China
Africa
Explore chapters and articles related to this topic
Uncertainties of IGRT for lung cancer
Published in Jing Cai, Joe Y. Chang, Fang-Fang Yin, Principles and Practice of Image-Guided Radiation Therapy of Lung Cancer, 2017
Irina Vergalasova, Guang Li, Chris R. Kelsey, Hong Ge, Long Huang, Jing Cai
PET imaging has been used as a guide to the physician during GTV delineation. Since their introduction, PET/CT scanners have become widely prevalent in oncology and are crucial for the diagnosis and staging of lung cancer [68]. PET can be particularly helpful to delineate gross disease from atelectasis, though the spatial resolution with PET is limited. Some lung cancer histology's, particularly adenocarcinoma in situ, often appear as ground glass opacities that are only mildly FDG-avid on PET. These are often more challenging to delineate given their less discrete borders. Identification of involved lymph nodes in the mediastinum is challenging and is ideally assessed both clinically (e.g., PET-CT) and pathologically (e.g., mediastinoscopy or endobronchial ultrasound). PET-CT, while more accurate than CT alone, has limitations related to histologic subtype, type of PET-CT scanner, dose of 18F-2-fluoro-deoxy-D-glucose (FDG), and so forth [69]. Lymph nodes known or suspected of harboring disease should be contoured using soft tissue (chest/abdomen) windows. Most mediastinal lymph node stations, except for the AP window (level 5), are easily visualized without intravenous contrast. On the other hand, involved lymph nodes in the hilum are often very difficult to distinguish from blood vessels on non-contrasted scans. Utilizing IV contrast is suggested when hilar lymph nodes are involved.
Decision Analysis
Published in Richard E. Neapolitan, Xia Jiang, Artificial Intelligence, 2018
Richard E. Neapolitan, Xia Jiang
The CT scan can detect mediastinal metastases. The test is not absolutely accurate. Rather, if we let MedMet be a variable whose values are present and absent depending on whether or not mediastinal metastases are present and CTest be a variable whose values are cpos and cneg depending on whether or not the CT scan is positive, we have P(CTest=cpos|MedMet=present)=.82P(CTest=cpos|MedMet=absent)=.19. $$ \begin{array}{l} P(CTest = cpos|MedMet = \,present)\, = \,.82 \hfill \\ P(CTest = cpos|MedMet\, = \,absent) = \,.19. \hfill \\ \end{array} $$
The cases
Published in Chris Schelvan, Annabel Copeman, Jacky Davis, Annmarie Jeanes, Jane Young, Paediatric Radiology for MRCPCH and FRCR, 2020
Chris Schelvan, Annabel Copeman, Jacky Davis, Annmarie Jeanes, Jane Young
Mediastinal masses have a wide differential, and it is useful to consider them according to their anatomical location—superior, anterior, middle, or posterior mediastinum. This can often be determined on the plain X-ray, but most children will proceed to CT or MRI for further assessment.
Few-View CT Image Reconstruction via Least-Squares Methods: Assessment and Optimization
Published in Nuclear Science and Engineering, 2023
Mónica Chillarón Pérez, Vicente E. Vidal, Gumersindo J. Verdú, Gregorio Quintana-Ortí
From the previous results, it was determined that every least-squares method tested provided similar image quality for the same reconstructions. Nevertheless, the results for these particular implementations will be analyzed to conclude which method is the best choice. Table VII presents the metrics results for both of the methods, using different numbers of views. These results correspond to the images shown in Fig. 5. The reference image is a thorax CT image selected from the dataset DeepLesion, focused on the mediastinum area. Even when a reduced number of views is used, the reconstructions obtained have decent quality, but the MSE varies two orders of magnitude from the worst to the best reconstruction, which is significant. In every case, the LSMR images have slightly worse quality, as shown in Table VII. It can be observed that the LSQR method obtains better PSNR and SSIM values in every case, although this is not easily perceived by the human eye as shown in the images of Fig. 5, where the reconstructed images are displayed. The images are displayed focusing on the Hounsfield units (HU) window specified by the dataset authors for this particular image, which goes from −175 to 275 HU. Looking to the zoomed-in regions of the images, it can be observed how in the reconstructions from a reduced number of views (32), some areas are blurred, and the internal structures of the original images are not well preserved. However, the results obtained using 60 or 75 views are very similar, and no loss can be seen.
Mediastinal lymph nodes segmentation using 3D convolutional neural network ensembles and anatomical priors guiding
Published in Computer Methods in Biomechanics and Biomedical Engineering: Imaging & Visualization, 2023
David Bouget, André Pedersen, Johanna Vanel, Haakon O. Leira, Thomas Langø
Using anatomical priors during the training process partly had the expected behaviour, whereby less false-positive segmentation was predicted over the oesophagus and azygos vein. At the same time, side-effects were also witnessed from lower predicted probabilities over lymph nodes close to those anatomical structures and with similar attenuation values. In order to purely favour instance detection recall, anatomical priors might be leveraged as well in post-processing. The larger number of voxels predicted with a high probability to belong to the lymph node class could be refined by applying a mask containing the location of every other know anatomical structure in the mediastinum. The number of false positives per patient would then be drastically reduced, the pixel-wise segmentation over the lymph nodes refined, and the overall patient-wise recall kept high. With such strategies, well-performing models are required for at least 15 to 20 anatomical structures. Considering a standalone model for each anatomical structure, the total processing time for a new CT patient would be forcibly longer yet not detrimental as real-time processing is not a requirement for this modality. This post-processing step could either be performed as a simple masking, or end-to-end through a shallow refinement network. Be it as it may, we believe there is strong potential in further investigating anatomical priors guiding, which would circumvent the need for any refinement or post-processing step. More training samples, and especially the knowledge of more than four other anatomical structures, appear mandatory to proceed.
Development and implementation of a time- and computationally-efficient methodology for reconstructing real-world crashes using finite element modeling to improve crash injury research investigations
Published in Computer Methods in Biomechanics and Biomedical Engineering, 2022
Casey Costa, James P. Gaewsky, Joel D. Stitzel, F. Scott Gayzik, Fang-Chi Hsu, R. Shayn Martin, Anna N. Miller, Ashley A. Weaver
Descriptive statistics were performed for kinematic scores. All injury metrics were assessed for normality using Shapiro-Wilk tests. Student’s t-tests and Wilcoxon rank sum tests were used for normally distributed and non-normally distributed data, respectively, to compare drivers with AIS 3+ thorax injury to those without AIS 3+ thorax injury. Univariate logistic regressions were performed with regional injury metrics and belt force used as independent variables and injury status for AIS 2+ thorax injury, AIS 3+ thorax injury, AIS 3+ lung and mediastinum injury, and AIS 3+ rib fractures used as dependent variables. For logistic regression models that had p-values <0.2, injury risk functions were developed using the logistic regression models. All statistical analyses were conducted in JMP Pro 13 (SAS Institute, Cary, NC).