DM
D. Maresca
info
Please Note
<p>This page displays the records of the person named above and is not linked to a unique person identifier. This record may need to be merged to a profile.</p>
2 records found
1
Ultrasound imaging is a non-invasive imaging method, which uses ultrasound waves to produce images of the internal organs in the body. Since sound waves can interfere with each other, the ultrasound images are diffraction limited. Ultrasound localization microscopy (ULM) is a processing technique, which is able to bypass this diffraction limit by localizing individual spatially isolated contrast agent microbubble (MB)s in the low resolution ultrasound frames. These MBs acts as a point source and appears as a blurry point in the ultrasound frame also known as Point spread function (PSF) whose centroids can be localized with a precision beyond the diffraction limit. By localizing these MBs and tracking their paths over thousands of consecutive ultrasound frames and accumulating their tracks, a super resolution image of the vasculature can be reconstructed. While these super resolution images significant benefits to biomedical applications, they require long acquisition times.
This thesis investigates whether the deep learning model DBlink, a bidirectional convolutional long short-term memory (LSTM) with a Convolutional Neural Networks (CNN) head can reduce the long acquisition time of ULM. An in silico rat renal arterial tree was simulated to provide the data required for training and evaluating the deep learning model. Two different input type were explored for the DBlink model: Localization maps (summed frames of super resolved localizations) and velocity tracks (maps containing super resolved velocity tracks) of the MB. The effect of different receptive field (RFd) sizes were also examined.
The performance of the DBlink model was compared to the conventional ULM method and showed a reduced acquisition time of 8.7 seconds for large radii vessels in silico. However, the reduction in acquisition time diminishes for small radii vessel, where the passage of MB is still limited by low blood flow rate. Although DBlink reduces acquisition time, it introduces hallucinations in the reconstruction of vessels, especially in regions containing dense small vessels.
Overall, this research highlights the use of the deep learning model DBlink in ULM and the use of different input type to reduce the acquisition time in ULM. However, further research is needed in order to apply this deep learning model in vivo.
...
This thesis investigates whether the deep learning model DBlink, a bidirectional convolutional long short-term memory (LSTM) with a Convolutional Neural Networks (CNN) head can reduce the long acquisition time of ULM. An in silico rat renal arterial tree was simulated to provide the data required for training and evaluating the deep learning model. Two different input type were explored for the DBlink model: Localization maps (summed frames of super resolved localizations) and velocity tracks (maps containing super resolved velocity tracks) of the MB. The effect of different receptive field (RFd) sizes were also examined.
The performance of the DBlink model was compared to the conventional ULM method and showed a reduced acquisition time of 8.7 seconds for large radii vessels in silico. However, the reduction in acquisition time diminishes for small radii vessel, where the passage of MB is still limited by low blood flow rate. Although DBlink reduces acquisition time, it introduces hallucinations in the reconstruction of vessels, especially in regions containing dense small vessels.
Overall, this research highlights the use of the deep learning model DBlink in ULM and the use of different input type to reduce the acquisition time in ULM. However, further research is needed in order to apply this deep learning model in vivo.
...
Ultrasound imaging is a non-invasive imaging method, which uses ultrasound waves to produce images of the internal organs in the body. Since sound waves can interfere with each other, the ultrasound images are diffraction limited. Ultrasound localization microscopy (ULM) is a processing technique, which is able to bypass this diffraction limit by localizing individual spatially isolated contrast agent microbubble (MB)s in the low resolution ultrasound frames. These MBs acts as a point source and appears as a blurry point in the ultrasound frame also known as Point spread function (PSF) whose centroids can be localized with a precision beyond the diffraction limit. By localizing these MBs and tracking their paths over thousands of consecutive ultrasound frames and accumulating their tracks, a super resolution image of the vasculature can be reconstructed. While these super resolution images significant benefits to biomedical applications, they require long acquisition times.
This thesis investigates whether the deep learning model DBlink, a bidirectional convolutional long short-term memory (LSTM) with a Convolutional Neural Networks (CNN) head can reduce the long acquisition time of ULM. An in silico rat renal arterial tree was simulated to provide the data required for training and evaluating the deep learning model. Two different input type were explored for the DBlink model: Localization maps (summed frames of super resolved localizations) and velocity tracks (maps containing super resolved velocity tracks) of the MB. The effect of different receptive field (RFd) sizes were also examined.
The performance of the DBlink model was compared to the conventional ULM method and showed a reduced acquisition time of 8.7 seconds for large radii vessels in silico. However, the reduction in acquisition time diminishes for small radii vessel, where the passage of MB is still limited by low blood flow rate. Although DBlink reduces acquisition time, it introduces hallucinations in the reconstruction of vessels, especially in regions containing dense small vessels.
Overall, this research highlights the use of the deep learning model DBlink in ULM and the use of different input type to reduce the acquisition time in ULM. However, further research is needed in order to apply this deep learning model in vivo.
This thesis investigates whether the deep learning model DBlink, a bidirectional convolutional long short-term memory (LSTM) with a Convolutional Neural Networks (CNN) head can reduce the long acquisition time of ULM. An in silico rat renal arterial tree was simulated to provide the data required for training and evaluating the deep learning model. Two different input type were explored for the DBlink model: Localization maps (summed frames of super resolved localizations) and velocity tracks (maps containing super resolved velocity tracks) of the MB. The effect of different receptive field (RFd) sizes were also examined.
The performance of the DBlink model was compared to the conventional ULM method and showed a reduced acquisition time of 8.7 seconds for large radii vessels in silico. However, the reduction in acquisition time diminishes for small radii vessel, where the passage of MB is still limited by low blood flow rate. Although DBlink reduces acquisition time, it introduces hallucinations in the reconstruction of vessels, especially in regions containing dense small vessels.
Overall, this research highlights the use of the deep learning model DBlink in ULM and the use of different input type to reduce the acquisition time in ULM. However, further research is needed in order to apply this deep learning model in vivo.
The use of ultrafast Ultrasound Localization Microscopy (uULM) is a promising technique for obtaining images with a very high resolution. This technique is based on the localization of subwavelength intravascular microbubbles resonating under ultrasound stimulus. Pre-clinical and post-clinical ultrasound applications, such as tumour, microvascular or stroke imaging, are numerous.
This research investigates the theoretical precision limit of uULM in localizing a moving subwavelength scatterer. A 1D transducer array and the scatterer are simulated with the Vantage Research Ultrasound Simulator (Verasonics, Kirkland, WA, USA). A large beam is transmitted through the medium by the transducer array and hits the scatterer. As a sub-wavelength scatterer radiates as an omni-directional pressure field, transducer arrays of finite apertures receive a part of this spherical wave back as parabolas (the radio-frequency data, or RF-data), from which beamformed images (the BF-data) can be reconstructed. The intensity of the RF-data is given in arbitrary units, while the intensity for the BF-data is indicated in decibels. Both the RF-data and BF-data are influenced by different sources of noise, such as jitter and false peaks, which can be modelled with zero-mean white Gaussian noise. In this manuscript, the z-axis is defined to be the direction away from the probe and the x-axis is defined to be colinear to the piezoelectric elements. In Figure 1, an illustration of this process, called Plane Wave Imaging, with the given directions is shown.
The localization precision in the x- and z-directions is computed for different intensities, different signal-to-noise ratios (SNRs), and different depths z for both the RF-data and the BF-data. For the RF-data, there is a clear relation between the depth of the scatterer and the localization precision: if the scatterer is moving further away, the minimum standard deviation decreases due to attenuation in tissues. For example, for a depth of 11 mm and an SNR of 29 dB, one is able to localize the microbubble with a precision of 10 nm in the x-direction and 50 nm in the z-direction based on the RF-data, while this decreases to respectively 1 μm and 0.5 mm for a depth of 21 mm.
The BF-based precision limits are less dependent on depth: for different depths, the limits remain approximately the same, being 0.5 μm in the x-direction and 1 μm in the z-direction for an intensity of 20 dB and an SNR of 29 dB. A remarkable result is that the localization using the radio-frequency data is more precise compared to the beamformed images if the scatterer is close to the transducer array. However, after a certain depth, the BF-based localization surpasses the RF-based one. This difference in precision is due to the beamforming process: to translate radio-frequency data into a readable image, one needs to sum the energy scattered back and select that value as the pixel intensity for the final image. This process, called beamforming, can be done on the fly or in post-process. In this research, it is complex to compare the initial values, being the SNR, the maximum intensity value, and the dependence on depth, between the RF- and BF-data.
The localization precision for the RF-data and for the BF-data reacts similarly to changes in the amount of noise. For a high SNR, the position of the scatterer can be determined more precisely compared to a situation with a low SNR. For example, the RF-data reveals a localization precision of 1 nm in the x-direction for an SNR of 33 dB, while an SNR of 23 dB results in a precision of 10 mm for the same intensity and depth. ...
This research investigates the theoretical precision limit of uULM in localizing a moving subwavelength scatterer. A 1D transducer array and the scatterer are simulated with the Vantage Research Ultrasound Simulator (Verasonics, Kirkland, WA, USA). A large beam is transmitted through the medium by the transducer array and hits the scatterer. As a sub-wavelength scatterer radiates as an omni-directional pressure field, transducer arrays of finite apertures receive a part of this spherical wave back as parabolas (the radio-frequency data, or RF-data), from which beamformed images (the BF-data) can be reconstructed. The intensity of the RF-data is given in arbitrary units, while the intensity for the BF-data is indicated in decibels. Both the RF-data and BF-data are influenced by different sources of noise, such as jitter and false peaks, which can be modelled with zero-mean white Gaussian noise. In this manuscript, the z-axis is defined to be the direction away from the probe and the x-axis is defined to be colinear to the piezoelectric elements. In Figure 1, an illustration of this process, called Plane Wave Imaging, with the given directions is shown.
The localization precision in the x- and z-directions is computed for different intensities, different signal-to-noise ratios (SNRs), and different depths z for both the RF-data and the BF-data. For the RF-data, there is a clear relation between the depth of the scatterer and the localization precision: if the scatterer is moving further away, the minimum standard deviation decreases due to attenuation in tissues. For example, for a depth of 11 mm and an SNR of 29 dB, one is able to localize the microbubble with a precision of 10 nm in the x-direction and 50 nm in the z-direction based on the RF-data, while this decreases to respectively 1 μm and 0.5 mm for a depth of 21 mm.
The BF-based precision limits are less dependent on depth: for different depths, the limits remain approximately the same, being 0.5 μm in the x-direction and 1 μm in the z-direction for an intensity of 20 dB and an SNR of 29 dB. A remarkable result is that the localization using the radio-frequency data is more precise compared to the beamformed images if the scatterer is close to the transducer array. However, after a certain depth, the BF-based localization surpasses the RF-based one. This difference in precision is due to the beamforming process: to translate radio-frequency data into a readable image, one needs to sum the energy scattered back and select that value as the pixel intensity for the final image. This process, called beamforming, can be done on the fly or in post-process. In this research, it is complex to compare the initial values, being the SNR, the maximum intensity value, and the dependence on depth, between the RF- and BF-data.
The localization precision for the RF-data and for the BF-data reacts similarly to changes in the amount of noise. For a high SNR, the position of the scatterer can be determined more precisely compared to a situation with a low SNR. For example, the RF-data reveals a localization precision of 1 nm in the x-direction for an SNR of 33 dB, while an SNR of 23 dB results in a precision of 10 mm for the same intensity and depth. ...
The use of ultrafast Ultrasound Localization Microscopy (uULM) is a promising technique for obtaining images with a very high resolution. This technique is based on the localization of subwavelength intravascular microbubbles resonating under ultrasound stimulus. Pre-clinical and post-clinical ultrasound applications, such as tumour, microvascular or stroke imaging, are numerous.
This research investigates the theoretical precision limit of uULM in localizing a moving subwavelength scatterer. A 1D transducer array and the scatterer are simulated with the Vantage Research Ultrasound Simulator (Verasonics, Kirkland, WA, USA). A large beam is transmitted through the medium by the transducer array and hits the scatterer. As a sub-wavelength scatterer radiates as an omni-directional pressure field, transducer arrays of finite apertures receive a part of this spherical wave back as parabolas (the radio-frequency data, or RF-data), from which beamformed images (the BF-data) can be reconstructed. The intensity of the RF-data is given in arbitrary units, while the intensity for the BF-data is indicated in decibels. Both the RF-data and BF-data are influenced by different sources of noise, such as jitter and false peaks, which can be modelled with zero-mean white Gaussian noise. In this manuscript, the z-axis is defined to be the direction away from the probe and the x-axis is defined to be colinear to the piezoelectric elements. In Figure 1, an illustration of this process, called Plane Wave Imaging, with the given directions is shown.
The localization precision in the x- and z-directions is computed for different intensities, different signal-to-noise ratios (SNRs), and different depths z for both the RF-data and the BF-data. For the RF-data, there is a clear relation between the depth of the scatterer and the localization precision: if the scatterer is moving further away, the minimum standard deviation decreases due to attenuation in tissues. For example, for a depth of 11 mm and an SNR of 29 dB, one is able to localize the microbubble with a precision of 10 nm in the x-direction and 50 nm in the z-direction based on the RF-data, while this decreases to respectively 1 μm and 0.5 mm for a depth of 21 mm.
The BF-based precision limits are less dependent on depth: for different depths, the limits remain approximately the same, being 0.5 μm in the x-direction and 1 μm in the z-direction for an intensity of 20 dB and an SNR of 29 dB. A remarkable result is that the localization using the radio-frequency data is more precise compared to the beamformed images if the scatterer is close to the transducer array. However, after a certain depth, the BF-based localization surpasses the RF-based one. This difference in precision is due to the beamforming process: to translate radio-frequency data into a readable image, one needs to sum the energy scattered back and select that value as the pixel intensity for the final image. This process, called beamforming, can be done on the fly or in post-process. In this research, it is complex to compare the initial values, being the SNR, the maximum intensity value, and the dependence on depth, between the RF- and BF-data.
The localization precision for the RF-data and for the BF-data reacts similarly to changes in the amount of noise. For a high SNR, the position of the scatterer can be determined more precisely compared to a situation with a low SNR. For example, the RF-data reveals a localization precision of 1 nm in the x-direction for an SNR of 33 dB, while an SNR of 23 dB results in a precision of 10 mm for the same intensity and depth.
This research investigates the theoretical precision limit of uULM in localizing a moving subwavelength scatterer. A 1D transducer array and the scatterer are simulated with the Vantage Research Ultrasound Simulator (Verasonics, Kirkland, WA, USA). A large beam is transmitted through the medium by the transducer array and hits the scatterer. As a sub-wavelength scatterer radiates as an omni-directional pressure field, transducer arrays of finite apertures receive a part of this spherical wave back as parabolas (the radio-frequency data, or RF-data), from which beamformed images (the BF-data) can be reconstructed. The intensity of the RF-data is given in arbitrary units, while the intensity for the BF-data is indicated in decibels. Both the RF-data and BF-data are influenced by different sources of noise, such as jitter and false peaks, which can be modelled with zero-mean white Gaussian noise. In this manuscript, the z-axis is defined to be the direction away from the probe and the x-axis is defined to be colinear to the piezoelectric elements. In Figure 1, an illustration of this process, called Plane Wave Imaging, with the given directions is shown.
The localization precision in the x- and z-directions is computed for different intensities, different signal-to-noise ratios (SNRs), and different depths z for both the RF-data and the BF-data. For the RF-data, there is a clear relation between the depth of the scatterer and the localization precision: if the scatterer is moving further away, the minimum standard deviation decreases due to attenuation in tissues. For example, for a depth of 11 mm and an SNR of 29 dB, one is able to localize the microbubble with a precision of 10 nm in the x-direction and 50 nm in the z-direction based on the RF-data, while this decreases to respectively 1 μm and 0.5 mm for a depth of 21 mm.
The BF-based precision limits are less dependent on depth: for different depths, the limits remain approximately the same, being 0.5 μm in the x-direction and 1 μm in the z-direction for an intensity of 20 dB and an SNR of 29 dB. A remarkable result is that the localization using the radio-frequency data is more precise compared to the beamformed images if the scatterer is close to the transducer array. However, after a certain depth, the BF-based localization surpasses the RF-based one. This difference in precision is due to the beamforming process: to translate radio-frequency data into a readable image, one needs to sum the energy scattered back and select that value as the pixel intensity for the final image. This process, called beamforming, can be done on the fly or in post-process. In this research, it is complex to compare the initial values, being the SNR, the maximum intensity value, and the dependence on depth, between the RF- and BF-data.
The localization precision for the RF-data and for the BF-data reacts similarly to changes in the amount of noise. For a high SNR, the position of the scatterer can be determined more precisely compared to a situation with a low SNR. For example, the RF-data reveals a localization precision of 1 nm in the x-direction for an SNR of 33 dB, while an SNR of 23 dB results in a precision of 10 mm for the same intensity and depth.