Stephen Z. Pinter

Detectability of Small Blood Vessels with High-Frequency Power Doppler and Selection of Wall Filter Cut-Off Velocity for Microvascular Imaging

Power Doppler imaging of physiologic and pathologic angiogenesis is widely used in preclinical studies to track normal development, disease progression and treatment efficacy but can be challenging given the presence of small blood vessels and slow flow velocities. Power Doppler images can be plagued with false-positive color pixels or undetected vessels, thereby complicating the interpretation of vascularity metrics such as color pixel density (CPD). As an initial step toward improved microvascular quantification, flow-phantom experiments were performed to establish relationships between vessel detection and various combinations of vessel size (160, 200, 250, 300 and 360 microm), flow velocity (4, 3, 2, 1 and 0.5 mm/s) and transducer frequency (30 and 40 MHz) while varying the wall filter cut-off velocity. Receiver operating characteristic (ROC) curves and areas under ROC curves indicate that good vessel detection performance can be achieved with a 40-MHz transducer for flow velocities > or =2 mm/s and with a 30-MHz transducer for flow velocities > or =1 mm/s. In the second part of the analysis, CPD was plotted as a function of wall filter cut-off velocity for each flow-phantom data set. Three distinct regions were observed: overestimation of CPD at low cut-offs, underestimation of CPD at high cut-offs and a plateau at intermediate cut-offs. The CPD at the plateau closely matched the phantoms vascular volume fraction and the length of the plateau corresponded with the flow-detection performance of the Doppler system assessed using ROC analysis. Color pixel density vs. wall filter cut-off curves from analogous in vivo experiments exhibited the same shape, including a distinct CPD plateau. The similar shape of the flow-phantom and in vivo curves suggests that the presence of a plateau in vivo can be used to identify the best-estimate CPD value that can be treated as a quantitative vascularity metric. The ability to identify the best CPD estimate is expected to improve quantification of angiogenesis and anti-vascular treatment responses with power Doppler.

Objective Selection of High-Frequency Power Doppler Wall Filter Cutoff Velocity for Regions of Interest Containing Multiple Small Vessels

High-frequency (> 20 MHz) power Doppler ultrasound is frequently used to quantify vascularity in preclinical studies of small animal angiogenic models, but quantitative images can be difficult to obtain in the presence of flow artifacts. To improve flow quantification, color pixel density (CPD) can be plotted as a function of wall filter cutoff velocity to produce a wall-filter selection curve that can be used to estimate actual vascular volume fraction. A mathematical model based on receiver operating characteristic statistics is developed to study the behavior of wall-filter selection curves. The model is compared to experimental data acquired with a 30-MHz transducer and a custom-designed multiple-vessel flow phantom capable of mimicking a range of blood vessel sizes (200-300 ¿m), blood flow velocities (1-10 mm/s), and blood vessel orientations. At high flow rates, wall-filter selection curves for multiple-vessel regions include a plateau whose CPD corresponds with the total vascular volume fraction. Conversely, the vascular volume fraction of a subset of vessels is obtained at low flow rates. Detection of the volume fraction of all vessels is ensured when a plateau is > 0.5 mm/s in length and begins at a wall filter cutoff < 2 mm/s.

Improved objective selection of power Doppler wall-filter cut-off velocity for accurate vascular quantification.

The wall-filter selection curve method is proposed to objectively identify a cut-off velocity that minimizes artifacts in power Doppler images. A selection curve, which is constructed by plotting the color pixel density (CPD) as a function of the cut-off velocity, exhibits characteristic intervals hypothesized to include the optimum cut-off velocity. This article presents an improved implementation of the method that automatically detects characteristic intervals in a selection curve and selects an operating point cut-off velocity along a characteristic interval. The method is applied to subregions within the Doppler image to adapt the cut-off velocity to local variations in vascularity. The methods performance is evaluated in 30-MHz power Doppler images of a four-vessel flow phantom. At high (>5 mm/s) flow velocities, qualitative improvements in vessel delineation are achieved and the CPD in the resulting images is accurate to within 3% of the vascular volume fraction of the phantom.

Understanding quantification of microvascularity with high-frequency power Doppler ultrasound

High-frequency power Doppler imaging of angiogenesis can be challenging given the presence of small blood vessels and slow flow velocities. In the presence of substantial Doppler artifacts such as false-positive color pixels or undetected vessels, color pixel density (CPD) and related vascularity metrics do not provide accurate estimates of vascular volume fraction. As a step towards improved microvascular quantification, flow-phantom experiments were performed to establish relationships between CPD and wall filter cut-off velocity for various combinations of vessel size (160, 200, 250, 300, and 360 μm), flow velocity (4, 3, 2, 1, and 0.5 mm/s), and transducer frequency (30 and 40 MHz). Three distinct regions were observed in plots of CPD versus wall filter cut-off velocity: overestimation of CPD at low cut-offs, underestimation of CPD at high cut-offs, and a plateau at intermediate cut-offs. The CPD at the plateau closely matched the phantoms actual vascular volume fraction. The length of the plateau corresponded with the flow-detection performance of the Doppler system, which was assessed using receiver operating characteristic analysis. Color pixel density versus wall filter cut-off curves from analogous in vivo experiments exhibited the same shape, including a distinct CPD plateau. The similar shape of the flow-phantom and in vivo curves suggests that the presence of a plateau can be used to identify the best-estimate CPD value in an in vivo experiment. The ability to identify the best CPD estimate is expected to improve quantification of angiogenesis and anti-angiogenic treatment responses with power Doppler.

A Method to Validate Quantitative High-Frequency Power Doppler Ultrasound With Fluorescence in Vivo Video Microscopy

Flow quantification with high-frequency (>20 MHz) power Doppler ultrasound can be performed objectively using the wall-filter selection curve (WFSC) method to select the cutoff velocity that yields a best-estimate color pixel density (CPD). An in vivo video microscopy system (IVVM) is combined with high-frequency power Doppler ultrasound to provide a method for validation of CPD measurements based on WFSCs in mouse testicular vessels. The ultrasound and IVVM systems are instrumented so that the mouse remains on the same imaging platform when switching between the two modalities. In vivo video microscopy provides gold-standard measurements of vascular diameter to validate power Doppler CPD estimates. Measurements in four image planes from three mice exhibit wide variation in the optimal cutoff velocity and indicate that a predetermined cutoff velocity setting can introduce significant errors in studies intended to quantify vascularity. Consistent with previously published flow-phantom data, in vivo WFSCs exhibited three characteristic regions and detectable plateaus. Selection of a cutoff velocity at the right end of the plateau yielded a CPD close to the gold-standard vascular volume fraction estimated using IVVM. An investigator can implement the WFSC method to help adapt cutoff velocity to current blood flow conditions and thereby improve the accuracy of power Doppler for quantitative microvascular imaging.

Improved method for objective selection of power Doppler wall filter cut‐off velocity for microvascular imaging.

The wall‐filter selection curve method has been enhanced to improve detection and interpretation of color pixel density (CPD) plateaus. The improved algorithm was developed by analyzing data acquired from three fields of view in a four‐vessel flow phantom using a 30‐MHz swept‐scan transducer. An N‐point maximum envelope peak search applied to the first difference of CPD detects selection curve plateaus by incorporating criteria that identify intervals of minimum variation in CPD. Selection curves for regions of interest (ROIs) containing multiple vessels can be difficult to interpret, so the algorithm subdivides the image into small ROIs, constructs selection curves for each ROI, and sums the resulting vascularity estimates. The lower limit on ROI size is constrained by a need to avoid ROIs that are completely filled by blood. A multiple‐step decision algorithm was designed that considers the number, length, and slope of each plateau to identify the cutoff velocity that yields the best vascularity estimate. At high (> 5 mm/s) flow velocities, the decision algorithm yielded a summed CPD that was within 5% of the vascular volume fraction in each field of view. These improvements are an initial step toward automating wall‐filter cutoff settings in a power Doppler system.

The wall‐filter selection curve method for objective tuning of power Doppler clutter filter cutoff velocity.

High‐frequency power Doppler ultrasound is commonly used to assess vascularity in small‐animal cancer models, but quantitative images can be difficult to obtain in the presence of clutter artifacts. To improve vascular quantification, the color pixel density (CPD) in a region of interest can be plotted as a function of wall‐filter cutoff velocity to produce the wall‐filter selection curve. A mathematical model based on receiver operating characteristic statistics was developed to guide the interpretation of wall‐filter selection curves. Mathematical predictions were tested using a VisualSonics Vevo 770 system with a 30‐MHz transducer and a flow phantom containing four 200–300‐μm‐diameter vessels. The phantom mimicked vessel configurations observed in micro‐CT images of a transgenic mouse prostate cancer model. Selection curves characteristically include a plateau whose CPD may correspond to either the total vascular volume fraction or to the volume fraction of a subset of vessels in the region. The flow‐p...

On the use of three‐dimensional Doppler acquisition for real‐time volume flow estimation.

Clinical volumetric blood flow estimation relies on several assumptions. Among them are cylindrical vessel geometry, symmetric flow profile, and Doppler angle. None of them are known well enough to obtain clinically relevant estimates. 3‐D color flow acquisition circumvents these assumptions, posing a viable tool for in vivo blood volume flow analysis. A 4‐D cardiac scanner operating a 2‐D array for real‐time 3‐D color flow imaging [GE Healthcare, Milwaukee, WI] was used. The array was positioned to fully intersect a 2‐cm‐diameter flow tube with the constant‐depth plane (CPlane). Blood mimicking fluid was circulated at up to 6 l/min using a cardiac bypass pump (60 and 80 beats/min). A trigger source synchronized the pump and scanner. Data volumes were acquired equally spaced throughout the cardiac cycle. Temporally resolved volume flow was derived from CPlane data integration using Doppler power partial volume correction. Results show less than 7% mean flow error for temporally resolved volume flow (100 p...

Estimating vascular volume fraction in a network of small blood vessels with high-frequency power Doppler ultrasound

Quantitative images of high-frequency (> 20 MHz) power Doppler ultrasound can be difficult to obtain in the presence of flow artifacts due to power Dopplers sensitivity to operator-dependent acquisition settings. To improve flow quantification, color pixel density (CPD) can be plotted as a function of wall filter cut-off velocity to produce a wall-filter selection curve that can be used to estimate vascular volume fraction by locating the plateau along the curve. The behavior of the wall-filter selection curve in a multiple-vessel region of interest is studied using a custom-designed multiple-vessel flow phantom. The flow phantom is capable of mimicking a range of blood vessel sizes (200–300 µm), blood flow velocities (1–10 mm/s), and blood vessel orientations (long-axis and transverse). At high flow rates, single-vessel wall-filter selection curves superimpose to produce a multiple-vessel curve where the CPD at the left-most plateau corresponds with the actual vascular volume fraction. However, interpretation of the multiple-vessel wall-filter selection curve is not straightforward when the flow rate in the vascular network is low.

12B-1 Detectability of Small Blood Vessels Using High-Frequency Power Doppler Ultrasound

Power Doppler imaging of physiological and pathological angiogenesis can be challenging given the presence of small blood vessels and slow flow velocities. Images can be plagued with false-positive color pixels or undetected vessels, thereby complicating the interpretation of vascularity metrics. This paper presents studies of blood vessel detectability using flow phantoms with various combinations of vessel size (160, 200, 250, 300, and 360 mum), flow velocity (4, 3, 2, 1, and 0.5 mm/s), and transducer frequency (30 and 40 MHz), while varying the wall filter cut-off velocity. Receiver operating characteristic (ROC) curves and areas under ROC curves indicate that good vessel detection performance can be achieved with a 40-MHz transducer for flow velocities ges 2 mm/s and with a 30-MHz transducer for flow velocities ges 1 mm/s.