Periodic Reporting for period 3 - MUSE (Multi-perspective Ultrasound Strain Imaging & Elastography)
Reporting period: 2021-02-01 to 2022-07-31
Cardiovascular disease is still the leading cause of death worldwide. In this project we focus on the heart and the aorta. Current clinical decision making is based on several biomedical measurements and imaging, yielding geometrical and functional parameters that are compared to large population statistics (evidence-based medicine). In our group, we strive to assist clinical decision making based on accurate, functional measurements, and patients-specific models of the diseased organ.
Ultrasound imaging is a widely used imaging technique, one of the workhorses in the clinic, that is frequently used for imaging of the cardiovascular system. Ultrasound (US) has several advantages, it is non-invasive, relatively cheap, can be used at the patient’s bedside, and most importantly: is known for its high temporal and spatial resolution. Unfortunately, US is also known for its drawbacks, which are a limited field-of-view and contrast. The latter is anisotropic, as well as the resolution. In this project, we aim to tackle these physical limitations by developing so-called multiperspective US imaging techniques.
The project has four objectives:
1) developing a novel imaging platform, including
2) new methods for functional measurements on organs;
3) create patient-specific models based on these new functional imaging techniques and
4) perform extensive experimental verification and in vivo validation of the methods proposed.
In multiperspective ultrasound imaging (MPUS), we use multiple transducers rather than a single one as is the case in conventional US imaging. Systems are developed that can perform MPUS in 2-D and 3D. Algorithms are developed that can fuse the image data obtained with both transducers and/or reconstruct high quality images (higher contrast, improved resolution), with a larger field-of-view. With these new systems it is crucial to know where the probes are, a problem that is tackled using both dedicated US acquisition and image analysis techniques.
Based on the improved MPUS images, new segmentation and elastography techniques have been developed. Thanks to the availability of MPUS data, segmentation and motion estimation have improved significantly or has become feasible at all (in case of the aorta). Moreover, the data can be used to improve patient-specific models of the aorta that are currently developed to assess the mechanical state of the vessel, parameters that cannot be assessed via measurements.
To translate these new methods from bench (design/development phase) to bedside (clinical pilots), simulations and setups were introduced to test and verify the performance of the imaging techniques proposed, and in vivo pilot studies were set up. A simulation framework that can produce realistic MPUS images with a known ground truth on geometry or motion was developed, as well as experimental setups that allow testing of the new MPUS systems under more realistic conditions (i.e. biological tissue). Finally, volunteer and patient pilot studies have been organized, that allow for in vivo validation.
Ultrasound imaging is a widely used imaging technique, one of the workhorses in the clinic, that is frequently used for imaging of the cardiovascular system. Ultrasound (US) has several advantages, it is non-invasive, relatively cheap, can be used at the patient’s bedside, and most importantly: is known for its high temporal and spatial resolution. Unfortunately, US is also known for its drawbacks, which are a limited field-of-view and contrast. The latter is anisotropic, as well as the resolution. In this project, we aim to tackle these physical limitations by developing so-called multiperspective US imaging techniques.
The project has four objectives:
1) developing a novel imaging platform, including
2) new methods for functional measurements on organs;
3) create patient-specific models based on these new functional imaging techniques and
4) perform extensive experimental verification and in vivo validation of the methods proposed.
In multiperspective ultrasound imaging (MPUS), we use multiple transducers rather than a single one as is the case in conventional US imaging. Systems are developed that can perform MPUS in 2-D and 3D. Algorithms are developed that can fuse the image data obtained with both transducers and/or reconstruct high quality images (higher contrast, improved resolution), with a larger field-of-view. With these new systems it is crucial to know where the probes are, a problem that is tackled using both dedicated US acquisition and image analysis techniques.
Based on the improved MPUS images, new segmentation and elastography techniques have been developed. Thanks to the availability of MPUS data, segmentation and motion estimation have improved significantly or has become feasible at all (in case of the aorta). Moreover, the data can be used to improve patient-specific models of the aorta that are currently developed to assess the mechanical state of the vessel, parameters that cannot be assessed via measurements.
To translate these new methods from bench (design/development phase) to bedside (clinical pilots), simulations and setups were introduced to test and verify the performance of the imaging techniques proposed, and in vivo pilot studies were set up. A simulation framework that can produce realistic MPUS images with a known ground truth on geometry or motion was developed, as well as experimental setups that allow testing of the new MPUS systems under more realistic conditions (i.e. biological tissue). Finally, volunteer and patient pilot studies have been organized, that allow for in vivo validation.
Systems have been developed to perform multiperspective US imaging using both high-end clinical scanners and so-called open-source US systems. The latter were used to develop new US acquisition schemes that allow for multi-perspective imaging at high frame rates. Experimental imaging was tested using two-dimensional and three-dimensional transducers, for imaging dynamic porcine ex-vivo aortas and hearts. Moreover, first multi-perspective ultrafast 3D US acquisitions have been performed using two matrix array transducers ex vivo and acquisition sequences are optimized for low depths. Probe tracking based on high-frame rate imaging was developed.
Algorithms for fully automated registration have been developed, image reconstruction and smart image fusion and spatio-temporal registration algorithms to benefit the most of the data available. A temporal and registration algorithm for multi-perspective cardiac images has been created and tested in 20 healthy volunteers, which registers the images based on their motion patterns that were extracted using a singular value decomposition-based method. This newly designed spatial registration algorithm has been adapted for application in multi-perspective AAA images obtained from different angles and was tested in volunteers and AAA patients. Moreover, the performance of the proximal-distal spatial registration algorithm was determined by comparing the segmentations of the fused AAA images to CT-derived segmentations in AAA patients.
3D segmentation algorithms for multi-perspective cardiac has been implemented using a priorly trained 2D deep learning model, which was extended towards 3D, in combination with a statistical shape model and 3D wavelets. For aorta/AAA imaging, a fully automated segmentation approach has been developed (based on a Kalman-star approach) and was applied to a large group of patient data, and to multi-perspective ultrasound images. The approach has recently been improved using a novel DL approach.
3D motion estimation has been optimized using improved motion estimation algorithms and smart acquisitions schemes to improve the 3D frame rate. 3D MPUS strain estimation has been compared to the gold standard (MRI) in healthy volunteers and will be extended to a small study in AoS patients. The potential of improved AAA wall motion tracking after temporal registration has been demonstrated and was compared to conventional 3D speckle tracking. Ultrafast 3D RF data obtained with multiple probes can be reconstructed at a fast pace to perform 3D RF speckle tracking. The ex vivo study shows the vast improvement in motion tracking using this approach in both porcine hearts and aortas.
Methods for geometry assessment and mechanical characterization of AAAs have been developed. A fully automated modeling pipeline for global assessment of mechanical properties of AAAs has been developed, tested on simulations, and is now extended to MPUS imaging. Multi-perspective RF and DICOM imaging was performed on 3D printed phantoms and in porcine aortas for in vitro validation. Pilot in vivo studies in volunteers and patients have been initiated and are currently ongoing for clinical introduction and validation of the methods proposed.
Algorithms for fully automated registration have been developed, image reconstruction and smart image fusion and spatio-temporal registration algorithms to benefit the most of the data available. A temporal and registration algorithm for multi-perspective cardiac images has been created and tested in 20 healthy volunteers, which registers the images based on their motion patterns that were extracted using a singular value decomposition-based method. This newly designed spatial registration algorithm has been adapted for application in multi-perspective AAA images obtained from different angles and was tested in volunteers and AAA patients. Moreover, the performance of the proximal-distal spatial registration algorithm was determined by comparing the segmentations of the fused AAA images to CT-derived segmentations in AAA patients.
3D segmentation algorithms for multi-perspective cardiac has been implemented using a priorly trained 2D deep learning model, which was extended towards 3D, in combination with a statistical shape model and 3D wavelets. For aorta/AAA imaging, a fully automated segmentation approach has been developed (based on a Kalman-star approach) and was applied to a large group of patient data, and to multi-perspective ultrasound images. The approach has recently been improved using a novel DL approach.
3D motion estimation has been optimized using improved motion estimation algorithms and smart acquisitions schemes to improve the 3D frame rate. 3D MPUS strain estimation has been compared to the gold standard (MRI) in healthy volunteers and will be extended to a small study in AoS patients. The potential of improved AAA wall motion tracking after temporal registration has been demonstrated and was compared to conventional 3D speckle tracking. Ultrafast 3D RF data obtained with multiple probes can be reconstructed at a fast pace to perform 3D RF speckle tracking. The ex vivo study shows the vast improvement in motion tracking using this approach in both porcine hearts and aortas.
Methods for geometry assessment and mechanical characterization of AAAs have been developed. A fully automated modeling pipeline for global assessment of mechanical properties of AAAs has been developed, tested on simulations, and is now extended to MPUS imaging. Multi-perspective RF and DICOM imaging was performed on 3D printed phantoms and in porcine aortas for in vitro validation. Pilot in vivo studies in volunteers and patients have been initiated and are currently ongoing for clinical introduction and validation of the methods proposed.
So far, most objectives have been reached and beyond. A fully automated pipeline for anatomical and functional measurements using MPUS has been established and has been tested extensively. 2D and 3D MPUS imaging shows vast improvements in contrast and field-of-view, as well as improved accuracy and precision of deformation measurements. In the final period of this project, all ex vivo analyses will be finalized and the pilots in AoS and AAA patients concluded. The MPUS data will be compared to single perspective US, and to the gold standard, tagged MRI and contrast-enhanced CT imaging, respectively. The project has moved beyond its own state-of-the-art by the development of a 3D ultrafast MPUS imaging system, including bistatic imaging functionality and probe localization, which improves the functionality of ultrasound even further. Final analyses on ex and in vivo ultrafast data are currently ongoing, but current results show vast improvements