Medical Physics

Degenerative conditions of the lumbar spine, such as neural foraminal narrowing, subarticular stenosis, and spinal canal stenosis, are common and can lead to chronic pain and disability. Detecting and assessing the severity of these conditions is a key task in radiology, often requiring considerable expertise. In this work, we aim to develop automated models that aid radiologists in diagnosing these conditions using lumbar spine MR images. We utilize state-of-the-art deep learning methods for both detection and classification tasks, focusing on severity classification across five intervertebral disc levels and three different imaging modalities. The dataset and methodology used in this study are part of the RSNA 2024 Lumbar Spine Degenerative Classification competition. Our approach involves a two-step process, combining object detection with YOLO v8 and severity classification using a DeepScoreNet architecture. Through careful cross-validation and model evaluation, we demonstrate the potential of these models to improve diagnostic accuracy and efficiency. The DeepScoreNet model, introduced in this article, comprises 2,787,651 parameters and achieved accuracies of 81% for score(severity) prediction in cases of Neural Foraminal Narrowing, 89% for Spinal Canal Stenosis, and 74% for Subarticular Stenosis. The code for this approach is available in the GitHub repository.(https://github.com/liamirpy/Lumbar-Spine-Degenerative-Classification)

BackgroundPreclinical investigations studies have shown that FLASH radiotherapy (FLASH-RT), delivering radiation in ultra-high dose rates (UHDR), preserves healthy tissue and reduces toxicity, all while maintaining an effective tumor response compared to conventional radiotherapy (CONV-RT), the combined biological benefit was termed as "FLASH effect". However, the mechanisms responsible for this effect remain unclear. Research demonstrated that oxygen concentration contributes to the FLASH effect, and it has been hypothesized that Fenton reaction might play a role in the "FLASH effect". PurposeWe propose to investigate the effect of ultra-high dose rate (UHDR), compared to conventional dose rates (CONV), on the Fenton reaction by studying the radiolysis of Fricke solution. The study will focus on how dose, dose rate, and initial oxygen concentration influence the activation of the Fenton reaction. Methods and MaterialsTOPAS-nBio version 2.0 was used to simulate the radiolysis of the Fricke system. A cubic water phantom of 3{micro}m side was irradiated by 300MeV protons on one of its edges. For UHDR, a proton field (1.5x1.5{micro}m2) was delivered in a single pulse of 1ns width. The protons were accumulated until reached 5Gy or 10Gy absorbed dose. For CONV, the independent history approach was used to mimic 60Co irradiation. For both dose-rates, oxygen concentrations representative of hypoxic and normoxic tissues (10- 250{micro}M) were simulated. The G-value for oxidant ions G(Fe3+) and {Delta}G-value of Fenton reaction (H2O2 + Fe2+[->] Fe3++*OH+OH-) were scored. The simulations ended after G(Fe3+) achieved steady-state, and calculated yields were compared with published data. ResultsFor CONV, G(Fe3+) agreed with ICRU-report 34 data by (0.97{+/-}0.1) %. For UHDR, G(Fe3+) agreed with ICRU data by (1.24{+/-}0.1)% and (0.92{+/-}0.1)% for 5Gy and 10Gy, respectively. Notably, UHDR at 10 Gy reduced the occurrence of Fenton reactions by (1.0{+/-}0.1)% and (11.5{+/-}0.1)% at initial oxygen concentrations of 250 {micro}M and 10 {micro}M, respectively. In consequence, UHDR decreased G(Fe3+) by (1.8{+/-}0.1)% and (12.5{+/-}0.1)% at these oxygen levels. Additionally, increasing the absorbed dose to 15 Gy and 20 Gy at low oxygen (10 {micro}M), UHDR further reduced the {Delta}G-value by (15.7{+/-}0.1)% and (18.6{+/-}0.1)%, respectively. The decrease was driven by intertrack effects present in UHDR pulses and its impact on the scavenging effect that oxygen had over hydrogen radicals. ConclusionsUHDR reduces the yield of Fe3+ (G(Fe3+)) and significantly impacts Fenton reactions, particularly at low oxygen concentrations, while showing minimal effects at higher oxygen levels. This effect becomes more pronounced at higher dose thresholds, such as 10-20 Gy. This emphasizes the important role of the initial oxygen concentration in UHDR and its influence on the Fenton reaction, a mechanism that may contribute to elucidate the FLASH effect.

Ultra-high-dose-rate therapy enhances the protection of normal tissues and reduces side effects while effectively controlling tumors. This biological phenomenon is called the FLASH effect, and when observed, therapy is called FLASH Radiotherapy (FLASH-RT). Various hypotheses have been proposed to explain how ultra-high dose rates achieve these effects under different conditions, with the impact of tissue oxygen perfusion still needing further investigation. FLASH-RT involves brief exposure to radiation, which results in fewer heartbeats occurring during the irradiation period, which could lead to reduced tissue oxygen perfusion occurring during the treatment timeframe. Therefore, we developed a compartmental model to simulate oxygen transfer and its interaction with radiation. The proposed model consists of three compartments: 1) the heart and arteries; 2) the irradiated brains blood vessels and capillaries; and 3) the irradiated brain tissue. We employed a system of differential equations, incorporating experimental data from in vivo oxygen measurements using the Oxyphor probe in the brain, to fit the model parameters to the experimental results. This model shows how dose rate and oxygen perfusion could influence chemical processes such as lipid peroxidation, potentially leading to differential biological effects. Our analysis of lipid peroxidation as a function of dose rate revealed a sigmoidal dose-rate-response curve that correlates well with several published biological response datasets. Our results indicate that the differential chemical effects of FLASH-RT compared with conventional dose rates may depend on factors such as oxygen perfusion, consumption, and tissue oxygen tension. This suggests that the temporal dynamics of oxygen could play a crucial role in enhancing the therapeutic window for FLASH-RT treatments. Furthermore, it suggests that the magnitude of some observed FLASH effects may vary across tissues or tumors and across experimental models, given differential oxygen dynamics.

PurposeThis study investigates the robustness of quantitative radiomic features derived from computed tomography (CT) images of a novel patient informed 3-D printed phantom, which captures the morphological heterogeneity of tumors and normal tissue observed on CT scans.\n\nMethodsUsing a novel voxel-based multi-material three-dimensional (3D) printer, an anthropomorphic phantom that was modeled after diseased tissue seen on 6 patient CT scans was manufactured. Four patients presented with pancreatic adenocarcinoma tumors (PDAC), 1 with non-small cell lung carcinoma (NSCLC) and 1 with advanced stage hepatic cirrhosis. The 5 tumors were segmented, extracted and then imbedded into CT images of the heterogenous portion of the cirrhotic liver. The composite scan of the implanted tumor within the background cirrhotic liver was then 3D printed. The resultant phantom was scanned sequentially, 30 times with a clinical CT scanner using a reference CT protocol. One hundred and four quantitative radiomic features were then extracted from images of each lesion to determine their repeatability. Repeatability of each radiomic feature was evaluated using the within subject coefficient of variation (wCV, %). A feature with a wCV (%) > 10% was considered as being unrepeatable. A subset of the repeatable features that were also found to be prognostic for lung and pancreatic cancers were then assessed for their percent deviation (pDV, %) from reference values. The reference values were those derived from the repeatability portion of this study. The assessment was conducted by re-scanning the phantom with 11 different clinically relevant sets of scanning parameters. Deviation of radiomic features derived from images of each tumor across all sets of scanning parameters was assessed using the percent deviation relative to the reference values.\n\nResultsTwenty nine of the 104 features presented with wCV (%) > 10%. The lack of repeatability was found to depend on tumor type. The only class of radiomic features with a wCV (%) < 10% were those calculated using the neighboring grey level dependence-based matrices (NGLDM). Notably, skewness, first information correlation, cluster shade, Haralick correlation, autocorrelation, busyness, complexity, high gray level zone emphasis, small area high gray level emphasis, large area low gray level emphasis, large area high gray level emphasis, short run high grey level emphasis, and valley radiomic features had wCV (%) values > 10% for select tumors within the phantom. Two radiomic features prognostic for NSCLC, energy and grey level non-uniformity, had pDVs (%) that exceeded 30% across all scanning techniques. The pDV (%) for the 4 radiomic features prognostic for PDAC tumors depended on tumor type and selected scanning parameter. Application of the lung kernel caused the largest pDVs (%). Scans acquired with the reduced tube current of 100 mA and reconstructed with the bone kernel yielded pDVs (%) within {+/-} 10%.\n\nConclusionWe demonstrated the feasibility with which patient informed 3D printed phantoms can be manufactured directly from lesions seen on CT scans, and demonstrate their potential use for the assessment of robust quantitative radiomic features.

PurposeGiven the high prevalence of vertebral fractures post-radiotherapy in patients with metastatic spine disease, accurate and rapid muscle segmentation could support efforts to quantify muscular changes due to disease or treatment and enable biomechanical modeling for assessments of vertebral loading to improve personalized evaluation of vertebral fracture risk. This study presents a deep-learning approach for segmenting the complete volume of the trunk muscles from clinical CT images trained using sparsely annotated data. Materials and Methodswe extracted 2,009 axial CT images at the midpoint of each vertebral level (T4 to L4) from clinical CT of 148 cancer patients. The key extensor and flexor muscles (up to 8 muscles per side) were manually contoured and labeled per image in the thoracic and lumbar regions. We first trained a 2D nnU-Net deep-learning model on these labels to segment key extensor and flexor muscles. Using these sparse annotations per spine, we trained the model to segment each muscles entire 3D volume ResultsThe proposed method achieved comparable performance to manual segmentations, as assessed by expert radiologists, with a mean Dice score above 0.769. Significantly, the model drastically reduced segmentation time, from 4.3-6.5 hours for manual segmentation of 14 single axial CT images to approximately 1 minute for segmenting the complete thoracic-abdominal 3D volume. ConclusionThe approach demonstrates high potential for automating 3D muscle segmentation, significantly reducing the manual intervention required for generating musculoskeletal models, and could be instrumental in enhancing clinical decision-making and patient care in radiation oncology. SummaryA deep learning 2D nnU-Net model, trained on a sparse set of 2D muscle annotations, successfully segmented the entire volume of 20 thoracolumbar muscles from cancer patients clinical CT data. The model showed a remarkable increase in segmentation efficacy and generalizability, achieving comparable performance to manual segmentations in delineating each muscle anatomy. Key Points{blacksquare} A deep learning model (2D nnU-Net), developed using a sparse set of single axial CT-slice at each mid-per vertebral level, containing manual image annotation of 20 thoracic and lumbar muscles, achieved comparable performance to manual segmentations, as assessed by expert radiologists, with a mean Dice score above 0.769. {blacksquare}The model drastically reduced segmentation time, from 4.3-6.5 hours for manual segmentation of 14 single axial CT images to approximately 1 minute for segmenting the complete thoracic-abdominal 3D volume. {blacksquare}Radiologist assessment based on a Likert scale (0-5) for clinical acceptability of the muscle anatomical segmentation showed strong model performance for a representative sample of clinical CT data (a (mean(SD) of 4.66 (0.73)) and external data (4.66 (0.73).

Iodine-based staining techniques are commonly used in histological imaging and micro-computed tomography ({micro}CT) due to iodines affinity for binding to specific molecules. However, the basis for tissue-specific contrast has not yet been sufficiently explored. In this study, we analyse the human proteome at four different levels: individual proteins, protein families, tissues with additional expression values for selected proteins, and organs as a distinct combination of different tissues. At each level, we try to identify proteins/groups with high potential for iodine binding, especially those rich in aromatic heterocyclic amino acids. Using bioinformatic methods, we evaluate the occurrence of aromatic/non-aromatic heterocyclic, carbocyclic, and the remaining 15 amino acids in 20,650 proteins, 1,487 families, 57 tissues, and 16 organs. At the protein level, structural proteins such as titin, nebulin, obscurin, mucin, filaggrin and hornerin have a high absolute number of aromatic heterocyclic amino acids, which could explain the high {micro}CT contrast in muscle, skin and mucosal tissues. At the next level, however, structural families (such as the Laminin-family) rank significantly lower in comparison. These results are reflected in tissues and organs for which protein expression is available. Here, no significant correlations between the enrichment of heterocycles and the intensity of iodine staining can be observed. Furthermore, the enrichment of amino acids in each tissue/organ is relatively similar and shows no significant difference. Our results provide a general basis for iodine-based tissue imaging and serve as a potential starting point for future research, e.g. for cross-species applications and for the structural and functional effects of iodination.

BackgroundAutomated organ segmentation from computed tomography (CT) images facilitates a number of clinical applications, including clinical diagnosis, monitoring of treatment response, quantification, radiation therapy treatment planning, and radiation dosimetry. PurposeTo develop a novel deep learning framework to generate multi-organ masks from CT images for 23 different body organs. MethodsA dataset consisting of 3106 CT images (649,398 axial 2D CT slices, 13,640 images/segment pairs) and ground-truth manual segmentation from various online available databases were collected. After cropping them to body contour, they were resized, normalized and used to train separate models for 23 organs. Data were split to train (80%) and test (20%) covering all the databases. A Res-UNET model was trained to generate segmentation masks from the input normalized CT images. The model output was converted back to the original dimensions and compared with ground-truth segmentation masks in terms of Dice and Jaccard coefficients. The information about organ positions was implemented during post-processing by providing six anchor organ segmentations as input. Our model was compared with the online available "TotalSegmentator" model through testing our model on their test datasets and their model on our test datasets. ResultsThe average Dice coefficient before and after post-processing was 84.28% and 83.26% respectively. The average Jaccard index was 76.17 and 70.60 before and after post-processing respectively. Dice coefficients over 90% were achieved for the liver, heart, bones, kidneys, spleen, femur heads, lungs, aorta, eyes, and brain segmentation masks. Post-processing improved the performance in only nine organs. Our model on the TotalSegmentator dataset was better than their models on our dataset in five organs out of 15 common organs and achieved almost similar performance for two organs. ConclusionsThe availability of a fast and reliable multi-organ segmentation tool leverages implementation in clinical setting. In this study, we developed deep learning models to segment multiple body organs and compared the performance of our models with different algorithms. Our model was trained on images presenting with large variability emanating from different databases producing acceptable results even in cases with unusual anatomies and pathologies, such as splenomegaly. We recommend using these algorithms for organs providing good performance. One of the main merits of our proposed models is their lightweight nature with an average inference time of 1.67 seconds per case per organ for a total-body CT image, which facilitates their implementation on standard computers.

PurposeThe relative biological effectiveness (RBE) of tumor control for proton beam therapy (PBT) compared to photon radiotherapy (RT) is typically assumed to be independent of fractionation. To test this, we modeled published PBT outcome results for early-stage non-small cell lung cancer (NSCLC) treatments across a range of fractionation schedules. Materials and MethodsAll published and analyzable cohorts were included (399 patients, 413 treated lesions). Two models were used to fit the data: a previously published tumor simulation model that fits photon RT results of NSCLC across all fractionation regimes and the Fowler LQ model with a kick-off time term. The treatment effect of each cohort was referenced to the photon equivalent dose through mechanistic model simulations in a 2 Gy/weekday scenario, with radiobiological parameters determined to simultaneously best-fit all fractionation results. The tumor control RBE of each published treatment schedule, compared to the modeled photon RT effect of the same schedule, was then estimated. ResultsFor cohorts whose treatments lasted less than three weeks (i.e., 12 fractions or less), the RBE of PBT was in the range of 1.08 to 1.11. However, for fractionated treatments stretching over four weeks or more (20-25 fractions), the relative effectiveness was much lower, with RBEs in the range of 0.82-0.89. This conclusion was unchanged using the simpler Fowler LQ + time model. ConclusionsThe proton RBE for hypo-fractionated schedules was 20-30% higher than for conventional schedules. The derived radiobiological parameters of PBT differ significantly from those of photon RT, indicating that PBT is influenced differentially by radiobiological mechanisms which require further investigation.

PurposeExternal beam radiotherapy (XRT) is a widely used cancer treatment, yet responses vary dramatically between patients. These differences are not accounted for in clinical practice, in part due to a lack of sensitive biomarkers of early response. In this work, we test the hypothesis that quantification of intratumor heterogeneity is a sensitive and robust biomarker of early response to XRT. A novel Magnetic Resonance Imaging (MRI) approach is proposed, utilizing histogram analysis of clinically-used T2 relaxation measurements to assess early changes in the tumor heterogeneity following irradiation in murine models of pancreatic cancer, indicative of radiotherapy response. Methods and MaterialsDynamic Magnetic Resonance T2 relaxation imaging was performed every 72h following 10 Gy dose XRT in two murine models of pancreatic cancer. Proposed biomarker of radiotherapy response was compared with tumor growth kinetics, and biological validation was performed through quantitative histology analysis. ResultsQuantification of tumor T2 interquartile range (IQR) as a measure of histogram width showed excellent sensitivity for detection of XRT-induced tumor changes as early as 72h after treatment, outperforming whole tumor T2 and Diffusion weighted MRI metrics. This response was observed both in quantitative T2 maps and in T2-weighted images that are routine in clinical practice. Histological comparison revealed the T2 IQR provides a measure of spatial heterogeneity in tumor cell density, related to radiation-induced necrosis. The early IQR changes were found to presage subsequent tumor volume changes in two distinct pancreatic models, suggesting promise for treatment response prediction. The metric showed excellent test-retest robustness. ConclusionsOur preclinical findings indicate that spatial heterogeneity analysis of T2 MRI can provide a sensitive and readily translatable method for early radiotherapy response assessment in pancreatic cancer. We propose that this will be useful in adaptive radiotherapy, specifically in MRI-guided treatment paradigms.

Microcomputed tomography (micro-CT) is a frequently used imaging tool for a wide spectrum of in-vivo mouse models in basic and translational research. To allow an accurate interpretation of micro-CT images, high spatial resolution is necessary. However, this may also lead to a high radiation exposure of the animals. Therefore, animal welfare requires exact information about the expected radiation doses for experimental planning. To gain this, a mouse cadaver was herein used for micro-CT analyses under different conditions. For each radiation dose measurement, the cadaver was labeled with thermoluminescent dosimeter chips around the thoracic skin surface. Micro-CT scans of the thorax were performed with spatial resolutions of 35 {micro}m, 18 {micro}m and 9 {micro}m in combination with Al0.5, Al1.0, CuAl and Cu filters. As a surrogate of image quality, the number of identifiable lung vessels was counted on a transversal micro-CT slice. Measured radiation doses varied from 0.09 Gy up to 5.18 Gy dependent on resolution and filter settings. A significant dose reduction of > 75% was achieved by a Cu filter when compared to an Al0.5 filter. However, this resulted in a markedly reduced image quality and interpretability of microstructures due to higher radiation shielding and lower spatial resolution. Thus, the right combination of distinct filters and several scan protocol settings adjusted to the individual requirements can significantly reduce the radiation dose of micro-CT leading to a higher animal welfare standard.

We propose a deep learning approach for detecting anterior cruciate ligament (ACL) tears from knee MRI using a dual-branch convolutional architecture. The model independently processes sagittal and coronal MRI sequences using EfficientNet-B2 backbones with spatial attention modules, followed by a late fusion classifier for binary prediction. MRI volumes are standardized to a fixed number of slices, and domain-specific normalization and data augmentation are applied to enhance model robustness. Trained on a stratified 80/20 split of the MRNet dataset, our best model--using the Adam optimizer and a learning rate of 1e-4--achieved a validation AUC of 0.98 and a test AUC of 0.93. These results show strong predictive performance while maintaining computational efficiency. This work demonstrates that accurate diagnosis is achievable using only two anatomical planes and sets the stage for further improvements through architectural enhancements and broader data integration.

ObjectiveTo validate the feasibility of AI Deep Learning Reconstruction for Coronary Artery Calcification Scoring in order to decrease radiation exposure on a 4cm detector CT scanner. This is the first such validation on devices that are most commonly utilized for this procedure. MethodsData from 105 consecutive patients referred for Coronary Artery Calcification Scoring (CACS) in 4 centers was reconstructed with Filtered Back Projection (FBP), Iterative Reconstruction (Hybrid-IR), and AI Deep Learning Reconstruction (AI DLR), and analyzed both quantitatively and qualitatively to determine if AI DLR can be routinely used for this purpose. Additional phantom testing was performed to determine if further dose reduction can be accomplished with AI DLR while maintaining or improving image quality compared to current Hybrid-IR reconstruction. ResultsQuantitively, there was excellent agreement between the three reconstructions (FBP, Hybrid IR and AI DLR) with an interclass coefficient of 0.99. The mean CACS for Filtered Back Projection Reconstructions was 111.05. The mean CACS for Hybrid-IR was 91.30. The mean CACS for AI Deep Learning Reconstructions was 93.50. Qualitatively, image quality was consistently better with AI DLR than with Hybrid-IR at both soft tissue and lung windowing. Based on our phantom experiments, AI DLR allows for dose reduction of at least a 37% without any image quality penalty compared to Hybrid-IR. ConclusionsThe use of AI DLR for use in CACS on 4 cm coverage CT scanner has been quantitatively and qualitatively validated for use for the first time. AI DLR produces qualitatively and quantitively better image quality than Hybrid-IR at the same dose level, and produces good agreement in categorization of Agatston scores. In vivo and in vitro evaluations show that AI DLR will allow for an at least a 37% further dose reduction on a 4 cm coverage CT scanner.

PurposeConvolutional neural networks (CNNs) have been proposed for super-resolution in CT, but training of CNNs requires high-resolution reference data. Higher spatial resolution can also be achieved using deconvolution, but conventional deconvolution approaches amplify noise. We develop a CNN that mitigates increasing noise and that does not require higher-resolution reference images. MethodsOur model includes a noise reduction CNN and a deconvolution CNN that are separately trained. The noise reduction CNN is a U-Net, similar to other noise reduction CNNs found in the literature. The deconvolution CNN uses an autoencoder, where the decoder is fixed and provided as a hyperparameter that represents the system point spread function. The encoder is trained to provide a deconvolution that does not amplify noise. Ringing can occur from deconvolution but is controlled with a difference of gradients loss function term. Our technique was demonstrated on a variety of patient images and on ex vivo kidney stones. ResultsThe noise reduction and deconvolution CNNs produced visually sharper images at low noise. In ex vivo mixed kidney stones, better visual delineation of the kidney stone components could be seen. ConclusionsA noise reduction and deconvolution CNN improves spatial resolution and reduces noise without requiring higher-resolution reference images.

Spectral CT has been increasingly implemented clinically for its better characterization and quantification of materials through its multi-energy results. It also facilitates calculation of physical density utilizing the Alvarez-Macovski model without approximations. These spectral physical density quantifications allow for non-invasive mass measurements and temperature evaluations by manipulating the definition of physical density and thermal volumetric expansion, respectively. To develop the model, original and parametrized versions of the Alvarez-Macovski model and electron density-physical density model were validated with a phantom. The best physical density model was then implemented on clinical spectral CT scans of ex vivo bovine muscle to determine the accuracy and effect of acquisition parameters on mass measurements. In addition, the relationship between physical density and changes in temperature was evaluated by scanning and subjecting the tissue to a range of temperatures. A linear fit utilizing the thermal volumetric expansion was performed to assess the correlation. The parametrized Alvarez-Macovski model performed best in both model development and validation with errors within {+/-}0.02 g/mL. As observed with muscle, physical density was not significantly affected by dose and acquisition mode but was slightly affected by collimation. These effects were also reflected in mass measurements, which demonstrated accuracy with a maximum percent error of 0.34%, further validating the physical density model. Furthermore, physical density was strongly correlated (R of 0.9781) to temperature changes through thermal volumetric expansion. Accurate and precise spectral physical density quantifications enable non-invasive mass measurements for pathological detection and temperature evaluation for thermal therapy monitoring in interventional oncology.

BackgroundLarge-scale CT-based reference standards for abdominal organ volume, incorporating age, sex, and body size, are limited. PurposeTo establish sex- and age-specific reference distributions for major abdominal organ volumes on non-contrast abdominopelvic CT in a nationwide Japanese cohort to provide a foundation for automated clinical assessment and dose optimization. Materials and MethodsIn this retrospective, multicenter study, using the Japan Medical Image Database, we identified all non-contrast abdominopelvic CT examinations performed in 2024. Unique adults with available data on age, sex, height, and weight were included in this study. The final sample comprised 49,764 examinations (26,456 men and 23,308 women) conducted at nine institutions. Automated segmentation (TotalSegmentator v2.10.0) was used to produce organ volumes, excluding hollow viscera. The sex-specific 10th, 25th, 50th, 75th, and 90th percentiles were calculated. Age-volume relationships of body surface area (BSA)-normalized volumes (mL/m2) were modeled using natural cubic splines (four degrees of freedom) separately by sex. ResultsMedian (mL) male vs female volumes were as follows: liver, 1194.7 vs 1024.0; pancreas, 63.6 vs 52.2; spleen, 118.1 vs 95.1; kidneys (total), 268.3 vs 221.2; adrenals (total), 6.6 vs 4.2; iliopsoas (total), 483.4 vs 317.7; prostate, 24.9 (men only). Age-volume relationships of BSA-normalized volumes showed convex patterns for the liver, pancreas, and kidneys in both sexes and for male adrenal glands; lower values in older age groups for the spleen and iliopsoas in both sexes; and higher values in older age groups for the prostate and female adrenal glands. ConclusionThis nationwide Japanese CT cohort provides sex- and age-resolved volumetric reference standards. These standards enable objective identification of abnormalities, support personalized medicine, and facilitate automated AI-based reporting to reduce radiologist workload and optimize radiation dose protocols. Key ResultsO_LIMedian volumes (men vs women, mL): liver 1195/1024; pancreas 64/52; spleen 118/95; kidneys 268/221; adrenals 6.6/4.2; iliopsoas 483/318; prostate 25. C_LIO_LIBody surface area-normalized age-volume relationships were convex for liver, pancreas, and kidneys in both sexes and for male adrenal glands. C_LIO_LISpleen and iliopsoas declined monotonically with age in both sexes, whereas prostate and female adrenal glands increased monotonically. C_LI

Computed tomography (CT) is an extensively used imaging modality capable of generating detailed images of a patients internal anatomy for diagnostic and interventional procedures. High-resolution volumes are created by measuring and combining information along many radiographic projection angles. In current medical practice, single and dual-view two-dimensional (2D) topograms are utilized for planning the proceeding diagnostic scans and for selecting favorable acquisition parameters, either manually or automatically, as well as for dose modulation calculations. In this study, we develop modified 2D to three-dimensional (3D) encoder-decoder neural network architectures to generate CT-like volumes from single and dual-view topograms. We validate the developed neural networks on synthesized topograms from publicly available thoracic CT datasets. Finally, we assess the viability of the proposed transformational encoder-decoder architecture on both common image similarity metrics and quantitative clinical use case metrics, a first for 2D-to-3D CT reconstruction research. According to our findings, both single-input and dual-input neural networks are able to provide accurate volumetric anatomical estimates. The proposed technology will allow for improved (i) planning of diagnostic CT acquisitions, (ii) input for various dose modulation techniques, and (iii) recommendations for acquisition parameters and/or automatic parameter selection. It may also provide for an accurate attenuation correction map for positron emission tomography (PET) with only a small fraction of the radiation dose utilized.

PurposeTo construct robust and validated radiomic predictive models, the development of a reliable method that can identify reproducible radiomic features robust to varying image acquisition methods and other scanner parameters should be preceded with rigorous validation. Due to the property of high correlation present between radiomic features, we hypothesize that reproducible radiomic features across different datasets that are obtained from different image acquisition settings preserve some level of connectivity between features in the form of a network.\n\nMethodsWe propose a regularized partial correlation network to identify robust and reproducible radiomic features. This approach was tested on two radiomic feature sets generated with two different reconstruction methods from a cohort of 47 lung cancer patients. The commonality of the resulting two networks was assessed. A largest common network component from the two networks was tested on phantom data consisting of 5 cancer samples. We further propose a novel K-means algorithm coupled with the optimal mass transport (OMT) theory to cluster samples. This approach following the regularized partial correlation analysis was tested on computed tomography (CT) scans from 77 head and neck cancer patients that were downloaded from The Cancer Imaging Archive (TCIA) and validated on CT scans from 83 head and neck cancer patients treated at our institution.\n\nResultsCommon radiomic features were found in relatively large network components between the resulting two partial correlation networks from a cohort of 47 lung cancer patients. The similarity of network components in terms of the common number of radiomic features was statistically significant. For phantom data, the Wasserstein distance on a largest common network component from the lung cancer data was much smaller than the Wasserstein distance on the same network using random radiomic features, implying the reliability of those radiomic features present in the network. Further analysis using the proposed Wasserstein K-means algorithm on TCIA head and neck cancer data showed that the resulting clusters separate tumor subsites and this was validated on our institution data.\n\nConclusionsWe showed that a network-based analysis enables identifying reproducible radiomic features. This was validated using phantom data and external data via the Wasserstein distance metric and the proposed Wasserstein K-means method.

BackgroundOropharyngeal cancer (OPC) is a widespread disease, with radiotherapy being a core treatment modality. Manual segmentation of the primary gross tumor volume (GTVp) is currently employed for OPC radiotherapy planning, but is subject to significant interobserver variability. Deep learning (DL) approaches have shown promise in automating GTVp segmentation, but comparative (auto)confidence metrics of these models predictions has not been well-explored. Quantifying instance-specific DL model uncertainty is crucial to improving clinician trust and facilitating broad clinical implementation. Therefore, in this study, probabilistic DL models for GTVp auto-segmentation were developed using large-scale PET/CT datasets, and various uncertainty auto-estimation methods were systematically investigated and benchmarked. MethodsWe utilized the publicly available 2021 HECKTOR Challenge training dataset with 224 co-registered PET/CT scans of OPC patients with corresponding GTVp segmentations as a development set. A separate set of 67 co-registered PET/CT scans of OPC patients with corresponding GTVp segmentations was used for external validation. Two approximate Bayesian deep learning methods, the MC Dropout Ensemble and Deep Ensemble, both with five submodels, were evaluated for GTVp segmentation and uncertainty performance. The segmentation performance was evaluated using the volumetric Dice similarity coefficient (DSC), mean surface distance (MSD), and Hausdorff distance at 95% (95HD). The uncertainty was evaluated using four measures from literature: coefficient of variation (CV), structure expected entropy, structure predictive entropy, and structure mutual information, and additionally with our novel Dice-risk measure. The utility of uncertainty information was evaluated with the accuracy of uncertainty-based segmentation performance prediction using the Accuracy vs Uncertainty (AvU) metric, and by examining the linear correlation between uncertainty estimates and DSC. In addition, batch-based and instance-based referral processes were examined, where the patients with high uncertainty were rejected from the set. In the batch referral process, the area under the referral curve with DSC (R-DSC AUC) was used for evaluation, whereas in the instance referral process, the DSC at various uncertainty thresholds were examined. ResultsBoth models behaved similarly in terms of the segmentation performance and uncertainty estimation. Specifically, the MC Dropout Ensemble had 0.776 DSC, 1.703 mm MSD, and 5.385 mm 95HD. The Deep Ensemble had 0.767 DSC, 1.717 mm MSD, and 5.477 mm 95HD. The uncertainty measure with the highest DSC correlation was structure predictive entropy with correlation coefficients of 0.699 and 0.692 for the MC Dropout Ensemble and the Deep Ensemble, respectively. The highest AvU value was 0.866 for both models. The best performing uncertainty measure for both models was the CV which had R-DSC AUC of 0.783 and 0.782 for the MC Dropout Ensemble and Deep Ensemble, respectively. With referring patients based on uncertainty thresholds from 0.85 validation DSC for all uncertainty measures, on average the DSC improved from the full dataset by 4.7% and 5.0% while referring 21.8% and 22% patients for MC Dropout Ensemble and Deep Ensemble, respectively. ConclusionWe found that many of the investigated methods provide overall similar but distinct utility in terms of predicting segmentation quality and referral performance. These findings are a critical first-step towards more widespread implementation of uncertainty quantification in OPC GTVp segmentation.

PurposeSingle-energy computed tomography (CT) often suffers from poor contrast, yet it remains critical for effec-tive radiotherapy treatment. Modern therapy systems are often equipped with both megavoltage (MV) and kilovoltage (kV) x-ray sources and thus already possess the hardware needed for dual-energy (DE) CT. There exists an unexplored potential for enhanced image contrast using MV-kV DE-CT in radiotherapy contexts. ApproachA toy model comprising a single-line integral through a two-material object was designed for computing basis material signal-to-noise ratio (SNR) using estimation theory. Five dose-matched spectra (three kV, two MV) and three variables were considered: spectral combination, spectral dose allocation, and object material composition. The single-line model was extended to a simulated fan-beam CT acquisition of an anthropomorphic phantom with and without a metal implant. Basis material sinograms were computed and synthesized into virtual monoenergetic images (VMIs). MV-kV and kV-kV VMIs were compared with single-energy images. ResultsThe 80kV-140kV pair typically yielded the best SNRs, but for bone thicknesses greater than 8 cm, the detunedMV-80kV pair surpassed it. Peak MV-kV SNR was achieved with approximately 90% dose allocated to the MV spectrum. For the CT simulations, MV-kV VMIs yielded a higher contrast-to-noise ratio (CNR) than single-energy CT at specific monoenergies. With the metal implant, MV-kV produced a higher maximum CNR and lower minimum root-mean-square-error than kV-kV. ConclusionsThis work quantitatively analyzes MV-kV DE-CT imaging and assesses its potential advantages. This technique may yield improved contrast and accuracy relative to dose-matched single-energy CT or kV-kV DE-CT, depending on object composition.

Purpose/Objective(s)Here we investigate an approach to develop and clinically validate auto-contouring models for lymph node levels and structures of deglutition and mastication in the head and neck. An objective of this work is to provide high quality resources to the scientific community to promote advancement of treatment planning, clinical trial management, and toxicity studies for the head and neck. Materials/MethodsCTs of 145 patients who were irradiated for a head and neck primary malignancy at MD Anderson Cancer Center were retrospectively curated. Data were contoured by radiation oncologists and a resident physician and divided into two separate cohorts. One cohort was used to analyze lymph node levels (IA, IB, II, III, IV, V, RP) and the other used to analyze 17 swallowing and chewing structures. Forty-seven patients were in the lymph node level cohort (training/testing = 32/15). All these patients received definitive radiotherapy without a nodal dissection to minimize anatomic perturbation of the lymph node levels. The remaining 98 patients formed the swallowing/chewing structures cohort (training/testing =78/20). Separate nnUnet models were trained and validated using the separate cohorts. For the lymph node levels, two double blinded studies were used to score preference and clinical acceptability (using a 5-point Likert scale) of AI vs human contours. For the swallowing and chewing structures, clinical acceptability was scored. Quantitative analyses of the test sets were performed for AI vs human contours for all structures using the Dice Similarity Coefficient (DSC) and the 95208 percentile Hausdorff distance (HD95th). ResultsAcross all lymph node levels (IA, IB, II, III, IV, V, RP), median DSC ranged from 0.77 to 0.89 for AI vs manual contours in the testing cohort. Across all lymph node levels, the AI contour was superior to or equally preferred to the manual contours at rates ranging from 75% to 91% in the first blinded study. In the second blinded study, physician preference for the manual vs AI contour was statistically different for only the RP contours (p < 0.01). Thus, there was not a significant difference in clinical acceptability for nodal levels I-V for manual versus AI contours. Across all physician-generated contours, 82% were rated as usable with stylistic to no edits, and across all AI-generated contours, 92% were rated as usable with stylistic to no edits. For the swallowing structures median DSC ranged from 0.86 to 0.96 and was greater than 0.90 for 11/17 structures types. Of the 340 contours in the test set, only 4% required minor edits. ConclusionsAn approach to generate clinically acceptable automated contours for lymph node levels and swallowing and chewing structures in the head and neck was demonstrated. For nodal levels I-V, there was no significant difference in clinical acceptability in manual vs AI contours. Of the two testing cohorts for lymph nodes and swallowing and chewing structures, only 8% and 4% of structures required minor edits, respectively. All testing and training data are being made publicly available on The Cancer Imaging Archive.