Periodic Reporting for period 2 - QUSTom (Quantitative Ultrasound Stochastic Tomography - Revolutionizing breast cancer diagnosis and screening with supercomputing-based radiation-free imaging.)
Berichtszeitraum: 2023-04-01 bis 2024-09-30
QUSTom (Quantitative Ultrasound Stochastic Tomography) is a new European project that aims to introduce a new medical imaging modality based for the first time on ultrasound and supercomputing, which will complement or even replace current techniques that use X-rays such as mammograms. This technology will be completely safe for patients as it does not use ionising radiation. It will also offer superior image quality and better monitoring of tumours, among other advantages.
The algorithms developed to obtain the medical images will offer two types of images simultaneously: the image of the patient's tissue and its associated uncertainty, which shows, pixel by pixel, how reliable the information is. The project also incorporates concepts such as multimodal imaging and real 3D imaging, an unprecedented combination in ultrasound breast imaging.
These algorithms, which will be developed using supercomputers within BSC, will be inspired by others that have proven effective in completely different research areas, such as the analysis of the earth's subsurface.
The project has five other partners: Karlsruher Institut für Technologie, Vall d'Hebron Research Institute, Arctur and the BSC and Imperial College London spin-off FrontWave Imaging, which is aligned with the objectives of this project, as well as Imperial College London itself as an associate partner. Therefore, physicists, engineers, operational experts and radiologists will work together to develop the next generation of radiation-free, accurate and scalable breast cancer diagnostic tools.
QUSTom poses an excellent opportunity to bring ultrasound imaging to the next level. Interestingly enough, the proposed revolution comes not just from an extraordinary imaging device but from the imaging algorithms used to generate unprecedented ultrasound images. Images that we can fairly compare to those obtained with MRI.
The process of understanding, interpreting and configuring the images as a new diagnostic tool will be carried out by Vall d'Hebron Hospital. This new technology will be developed within the framework of the multimodal assessment that the team performs in the clinical care process as an integral part of the diagnosis and monitoring of breast cancer, all in the context of the Breast Pathology Unit.
The algorithms developed to obtain the medical images will offer two types of images simultaneously: the image of the patient's tissue and its associated uncertainty, which shows, pixel by pixel, how reliable the information is. The project also incorporates concepts such as multimodal imaging and real 3D imaging, an unprecedented combination in ultrasound breast imaging.
These algorithms, which will be developed using supercomputers within BSC, will be inspired by others that have proven effective in completely different research areas, such as the analysis of the earth's subsurface.
The project has five other partners: Karlsruher Institut für Technologie, Vall d'Hebron Research Institute, Arctur and the BSC and Imperial College London spin-off FrontWave Imaging, which is aligned with the objectives of this project, as well as Imperial College London itself as an associate partner. Therefore, physicists, engineers, operational experts and radiologists will work together to develop the next generation of radiation-free, accurate and scalable breast cancer diagnostic tools.
QUSTom poses an excellent opportunity to bring ultrasound imaging to the next level. Interestingly enough, the proposed revolution comes not just from an extraordinary imaging device but from the imaging algorithms used to generate unprecedented ultrasound images. Images that we can fairly compare to those obtained with MRI.
The process of understanding, interpreting and configuring the images as a new diagnostic tool will be carried out by Vall d'Hebron Hospital. This new technology will be developed within the framework of the multimodal assessment that the team performs in the clinical care process as an integral part of the diagnosis and monitoring of breast cancer, all in the context of the Breast Pathology Unit.
Our consortium has worked on the validity of variance images obtained with UQ-FWI, performing in-silico and in-vitro experiments to quantitatively control whether high variance values result from highly inaccurate images. The tests have been successful and allow us to proceed with the application for in-vivo tests.
We have also developed novel numerical methods to simulate ultrasound wave propagation in multi-parameter acoustic bodies. The method developed poses full compatibility with the background technology of the project. Initial experiments are also in place to characterise finite-size transducers, such as the piezoelectric components of the 3D-USCT-III device. The usual isotropic radiation approximation is inaccurate for such components, so revising the underlying assumptions is necessary to enable wider azimuth receivers for each transmitting element.
Other activities are related to improving the algorithmic efficiency of our imaging algorithm's wave propagation simulations (i.e. kernel). The kernel accounts for more than 90% of our method's time (cost) of image reconstruction. The software has been upgraded, allowing computing images at more than 5x the original compute time. In addition, the imaging reconstruction algorithm has been successfully deployed in MareNostrum4 and MareNostrum5, two of the fastest supercomputers in Europe. Moreover, the kernels can now run on GPUs, the fastest commodity computing hardware available.
Regarding the USCT3D-III device, we have resorted to using a pre-existing device, which has undergone strict testing prior to being shipped to the hospital to start the clinical validation. A new device will be assembled and potentially include improvements from the feedback obtained from the old one. The testing of the device has included biocompatibility and safety checks in collaboration with certified test labs. Also, we have completed the required documentation for the Spanish Drug Agency (AEMPS) to undergo the validation study in clinical premises. Several in-vitro and in-vivo experiments have been carried out to help assess optimal data acquisition configurations. Those experiments include central frequencies ranging from 0.9 to 2.5 MHz, emission averaging 1 to 1024 repetitions and either 1 or 2 motor positions. Initial simulations have been performed with the 3D reconstruction algorithm to assess the computational feasibility of the challenge we foresee for full-azimuth 3D FWI images of USCT data with no red flags.
After approval from the AEMPS we have gathered data from 59 volunteers at the Vall d'Hebron hospital. The subjects are part of the regular screening program and thus radiologists will have a mammogram for comparison in their evaluation. The data has been safely transferred to BSC storage system for additional processing, the generation of alternative ultrasound images, and the final FWI reconstruction.
In addition, the UQ principles have been applied to X-ray data with promising results, whereas requirements for an upcoming USCT-IV device have been outlined towards improving the device characteristics. One of our partners (FrontWave Imaging) has been selected as one of only 12 finalists of the Innovation Radar Prize 2024 partly because of their QUSTom developments. Last but not least an important effort has been made in presenting the work to the general public by means of communication activities.
We have also developed novel numerical methods to simulate ultrasound wave propagation in multi-parameter acoustic bodies. The method developed poses full compatibility with the background technology of the project. Initial experiments are also in place to characterise finite-size transducers, such as the piezoelectric components of the 3D-USCT-III device. The usual isotropic radiation approximation is inaccurate for such components, so revising the underlying assumptions is necessary to enable wider azimuth receivers for each transmitting element.
Other activities are related to improving the algorithmic efficiency of our imaging algorithm's wave propagation simulations (i.e. kernel). The kernel accounts for more than 90% of our method's time (cost) of image reconstruction. The software has been upgraded, allowing computing images at more than 5x the original compute time. In addition, the imaging reconstruction algorithm has been successfully deployed in MareNostrum4 and MareNostrum5, two of the fastest supercomputers in Europe. Moreover, the kernels can now run on GPUs, the fastest commodity computing hardware available.
Regarding the USCT3D-III device, we have resorted to using a pre-existing device, which has undergone strict testing prior to being shipped to the hospital to start the clinical validation. A new device will be assembled and potentially include improvements from the feedback obtained from the old one. The testing of the device has included biocompatibility and safety checks in collaboration with certified test labs. Also, we have completed the required documentation for the Spanish Drug Agency (AEMPS) to undergo the validation study in clinical premises. Several in-vitro and in-vivo experiments have been carried out to help assess optimal data acquisition configurations. Those experiments include central frequencies ranging from 0.9 to 2.5 MHz, emission averaging 1 to 1024 repetitions and either 1 or 2 motor positions. Initial simulations have been performed with the 3D reconstruction algorithm to assess the computational feasibility of the challenge we foresee for full-azimuth 3D FWI images of USCT data with no red flags.
After approval from the AEMPS we have gathered data from 59 volunteers at the Vall d'Hebron hospital. The subjects are part of the regular screening program and thus radiologists will have a mammogram for comparison in their evaluation. The data has been safely transferred to BSC storage system for additional processing, the generation of alternative ultrasound images, and the final FWI reconstruction.
In addition, the UQ principles have been applied to X-ray data with promising results, whereas requirements for an upcoming USCT-IV device have been outlined towards improving the device characteristics. One of our partners (FrontWave Imaging) has been selected as one of only 12 finalists of the Innovation Radar Prize 2024 partly because of their QUSTom developments. Last but not least an important effort has been made in presenting the work to the general public by means of communication activities.
We have proved the potential of ultrasound tomography for breast cancer diagnosis, by merging innovation in data acquisition apparatus, algorithms and compute infrastructures. Inexpensive UQ has been validated as a novel capability in medical imaging, a clincal validation has been carried out and we are only missing definitive results regarding the reconstructions and their comparison with mammograms. Towards further uptake the validation would need to be finished and the results quantified, beisdes streamlining each of the technologies independently and/or in combination. This requires additional R&D funding as well as support towards commercialization and additional regulatory support (see next chapter)