Augmented patient-specific functional medical imaging by implicit manifold learning

This paper uses machine learning to enrich magnetic resonance angiography and magnetic resonance imaging acquisitions. A convolutional neural network is built and trained over a synthetic database linking geometrical parameters and mechanical characteristics of the arteries to blood flow rates and pressures in an arterial network. Once properly trained, the resulting neural network can be used in order to predict blood pressure in cerebral arteries noninvasively in nearly real-time. One challenge here is that not all input variables present in the synthetic database are known from patient-specific medical data. To overcome this challenge, a learning technique, which we refer to as implicit manifold learning, is employed: in this view, the input and output data of the neural network are selected based on their availability from medical measurements rather than being defined from the mechanical description of the arterial system. The results show the potential of the method and that machine learning is an alternative to costly ensemble based inversion involving sophisticated fluid structure models.

Effect of Pulse Shaping on Subharmonic Aided Pressure Estimation In Vitro and In Vivo

OBJECTIVES

Subharmonic imaging (SHI) is a technique that uses the nonlinear oscillations of microbubbles when exposed to ultrasound at high pressures transmitting at the fundamental frequency ie, f_o and receiving at half the transmit frequency (ie, f_o /2). Subharmonic aided pressure estimation (SHAPE) is based on the inverse relationship between the subharmonic amplitude of the microbubbles and the ambient pressure change.

METHODS

Eight waveforms with different envelopes were optimized with respect to acoustic power at which the SHAPE study is most sensitive. The study was run with four input transmit cycles, first in vitro and then in vivo in three canines to select the waveform that achieved the best sensitivity for detecting changes in portal pressures using SHAPE. A Logiq 9 scanner with a 4C curvi-linear array was used to acquire 2.5 MHz radio-frequency data. Scanning was performed in dual imaging mode with B-mode imaging at 4 MHz and a SHI contrast mode transmitting at 2.5 MHz and receiving at 1.25 MHz. Sonazoid, which is a lipid stabilized gas filled bubble of perfluorobutane, was used as the contrast agent in this study.

RESULTS

A linear decrease in subharmonic amplitude with increased pressure was observed for all waveforms (r from -0.77 to -0.93; P < .001) in vitro. There was a significantly higher correlation of the SHAPE gradient with changing pressures for the broadband pulses as compared to the narrowband pulses in both in vitro and in vivo results. The highest correlation was achieved with a Gaussian windowed binomial filtered square wave with an r-value of -0.95. One of the three canines was eliminated for technical reasons, while the other two produced very similar results to those obtained in vitro (r from -0.72 to -0.98; P <.01). The most consistent in vivo results were achieved with the Gaussian windowed binomial filtered square wave (r = -0.95 and -0.96).

CONCLUSIONS

Using this waveform is an improvement to the existing SHAPE technique (where a square wave was used) and should make SHAPE more sensitive for noninvasively determining portal hypertension.

Aortic relative pressure components derived from four-dimensional flow cardiovascular magnetic resonance

Purpose

To describe the assessment of the spatiotemporal distribution of relative aortic pressure quantifying the magnitude of its three major components.

Methods

Nine healthy volunteers and three patients with aortic disease (bicuspid aortic valve, dissection, and Marfan syndrome) underwent 4D-flow CMR. Spatiotemporal pressure maps were computed from the CMR flow fields solving the pressure Poisson equation. The individual components of pressure were separated into time-varying inertial (“transient”), spatially varying inertial (“convective”), and viscous components.

Results

Relative aortic pressure is primarily caused by transient effects followed by the convective and small viscous contributions (64.5, 13.6, and 0.3 mmHg/m, respectively, in healthy subjects), although regional analysis revealed prevalent convective effects in specific contexts, e.g., Sinus of Valsalva and aortic arch at instants of peak velocity. Patients showed differences in peak transient values and duration, and localized abrupt convective changes explained by abnormalities in aortic geometry, including the presence of an aneurysm, a pseudo-coarctation, the inlet of a dissection, or by complex flow patterns.

Conclusion

The evaluation of the three components of relative pressure enables the quantification of mechanistic information for understanding and stratifying aortic disease, with potential future implications for guiding therapy. Magn Reson Med 72:1162–1169, 2014. © 2013 The Authors. Magnetic Resonance in Medicine published by Wiley Periodicals, Inc. on behalf of International Society for Magnetic Resonance in Medicine. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

On the implementation of an automated acoustic output optimization algorithm for subharmonic aided pressure estimation

Incident acoustic output (IAO) dependent subharmonic signal amplitudes from ultrasound contrast agents can be categorized into occurrence, growth or saturation stages. Subharmonic aided pressure estimation (SHAPE) is a technique that utilizes growth stage subharmonic signal amplitudes for hydrostatic pressure estimation. In this study, we developed an automated IAO optimization algorithm to identify the IAO level eliciting growth stage subharmonic signals and also studied the effect of pulse length on SHAPE. This approach may help eliminate the problems of acquiring and analyzing the data offline at all IAO levels as was done in previous studies and thus, pave the way for real-time clinical pressure monitoring applications. The IAO optimization algorithm was implemented on a Logiq 9 (GE Healthcare, Milwaukee, WI) scanner interfaced with a computer. The optimization algorithm stepped the ultrasound scanner from 0% to 100% IAO. A logistic equation fitting function was applied with the criterion of minimum least squared error between the fitted subharmonic amplitudes and the measured subharmonic amplitudes as a function of the IAO levels and the optimum IAO level was chosen corresponding to the inflection point calculated from the fitted data. The efficacy of the optimum IAO level was investigated for in vivo SHAPE to monitor portal vein (PV) pressures in 5 canines and was compared with the performance of IAO levels, below and above the optimum IAO level, for 4, 8 and 16 transmit cycles. The canines received a continuous infusion of Sonazoid microbubbles (1.5 μl/kg/min; GE Healthcare, Oslo, Norway). PV pressures were obtained using a surgically introduced pressure catheter (Millar Instruments, Inc., Houston, TX) and were recorded before and after increasing PV pressures. The experiments showed that optimum IAO levels for SHAPE in the canines ranged from 6% to 40%. The best correlation between changes in PV pressures and in subharmonic amplitudes (r=-0.76; p=0.24), and between the absolute PV pressures and the subharmonic amplitudes (r=-0.89; p<0.01) were obtained for the optimized IAO and 4 transmit cycles. Only for the optimized IAO and 4 transmit cycles did the subharmonic amplitudes differ significantly (p<0.01) before and after increasing PV pressures. A new algorithm to identify optimum IAO levels for SHAPE has been developed and validated with the best results being obtained for 4 transmit cycles. The work presented in this study may pave the way for real-time clinical applications of estimating pressures using the subharmonic signals from ultrasound contrast agents.

Investigating the efficacy of subharmonic aided pressure estimation for portal vein pressures and portal hypertension monitoring

The efficacy of using subharmonic emissions from Sonazoid microbubbles (GE Healthcare, Oslo, Norway) to track portal vein pressures and pressure changes was investigated in 14 canines using either slow- or high-flow models of portal hypertension (PH). A modified Logiq 9 scanner (GE Healthcare, Milwaukee, WI, USA) operating in subharmonic mode (f(transmit): 2.5 MHz, f(receive): 1.25 MHz) was used to collect radiofrequency data at 10-40% incident acoustic power levels with 2-4 transmit cycles (in triplicate) before and after inducing PH. A pressure catheter (Millar Instruments, Inc., Houston, TX, USA) provided reference portal vein pressures. At optimum insonification, subharmonic signal amplitude changes correlated with portal vein pressure changes; r ranged from -0.82 to -0.94 and from -0.70 to -0.73 for PH models considered separately or together, respectively. The subharmonic signal amplitudes correlated with absolute portal vein pressures (r: -0.71 to -0.79). Statistically significant differences between subharmonic amplitudes, before and after inducing PH, were noted (p ≤ 0.01). Portal vein pressures estimated using subharmonic aided pressure estimation did not reveal significant differences (p > 0.05) with respect to the pressures obtained using the Millar pressure catheter. Subharmonic-aided pressure estimation may be useful clinically for portal vein pressure monitoring.