/CT datasets, they focus on segmentations of pathological tissues but ignore healthy anatomical regions.\u003c/p\u003e\u003cp\u003eIn the present study, we address the limited availability of open source PET/CT images with segmented tissues as part of our ENHANCE.PET\u003csup\u003e\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e\u003c/sup\u003e initiative, which aims to facilitate the sharing of open-source tools and datasets to support research within the PET community. We curated a large [18F]FDG PET/CT dataset with anatomical segmentations fully verified by \u003cem class=\"Highlight ht71194251-f7a6-4c2d-a145-3d9f25b46662\" highlight=\"true\" htmatch=\"human*\" htloopnumber=\"264975710\" style=\"font-style: inherit;\"\u003ehuman\u003c/em\u003e readers. This dataset includes 1,597 whole-body and total-body PET/CT scans, along with corresponding CT-derived segmentations of 130 non-pathological tissues per scan. The initial segmentations were generated using our in-house tool, MOOSE\u003csup\u003e\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e\u003c/sup\u003e, for automatic CT segmentation and were manually verified and corrected using 3D Slicer\u003csup\u003e\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e\u003c/sup\u003e, a software platform for image analysis. The data include contributions from the LuCaPET consortium (grant number ERAPerMed_324, “\u003cem\u003eClinical decision support for predicting cachexia in cancer patients using hybrid PET/CT imaging\u003c/em\u003e”) and from the AutoPET Challenge\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e, whose images and lesion segmentations were already available as open-source on The Cancer Imaging Archive\u003csup\u003e\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e\u003c/sup\u003e. Focused mainly on the oncological \u003cem class=\"Highlight ht71194251-f7a6-4c2d-a145-3d9f25b46662\" highlight=\"true\" htmatch=\"case*\" htloopnumber=\"264975710\" style=\"font-style: inherit;\"\u003ecases\u003c/em\u003e of lung cancer, melanoma, and lymphoma (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e), the ENHANCE.PET 1.6k dataset also includes participants without known disease. The dataset is provided in anonymized NIfTI format to ensure \u003cem class=\"Highlight ht71194251-f7a6-4c2d-a145-3d9f25b46662\" highlight=\"true\" htmatch=\"patient\" htloopnumber=\"264975710\" style=\"font-style: inherit;\"\u003epatient\u003c/em\u003e privacy, along with demographic details and CT and PET acquisition parameters as non-imaging metadata.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eCompared to other \u003cem class=\"Highlight ht71194251-f7a6-4c2d-a145-3d9f25b46662\" highlight=\"true\" htmatch=\"publicly available\" htloopnumber=\"264975710\" style=\"font-style: inherit;\"\u003epublicly available\u003c/em\u003e datasets, ENHANCE.PET 1.6k uniquely focuses on the segmentations of organ volumes while avoiding pathological tissues (e.g., tumors, Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). This dataset is particularly suited for training models aimed at automatic image segmentation and identification of healthy tissues. We believe that the open availability of this dataset will advance the differential understanding of healthy and pathological tissues in computational medicine. This comprehensive resource is now available to facilitate future research and advanced data analysis in whole-body PET/CT imaging. We anticipate that applications such as developing and \u003cem class=\"Highlight ht4fc55b9d-f515-4fa8-9e5d-6731d62f45ba\" highlight=\"true\" htmatch=\"validat*\" htloopnumber=\"264975710\" style=\"font-style: inherit;\"\u003evalidating\u003c/em\u003e deep learning algorithms for automated data analysis and studies on disease-related systemic abnormalities will greatly benefit from this high-quality data collection.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003e\u003cb\u003eData collection\u003c/b\u003e\u003c/p\u003e\u003cp\u003e The ENHANCE.PET 1.6k dataset was acquired in accordance with the guidelines set forth in the \u003cem class=\"Highlight ht29216696-c42e-4f00-932a-aea34347df6a\" highlight=\"true\" htmatch=\"declaration of helsinki\" htloopnumber=\"264975710\" style=\"font-style: inherit;\"\u003eDeclaration of Helsinki\u003c/em\u003e. Images were acquired between 1999 and 2022 from various institutions and studies, summarized in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e: the open-source dataset AutoPET\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e, the University Hospital Leipzig in Germany (\u003cem class=\"Highlight htf340ff0d-a602-4893-ae8d-b60ea075e112\" highlight=\"true\" htmatch=\"irb\" htloopnumber=\"264975710\" style=\"font-style: inherit;\"\u003eIRB\u003c/em\u003e: 259/18-ek) and the Azienda Ospedaliero Universitaria Careggi in Italy (\u003cem class=\"Highlight htf340ff0d-a602-4893-ae8d-b60ea075e112\" highlight=\"true\" htmatch=\"irb\" htloopnumber=\"264975710\" style=\"font-style: inherit;\"\u003eIRB\u003c/em\u003e: 21306_oss) as part of the LuCaPET consortium.\u003c/p\u003e\u003cp\u003eParticipant demographics across the different clinical conditions are summarized in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e\u003cdiv class=\"gridtable\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\"±\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\"±\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\"±\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eDemographics and clinical details of the participants included in the study.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"7\"\u003e\u003c/colgroup\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePartner University\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eClinical Condition\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003e#\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eSex\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eAge [years]\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c6\"\u003e\u003cp\u003eWeight [kg]\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c7\"\u003e\u003cp\u003eHeight [cm]\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAzienda Ospedaliero Universitaria Careggi, IT\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLung Cancer\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e199\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e72F / 127M\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"±\" colname=\"c5\"\u003e\u003cp\u003e71 ± 10\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"±\" colname=\"c6\"\u003e\u003cp\u003e72 ± 15\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"±\" colname=\"c7\"\u003e\u003cp\u003e168 ± 9\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eUniversity Hospital Leipzig, DE\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLung Cancer\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e384\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e110F / 274M\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"±\" colname=\"c5\"\u003e\u003cp\u003e65 ± 11\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"±\" colname=\"c6\"\u003e\u003cp\u003e76 ± 16\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"±\" colname=\"c7\"\u003e\u003cp\u003e172 ± 9\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eOpen-source dataset AutoPET\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eLung Cancer\u003c/p\u003e\u003cp\u003eLymphoma\u003c/p\u003e\u003cp\u003eMelanoma\u003c/p\u003e\u003cp\u003eNegative Findings\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e168\u003c/p\u003e\u003cp\u003e145\u003c/p\u003e\u003cp\u003e188\u003c/p\u003e\u003cp\u003e513\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e65F / 103M\u003c/p\u003e\u003cp\u003e69F / 76M\u003c/p\u003e\u003cp\u003e77F / 111M\u003c/p\u003e\u003cp\u003e233F / 280M\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"±\" colname=\"c5\"\u003e\u003cp\u003e60 ± 15\u003c/p\u003e\u003cp\u003e57 ± 18\u003c/p\u003e\u003cp\u003e60 ± 16\u003c/p\u003e\u003cp\u003e60 ± 16\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"±\" colname=\"c6\"\u003e\u003cp\u003e80 ± 18\u003c/p\u003e\u003cp\u003e79 ± 19\u003c/p\u003e\u003cp\u003e79 ± 20\u003c/p\u003e\u003cp\u003e79 ± 18\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"±\" colname=\"c7\"\u003e\u003cp\u003e173 ± 10\u003c/p\u003e\u003cp\u003e173 ± 11\u003c/p\u003e\u003cp\u003e172 ± 9\u003c/p\u003e\u003cp\u003e171 ± 11\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/table\u003e\u003c/div\u003e\u003cp\u003e\u003cb\u003eImaging Protocols\u003c/b\u003e\u003c/p\u003e\u003cp\u003eThree different PET/CT systems were used for image acquisition at the participating medical centres: Siemens Biograph mCT (N = 1398), Philips Gemini TF (N = 179), and GE Healthcare Discovery MI (N = 20). At all three sites, diagnostic CT scans were acquired with X-ray tube voltages between 100 kVp and 140 kVp, and CT data were reconstructed with a slice thickness between 1 mm and 5 mm. Details on the CT reconstruction parameters are provided in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e\u003cdiv class=\"gridtable\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eSummary of CT systems and reconstruction parameters of the ENHANCE.PET 1.6k dataset.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003c/colgroup\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePartner University\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eAzienda Ospedaliero Universitaria Careggi, IT\u003c/p\u003e\u003cp\u003eN = 199\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eUniversity Hospital Leipzig, DE\u003c/p\u003e\u003cp\u003eN = 384\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eOpen-source dataset AutoPET\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e\u003cp\u003eN = 1014\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePET/CT System Manufacturer\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eGE Medical System (19)\u003c/p\u003e\u003cp\u003ePhilips (180)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eSiemens\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eSiemens\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSystem Model\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eDiscovery MI (19)\u003c/p\u003e\u003cp\u003eGemini TF TOF 16 (180)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eBiograph 16 (207)\u003c/p\u003e\u003cp\u003eSensation 16 (177)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eBiograph 128 (900)\u003c/p\u003e\u003cp\u003eSOMATOM\u003c/p\u003e\u003cp\u003e Definition (114)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ekVp\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e120 (196)\u003c/p\u003e\u003cp\u003e140 (3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e120\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e100 (2)\u003c/p\u003e\u003cp\u003e120 (938)\u003c/p\u003e\u003cp\u003e140 (74)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFilter Type\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eN/A\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eNONE\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eFLAT\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eConvolutional Kernel\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eN/A\u003c/p\u003e\u003cp\u003eSTANDARD (19)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003eB10f (325)\u003c/p\u003e\u003cp\u003eB31f (56)\u003c/p\u003e\u003cp\u003eB40f (3)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003eI30f (31)\u003c/p\u003e\u003cp\u003eI31f (471)\u003c/p\u003e\u003cp\u003eB30f (24)\u003c/p\u003e\u003cp\u003eB31f (488)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAxial Pixel Size (mm)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.98–1.37\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.98\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.69–0.98\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSlice Thickness (mm)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e3.75 (19)\u003c/p\u003e\u003cp\u003e5 (180)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e2 (36)\u003c/p\u003e\u003cp\u003e3 (347)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1 (52)\u003c/p\u003e\u003cp\u003e2 (141)\u003c/p\u003e\u003cp\u003e3 (821)\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFocal Spot Size (mm)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eN/A\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.7 (330)\u003c/p\u003e\u003cp\u003e1.2 (54)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e1.2\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/table\u003e\u003c/div\u003e\u003cp\u003eParticipants were asked to fast for 6 hours before the examinations and were scanned in the supine position, with arms up. Each subject underwent a static PET acquisition following an intravenous injection of [18F]FDG (314 ± 48 MBq). Uptake times varied across the three sites, with an average of (68 ± 21) minutes post-injection. PET images were reconstructed with attenuation and scatter corrections applied using the corresponding CT data.\u003c/p\u003e\u003cp\u003eDetails on the CT and PET acquisition parameters are reported for each participant as non-imaging parameters in the available spreadsheet files. The download link is provided in the Data Records section.\u003c/p\u003e\u003cp\u003e\u003cb\u003eSegmentations and Data Processing\u003c/b\u003e\u003c/p\u003e\u003cp\u003ePET/CT images were retrieved in anonymized DICOM format from the participants and centralized at the Medical University of Vienna. The metadata were used to extract relevant information about the CT and PET acquisition protocols as well as essential demographic details of the participants. For subsequent analysis and segmentation, all data were converted to NIfTI format using the \u003cem\u003edcm2niix\u003c/em\u003e DICOM to NIfTI converter\u003csup\u003e\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e\u003c/sup\u003e. To ensure that participants could not be visually identified from their CT images\u003csup\u003e\u003cspan additionalcitationids=\"CR44\" citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e–\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e\u003c/sup\u003e, both the PET and CT images from the Azienda Ospedaliero Universitaria Careggi and the University Hospital Leipzig were edited: in the PET images, voxels between the upper part of the skull segmentation and the bottom of the brain, within a cylinder of 16 voxels in the z-direction, were set to zero (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eB). Similarly, the corresponding CT region was set to -1000 Hounsfield Units to simulate air. The images from the AutoPET challenge were left unedited as per their version available online.\u003c/p\u003e\u003cp\u003eTo maximize efficiency and accelerate the workflow, the processing of the entire ENHANCE.PET 1.6k dataset was done serially: automatic segmentation of the CT images, manual refinement of the derived labels, and retraining of the original segmentation models, according to the following scheme. Our in-house developed software, MOOSE\u003csup\u003e\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e\u003c/sup\u003e, was first used for the automatic segmentation of 384 lung cancer images from the University Hospital Leipzig. The resulting segmentations were manually refined by 10 medical students using the 3D Slicer image analysis software\u003csup\u003e\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e\u003c/sup\u003e. For each dataset, a student was randomly assigned to verify and correct the segmentations, addressing possible systematic errors such as inaccuracies at anatomical borders of target regions, mislabelling between left and right regions, or misclassification of regions with similar intensities on CT. A second student was then tasked with reviewing the first student's work and correcting any remaining mistakes. Once both students agreed on the final version of the dataset, it was reviewed by a radiology resident and a nuclear medicine resident. The PET images overlapped with the corresponding CT were used to exclude tumor volumes from the segmentation of various abdominal organs.\u003c/p\u003e\u003cp\u003eThese preliminary segmentations were divided into seven different classes, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e: organs, cardiac, muscles, ribs, peripheral bones, vertebrae, and body composition around the L3 vertebra area. These masks were then used to retrain our models using nnU-Net\u003csup\u003e\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e\u003c/sup\u003e. Details on the retraining process are provided in the Technical \u003cem class=\"Highlight ht4fc55b9d-f515-4fa8-9e5d-6731d62f45ba\" highlight=\"true\" htmatch=\"validat*\" htloopnumber=\"264975710\" style=\"font-style: inherit;\"\u003eValidation\u003c/em\u003e section.\u003c/p\u003e\u003cp\u003eThe newly trained models were subsequently used for the segmentation of the second dataset, originating from Azienda Ospedaliero Universitaria Careggi, and again underwent manual refinement and quality control described above. The same workflow of automatic segmentation, manual refinement, and model retraining was then applied to the remaining images from the AutoPET\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e open-source dataset. In the \u003cem class=\"Highlight ht71194251-f7a6-4c2d-a145-3d9f25b46662\" highlight=\"true\" htmatch=\"case*\" htloopnumber=\"264975710\" style=\"font-style: inherit;\"\u003ecase\u003c/em\u003e of the AutoPET data, since the corresponding lesion segmentations were available online, they were taken as ground truth and directly subtracted from the organ segmentations without the need for manual correction.\u003c/p\u003e\u003cp\u003eTo maximize data variability and ensure robust performance across different scanning conditions, we included in our training dataset 86 additional total-body CT images (30F/56M, 53 ± 15 years, 87 ± 19 kg, 172 ± 9 cm) from the University of California Davis, California (USA), acquired on a United Imaging uEXPLORER PET/CT system. The segmentations were generated using the same workflow as for the other datasets. Their inclusion helped account for different acquisition protocols (arms-down positioning) and provided broader representation of physiological variations, as they covered various inflammatory and pathological conditions, such as head and neck cancer (N = 5), arthritis (N = 43), genito-urinary cancer (N = 9), and healthy controls (N = 29). While these \u003cem class=\"Highlight ht71194251-f7a6-4c2d-a145-3d9f25b46662\" highlight=\"true\" htmatch=\"case*\" htloopnumber=\"264975710\" style=\"font-style: inherit;\"\u003ecases\u003c/em\u003e cannot currently be shared due to privacy restrictions, we are working to make them \u003cem class=\"Highlight ht71194251-f7a6-4c2d-a145-3d9f25b46662\" highlight=\"true\" htmatch=\"publicly available\" htloopnumber=\"264975710\" style=\"font-style: inherit;\"\u003epublicly available\u003c/em\u003e in the future.\u003c/p\u003e\u003cp\u003eAt each stage of retraining, the size of the training data increased, improving model performance. This iterative process allowed for efficient verification and faster corrections by the medical students without compromising precision, especially in regions where systematic errors had been identified and were hindering the manual correction process. Segmentation of the fingers and hand bones showed the most significant improvement as the training dataset grew (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e). In the first round of training, several instances of left-right misclassification were identified, especially when the hands were crossed over the abdomen or above the head. However, this issue progressively improved with each retraining step. Another improvement achieved through more extensive training data was the automatic inclusion of the quadratus lumborum muscle in the \"skeletal muscle\" label for body composition, which had previously been missing and required manual correction (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e).\u003c/p\u003e\u003cp\u003e\u003cb\u003eData Records\u003c/b\u003e\u003c/p\u003e\u003cp\u003eThe PET/CT images and corresponding segmentations from Azienda Ospedaliero Universitaria Careggi, University Hospital Leipzig, and the AutoPET\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e dataset are stored on Amazon Web Services (AWS, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://aws.amazon.com/\u003c/span\u003e\u003cspan address=\"https://aws.amazon.com/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) and can be downloaded either directly following MOOSE\u003csup\u003e\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e\u003c/sup\u003e installation via command line, as described in the Code Availability section, or at the following link:\u003c/p\u003e\u003cp\u003e\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://enhance-pet.s3.eu-central-1.amazonaws.com/enhance-pet-1_6k/enhance-pet-1_6k.zip\u003c/span\u003e\u003cspan address=\"https://enhance-pet.s3.eu-central-1.amazonaws.com/enhance-pet-1_6k/enhance-pet-1_6k.zip\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp\u003eThe imaging data are stored in three separate folders containing the CT images, PET images, and ground truth segmentations, respectively. Within the segmentations folder, there are seven subfolders corresponding to different segmentation classes: \"Body-Composition,\" \"Cardiac,\" \"Muscles,\" \"Organs,\" \"Peripheral Bones,\" \"Ribs,\" and \"Vertebrae.\" Each folder contains NIfTI data files, named sequentially from 0001.nii.gz to 1597.nii.gz. The directory structure of the ENHANCE.PET 1.6k dataset is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e.\u003c/p\u003e\u003cp\u003eA JSON file containing the complete list of segmentations and their corresponding intensities within the multi-class files is available for download at the following link:\u003c/p\u003e\u003cp\u003e\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://enhance-pet.s3.eu-central-1.amazonaws.com/enhance-pet-1_6k/labels.json\u003c/span\u003e\u003cspan address=\"https://enhance-pet.s3.eu-central-1.amazonaws.com/enhance-pet-1_6k/labels.json\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\u003cp\u003eIn addition to the imaging data, non-imaging information is provided in two spreadsheet files. The \u003cem\u003eCT-details.xlsx\u003c/em\u003e file contains details on the CT acquisition parameters (e.g., PET/CT system manufacturer and model, kVp, filter type, convolutional kernel, axial pixel size, slice thickness, and focal spot size) for each participant. The \u003cem\u003ePT-details.xlsx\u003c/em\u003e file provides the corresponding demographic information (e.g., clinical indication, sex, age, weight, height) as well as PET acquisition parameters (e.g., injected activity, acquisition date and time, radioactivity injection details, image units, slope, intercept, system model and manufacturer).\u003c/p\u003e\u003cp\u003e\u003cb\u003eTechnical \u003cem class=\"Highlight ht4fc55b9d-f515-4fa8-9e5d-6731d62f45ba\" highlight=\"true\" htmatch=\"validat*\" htloopnumber=\"264975710\" style=\"font-style: inherit;\"\u003eValidation\u003c/em\u003e\u003c/b\u003e\u003c/p\u003e\u003cp\u003eWe used the ENHANCE.PET 1.6k dataset with the additional 86-image contribution from the University of California, Davis to develop a deep learning-based method for the automatic segmentation of CT scans: 80% of the images and corresponding labels were sampled from the total with a stratified sampling, thus ensuring that the original proportion of data by medical facility and clinical condition remained unchanged. The 1346 selected imaging data served as a training dataset for multiple segmentation models targeting different anatomical regions, including bones of the limbs and skull, thoracic cage bones, vertebrae and sacrum, major abdominal organs, lower back muscles, cardiac tissues, and body composition around the L3 vertebra. A detailed list of regions segmented by each model is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e.\u003c/p\u003e\u003cp\u003ePrior to training, all images and labels were resampled with SimpleITK (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://simpleitk.org/about.html\u003c/span\u003e\u003cspan address=\"https://simpleitk.org/about.html\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) from the original resolution to a voxel spacing of 1.5 x 1.5 x 1.5 mm using B-spline interpolation. This provided a resolution high enough to segment fine structures on the CT images, while considering the computational burden of training. The models were trained using nnU-Net\u003csup\u003e\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e\u003c/sup\u003e, a state-of-the-art self-configuring framework based on the U-Net architecture for semantic segmentation. The training process was performed over 2000 epochs.\u003c/p\u003e\u003cp\u003eTo assess the model performance, we tested all segmentation models on the remaining 20% of the ENHANCE.PET 1.6k. Segmentation accuracy was evaluated using the Dice Similarity Coefficient (DSC) to quantify the overlap between predicted segmentations and the reference labels and with the Average Symmetric Surface Distance (ASSD)\u003csup\u003e\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e\u003c/sup\u003e to estimate the average distance between surface voxels of the reference labels and the automated segmentation. The averaged results for the generated models are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e, and the metrics for each label are reported in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e4\u003c/span\u003e.\u003c/p\u003e\u003cdiv class=\"gridtable\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eMean Dices and Average Symmetric Surface Distance (ASSD) per segmented volume between the reference labels of the test dataset (N = 337, 20% of the total ENHANCE.PET 1.6k) and the labels resulting from the AI prediction. Left and right regions were merged for this analysis.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003c/colgroup\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eModel\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRegions\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eMean Dice ± St Dev\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eMean ASSD ± St Dev\u003c/p\u003e\u003cp\u003e[mm]\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eOrgans\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eAdrenal Glands\u003c/p\u003e\u003cp\u003eBladder\u003c/p\u003e\u003cp\u003eBrain\u003c/p\u003e\u003cp\u003eGallbladder\u003c/p\u003e\u003cp\u003eKidney\u003c/p\u003e\u003cp\u003eLiver\u003c/p\u003e\u003cp\u003eLung Lower Lobe\u003c/p\u003e\u003cp\u003eLung Middle Lobe\u003c/p\u003e\u003cp\u003eLung Upper Lobe\u003c/p\u003e\u003cp\u003ePancreas\u003c/p\u003e\u003cp\u003eSpleen\u003c/p\u003e\u003cp\u003eStomach\u003c/p\u003e\u003cp\u003eThyroid\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.82 ± 0.12\u003c/p\u003e\u003cp\u003e0.90 ± 0.18\u003c/p\u003e\u003cp\u003e0.96 ± 0.13\u003c/p\u003e\u003cp\u003e0.88 ± 0.15\u003c/p\u003e\u003cp\u003e0.96 ± 0.05\u003c/p\u003e\u003cp\u003e0.98 ± 0.03\u003c/p\u003e\u003cp\u003e0.95 ± 0.12\u003c/p\u003e\u003cp\u003e0.94 ± 0.09\u003c/p\u003e\u003cp\u003e0.97 ± 0.05\u003c/p\u003e\u003cp\u003e0.90 ± 0.08\u003c/p\u003e\u003cp\u003e0.96 ± 0.08\u003c/p\u003e\u003cp\u003e0.95 ± 0.06\u003c/p\u003e\u003cp\u003e0.87 ± 0.19\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.5 ± 0.5\u003c/p\u003e\u003cp\u003e1.0 ± 2.0\u003c/p\u003e\u003cp\u003e0.6 ± 0.4\u003c/p\u003e\u003cp\u003e0.8 ± 1.7\u003c/p\u003e\u003cp\u003e0.5 ± 0.7\u003c/p\u003e\u003cp\u003e0.6 ± 0.5\u003c/p\u003e\u003cp\u003e0.6 ± 1.4\u003c/p\u003e\u003cp\u003e0.8 ± 1.8\u003c/p\u003e\u003cp\u003e0.9 ± 0.6\u003c/p\u003e\u003cp\u003e0.7 ± 0.7\u003c/p\u003e\u003cp\u003e0.7 ± 0.8\u003c/p\u003e\u003cp\u003e0.7 ± 1.0\u003c/p\u003e\u003cp\u003e0.6 ± 0.8\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eCardiac\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eMyocardium\u003c/p\u003e\u003cp\u003eAtrium\u003c/p\u003e\u003cp\u003eVentricle\u003c/p\u003e\u003cp\u003eAorta\u003c/p\u003e\u003cp\u003eIliac Artery\u003c/p\u003e\u003cp\u003eIliac Vena\u003c/p\u003e\u003cp\u003eInferior Vena Cava\u003c/p\u003e\u003cp\u003ePortal Splenic Vein\u003c/p\u003e\u003cp\u003ePulmonary Artery\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.92 ± 0.05\u003c/p\u003e\u003cp\u003e0.96 ± 0.03\u003c/p\u003e\u003cp\u003e0.96 ± 0.03\u003c/p\u003e\u003cp\u003e0.95 ± 0.02\u003c/p\u003e\u003cp\u003e0.90 ± 0.09\u003c/p\u003e\u003cp\u003e0.92 ± 0.07\u003c/p\u003e\u003cp\u003e0.92 ± 0.06\u003c/p\u003e\u003cp\u003e0.82 ± 0.18\u003c/p\u003e\u003cp\u003e0.94 ± 0.05\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.5 ± 0.4\u003c/p\u003e\u003cp\u003e0.5 ± 0.4\u003c/p\u003e\u003cp\u003e0.5 ± 0.4\u003c/p\u003e\u003cp\u003e0.5 ± 0.2\u003c/p\u003e\u003cp\u003e0.6 ± 1.9\u003c/p\u003e\u003cp\u003e0.8 ± 1.8\u003c/p\u003e\u003cp\u003e0.6 ± 0.5\u003c/p\u003e\u003cp\u003e0.9 ± 1.8\u003c/p\u003e\u003cp\u003e0.8 ± 0.6\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMuscles\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eAutochthon\u003c/p\u003e\u003cp\u003eGluteus Maximus\u003c/p\u003e\u003cp\u003eGluteus Medius\u003c/p\u003e\u003cp\u003eGluteus Minimus\u003c/p\u003e\u003cp\u003eIliopsoas\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.98 ± 0.01\u003c/p\u003e\u003cp\u003e0.98 ± 0.01\u003c/p\u003e\u003cp\u003e0.98 ± 0.01\u003c/p\u003e\u003cp\u003e0.96 ± 0.03\u003c/p\u003e\u003cp\u003e0.97 ± 0.02\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.3 ± 0.2\u003c/p\u003e\u003cp\u003e0.3 ± 0.2\u003c/p\u003e\u003cp\u003e0.3 ± 0.2\u003c/p\u003e\u003cp\u003e0.3 ± 0.2\u003c/p\u003e\u003cp\u003e0.3 ± 0.2\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRibs\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRib 1\u003c/p\u003e\u003cp\u003eRib 2\u003c/p\u003e\u003cp\u003eRib 3\u003c/p\u003e\u003cp\u003eRib 4\u003c/p\u003e\u003cp\u003eRib 5\u003c/p\u003e\u003cp\u003eRib 6\u003c/p\u003e\u003cp\u003eRib 7\u003c/p\u003e\u003cp\u003eRib 8\u003c/p\u003e\u003cp\u003eRib 9\u003c/p\u003e\u003cp\u003eRib 10\u003c/p\u003e\u003cp\u003eRib 11\u003c/p\u003e\u003cp\u003eRib 12\u003c/p\u003e\u003cp\u003eSternum\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.86 ± 0.12\u003c/p\u003e\u003cp\u003e0.88 ± 0.12\u003c/p\u003e\u003cp\u003e0.88 ± 0.11\u003c/p\u003e\u003cp\u003e0.90 ± 0.10\u003c/p\u003e\u003cp\u003e0.90 ± 0.09\u003c/p\u003e\u003cp\u003e0.91 ± 0.09\u003c/p\u003e\u003cp\u003e0.91 ± 0.09\u003c/p\u003e\u003cp\u003e0.91 ± 0.09\u003c/p\u003e\u003cp\u003e0.90 ± 0.10\u003c/p\u003e\u003cp\u003e0.90 ± 0.11\u003c/p\u003e\u003cp\u003e0.89 ± 0.14\u003c/p\u003e\u003cp\u003e0.86 ± 0.18\u003c/p\u003e\u003cp\u003e0.94 ± 0.03\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.5 ± 0.6\u003c/p\u003e\u003cp\u003e0.4 ± 0.8\u003c/p\u003e\u003cp\u003e0.4 ± 0.6\u003c/p\u003e\u003cp\u003e0.4 ± 0.9\u003c/p\u003e\u003cp\u003e0.5 ± 2.0\u003c/p\u003e\u003cp\u003e0.5 ± 1.8\u003c/p\u003e\u003cp\u003e0.6 ± 2.4\u003c/p\u003e\u003cp\u003e0.6 ± 2.3\u003c/p\u003e\u003cp\u003e0.7 ± 1.9\u003c/p\u003e\u003cp\u003e0.8 ± 2.7\u003c/p\u003e\u003cp\u003e0.9 ± 3.2\u003c/p\u003e\u003cp\u003e0.9 ± 3.7\u003c/p\u003e\u003cp\u003e0.5 ± 1.1\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eVertebrae\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eVertebra C1\u003c/p\u003e\u003cp\u003eVertebra C2\u003c/p\u003e\u003cp\u003eVertebra C3\u003c/p\u003e\u003cp\u003eVertebra C4\u003c/p\u003e\u003cp\u003eVertebra C5\u003c/p\u003e\u003cp\u003eVertebra C6\u003c/p\u003e\u003cp\u003eVertebra C7\u003c/p\u003e\u003cp\u003eVertebra T1\u003c/p\u003e\u003cp\u003eVertebra T2\u003c/p\u003e\u003cp\u003eVertebra T3\u003c/p\u003e\u003cp\u003eVertebra T4\u003c/p\u003e\u003cp\u003eVertebra T5\u003c/p\u003e\u003cp\u003eVertebra T6\u003c/p\u003e\u003cp\u003eVertebra T7\u003c/p\u003e\u003cp\u003eVertebra T8\u003c/p\u003e\u003cp\u003eVertebra T9\u003c/p\u003e\u003cp\u003eVertebra T10\u003c/p\u003e\u003cp\u003eVertebra T11\u003c/p\u003e\u003cp\u003eVertebra T12\u003c/p\u003e\u003cp\u003eVertebra L1\u003c/p\u003e\u003cp\u003eVertebra L2\u003c/p\u003e\u003cp\u003eVertebra L3\u003c/p\u003e\u003cp\u003eVertebra L4\u003c/p\u003e\u003cp\u003eVertebra L5\u003c/p\u003e\u003cp\u003eHip\u003c/p\u003e\u003cp\u003eSacrum\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.89 ± 0.10\u003c/p\u003e\u003cp\u003e0.92 ± 0.09\u003c/p\u003e\u003cp\u003e0.91 ± 0.09\u003c/p\u003e\u003cp\u003e0.90 ± 0.09\u003c/p\u003e\u003cp\u003e0.90 ± 0.11\u003c/p\u003e\u003cp\u003e0.90 ± 0.09\u003c/p\u003e\u003cp\u003e0.91 ± 0.09\u003c/p\u003e\u003cp\u003e0.93 ± 0.09\u003c/p\u003e\u003cp\u003e0.93 ± 0.10\u003c/p\u003e\u003cp\u003e0.92 ± 0.10\u003c/p\u003e\u003cp\u003e0.93 ± 0.09\u003c/p\u003e\u003cp\u003e0.93 ± 0.07\u003c/p\u003e\u003cp\u003e0.93 ± 0.08\u003c/p\u003e\u003cp\u003e0.93 ± 0.09\u003c/p\u003e\u003cp\u003e0.94 ± 0.08\u003c/p\u003e\u003cp\u003e0.94 ± 0.08\u003c/p\u003e\u003cp\u003e0.94 ± 0.10\u003c/p\u003e\u003cp\u003e0.94 ± 0.11\u003c/p\u003e\u003cp\u003e0.94 ± 0.12\u003c/p\u003e\u003cp\u003e0.93 ± 0.13\u003c/p\u003e\u003cp\u003e0.93 ± 0.14\u003c/p\u003e\u003cp\u003e0.93 ± 0.15\u003c/p\u003e\u003cp\u003e0.93 ± 0.16\u003c/p\u003e\u003cp\u003e0.90 ± 0.20\u003c/p\u003e\u003cp\u003e0.97 ± 0.08\u003c/p\u003e\u003cp\u003e0.96 ± 0.08\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.5 ± 0.7\u003c/p\u003e\u003cp\u003e0.4 ± 0.4\u003c/p\u003e\u003cp\u003e0.4 ± 0.4\u003c/p\u003e\u003cp\u003e0.4 ± 0.5\u003c/p\u003e\u003cp\u003e0.4 ± 0.4\u003c/p\u003e\u003cp\u003e0.4 ± 0.4\u003c/p\u003e\u003cp\u003e0.4 ± 0.6\u003c/p\u003e\u003cp\u003e0.4 ± 0.7\u003c/p\u003e\u003cp\u003e0.4 ± 0.9\u003c/p\u003e\u003cp\u003e0.4 ± 0.9\u003c/p\u003e\u003cp\u003e0.4 ± 0.9\u003c/p\u003e\u003cp\u003e0.4 ± 0.5\u003c/p\u003e\u003cp\u003e0.4 ± 0.6\u003c/p\u003e\u003cp\u003e0.4 ± 1.0\u003c/p\u003e\u003cp\u003e0.4 ± 0.9\u003c/p\u003e\u003cp\u003e0.5 ± 1.0\u003c/p\u003e\u003cp\u003e0.5 ± 1.2\u003c/p\u003e\u003cp\u003e0.5 ± 1.4\u003c/p\u003e\u003cp\u003e0.6 ± 1.7\u003c/p\u003e\u003cp\u003e0.7 ± 2.2\u003c/p\u003e\u003cp\u003e0.7 ± 2.4\u003c/p\u003e\u003cp\u003e0.7 ± 2.6\u003c/p\u003e\u003cp\u003e0.8 ± 2.6\u003c/p\u003e\u003cp\u003e0.6 ± 2.0\u003c/p\u003e\u003cp\u003e0.3 ± 0.2\u003c/p\u003e\u003cp\u003e0.3 ± 0.2\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003ePeripheral Bones\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eCarpal\u003c/p\u003e\u003cp\u003eClavicle\u003c/p\u003e\u003cp\u003eFemur\u003c/p\u003e\u003cp\u003eFibula\u003c/p\u003e\u003cp\u003eFingers\u003c/p\u003e\u003cp\u003eHumerus\u003c/p\u003e\u003cp\u003eMetacarpal\u003c/p\u003e\u003cp\u003eMetatarsal\u003c/p\u003e\u003cp\u003ePatella\u003c/p\u003e\u003cp\u003eRadius\u003c/p\u003e\u003cp\u003eScapula\u003c/p\u003e\u003cp\u003eSkull\u003c/p\u003e\u003cp\u003eTarsal\u003c/p\u003e\u003cp\u003eTibia\u003c/p\u003e\u003cp\u003eToes\u003c/p\u003e\u003cp\u003eUlna\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.76 ± 0.33\u003c/p\u003e\u003cp\u003e0.91 ± 0.16\u003c/p\u003e\u003cp\u003e0.96 ± 0.15\u003c/p\u003e\u003cp\u003e0.93 ± 0.19\u003c/p\u003e\u003cp\u003e0.55 ± 0.39\u003c/p\u003e\u003cp\u003e0.94 ± 0.15\u003c/p\u003e\u003cp\u003e0.71 ± 0.33\u003c/p\u003e\u003cp\u003e0.92 ± 0.17\u003c/p\u003e\u003cp\u003e0.92 ± 0.19\u003c/p\u003e\u003cp\u003e0.66 ± 0.42\u003c/p\u003e\u003cp\u003e0.91 ± 0.16\u003c/p\u003e\u003cp\u003e0.90 ± 0.16\u003c/p\u003e\u003cp\u003e0.95 ± 0.17\u003c/p\u003e\u003cp\u003e0.95 ± 0.16\u003c/p\u003e\u003cp\u003e0.90 ± 0.15\u003c/p\u003e\u003cp\u003e0.68 ± 0.40\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.8 ± 0.6\u003c/p\u003e\u003cp\u003e0.5 ± 2.1\u003c/p\u003e\u003cp\u003e0.3 ± 0.6\u003c/p\u003e\u003cp\u003e0.3 ± 0.5\u003c/p\u003e\u003cp\u003e10.0 ± 37.5\u003c/p\u003e\u003cp\u003e0.4 ± 1.0\u003c/p\u003e\u003cp\u003e5.2 ± 22.7\u003c/p\u003e\u003cp\u003e0.8 ± 4.1\u003c/p\u003e\u003cp\u003e0.5 ± 1.0\u003c/p\u003e\u003cp\u003e0.5 ± 0.8\u003c/p\u003e\u003cp\u003e0.3 ± 0.4\u003c/p\u003e\u003cp\u003e0.5 ± 0.6\u003c/p\u003e\u003cp\u003e1.4 ± 4.5\u003c/p\u003e\u003cp\u003e0.3 ± 0.3\u003c/p\u003e\u003cp\u003e0.6 ± 1.3\u003c/p\u003e\u003cp\u003e0.5 ± 0.6\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eBody Composition\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eSkeletal Muscle\u003c/p\u003e\u003cp\u003eSubcutaneous Fat\u003c/p\u003e\u003cp\u003eVisceral Fat\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.85 ± 0.08\u003c/p\u003e\u003cp\u003e0.87 ± 0.08\u003c/p\u003e\u003cp\u003e0.87 ± 0.09\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e2.0 ± 1.5\u003c/p\u003e\u003cp\u003e2.4 ± 1.8\u003c/p\u003e\u003cp\u003e1.5 ± 1.6\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/table\u003e\u003c/div\u003e\u003cp\u003eAll models achieved high accuracy, with mean DSC values exceeding 0.85 across most regions and mean ASSD values below 3 mm in all regions. The \"Muscles\" model achieved the highest overlap and the lowest prediction error, with an average DSC of 0.97 ± 0.02 and an ASSD of 0.3 ± 0.2 mm (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003e). Cardiac, organ, and vertebrae models also achieved high average DSC values, exceeding 0.90. The peripheral bones model had the lowest performance, with an average DSC of 0.86 ± 0.27 and the highest variation in ASSD, at 0.8 ± 7.2 mm. The lower performance in these regions is likely due to their small size and thin anatomical structures: the digits of the hand had the most significant negative impact on model performance, with some \u003cem class=\"Highlight ht71194251-f7a6-4c2d-a145-3d9f25b46662\" highlight=\"true\" htmatch=\"case*\" htloopnumber=\"264975710\" style=\"font-style: inherit;\"\u003ecases\u003c/em\u003e of left/right misclassification identified (especially when patients underwent imaging with their hands crossed over the abdomen), resulting in an average DSC of 0.55 ± 0.39 and an ASSD of 10 ± 37 mm (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e4\u003c/span\u003e). Similarly, the segmentation of the metacarpals yielded a DSC of 0.71 ± 0.33 and an ASSD of 5 ± 23 mm. Other regions with lower overlap included the portal and splenic veins (DSC: 0.82 ± 0.18) and the adrenal glands (DSC: 0.82 ± 0.12), most likely due to their low contrast resolution in CT imaging of the test dataset, which makes delineation more challenging. The ribs and body composition models also showed higher variation in ASSD, at 0.6 ± 2.1 mm and 2.0 ± 1.7 mm, respectively.\u003c/p\u003e\u003cp\u003eThe ENHANCE.PET 1.6k dataset proved to be satisfactory for training models for automated CT image segmentation. This dataset has the potential to contribute significantly to further advancements in deep learning-based algorithms, including attempts to improve segmentation models performance or the addition of new volumes of interest not covered in the present study. The dual availability of both CT and PET images, together with the inclusion of segmentations for multiple anatomical regions, makes the ENHANCE.PET 1.6k dataset particularly valuable for research focused on diseases that affect multiple organs or systems, such as metabolic disorders or systemic inflammatory diseases\u003csup\u003e\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e,\u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e\u003c/sup\u003e, or for studies on normal glucose metabolism in healthy tissues. As a limitation, since the segmentations were derived from the CT images, the alignment with the corresponding PET images may be compromised in \u003cem class=\"Highlight ht71194251-f7a6-4c2d-a145-3d9f25b46662\" highlight=\"true\" htmatch=\"case*\" htloopnumber=\"264975710\" style=\"font-style: inherit;\"\u003ecases\u003c/em\u003e of significant \u003cem class=\"Highlight ht71194251-f7a6-4c2d-a145-3d9f25b46662\" highlight=\"true\" htmatch=\"patient\" htloopnumber=\"264975710\" style=\"font-style: inherit;\"\u003epatient\u003c/em\u003e motion, which was not systematically assessed in this study.\u003c/p\u003e\u003cp\u003eWe hope that this open-source dataset will accelerate developments in medical imaging, ultimately contributing to the advancement of personalized medicine and more effective clinical decision-making.\u003c/p\u003e\u003cp\u003e\u003cb\u003eUsage Notes\u003c/b\u003e\u003c/p\u003e\u003cp\u003eThe contributions of the Azienda Ospedaliero-Universitaria Careggi and the University Hospital Leipzig to the ENHANCE.PET 1.6k dataset are licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0). Data from the AutoPET\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e Challenge are licensed under the Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0). All imaging data are presented in NIfTI format, ensuring participants’ privacy while allowing for easy use in further analysis. This format can be opened with most visualization software, including 3D Slicer (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.slicer.org/\u003c/span\u003e\u003cspan address=\"https://www.slicer.org/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003e)\u003c/span\u003e and ITK-SNAP (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://www.itksnap.org/pmwiki/pmwiki.php\u003c/span\u003e\u003cspan address=\"http://www.itksnap.org/pmwiki/pmwiki.php\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003e).\u003c/span\u003e DICOM to NIfTI conversion was performed using \u003cem\u003edcm2nii\u003c/em\u003e\u003csup\u003e\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e\u003c/sup\u003e, and all image processing was conducted using Python.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003ch2\u003eCompeting interests\u003c/h2\u003e\u003cp\u003eThe authors declare that there are no conflicts of interest related to this project. L.K.S.S and T.B. are co-founders of Zenta GmbH. R.D.B has received research support from Lilly and from United Imaging Healthcare during the course of this study. UC Davis has a revenue-sharing agreement with United Imaging Healthcare.\u003c/p\u003e\u003c/p\u003e\u003ch2\u003eAuthor contributions\u003c/h2\u003e\u003cp\u003eD.F., M.P., S.G., T.B., L.K.S.S.: data analysis and processing, project conceptualization, manuscript writing, development and maintenance of the software MOOSE\u003csup\u003e39\u003c/sup\u003e.\u003c/p\u003e\u003ch2\u003eAcknowledgements\u003c/h2\u003e\u003cp\u003eThis research was funded in whole or in part by the Austrian Science Fund (FWF) (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.55776/I5902\u003c/span\u003e\u003cspan address=\"10.55776/I5902\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e), under the ERA-NET Cofund scheme of the Horizon 2020 Research and Innovation Framework Programme of the European Commission Research Directorate-General, Grant Agreement No. 779282, which includes national funding for the partners by the FWF (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.55776/I5902\u003c/span\u003e\u003cspan address=\"10.55776/I5902\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e), Innovation Fund Denmark, Regione Toscana, and Saxon State Ministry for Science, Culture and Tourism (Germany). For open access purposes, the author has applied a CC BY public copyright license to any author accepted manuscript version arising from this submission. Sebastian Gutschmayer worked in parts under a research agreement between Siemens Healthineers and the Medical University Vienna. Armin Frille was supported by the postdoctoral fellowship \u0026lsquo;MetaRot program\u0026rsquo; (clinician scientist program) from the Federal Ministry of Education and Research (BMBF), Germany (FKZ 01EO1501, IFB Adiposity Diseases), a research grant from the \u0026lsquo;Mitteldeutsche Gesellschaft f\u0026uuml;r Pneumologie (MDGP) e.V.\u0026rsquo; (2018-MDGP-PA-002), a junior research grant from the Medical Faculty, University of Leipzig (934100-012), and a graduate fellowship from the \u0026lsquo;Novartis Foundation\u0026rsquo;. UC Davis data were acquired using support from the grants: National Psoriasis Foundation 19-4200:NPF, NIH R01AR076088, NIH R01AR085314, NIH K12CA138464, NIH P30CA093373, NIH R01CA206187.\u003c/p\u003e\u003ch2\u003eCode Availability\u003c/h2\u003e\u003cp\u003eThe segmentation software MOOSE and the presented open-source dataset are part of the ENHANCE.PET (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://enhance.pet/\u003c/span\u003e\u003cspan address=\"https://enhance.pet/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) initiative for facilitating data sharing and collaboration within the PET community. In case of downloading and usage of the dataset, please cite the ENHANCE.PET initiative and website.\u003c/p\u003e\u003cp\u003eMOOSE code is open-source and available online with extensive documentation, and can be accessed on GitHub at \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/ENHANCE-PET/MOOSE\u003c/span\u003e\u003cspan address=\"https://github.com/ENHANCE-PET/MOOSE\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Following MOOSE installation within a Python environment, the dataset can be downloaded via the command line:\u003c/p\u003e\u003cp\u003e\u0026lt; moosez -dtd -dd path/to/download/ \u0026gt;\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eZhang S, Metaxas D (2016) Large-Scale medical image analytics: Recent methodologies, applications and Future directions. Med Image Anal 33:98\u0026ndash;101\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePinto-Coelho L (2023) How artificial intelligence is shaping medical imaging technology: A survey of innovations and applications. Bioeng (Basel) 10\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKhalifa M, Albadawy M (2024) AI in diagnostic imaging: Revolutionising accuracy and efficiency. Comput Methods Programs Biomed Update 5:100146\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eShen C et al (2020) An introduction to deep learning in medical physics: advantages, potential, and challenges. Phys Med Biol 65:05TR01\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBeyer T et al (2000) A combined PET/CT scanner for clinical oncology. J Nucl Med 41:1369\u0026ndash;1379\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTownsend DW (2008) Dual-modality imaging: combining anatomy and function. J Nucl Med 49:938\u0026ndash;955\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWechalekar K, Sharma B, Cook G (2005) PET/CT in oncology\u0026mdash;a major advance. Clin Radiol 60:1143\u0026ndash;1155\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLarson DB, Magnus DC, Lungren MP, Shah NH, Langlotz CP (2020) Ethics of using and sharing clinical imaging data for artificial intelligence: A proposed framework. Radiology 295:675\u0026ndash;682\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGatidis S et al (2022) A whole-body FDG-PET/CT Dataset with manually annotated Tumor Lesions. Sci Data 9:601\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eOreiller V et al (2022) Head and neck tumor segmentation in PET/CT: The HECKTOR challenge. Med Image Anal 77:102336\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMa J et al (2024) Segment anything in medical images. Nat Commun 15:654\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWinder C et al (2024) Automated extraction of body composition metrics from abdominal CT or MR imaging: A scoping review. Eur J Radiol 181:111764\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJones T, Townsend D (2017) History and future technical innovation in positron emission tomography. J Med Imaging (Bellingham) 4:011013\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBadawi RD et al (2019) First human imaging studies with the EXPLORER total-body PET scanner. J Nucl Med 60:299\u0026ndash;303\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePrenosil GA et al (2022) Performance Characteristics of the Biograph Vision Quadra PET/CT System with a Long Axial Field of View Using the NEMA NU 2-2018 Standard. J Nucl Med 63:476\u0026ndash;484\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eShiyam Sundar LK, Hacker M, Beyer T, Whole-Body PET (2022) Imaging: A Catalyst for Whole-Person Research? J Nucl Med 64:197\u0026ndash;199\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSun T et al (2022) Identifying the individual metabolic abnormities from a systemic perspective using whole-body PET imaging. Eur J Nucl Med Mol Imaging 49:2994\u0026ndash;3004\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGutschmayer S, Muzik O, Hacker M, Ferrara D, Zuehlsdorff S, Newiger H, Beyer T, Sundar LS (2022) Towards holistic assessment of human physiology: Fully-automated construction of tracer independent total-body PET/CT normative database using diffeomorphisms [abstract]. in \u003cem\u003eAnnual Congress of the European Association of Nuclear Medicine October 15\u0026ndash;19\u003c/em\u003e, Spain vol. 49 225\u0026ndash;226 (Springer Science and Business Media LLC, 2022)\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003evan Ginneken B, Schaefer-Prokop CM, Prokop M (2011) Computer-aided diagnosis: how to move from the laboratory to the clinic. Radiology 261:719\u0026ndash;732\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBednarz B (2023) Theranostics and patient-specific dosimetry. Semin Radiat Oncol 33:317\u0026ndash;326\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePoel R et al (2021) The predictive value of segmentation metrics on dosimetry in organs at risk of the brain. Med Image Anal 73:102161\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRister B, Yi D, Shivakumar K, Nobashi T, Rubin DL (2020) CT-ORG, a new dataset for multiple organ segmentation in computed tomography. Sci Data 7:381\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLuo X et al (2022) A large scale dataset, benchmark and clinical applicable study for abdominal organ segmentation from CT image. Med Image Anal 82:102642\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKoitka S, SAROS et al (2024) A dataset for whole-body region and organ segmentation in CT imaging. Sci Data 11:483\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMa J et al (2020) AbdomenCT-1K: Is abdominal organ segmentation A solved problem? \u003cem\u003earXiv [cs.CV]\u003c/em\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWasserthal J et al (2023) TotalSegmentator: Robust Segmentation of 104 Anatomic Structures in CT Images. Radiology: Artif Intell 5:e230024\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChen X et al (2021) A deep learning-based auto-segmentation system for organs-at-risk on whole-body computed tomography images for radiation therapy. Radiother Oncol 160:175\u0026ndash;184\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHess S, Hansson SH, Pedersen KT, Basu S, H\u0026oslash;ilund-Carlsen P (2014) F. FDG-PET/CT in infectious and inflammatory diseases. PET Clin 9:497\u0026ndash;519\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYao J et al (2022) The imaging quantification of multiple organs by dynamic 18F-FDG PET/CT in discharged COVID-19 patients: A prospective pilot study. Int J Med Sci 19:1539\u0026ndash;1547\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePatel NH et al (2021) Heightened splenic and bone marrow uptake of 18F-FDG PET/CT is associated with systemic inflammation and subclinical atherosclerosis by CCTA in psoriasis: An observational study. Atherosclerosis 339:20\u0026ndash;26\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBang J-I et al (2020) Blood pool activity on F-18 FDG PET/CT as a possible imaging biomarker of metabolic syndrome. Sci Rep 10:17367\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJiang Y et al (2022) Abnormal [18F]FDG uptake in liver and adipose tissue: a potential imaging biomarker for cancer-associated cachexia. Eur Radiol. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00330-022-09226-z\u003c/span\u003e\u003cspan address=\"10.1007/s00330-022-09226-z\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFerrara D et al (2024) Detection of cancer-associated cachexia in lung cancer patients using whole-body [18F]FDG-PET/CT imaging: A multi-centre study. J Cachexia Sarcopenia Muscle. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1002/jcsm.13571\u003c/span\u003e\u003cspan address=\"10.1002/jcsm.13571\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYu J et al (2024) Systemic metabolic and volumetric assessment via whole-body [18F]FDG-PET/CT: Pancreas size predicts cachexia in head and neck squamous cell carcinoma. Cancers (Basel) 16:3352\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePak K et al (2024) Glucose metabolism and radiodensity of abdominal adipose tissue: A 5-year longitudinal study in a large PET cohort. Clin Endocrinol (Oxf) 101:623\u0026ndash;630\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eIshii S et al (2024) Investigation of factors affecting CT attenuation and glucose metabolism of bone marrow as seen on PET/CT scans. Am J Nucl Med Mol Imaging 14:22\u0026ndash;30\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSharma P, Chatterjee P, Alvarado LA, Dwivedi AK (2023) Standardized uptake value of normal organs on routine clinical [18F]FDG PET/CT: impact of tumor metabolism and patient-related factors. Nucl Med Rev Cent East Eur 26:1\u0026ndash;10\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSundar LKS et al (2023) The ENHANCE-PET Framework: An initiative to engage the imaging community in Advancements in Total-Body PET Analysis. EANM\u0026rsquo;23 Abstract Book Congress Sep 9\u0026ndash;13, 2023. Eur J Nucl Med Mol Imaging 50:1\u0026ndash;898\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSundar LKS et al (2022) Fully Automated, Semantic Segmentation of Whole-Body 18F-FDG PET/CT Images Based on Data-Centric Artificial Intelligence. J Nucl Med 63:1941\u0026ndash;1948\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFedorov A et al (2012) 3D Slicer as an image computing platform for the Quantitative Imaging Network. Magn Reson Imaging 30:1323\u0026ndash;1341\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eClark K et al (2013) The Cancer Imaging Archive (TCIA): maintaining and operating a public information repository. J Digit Imaging 26:1045\u0026ndash;1057\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLi X, Morgan PS, Ashburner J, Smith J, Rorden C (2016) The first step for neuroimaging data analysis: DICOM to NIfTI conversion. J Neurosci Methods 264:47\u0026ndash;56\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSelfridge AR et al (2023) Facial anonymization and privacy concerns in total-body PET/CT. J Nucl Med 64:1304\u0026ndash;1309\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSchwarz CG et al (2019) Identification of anonymous MRI research participants with face-recognition software. N Engl J Med 381:1684\u0026ndash;1686\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eClunie DA et al (2023) Report of the medical image DE-identification (MIDI) task group - best practices and recommendations. \u003cem\u003eArXiv\u003c/em\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eIsensee F, Jaeger PF, Kohl SAA, Petersen J, Maier-Hein KH (2021) nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation. Nat Methods 18:203\u0026ndash;211\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYeghiazaryan V, Voiculescu I (2018) Family of boundary overlap metrics for the evaluation of medical image segmentation. J Med Imaging (Bellingham) 5:015006\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKatal S, Eibschutz LS, Saboury B, Gholamrezanezhad A, Alavi A (2022) Advantages and applications of total-body PET scanning. Diagnostics (Basel) 12:426\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNakamoto R et al (2019) Diffusely Decreased Liver Uptake on FDG PET and Cancer-Associated Cachexia With Reduced Survival. Clin Nucl Med 44:634\u0026ndash;642\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[{"identity":"1611b136-fe84-469a-839f-930cce46b3f4","identifier":"10.13039/501100002428","name":"Austrian Science Fund","awardNumber":"10.55776/I5902","order_by":0}],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"FWF Austrian Science Fund","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"[18F]FDG-PET/CT images, anatomical segmentations, open-sourcing","lastPublishedDoi":"10.21203/rs.3.rs-7169062/v2","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7169062/v2","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eWe present a large whole-body and total-body curated dataset of dual-modality 2-deoxy-2-[18F]fluoro-D-glucose (FDG)-Positron Emission Tomography/Computed Tomography (PET/CT) studies, consisting of 1,597 PET/CT images and the corresponding CT-derived segmentations of 130 target regions. This multi-center dataset includes images from individuals without overt disease and patients with different pathologies (lung cancer, lymphoma, and melanoma). Target regions were first automatically segmented from CT images using an in-house software, and subsequently verified and corrected by physicians-in-training. In total, the segmented regions encompass 130 volumes, including abdominal organs, muscles, bones, cardiac subregions, vessels, adipose tissue, and skeletal muscle around the third lumbar vertebra. PET/CT images and corresponding CT-derived segmentations are provided in anonymized NIfTI format. The dataset can be used for deep learning training, validation, or multi-modality image analysis and thus fills an important gap in available resources to advance the use of PET/CT data in clinical management.\u003c/p\u003e","manuscriptTitle":"Sharing a whole-/total-body [18F]FDG-PET/CT dataset with CT-derived segmentations: an ENHANCE.PET initiative","msid":"","msnumber":"","nonDraftVersions":[{"code":2,"date":"2025-08-05 17:25:56","doi":"10.21203/rs.3.rs-7169062/v2","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}},{"code":1,"date":"2025-07-23 08:47:49","doi":"10.21203/rs.3.rs-7169062/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"ce0fc6fb-029f-4e82-afe2-3fdae3ec6285","owner":[],"postedDate":"August 5th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":52131358,"name":"Medical Physics"}],"tags":[],"updatedAt":"2025-07-23T08:47:49+00:00","versionOfRecord":[],"versionCreatedAt":"2025-08-05 17:25:56","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v2","identity":"rs-7169062","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7169062","identity":"rs-7169062","version":["v2"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}