Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Image Segmentation”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 703 records · Page 39

TU-AB-BRA-06: Texture Feature Reproducibility Between PET/CT and PET/MR Imaging Modalities

Purpose: Radiomics involves the extraction of texture features from different imaging modalities with the purpose of developing models to predict patient treatment outcomes. The purpose of this study is to investigate texture feature reproducibility across [18F]FDG PET/CT and [18F]FDG PET/MR imaging in patients with primary malignancies. Methods: Twenty five prospective patients with solid tumors underwent clinical [18F]FDG PET/CT scan followed by [18F]FDG PET/MR scans. In all patients the lesions were identified using nuclear medicine reports. The images were co-registered and segmented using an in-house auto-segmentation method. Fifty features, based on the intensity histogram, second and high order matrices, were extracted from the segmented regions from both image data sets. One-way random-effects ANOVA model of the intra-class correlation coefficient (ICC) was used to establish texture feature correlations between both data sets. Results: Fifty features were classified based on their ICC values, which were found in the range from 0.1 to 0.86, in three categories: high, intermediate, and low. Ten features extracted from second and high-order matrices showed large ICC ≥ 0.70. Seventeen features presented intermediate 0.5 ≤ ICC ≤ 0.65 and the remaining twenty three presented low ICC ≤ 0.45. Conclusion: Features with large ICC values could be reliable candidates for quantification as they lead to similar results from both imaging modalities. Features with small ICC indicates a lack of correlation. Therefore, the use of these features as a quantitative measure will lead to different assessments of the same lesion depending on the imaging modality from where they are extracted. This study shows the importance of the need for further investigation and standardization of features across multiple imaging modalities.

60 APPLIED LIFE SCIENCES↗

Summary Statistics for Fun Dough Data Acquired at LLNL

Using x-ray computerized tomography (CT), we have characterized the x-ray linear attenuation coefficients (LAC) of a Play Dough{trademark}-like product, Fun Dough{trademark}, designated as PD. Table 1 gives the first-order statistics for each of four CT measurements, estimated with a Gaussian kernel density estimator (KDE) analysis. The mean values of the LAC range from a high of about 2100 LMHU{sub D} at 100kVp to a low of about 1100 LMHU{sub D} at 300kVp. The standard deviation of each measurement is around 1% of the mean. The entropy covers the range from 3.9 to 4.6. Ordinarily, we would model the LAC of the material and compare the modeled values to the measured values. In this case, however, we did not have the composition of the material and therefore did not model the LAC. Using a method recently proposed by Lawrence Livermore National Laboratory (LLNL), we estimate the value of the effective atomic number, Z{sub eff}, to be near 8.5. LLNL prepared about 50mL of the Fun Dough{trademark} in a polypropylene vial and firmly compressed it immediately prior to the x-ray measurements. Still, layers can plainly be seen in the reconstructed images, indicating that the bulk density of the material in the container is affected by voids and bubbles. We used the computer program IMGREC to reconstruct the CT images. The values of the key parameters used in the data capture and image reconstruction are given in this report. Additional details may be found in the experimental SOP and a separate document. To characterize the statistical distribution of LAC values in each CT image, we first isolated an 80% central-core segment of volume elements ('voxels') lying completely within the specimen, away from the walls of the polypropylene vial. All of the voxels within this central core, including those comprised of voids and inclusions, are included in the statistics. We then calculated the mean value, standard deviation and entropy for (a) the four image segments and for (b) their digital gradient images. (A digital gradient image of a given image was obtained by taking the absolute value of the difference between the initial image and that same image offset by one voxel horizontally, parallel to the rows of the x-ray detector array.) The statistics of the initial image of LAC values are called 'first order statistics;' those of the gradient image, 'second order statistics.'

99 GENERAL AND MISCELLANEOUS↗

Summary Statistics for Homemade ?Play Dough? -- Data Acquired at LLNL

Using x-ray computerized tomography (CT), we have characterized the x-ray linear attenuation coefficients (LAC) of a homemade Play Dough{trademark}-like material, designated as PDA. Table 1 gives the first-order statistics for each of four CT measurements, estimated with a Gaussian kernel density estimator (KDE) analysis. The mean values of the LAC range from a high of about 2700 LMHU{sub D} 100kVp to a low of about 1200 LMHUD at 300kVp. The standard deviation of each measurement is around 10% to 15% of the mean. The entropy covers the range from 6.0 to 7.4. Ordinarily, we would model the LAC of the material and compare the modeled values to the measured values. In this case, however, we did not have the detailed chemical composition of the material and therefore did not model the LAC. Using a method recently proposed by Lawrence Livermore National Laboratory (LLNL), we estimate the value of the effective atomic number, Z{sub eff}, to be near 10. LLNL prepared about 50mL of the homemade 'Play Dough' in a polypropylene vial and firmly compressed it immediately prior to the x-ray measurements. We used the computer program IMGREC to reconstruct the CT images. The values of the key parameters used in the data capture and image reconstruction are given in this report. Additional details may be found in the experimental SOP and a separate document. To characterize the statistical distribution of LAC values in each CT image, we first isolated an 80% central-core segment of volume elements ('voxels') lying completely within the specimen, away from the walls of the polypropylene vial. All of the voxels within this central core, including those comprised of voids and inclusions, are included in the statistics. We then calculated the mean value, standard deviation and entropy for (a) the four image segments and for (b) their digital gradient images. (A digital gradient image of a given image was obtained by taking the absolute value of the difference between the initial image and that same image offset by one voxel horizontally, parallel to the rows of the x-ray detector array.) The statistics of the initial image of LAC values are called 'first order statistics;' those of the gradient image, 'second order statistics.'

99 GENERAL AND MISCELLANEOUS↗

SU-F-J-220: Micro-CT Based Quantification of Mouse Brain Vasculature: The Effects of Acquisition Technique and Contrast Material

Purpose: Cognitive impairment from radiation therapy to the brain may be linked to the loss of total blood volume in the brain. To account for brain injury, it is crucial to develop an understanding of blood volume loss as a result of radiation therapy. This study investigates µCT based quantification of mouse brain vasculature, focusing on the effect of acquisition technique and contrast material. Methods: Four mice were scanned on a µCT scanner (Siemens Inveon). The reconstructed voxel size was 18µm3 and all protocols were Hounsfield Unit (HU) calibrated. The mice were injected with 40mg of gold nanoparticles (MediLumine) or 100µl of Exitron 12000 (Miltenyi Biotec). Two acquisition techniques were also performed. A single kVp technique scanned the mouse once using an x-ray beam of 80kVp and segmentation was completed based on a threshold of HU values. The dual kVp technique scanned the mouse twice using 50kVp and 80kVp, this segmentation was based on the ratio of the HU value of the two kVps. After image reconstruction and segmentation, the brain blood volume was determined as a percentage of the total brain volume. Results: For the single kVp acquisition at 80kVp, the brain blood volume had an average of 3.5% for gold and 4.0% for Exitron 12000. Also at 80kVp, the contrast-noise ratio was significantly better for images acquired with the gold nanoparticles (2.0) than for those acquired with the Exitron 12000 (1.4). The dual kVp acquisition shows improved separation of skull from vasculature, but increased image noise. Conclusion: In summary, the effects of acquisition technique and contrast material for quantification of mouse brain vasculature showed that gold nanoparticles produced more consistent segmentation of brain vasculature than Exitron 12000. Also, dual kVp acquisition may improve the accuracy of brain vasculature quantification, although the effect of noise amplification warrants further study.

60 APPLIED LIFE SCIENCES↗

Level-set segmentation of pulmonary nodules in megavolt electronic portal images using a CT prior

Purpose: Pulmonary nodules present unique problems during radiation treatment due to nodule position uncertainty that is caused by respiration. The radiation field has to be enlarged to account for nodule motion during treatment. The purpose of this work is to provide a method of locating a pulmonary nodule in a megavolt portal image that can be used to reduce the internal target volume (ITV) during radiation therapy. A reduction in the ITV would result in a decrease in radiation toxicity to healthy tissue. Methods: Eight patients with nonsmall cell lung cancer were used in this study. CT scans that include the pulmonary nodule were captured with a GE Healthcare LightSpeed RT 16 scanner. Megavolt portal images were acquired with a Varian Trilogy unit equipped with an AS1000 electronic portal imaging device. The nodule localization method uses grayscale morphological filtering and level-set segmentation with a prior. The treatment-time portion of the algorithm is implemented on a graphical processing unit. Results: The method was retrospectively tested on eight cases that include a total of 151 megavolt portal image frames. The method reduced the nodule position uncertainty by an average of 40% for seven out of the eight cases. The treatment phase portion of the method has a subsecond execution time that makes it suitable for near-real-time nodule localization. Conclusions: A method was developed to localize a pulmonary nodule in a megavolt portal image. The method uses the characteristics of the nodule in a prior CT scan to enhance the nodule in the portal image and to identify the nodule region by level-set segmentation. In a retrospective study, the method reduced the nodule position uncertainty by an average of 40% for seven out of the eight cases studied.

60 APPLIED LIFE SCIENCES↗

Gamma-ray imaging with germanium detectors

Externally segmented germanium detectors promise a breakthrough in gamma-ray imaging capabilities while retaining the superb energy resolution of germanium spectrometers. By combining existing position-sensitive detectors with an appropriate code aperture, two-dimensional imaging with 0.2-deg angular resolution becomes practical for a typical balloon experiment. Much finer resolutions are possible with larger separations between detectors and the coded aperture as would be applicable for space-based or lunar-based observatories. Two coaxial germanium detectors divided into five external segments have been fabricated and have undergone extensive performance evaluation and imaging testing in our laboratory. These tests together with detailed Monte Carlo modeling calculations have demonstrated the great promise of this sensor technology for future gamma-ray missions.

Mahoney, W. A.↗

End-to-end Modeling of Lyot Coronagraph for LUVOIR: Sensitivity to Segment-Level Aberrations

Direct imaging and spectroscopic characterization of atmospheres of Earth-like exoplanets require starlight suppression to detect a faint planetary signal at small angular separations. Future space telescopes will require segmentation to survey a statistically significant population of nearby stars systems by increasing aperture size beyond the limits set by traditional monolithic apertures; however, segmentation also represents an additional challenge due to the introduction of temporal pupil discontinuities in terms of both phase and amplitude. For this study, a baseline Lyot coronagraph is used with the Large UV / Optical /Infrared (LUVOIR) mission’s 15-m on-axis, segmented optical telescope assembly (OTA). We present a high-fidelity, end-to-end modeling approach developed as part of the Segmented Coronagraph Design and Analysis (SCDA) group that includes diffractive optical propagations at the 1e-10 contrast level across the entire optical train including the OTA and intermediate re-imaging optics inside the coronagraph instrument. We perform a tolerancing analysis to assess the contrast sensitivity to wavefront perturbations for a set of key telescope terms including segment-level low-order and mid-spatial frequency aberrations. As an input to the end-to-end telescope model, we use two industry-provided segment-level wavefront aberration datasets pertaining: (1) thermal gradients and accelerations (courtesy of Ball-Aerospace and L3-Harris), and (2) telescope dynamics including line-of-sight and segment jitter (courtesy of Lockheed-Martin). Contrast performance outputs of the coronagraph are fed into a mission design reference simulator to assess the scientific yield sensitivity. End-to-end modeling will incorporate performance improvements due to active wavefront sensing and control and/or segment-level metrology to improve system performance.

Optics↗

Image wavelet decomposition and applications

The general problem of computer vision has been investigated for more that 20 years and is still one of the most challenging fields in artificial intelligence. Indeed, taking a look at the human visual system can give us an idea of the complexity of any solution to the problem of visual recognition. This general task can be decomposed into a whole hierarchy of problems ranging from pixel processing to high level segmentation and complex objects recognition. Contrasting an image at different representations provides useful information such as edges. An example of low level signal and image processing using the theory of wavelets is introduced which provides the basis for multiresolution representation. Like the human brain, we use a multiorientation process which detects features independently in different orientation sectors. So, images of the same orientation but of different resolutions are contrasted to gather information about an image. An interesting image representation using energy zero crossings is developed. This representation is shown to be experimentally complete and leads to some higher level applications such as edge and corner finding, which in turn provides two basic steps to image segmentation. The possibilities of feedback between different levels of processing are also discussed.

Treil, N.↗

Left ventricular endocardial surface detection based on real-time 3D echocardiographic data

OBJECTIVE: A new computerized semi-automatic method for left ventricular (LV) chamber segmentation is presented. METHODS: The LV is imaged by real-time three-dimensional echocardiography (RT3DE). The surface detection model, based on level set techniques, is applied to RT3DE data for image analysis. The modified level set partial differential equation we use is solved by applying numerical methods for conservation laws. The initial conditions are manually established on some slices of the entire volume. The solution obtained for each slice is a contour line corresponding with the boundary between LV cavity and LV endocardium. RESULTS: The mathematical model has been applied to sequences of frames of human hearts (volume range: 34-109 ml) imaged by 2D and reconstructed off-line and RT3DE data. Volume estimation obtained by this new semi-automatic method shows an excellent correlation with those obtained by manual tracing (r = 0.992). Dynamic change of LV volume during the cardiac cycle is also obtained. CONCLUSION: The volume estimation method is accurate; edge based segmentation, image completion and volume reconstruction can be accomplished. The visualization technique also allows to navigate into the reconstructed volume and to display any section of the volume.

Non-NASA Center↗

Validation of a novel robot-assisted 3DUS system for real-time planning and guidance of breast interstitial HDR brachytherapy

Purpose: In current clinical practice, there is no integrated 3D ultrasound (3DUS) guidance system clinically available for breast brachytherapy. In this study, the authors present a novel robot-assisted 3DUS system for real-time planning and guidance of breast interstitial high dose rate (HDR) brachytherapy treatment. Methods: For this work, a new computer controlled robotic 3DUS system was built to perform a hybrid motion scan, which is a combination of a 6 cm linear translation with a 30° rotation at both ends. The new 3DUS scanner was designed to fit on a modified Kuske assembly, keeping the current template grid configuration but modifying the frame to allow the mounting of the 3DUS system at several positions. A finer grid was also tested. A user interface was developed to perform image reconstruction, semiautomatic segmentation of the surgical bed as well as catheter reconstruction and tracking. A 3D string phantom was used to validate the geometric accuracy of the reconstruction. The volumetric accuracy of the system was validated with phantoms using magnetic resonance imaging (MRI) and computed tomography (CT) images. In order to accurately determine whether 3DUS can effectively replace CT for treatment planning, the authors have compared the 3DUS catheter reconstruction to the one obtained from CT images. In addition, in agarose-based phantoms, an end-to-end procedure was performed by executing six independent complete procedures with both 14 and 16 catheters, and for both standard and finer Kuske grids. Finally, in phantoms, five end-to-end procedures were performed with the final CT planning for the validation of 3DUS preplanning. Results: The 3DUS acquisition time is approximately 10 s. A paired Student t-test showed that there was no statistical significant difference between known and measured values of string separations in each direction. Both MRI and CT volume measurements were not statistically different from 3DUS volume (Student t-test: p > 0.05) and they were significantly correlated to 3DUS measurement (Pearson test: MRI p < 0.05 and CT p < 0.001). The mean angular separation distance between catheter trajectories segmented from 3DUS and CT images was 0.42° ± 0.24°, while the maximum and mean trajectory separations were 0.51 ± 0.19 and 0.37 ± 0.17 mm, respectively. Overall, the new finer grid has performed significantly better in terms of dosimetric indices. The planning target volume dosimetric indices were not found statistically different between 3DUS and CT planning (Student t-test, p > 0.05). Both the skin and the pectoral muscle dosimetric indices were within ABS guidelines. Conclusions: A novel robot-assisted 3DUS system was designed and validated. To their knowledge, this is the first system capable of performing real-time guidance and planning of breast multicatheter HDR brachytherapy treatments. Future investigation will test the feasibility of using the system in the clinic and for permanent breast brachytherapy.

60 APPLIED LIFE SCIENCES↗

PTT System Performance Evaluation in 3-D Imaging of AM Components

Additive manufacturing (AM, or 3D printing) for commercial nuclear energy applications is an emerging method for cost-efficient manufacturing aimed at replacing aging nuclear reactor parts and reducing costs for new construction. Because of the geometry of metallic structures of interest for nuclear applications, which consist of planar primitives with no symmetry of revolution, limited options are available for non-destructive evaluation (NDE) either during or post manufacturing. Known material flaws in AM include low density regions consisting of nonsintered powder, which have to be detected to ensure the safety of long-term performance nuclear reactor components. As a solution to NDE of AM, we are developing pulsed thermal tomography (PTT) models and depth inversion algorithms for 3D imaging. PTT has many advantages because the method is non-contact and allows for in-service, NDE of AM nuclear reactor parts. By analyzing transients of surface temperature response due to internal thermal resistances, one can obtain 3D reconstructions of material effusivity using a unique inversion algorithm developed at Argonne. This study investigates the capability of PTT in imaging several AM structures fabricated from Inconel 718 (IN718) feedstock. The AM structures have complex geometry, but do not have calibrated internal defects. The objective is to determine PTT settings, such as the total integration time, for imaging of representative AM structures. A new FLIR X8500sc IR camera with higher spatial resolution (1280x1024 at 181Hz frame rate) was integrated into the PTT laboratory setup. Software tools were developed to ensure compatibility of new camera files with legacy effusivity reconstruction codes. The new camera was used to obtain 3-D reconstructions of IN718 nozzle plate and a particle filter plate. The results demonstrate the capability of PTT in 3- D imaging of complex geometry AM structures. Further, increased resolution imaging is achieved compared to prior results using older FLIR SC4000 IR camera. In addition, preliminary studies were conducted on developing image processing algorithms for automated detection of flaws. Initial demonstrations are shown for segmentation of PTT images of flat bottom hole (FBH) specimens. Imaging of FBH specimens was described in a prior report. Intermediate results demonstrated in this report will be developed further in detection of flaws in AM specimens.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

Automatic digital image registration

This paper introduces a general procedure for automatic registration of two images which may have translational, rotational, and scaling differences. This procedure involves (1) segmentation of the images, (2) isolation of dominant objects from the images, (3) determination of corresponding objects in the two images, and (4) estimation of transformation parameters using the center of gravities of objects as control points. An example is given which uses this technique to register two images which have translational, rotational, and scaling differences.

Goshtasby, A.↗

Technical Note: SCUDA: A software platform for cumulative dose assessment

Purpose: Accurate tracking of anatomical changes and computation of actually delivered dose to the patient are critical for successful adaptive radiation therapy (ART). Additionally, efficient data management and fast processing are practically important for the adoption in clinic as ART involves a large amount of image and treatment data. The purpose of this study was to develop an accurate and efficient Software platform for CUmulative Dose Assessment (SCUDA) that can be seamlessly integrated into the clinical workflow. Methods: SCUDA consists of deformable image registration (DIR), segmentation, dose computation modules, and a graphical user interface. It is connected to our image PACS and radiotherapy informatics databases from which it automatically queries/retrieves patient images, radiotherapy plan, beam data, and daily treatment information, thus providing an efficient and unified workflow. For accurate registration of the planning CT and daily CBCTs, the authors iteratively correct CBCT intensities by matching local intensity histograms during the DIR process. Contours of the target tumor and critical structures are then propagated from the planning CT to daily CBCTs using the computed deformations. The actual delivered daily dose is computed using the registered CT and patient setup information by a superposition/convolution algorithm, and accumulated using the computed deformation fields. Both DIR and dose computation modules are accelerated by a graphics processing unit. Results: The cumulative dose computation process has been validated on 30 head and neck (HN) cancer cases, showing 3.5 ± 5.0 Gy (mean±STD) absolute mean dose differences between the planned and the actually delivered doses in the parotid glands. On average, DIR, dose computation, and segmentation take 20 s/fraction and 17 min for a 35-fraction treatment including additional computation for dose accumulation. Conclusions: The authors developed a unified software platform that provides accurate and efficient monitoring of anatomical changes and computation of actually delivered dose to the patient, thus realizing an efficient cumulative dose computation workflow. Evaluation on HN cases demonstrated the utility of our platform for monitoring the treatment quality and detecting significant dosimetric variations that are keys to successful ART.

60 APPLIED LIFE SCIENCES↗

GEONEX: Progressive Conditional Generative Adversarial Training Using Transfer learning

Obtaining accurate segmentation on large scale images is an open problem in deep learning. The main problem is the amount of labeled data that exists for large scale images. Traditionally, the common solution to this problem is to crop the large images into smaller images to increase the amount of available data and train a Conditional Generative Adversarial Network (CGAN). CGANs are currently the state of the art in image to image translation and provide better accuracy than the traditional method of training a encoder based conv-net architecture to minimize the loss at each pixel. This method can produce noisy and discontinuous images with inaccurate results. We seek to solve this problem by utilizing the concepts of transfer learning and progressive training to create a CGAN that can segment large scale images with a limited amount of labeled data. In transfer learning we recognize that many learned features are applicable to many classes from multiple domains. This introduces the concept of feature reusability, which is the basis for finetuning. Progressive training got its start in training models on the same images at different resolutions. In this work we instead train a GAN on increasing image scales by transferring the weights from the smaller scales to the larger scales. The learned features at the smaller scales are continually reused and applied to larger scales to create a CGAN that can perform accurate segmentation on large scale images. We apply this method to detect building footprints on very high-resolution overhead imagery (e.g Digital Globe and high resolution airborne platforms).

GEONEX↗

SU-C-201-06: Utility of Quantitative 3D SPECT/CT Imaging in Patient Specific Internal Dosimetry of 153-Samarium with GATE Monte Carlo Package

Purpose: Patient-specific 3-dimensional (3D) internal dosimetry in targeted radionuclide therapy is essential for efficient treatment. Two major steps to achieve reliable results are: 1) generating quantitative 3D images of radionuclide distribution and attenuation coefficients and 2) using a reliable method for dose calculation based on activity and attenuation map. In this research, internal dosimetry for 153-Samarium (153-Sm) was done by SPECT-CT images coupled GATE Monte Carlo package for internal dosimetry. Methods: A 50 years old woman with bone metastases from breast cancer was prescribed 153-Sm treatment (Gamma: 103keV and beta: 0.81MeV). A SPECT/CT scan was performed with the Siemens Simbia-T scanner. SPECT and CT images were registered using default registration software. SPECT quantification was achieved by compensating for all image degrading factors including body attenuation, Compton scattering and collimator-detector response (CDR). Triple energy window method was used to estimate and eliminate the scattered photons. Iterative ordered-subsets expectation maximization (OSEM) with correction for attenuation and distance-dependent CDR was used for image reconstruction. Bilinear energy mapping is used to convert Hounsfield units in CT image to attenuation map. Organ borders were defined by the itk-SNAP toolkit segmentation on CT image. GATE was then used for internal dose calculation. The Specific Absorbed Fractions (SAFs) and S-values were reported as MIRD schema. Results: The results showed that the largest SAFs and S-values are in osseous organs as expected. S-value for lung is the highest after spine that can be important in 153-Sm therapy. Conclusion: We presented the utility of SPECT-CT images and Monte Carlo for patient-specific dosimetry as a reliable and accurate method. It has several advantages over template-based methods or simplified dose estimation methods. With advent of high speed computers, Monte Carlo can be used for treatment planning on a day to day basis.

60 APPLIED LIFE SCIENCES↗

Block segmentation in feature space for realtime object detection in high granularity images

Computer vision has applications in object detection, image recognition and classification, and object tracking. One of the challenges of computer vision is the presence of useful information at multiple distance scales. Filtering techniques may sacrifice details at small scales in order to prioritize the analysis of large-scale features of the image. We present a strategy for coarse-graining multidimensional data while maintaining fine-grained detail for subsequent analysis. The algorithm is based on fixed-size block segmentation in the feature space. We apply this strategy to solve the long-standing challenge of detecting particle trajectories at the Large Hadron Collider in real time.

Computer vision↗

Using AI Segmentation Models to Improve Foreign Body Detection and Triage from Ultrasound Images

Medical imaging can be a critical tool for triaging casualties in trauma situations. In remote or military medicine scenarios, triage is essential for identifying how to use limited resources or prioritize evacuation for the most serious cases. Ultrasound imaging, while portable and often available near the point of injury, can only be used for triage if images are properly acquired, interpreted, and objectively triage scored. Here, we detail how AI segmentation models can be used for improving image interpretation and objective triage evaluation for a medical application focused on foreign bodies embedded in tissues at variable distances from critical neurovascular features. Ultrasound images previously collected in a tissue phantom with or without neurovascular features were labeled with ground truth masks. These image sets were used to train two different segmentation AI frameworks: YOLOv7 and U-Net segmentation models. Overall, both approaches were successful in identifying shrapnel in the image set, with U-Net outperforming YOLOv7 for single-class segmentation. Both segmentation models were also evaluated with a more complex image set containing shrapnel, artery, vein, and nerve features. YOLOv7 obtained higher precision scores across multiple classes whereas U-Net achieved higher recall scores. Using each AI model, a triage distance metric was adapted to measure the proximity of shrapnel to the nearest neurovascular feature, with U-Net more closely mirroring the triage distances measured from ground truth labels. Overall, the segmentation AI models were successful in detecting shrapnel in ultrasound images and could allow for improved injury triage in emergency medicine scenarios.

47 OTHER INSTRUMENTATION↗