A Measurement System to Realize 3-D Carotid Occlusion Measurement From 2-D Conventional Ultrasonography

IEEE SENSORS JOURNAL, VOL. 14, NO. 3, MARCH 2014

747

A Measurement System to Realize 3-D Carotid Occlusion Measurement from 2-D Conventional Ultrasonography Domenico Capriglione, Member, IEEE, Luigi Ferrigno, Member, IEEE, Gianfranco Miele, Member, IEEE, Vincenzo Paciello, Member, IEEE, Alfredo Paolillo, Member, IEEE, and Paolo Sommella, Member, IEEE

Abstract— A novel measurement system for the 3-D reconstruction of occlusions of the carotid artery is presented. The proposed system can be added onto a conventional diagnostic ultrasonographic system, so that it can be considered as an upgrade for preexisting equipments. It retrieves information from a set of ordinary 2-D ultrasound images and spatial sensors in order to build a 3-D representation of the vessel and of a possible occlusion and to extract measurements of important medical parameters from them. The geometrical modeling of carotid is dealt with. This paper is also concerned with a metrological characterization of the measurement procedure, including the development of a model for the estimation of the uncertainty according to international standards. Experimental results are also reported and discussed. Index Terms— Biomedical measurements, measurement uncertainty, thickness measurement, volume measurement, biomedical image processing, ultrasonic imaging.

I. I NTRODUCTION

T

HE artherosclerosis is a kind of cerebrovascular and cardiovascular disease (CVD) whose consequences (cerebral infarction, embolus, ictus, ischaemia) seriously cause human morbidity and mortality. According to World Health Organization, CVDs represent the third leading cause of death in the world [1]. More in details, the atherosclerotic process refers to the degeneration of the arterial wall and the deposition of lipids and other bloodborne material within the arterial wall of almost all vascular tissues. Several studies evidenced the relationships between the carotid artery wall status and CVDs [2]; specifically, an increased Intima – Media Thickness (IMT) is correlated to an augmented risk of brain infarction or cardiac attack. Moreover, the presence of a carotid plaque has been correlated to degenerative pathologies like vascular dementia and Alzheimer’s disease [3]. Hence, the inspection of the carotid wall is crucial

Manuscript received September 20, 2013; accepted October 19, 2013. Date of publication October 24, 2013; date of current version January 10, 2014. The associate editor coordinating the review of this paper and approving it for publication was Prof. Aime Lay-Ekuakille. D. Capriglione, L. Ferrigno, and G. Miele are with the Department of Electrical and Information Engineering, University of Cassino and Southern Lazio, Cassino 03043, Italy (e-mail: [email protected]; [email protected]; [email protected]). V. Paciello, A. Paolillo, and P. Sommella are with the Department of Industrial Engineering, University of Salerno, Fisciano 84084, Italy (e-mail: [email protected]; [email protected]; [email protected]). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/JSEN.2013.2287298

for the early assessment of pathology, as well as for the monitoring of the progression of a degenerative phenomenon. In clinical practice, the Ultra-Sound (US) techniques (also including B-mode, Color Doppler, Color Power Angio) represent the most used diagnostic tools for assessing CVDs (with respect to other modalities such as Magnetic Resonance Imaging and Computed Tomography Angiography), thanks to the following key advantages: i) real-time quick examination; ii) low costs; iii) repeatability and reliability; and iv) absence of radiations and non-invasiveness. Obviously, a major problem of the clinical examination based on US imaging is the need for quantitative measurements. In order to overcome the drawbacks of image segmentation and IMT measurement manually performed by operators, investigation efforts are being addressed to the development and characterization of fully automatic (user independent) systems able to detect the carotid layers and boundaries. In [4] an interesting method is presented, focused on the segmentation of US carotid images based on active snakes and on its classification as either healthy or diseased with a support vector machine (SVM) classifier. The proposed classifier operates on single images, thus the results are not representative of the volumetric arrangement of the carotid interfaces, but depend on the specific selection of that image. An interesting review about main segmentation algorithms is reported in [5], where dynamic programming, deformable snakes, Hough Transforms and classification approaches are compared. On the other hand, IMT it is not sensitive to the changes in plaque burden, which turned out to be a stronger predictor of cardiovascular events [6]. Consequently, in most recent years an increased interest moved towards area and volumetric measurements of carotid plaque burden such as total plaque area (TPA), total plaque volume (TPV) [7], vessel wall volume (VWV) [8], and vessel wall thickness maps (VWT maps) [9]. The hypothesis is that the area and volumetric measurements are more reflective of the plaque burden so that they may provide complementary information to IMT. Hence, two different approaches have been extensively investigated: i ) revisiting conventional method based on 2-D B-mode imaging [10], [11] and ii) exploring the 3-D ultrasound potential to provide comprehensive quantification of plaque [12]–[14]. The majority of these studies developed

1530-437X © 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

748

IEEE SENSORS JOURNAL, VOL. 14, NO. 3, MARCH 2014

algorithms for 3-D reconstruction of a geometric model of the lumen and wall of coronary arteries from US images able to achieve a good geometric accuracy, if compared with results based on manual segmentation performed by an expert physician. Following the trends in surgery and radiology, also for the US image-based reconstruction, an increasing interest is addressed to the experimental estimation of measurement error in comparison with objective references [15]. More in detail, a phantom presenting 3D features of plaques could be realized using stereolitography [16] and employing new tissue mimicking materials suitable for ultrasound studies [17]. However, yet in few examples [18]–[19] the volume measurements provided by 3-D model have been compared with pulsatile phantoms of the carotid bifurcation; moreover the experimental set-up typically includes a limited series of operating conditions. Thus, the quantitative assessment of the plaque and lumen has to be further explored to provide an accurate clinical tool. Starting from previous studies about longitudinally-oriented 2D images for automatic IMT measurement [20], the authors aim to introduce an accurate and low-cost real-time system for 3-D reconstruction of the carotid plaque. The principal contribution of the presented work is an extensive metrological characterization of the 3-D image system based on a rigorous analytical approach according to international guideline. The direct extension of the provided experimental set-up may also represent an effective methodology to assess the quantitative estimations provided by a wide class of ultrasound 3-D-reconstruction systems. The paper is organized as follows: the collocation of the work in the context of 3-D US carotid imaging is motivated in Section II. The ultrasound 3-D reconstruction system is described in Section III in terms of both the hardware components and algorithms for image processing. Then, the preliminary characterization and tuning of the systems are described in Section IV as well as the analytical approach adopted for the estimation of the measurement uncertainty. Section V reports the experimental results achieved by applying the suggested methodology during the clinical activity. Finally, some conclusions are drawn in section VI. II. M OTIVATION OF THE W ORK Different 3-D ultrasound instruments are present on the market, but most of them either are able to make only qualitative measurements or are highly expensive. They can be classified into two main classes: i) based on “add-on” devices, and ii) based on sensor arrays. 1) Systems based on “add-on” devices exploit typical 2-D ultrasound probes, together with additional hardware which is needed to locate each image of the set of acquired images in the 3-D space. The volume of interest is analyzed by suitably spacing the different image planes. In order to avoid unacceptable errors in the 3-D reconstruction, the position and the orientation of the probe should be evaluated with an adequate accuracy. 2) Systems based on sensor array employ special probes, containing arrays of ultrasound sources/receivers for

Fig. 1. Functional blocks of a 3-D measurement system based on 2-D images and add-on devices (the units for 3-D measurements are in red).

which the image scanning plane can be changed electronically and without particular work by the operator. The major disadvantage of such systems is the cost. Comparing these solutions, the former offers better cheapness and flexibility, while the latter is generally characterized by a better easiness of use. A common limitation for both classes of instruments is that the area and volume measurements provided by these instruments do not include any information about their uncertainty. In this paper, the problem of the evaluation of the uncertainty of the 3-D reconstruction based on US imaging has been dealt with, in terms of the identification of the main uncertainty sources and of its propagation. III. T HE P ROPOSED M EASUREMENT S YSTEM Due to the very wide diffusion of 2-D systems, the first kind of instruments has been considered for the prototype design and implementation. Focusing the attention on such a type of instruments [21]–[22], they can be schematized as reported in Fig. 1. The first block is composed by the hardware for capturing the 2-D images (probe) and for locating the probe. The acquired 2-D images are processed in order to extract the information required for the 3-D reconstruction. Finally the 3-D image is processed to achieve the wanted measurements. An important aspect to be considered in the system design and realization is that the carotid volume varies with the heart activity thus giving a very low repeatability in the measurements. To overcome such a problem, the 2-D image acquisitions have to be synchronized with the heart activity obtaining the same experimental conditions for each image. In Fig. 2, the schematic of the proposed prototype is shown. In the following, the hardware, the system operation and the main software sections are described in detail. A. The Hardware As described above, the goal of the proposal is the realization of a low cost 3D ultrasound measurement system able to operate with common 2D ultrasounds by the adoption of a suitable add-on measurement system. Required features are the capability of operating with most of the existing 2D ultrasound systems without the need for any hardware or software change in the 2D ultrasound instrument. To this aim, some hardware components are required. They concern with systems for the acquisition of the 2D images from the existing ultrasound,

CAPRIGLIONE et al.: MEASUREMENT SYSTEM TO REALIZE 3-D CAROTID OCCLUSION MEASUREMENT

Fig. 2.

Simplified block diagram of the proposed prototype.

systems for the correct localization of the 2D probe in the 3D space and suitable processing units. Fig. 2 schematically highlights these components: (a) the traditional 2D ultrasound system, (b) the frame grabber, (c) the processing unit, and (d) the add-on package. a) A traditional 2-D ultrasound system by Philips has been adopted for the production of the 2D ultrasound images. It is the ATL 5000T M , with 7-12 MHz changeable pulse ultrasound frequency. It allows to employ both B-mode and Color Power Angio (CPA) images which highlight also slow blood fluxes to be evidenced whichever the ultrasound orientation with respect to the flux direction. In particular, 10 MHz frequency was chosen, since it allows up to 39 mm analysis depth with a good resolution. Consequently, a very wide variety of carotid arteries can be analyzed (generally the carotid is modeled like a tube with a medium circular section of about 1 cm of radius and with the center placed under the skin at 12-20 mm depth). b) A PAL/NTSC compatible frame grabber device DFG1394-1e by Total Turn Key Solutions was used. It grabs the current display of the screen of the ultrasound machine and sends it to the PC-based processing unit through the Firewire bus. It allows the connection of the add-on system with the most widespread ultrasound machines available on the market. c) The processing unit is based on a PC equipped with a PCMCIA Data Acquisition Board NI 6062E, and with RS-232 and FirewireTM interfaces. The DAQ board is employed for data acquisition from the sensors at a sampling frequency of 500 samples/s. The RS 232 data interface bus is used for the communication with the sub-system for the

749

position tracking, and the Firewire interface for the acquisition of the 2-D images coming from the frame grabber. d) The add-on package is composed of three main hardware blocks: d.1) the probe angle measurement sub-system, d.2) the probe position tracking system, d.3) the synchronization unit. d.1) The angle of incidence, namely the angle between the start analysis plane and the current scanning plane, is measured by a specific sub-system based on a tri-axial MEMS accelerometer, LIS352AX by STMicroelectronics. The LIS352AX has an acceleration range of ±2g, a sensitivity of 363 mV/g and total non-linearity of ±0.5% of the full scale. In detail, the angle of incidence is measured as the angle by which the gravitational acceleration vector rotates during the scanning as seen in a coordinate reference system fixed onto the probe. Due to the slow dynamics involved, a simple low-pass RCfilter with a cut-off frequency of 25 Hz has been adopted to limit the signal bandwidth. To improve the system stability the device has been powered by a suitable voltage regulator. d.2) The probe position tracking is made by using a gyroscopic-based device, the Micro-ISU BP3010 by Bulmer Electronics & Control. It includes gyroscopes and accelerometers as well as internal circuits for power supply regulation, three microprocessors (for data compensation and communication management) and a RS-232 communication module for interfacing with the external PC. Through this communication port the device sends, with a maximum update rate of 64 Hz, the incremental velocity on each axis with a resolution of 1 mm/s over a 16 bits length word. d.3) It is well known that the cardiac activity might cause artifacts in the ultrasound images of the carotid artery. Images of the same carotid, acquired with the same orientation of the ultrasound probe, might differ since the carotid expansion due to the cardiac systolic events. Thus, most of the ultrasound systems adopt an ECG synchronization. In the proposed solution the synchronization unit is based on ECG signals provided by an ad-hoc realized device. Aim of this device is only the detection of the systolic event of the heart activity and not a complete analysis of the cardiac trace. For these reasons, a simplified version that employs only two electrodes (typical of ECG systems) enveloping the patient’s wrists is adopted. The differential voltage detected by the electrodes is suitably amplified by a two-stage module, which is constituted by a differential instrumentation amplifier followed by an operational amplifier in non-inverting configuration. To prevent the ECG artifacts due to the 50 Hz noise of the power supply, and due to the low frequency wandering, a bandpass filter with cutoff frequencies of 0.3 Hz and 33 Hz has been arranged after the amplifier module. B. General Operation of the System The intended simplicity of the realized system allows the diagnostician to carry out a clinical exam in exactly the same

750

IEEE SENSORS JOURNAL, VOL. 14, NO. 3, MARCH 2014

Fig. 4.

Fig. 3.

The add-on package.

manner of a common 2D ultrasound exam. In particular, once mounted the add-on package on the 2D ultrasound probe (Fig. 3) the diagnostician moves and tilts the probe until he finds a satisfying view of the segment of the carotid artery. Then, he pushes a button while holding the probe. This is a procedure similar to that adopted by a diagnostician to grab and print a photo of the ultrasound exam. This event is captured by the Trigger Detection For Data Storing module (described in the following subsection), which is synchronized by the ECG signal. This module acquires the probe angles and positions and selects the appropriate image from the frame grabber. In this way, although the frame gabber continuously acquires 2-D images, only the images acquired after a programmable delay triggered by the diastolic movement of the heart are stored on an external hard disk and successively employed for 3-D reconstruction. This step is repeated for each angle of incidence chosen by the diagnostician. Typically, a range of ±5° with a step of one degree for the angle of incidence allows to achieve an acceptable 3-D reconstruction [23], [24]. Once the data acquisition is completed, the diagnostician has to select a box (Region of Interest, ROI) on any image. This box is the area of the image that will be processed in order to extract the 2-D outline of the relevant interfaces between different tissues in the image. A useful feature of the proposed prototype is that if the edge detection fails on a single image, the diagnostician can choose between two solutions: whether to discard this image or to launch an edge correction routine.

Carotid artery transversal (a) and longitudinal (b) views.

As for the 3-D reconstruction, different solutions could be adopted such as to use transversal images (Fig. 4a), longitudinal images (Fig. 4b) or mixed solutions. Many experimental tests were carried out and they proved that the optimal choice in terms of image quality and 3-D reconstruction is achieved using longitudinal images; moreover the best longitudinal images are obtained by placing the probe in a back-lateral position [23], [24]. Finally, the software provides the 3-D reconstruction of the plaque and of the artery, as well as the percentage of occlusion and plaque volume together with related measurement uncertainties. C. The Trigger Detection for Data Storing Module This is a very important software module that allows a reliable measurement process. It has a double aim. On one hand, it has to guarantee the selection of 2D images synchronized with respect to the systolic cardiac event. On other hand, it must allow the synchronization of the measurement information coming from the angle and movement sensors with the acquired 2D ultrasound image. This module receives as inputs the signals acquired by the DAQ board (i.e. the ECG, the accelerometer, and the push button) and those coming from the data communication busses, namely the images coming from the FirewireTM bus and the movement data coming from the RS 232 bus. It has been developed in the LabViewTM environment and operates as follows. When the diagnostician selects an image, and pushes the acquisition button, the Trigger Detection For Data Storing module selects the first 2D ultrasound image that occurs after a suitable delay after the trigger event generated by the ECG signal. In the same time the software collects measurement data coming from the add-on package by means of a mobile acquisition window that has a length of 100 ms. This solution averages the angular and translation measurements, allowing the outlier filtering and the reduction of noise. As far as the software implementation is concerned, it can be schematized as the parallel execution of three synchronized loops. The first collects data coming from the DAQ board, the second from the movement probe and the third from the frame

CAPRIGLIONE et al.: MEASUREMENT SYSTEM TO REALIZE 3-D CAROTID OCCLUSION MEASUREMENT

Fig. 5.

Examples of outputs of the segmentation procedure.

grabber. The first loop executes a point by point acquisition and collects data into three vectors, one for each acquired quantity, managed as a sliding window of 100 ms length. A software routine detects when the ECG signal reaches the desired level and slope and generates an interrupt for the acquisition of the last 100 ms of data about the trigger event. In the same way, the second loop polls the RS232 bus at a baud rate of 38400 bps and collects data from the movement probe. This adopted Micro-ISU BP3010 sensor acquires data at a frequency of 64 Hz, which implies about 6 samples in the considered 100 ms window. Finally, the third loop manages frames coming from the frame grabber and selects the first image that occurs after the trigger event. D. Software: 2-D Edge Detection As for the 3-D image reconstruction phase, it is realized by segmenting each 2-D image, in the current plane, to extract the object contours, which in turn are used to obtain the object surface [21], [25]. Compared with other approaches available in literature, the main advantage of such a solution is related to the amount of data to be stored and managed because only the interesting points are processed. Consequently it allows a fast and efficient 3-D rendering to be achieved. Each 2-D image is processed in order to extract the three interfaces: intima-lumen in near wall, intima-lumen in far wall, lumen-plaque. The segmentation is completely automatic [20]. The technician is asked only to select the ROI for the image analysis, aiming at simplifying the elaboration software and at reducing the elaboration time. The contour detection algorithm is based on the recursive application of suitable cost functions. Namely, three cost functions are used, one for each interface. They are composed of some terms taking into account the characteristics of the investigated contours, such as, for example, gray level above and below the contour or the contour regularity. These terms are weighted by coefficients empirically evaluated during a training phase performed on different carotid CPA images. In Fig. 5 the outputs of the segmentation procedure for two CPA images are shown. E. Software: 3-D Volume Reconstruction The three interfaces intima-lumen in near wall, lumenintima, and intima-media in far wall are extracted from the 2-D image processing [23], [24]. Then, the 3-D reconstruction algorithm locates each point of each interface in the 3-D space with respect to a given reference system (O, x, y, z), based

Fig. 6.

751

A position example of the contour lines.

on the knowledge of its pixel coordinates (px, py) and on the orientation and position values (θx , θ y , d x , d y ) of the probe:       x y z

cos (θx ) − cos θy · sin (θx ) dx + px · tan (θx )·sin (θx ) px ·sin (θx ) dy px · py (1) 1

= − sin (θx ) cos θ y · cos  (θx ) 0 sin θ y  

Where px is the distance along the x-axis of the center of the selected image from the origin of the x axis. By processing each point of the three interfaces of each image, the 3-D reconstruction of the carotid can be displayed and saved. This process is illustrated in Fig. 6. In particular Fig. 6a) shows the interfaces related to a single 2-D image, while in Fig. 6b) the positioning of the interface pixels of 2-D images corresponding to all angles of incidence is reported. F. Software: Percentage Occlusion Measurement In order to evaluate the percentage of carotid occlusion, O% , the artery volume, VL , and the plaque volume, VP , have to be measured: VP O% = · 100. (2) VL To this aim the reconstructed 3-D carotid (see Fig. 7a) is sliced in the x direction, with a x-step, x, equal to one pixel, obtaining for each cross section in the z − y plane the polygons representing the lumen and the plaque, respectively (Fig. 7b and Fig. 7c). The lumen polygon is defined by the interpolated interface between intima-lumen interfaces (in the near and far wall), while the plaque polygon is delimited by the interpolated lumen-plaque and far wall lumen-intima interfaces. The volumes are obtained summing the single slice volumes, said ALj and APj the area of each polygon of the lumen and plaque, respectively (Fig. 7d): VL =

NL  j =1

ALj · x, VP =

NP 

APj · x.

(3)

j =1

Each area is computed by dividing it into adjacent triangles with consecutive points of the section contour:     Mkj   1 Akj = ·  (yi · zi+1 − yi+1 · zi ) , (4) 2   i=1 considering all the Mkj points describing the polygon (see Fig. 7d).

752

IEEE SENSORS JOURNAL, VOL. 14, NO. 3, MARCH 2014

Fig. 8.

Measured biases for the sensor x-axis.

As an example, in Fig. 8 the mean errors between the imposed angle of incidence and the measured one are reported for the x-axis of the sensor. As a result, the selected device exhibits a mean error over the range of interest (−10°, +10°) that can be modeled with a uniform pdf within the range [−0.6°, +0.6°]. Similar results have been obtained for y-axis and z-axis. B. Sub-System for the Measurement of the Probe Position

Fig. 7. Main steps for the percentage occlusion measurement: a) reconstructed 3-D carotid, b) 3-D interfaces of plaque volume, c) 3-D interfaces lumen volume, d) area evaluation, e) single slice volume.

The plaque volume measurements can be useful for the correct medical diagnosis, but, after processing the images and applying these relationships the measurements are still expressed in pixels. If one knows the pixel-to-millimeter conversion factor, k, which depends on the specific probe and machine used, then: 3

= k3 Vp . Vmm p

(5)

IV. P RELIMINARY C HARACTERIZATION AND T UNING OF THE M EASUREMENT S YSTEM A deep experimental characterization has been carried out to assess the prototype performance and to correct possible systematic effects. A. Sub-System for the Measurement of the Angle of incidence The subsystem for the measurement of the angle of incidence has been characterized by a comparison with a computer controlled pan-tilt unit, namely the PTU-46 by Directed Perception. This unit has an angular resolution equal to 0.013° and an overall accuracy of 0.039°. The experiment has been conducted on the biaxial accelerometer fixing the measurement subsystem for the angle of incidence onto the pan-tilt unit, and varying for each axis the pan-tilt angle in the +/−30° range with a resolution step of 0.5°. For each step, 200 consecutive measurements have been executed.

The subsystem for the measurement of the probe position has been characterized by a comparison with a 4 axes Adept Cobra 600TM robot, acting as a reference instrument. Such a system achieves a repeatability on the XY plane equal to 0.017 mm, whereas the repeatability for the Z-plane is 0.003 mm. The robot arm can be programmed into given positions by sending commands written with a specific syntax through a RS-232 serial port. To this purpose, a LabView software, which controls the robot arm and has a graphical user interface, has been implemented. A preliminary test has been carried out to assess the performance of the probe position measurement sub-system in absence of any motion or vibration. In this way, the intrinsic noise related to the accelerometer responses can be highlighted. Results have been obtained considering 9000 consecutive measurements over an interval equal to 15 minutes. The obtained results have shown an intrinsic repeatability, estimated as the experimental standard deviation referred to the estimated mean value of 0.012% for all the directions. Furthermore some tests with different probe speeds have been carried out over a path 15 mm-long for each one of the considered directions. The following speeds have been considered: 20 mm/s, and 10 mm/s. Each path has been covered in the direct and reverse direction (D and R, respectively) and for each path 100 consecutive measurements have been made. Two figures of merit have been considered: i) % defined as the mean percentage bias between the imposed and estimated distance; ii) σ % defined as the percentage experimental standard deviation of % with respect to the imposed distance. Obtained results are reported in Table I. Some considerations can be made: - the performances of x-axis and y-axis are generally better than those obtained with the z-axis;

CAPRIGLIONE et al.: MEASUREMENT SYSTEM TO REALIZE 3-D CAROTID OCCLUSION MEASUREMENT

753

TABLE I R ESULTS FOR T EST ON THE S UB -S YSTEM FOR THE M EASUREMENT OF THE

P ROBE P OSITION

Fig. 9. Examples of simulated arteries with plaques: a) two cases of plaques with different volumes; b) cases with different skin-artery distances, where the directions of incidence are reported. TABLE III T HE E RRORS IN P LAQUE V OLUME M EASUREMENTS

TABLE II A PPLIED C ORRECTION G AINS IN THE E STIMATION OF THE

P ROBE M OVEMENTS

- for the x and y axes the overall accuracy is bounded within the 10% with a good repeatability while the z-axis shows a worse performance, particularly for the reverse movements. Taking into account the obtained results, suitable mean correction gains have been applied to the estimated x, y, and z distances both for the direct movement and for the reverse side. The adopted movement correction gains are reported in Table II. C. Sub-System for the 3-D Volume Reconstruction The characterization steps have been carried out in an emulation environment. To this aim the software prototype was tested on a number of artificial images reproducing the human carotid, obtained in a 3D CAD environment (Autocad 2000™). The artery is modeled like a cylinder and the plaque, represented as a section of a spherical shape, is positioned inside it. Several plaques of different nominal volumes (V p,R E F ) were simulated and different distances (d) of the carotid with respect to the skin were considered (see Fig. 9). The 2-D acquisition process is emulated slicing the simulated artery: for each angle of incidence, a corresponding plane is built; the volume sections projected onto the selected planes give the 2-D images. Five planes rotating in the interval [−5°, +5°] around a pivot axis were considered.

The prototype software processes the obtained artificial 2-D images in order to reconstruct the plaque and to evaluate 3 the volume measurements (V pmm ). Then, the error (e%) of the measurement system on the carotid plaque volume is 3 calculated as e% = 100 · (V pmm − V p,R E F )/V p,R E F . The obtained results are summarized in Table III. A systematic effect is highlighted: the measurement error is negative for each test (the system underestimates the volume). Besides, the results show that e% increases with the distance d. Thus, the corrected volume of the plaque is evaluated as: 3

V Pc = c · V Pmm = c · k 3 · V P

(6)

where c is the correction factor. The error value e% depends on both carotid position and plaque dimension; consequently, the correction factor c should assume different values according with these parameters. Nevertheless, for both plaque and stenosis volume calculation, the error variation is always contained in a 1% range, and can be judged negligible with respect to all uncertainty sources. For this reason, a c mean value is used (c =1.032), calculated from the average of the percentage errors e%. The uncertainty of the correction factor c was computed from the standard deviation of e% (as reported in the following sub-section). D. Uncertainty Estimation The measurement uncertainty is evaluated by a-priori analysis, applying the uncertainty propagation law, suggested by the ISO-GUM [26], on the relationships implemented in the software and taking into account the metrological characterization made in the previous sub-section. In particular, considering the two volume measurements uncorrelated, the occlusion measurement uncertainty is

754

IEEE SENSORS JOURNAL, VOL. 14, NO. 3, MARCH 2014

evaluated as:

 ∂O% 2 2 ∂O% 2 2 uVp + uVL ∂Vp ∂VL 2   2 1 VP 2 = uVp + − 2 u2VL . VL VL

TABLE IV U NCERTAINTY C OMPONENTS



u2O% =

(7)

The uncertainties on the volume estimations depend on both the image processing and the position measurement as detailed in the following. In particular using eq. (3), it is obtained:   ∂V 2  u2Aj = x2 u2Aj (8) u2V = j ∂Aj j In order to evaluate the uncertainty of each jth cross section area, the ISO-GUM law is applied to eq.(4), assuming uncorrelated the 3-D coordinates of the plaque contour pixels: Mj   ∂Aj 2



2

 ∂Aj ∂Aj  · u yi , zi ∂yi ∂yi ∂zi i=1 (9) where u(yi , zi ) is the correlation between the z and y 3D coordinates of each contour point and: u2Aj =

u2yi +

∂Aj ∂zi

u2zi +2

∂Aj ∂Aj = zi+1 − zi−1 ; = yi+1 − yi−1. ∂yi ∂zi

(12)

The uncertainty component on the edge position, along the py direction, u py mainly depends on the image noise, on the edge detection algorithm and so on [18], and it does not depend on the add-on system. It was experimentally evaluated as: 5 pi xels = 2.9 pi xels (13) u py = √ 3 The uncertainty component due to d y was evaluated by considering the experimental variability of the correction factors described in the sub-section B. In particular, five different speeds (5 mm/s, 7 mm/s, 10 mm/s, 12 mm/s, 15 mm/s) have been considered. The resulting standard deviation of the correction factors was equal to 0.5 mm, then: ud y = 0.5mm

0.6◦ uθ = √ = 0.35◦ 3

(14)

Finally, the uncertainty on the angle of incidence depends on the error of the package used in evaluating the angles. Considering a maximum absolute error equal to 0.6° (in the range of interest), and assuming a uniform distribution,

(15)

In the uncertainty evaluation of the plaque volume (see eq. (5)) also the uncertainty of the k factor and of the correction factor has to be considered:  c 2  c 2  c 2 ∂ VP ∂ VP ∂ VP u 2C + u 2k + u 2VP u 2V c = P ∂c ∂k ∂ VP = k 6 · V P2 · u 2C + 9 · k 4 · c2 · V P2 · u 2k + k 6 · c2 · u 2VP . (16) In eq. (16) u k depends on the ultrasound probe specifications; typically a good approximation [21] for many probes is

(10)

The uncertainty and the correlation of the coordinates y and z of a generic contour point P(x, y, z) can be calculated considering eq. (1), and considering negligible the uncertainty on px (namely the abscissa of the image). Therefore, we obtain:    ∂y 2 2 ∂y 2 2 ∂y 2 2 2 uy = · uθ x + · uθy + · u py ∂θx ∂θ y ∂ py    ∂z 2 2 ∂z 2 2 ∂z 2 2 u2z = · uθy + · udy + · u py (11) ∂θ y ∂d y ∂ py ∂y ∂z 2 · u u(y, z) = ∂θ y ∂θ y θ y

an uncertainty contribution due to the angle measurement (for each axis) can be estimated as:

u k = 0.2 · 10−3

mm . pixel

(17)

The term u c is estimated as standard deviation of the correction factors reported in Table III. In particular, a value of u c = 0.005 has been found. Summarizing, Table IV reports the uncertainty components to be considered in the overall uncertainty estimation. V. E XPERIMENTAL R ESULTS In this Section, experimental results of tests carried out on two patients in different hospitals are shown. The first test is related to a patient under the age of thirty years and without evident diseases in the carotid artery. The second test is carried out on a patient with carotid atherosclerosis. The first test was conducted in a hospital clinic, namely “Villa dei Fiori” in Acerra, Italy. The probe used for this type of testing is the vascular and cardiac probe L12-5 38. It emits ultrasounds in a frequency interval ranging from 5 to 12 MHz, and is connected to the ultrasound Philips ATL HDI 5000. The second test was carried out in the hospital “Santa Maria Incoronata dell’Olmo” in Cava de’ Tirreni, Italy, using the probe for HST (Hemispheric Sound Technology) vascular analysis connected to the ultrasound Aloka Prosound 5000SV. In detail, in the first test the probe has a resolution of 0.066 mm/pixel, while in the second test a resolution of 0.1 mm/pixel is experienced. The use of different probes shows the suitability of the proposed measurement system to different ultrasound equipment. The two considered tests differ also in the plaque volume: in the first test a patient with a very small (practically negligible) plaque has been considered, whereas, in the second test an appreciable plaque volume has been involved.

CAPRIGLIONE et al.: MEASUREMENT SYSTEM TO REALIZE 3-D CAROTID OCCLUSION MEASUREMENT

TABLE V P ROBE P OSITIONS AND A NGLES

Fig. 10.

755

TABLE VI P ROBE P OSITIONS AND A NGLES

The 2-D edge detection on the #1 acquired B-mode image. Fig. 12.

The 2-D edge detection on images #1 and #2.

The obtained results as for the volume measurement and the stenosis percentage together with their uncertainty are reported in the following.

c O% = 6.6 % V P = 200 mm3 u O% = 1.5 % u v = 40 mm3 Looking at the obtained results, it is possible to highlight that the relative uncertainty is about 20% for both measurements. B. Results of the Second Test

Fig. 11.

The reconstructed 3-D carotid model.

A. Results of the First Test Five B-mode images were considered. The values measured as for the x and y angles and directions are reported in the Table V. In addition, no movement on both the z-axis angle and on the z direction was considered. In the post processing stage, the selection of the ROI is carried out and the 2-D edge detection is performed. As an example, Fig. 10 depicts the output of this stage for the image #1. Fig. 11 sketches the reconstructed 3-D carotid model achieved by processing the five images with eq. (1). As for the uncertainty estimation, taking into account the components reported in Table IV and eqs. (7) and (16), we obtain the overall uncertainty for both the stenosis percentage and the plaque volume (u O% and u v , respectively).

Ten B-mode images were considered. Also in this case, no movement on both the z-axis angle and on the z direction was performed. The values measured as for the x and y angles and directions are reported in Table VI, while Fig. 12 sketches the 2-D edge detections for the image #1 and #2. Once again, the overall uncertainty is evaluated by considering the components reported in Table IV and eqs. (7) and (16). The obtained results as for the volume measurements and the stenosis percentage together with their uncertainty are reported in the following.

c O% = 16% V P = 340 mm3 u O% = 2 % u v = 46 mm3 In this case, the plaque volume is greater than the one of the previous test, thus the relative measurement uncertainty related to the stenosis percentage is smaller (about 12%). VI. C ONCLUSION A low-cost measurement system for the three-dimensional analysis of occlusions within the carotid artery has been introduced. The proposed system can be added onto a conventional diagnostic ultrasonographic system in the shape of an add-on package, in order to upgrade preexisting equipment. The paper

756

IEEE SENSORS JOURNAL, VOL. 14, NO. 3, MARCH 2014

has described both the adopted hardware devices (subsystems for the probe tracking, the measurement of the angle of incidence and 2D image synchronization) and the processing routines (automatic 2D image segmentation and geometrical modeling), aiming to the 3-D reconstruction of a length of the carotid artery together with a possible occlusion, as well as to the measurements of critical parameters. The proposed solution is completed by a metrological characterization of the measurement system, including the development of a model for the estimation of the uncertainty according to international standards. The main uncertainty sources are disclosed (with reference to both hardware and software features) and the corresponding contributions are analytically analyzed. The reported experimental results show the effectiveness of the proposed system. A reduced relative uncertainty (lower than 20%) results in 3D stenosis measurements, which is valuable in supporting the medical diagnosis from reliable data. Future work will be concerned with the hardware simplification of the proposed measurement system in terms of adoption of wireless busses for more flexibility. Regarding the methodological approach, the characterization set-up and the uncertainty propagation will be applied to the comparison of US reconstruction systems based on different hardware devices and processing routines (2D segmentation and volume reconstruction). R EFERENCES [1] W. H. Organization. (2007). Cardiovascular Disease [Online]. Available: http://www.who.int/cardiovasculardiseases/en/ [2] J. P. Touboul, J. Labreuche, E. Vicaut, and P. Amarenco, “Carotid intimamedia thickness, plaques, and Framingham risk score as independent determinants of stroke risk,” Stroke, vol. 36 no. 8, pp. 1741–1745, Jul. 2005. [3] T. Watanabe, S. Koba, M. Kawamura, M. Itokawa, T. Idei, Y. Nakagawa, et al., “Small dense low-density lipoprotein and carotid atherosclerosis in relation to vascular dementia,” Methabolism, vol. 53 no. 4, pp. 476–482, Apr. 2004. [4] A. Chaudhry, M. Hassan, A. Khan, and J. Y. Kim, “Automatic active contour-based segmentation and classification of carotid artery ultrasound images,” J. Digit. Imaging, pp. 1–11, Feb. 2013, doi: 10.1007/s10278-012-9566-3. [5] X. Yang, J. Jin, M. Yuchi, and M. Ding, “Ultrasound carotid artery intima-media thickness (IMT) segmentation review,” in Proc. ICBMI, Dec. 2011, pp. 97–100. [6] S. H. Johnsen and E. B. Mathiesen, “Carotid plaque compared with intima-media thickness as a predictor of coronary and cerebrovascular disease,” Current Cardiol. Rep., vol. 11, no. 1, pp. 21–27, Jan. 2009. [7] C. D. Ainsworth, C. C. Blake, A. Tamayo, V. Beletsky, A. Fenster, and J. D. Spence, “Three-dimensional ultrasound measurement of change in carotid plaque volume: A tool for quickly measuring effects of treatment on atherosclerosis,” Stroke, vol. 36, no. 9, pp. 1904–1909, 2005. [8] M. Egger, J. D. Spence, A. Fenster, and G. Parraga, “Validation of three-dimensional ultrasound vessel wall volume: An imaging phenotype of carotid atherosclerosis,” Ultrasound Med. Biol., vol. 33, no. 6, pp. 905–914, 2007. [9] B. Chiu, M. Egger, J. D. Spence, G. Parraga, and A. Fenster, “Quantification of carotid vessel wall and plaque thickness change using 3D ultrasound images,” Med. Phys., vol. 35, pp. 3691–3710, Jul. 2008. [10] H. G. Beebe, S. X. Salles-Cunha, R. P. Scissons, S. M. Dosick, R. C. Whalen, S. S. Gale, et al., “Carotid arterial ultrasound scan imaging: A direct approach to stenosis measurement,” J. Vascular Surgery, vol. 29, no. 5, pp. 838–844, 1999. [11] A. H. Rotstein, R. N. Gibson, and P. M. King, “Direct B-mode NASCET style stenosis measurement and Doppler ultrasound as parameters for assessment of internal carotid artery stenosis,” Australasian Radiol., vol. 46, no. 1, pp. 52–56, Mar. 2002.

[12] A. Landry, J. D. Spence, and A. Fenster, “Measurement of carotid plaque volume by 3-dimensional ultrasound,” Stroke, vol. 35, no. 4, pp. 864–869, Mar. 2004. [13] E. C. Kyriacou, C. Pattichis, M. Pattichis, C. Loizou, C. Christodoulou, S. K. Kakkos, et al., “A review of noninvasive ultrasound image processing methods in the analysis of carotid plaque morphology for the assessment of stroke risk,” IEEE Trans. Inf. Technol. Biomed., vol. 14, no. 4, pp. 1027–1038, Jul. 2010. [14] E. Ukwatta, J. Awad, A. D. Ward, D. Buchanan, G. Parraga, and A. Fenster, “Coupled level set approach to segment carotid arteries from 3D ultrasound images,” in Proc. IEEE ISBI, Apr. 2011, pp. 37–40. [15] F. Conversano, E. Casciaro, R. Franchini, A. Lay-Ekuakille, and S. Casciaro, “A quantitative and automatic echographic method for real-time localization of endovascular devices,” IEEE Trans. Ultrason. Ferroelectr. Freq. Control, vol. 58, no. 10, pp. 2107–2117, Oct. 2011. [16] F. Conversano, R. Franchini, C. Demitri, L. Massoptier, F. Montagna, A. Maffezzoli, et al., “Hepatic vessel segmentation for 3D planning of liver surgery: Experimental evaluation of a new fully automatic algorithm,” Acad. Radiol., vol. 18 no. 4, pp. 461–470, Apr. 2011. [17] S. Casciaro, F. Conversano, S. Musio, E. Casciaro, C. Demitri, and A. Sannino, “Full experimental modelling of a liver tissue mimicking phantom for medical ultrasound studies employing different hydrogels,” J. Mater. Sci. Mater. Med., vol. 20 no. 4, pp. 983–989, Apr. 2009. [18] D. C. Barrat, B. B. Ariff, K. N. Humphries, S. A. M. G. Thom, and A. D. Hughes, “Reconstruction and quantification of the carotid artery bifurcation from 3-D ultrasound images,” IEEE Trans. Med. Imaging, vol. 23, no. 5, pp. 567–583, May 2004. [19] A. Fenster, B. Chiu, A. Landry, J. D. Spence, and G. E. Parraga, “3D ultrasound system for analysis of carotid plaque progression and regression,” in Proc. 14th ACSSC, Nov. 2006, pp. 1966–1970. [20] C. Liguori, A. Paolillo, and A. Pietrosanto, “An automatic measurement system for the evaluation of carotid intima-media thickness,” IEEE Trans. Instrum. Meas., vol. 50, no. 6, pp. 1684–1691, Dec. 2001. [21] A. Fenster and D. B. Downey, “3-D ultrasound imaging: A review,” IEEE Eng. Med. Biol., vol. 15, no. 6, pp. 41–51, Dec. 1996. [22] J. A. Hossack, T. S. Sumanaweera, and S. Napel, “Quantitative 3D ultrasound imaging using an automated image tracking technique,” in Proc. IEEE Ultrason. Symp., vol. 2. Oct. 2000, pp. 1593–1596. [23] D. Capriglione, L. Ferrigno, C. Liguori, and A. Paolillo, “Volumetric carotid plaque measurements based on ultrasound images: A preliminary approach,” in Proc. 12th IMEKO, 2002, pp. 255–260. [24] L. Ferrigno, V. Paciello, and A. Paolillo, “A low cost measurement system for the 3-D evaluation of carotid plaque based on ultrasound images,” in Proc. 13th IMEKO, 2004, pp. 468–474. [25] J. E. Wilhjelm, M.-L. M. Gronholdt, S. Rasmussen, K. Martinsen, and H. Sillesen, “Estimation of plaque contents with multi-angle 3D compound imaging,” in Proc. IEEE Ultrason. Symp., Apr. 1996, pp. 1077–1080. [26] Evaluation of Measurement Data—Guide to the Expression of Uncertainty in Measurement, JCGM Standard 100:2008, 1999.

Domenico Capriglione (M’04) was born in Cava de’ Tirreni, Italy, in 1975. He received the M.S. degree (cum laude) in electronic engineering from the University of Salerno, Salerno, Italy, in 2000. Since 2001, he has been an Assistant Professor of electrical and electronic measurements with the Department of Electrical and Information Engineering, University of Cassino and Southern Lazio, Cassino, Italy. His current research interests include measurement systems based on face recognition, DSP-based measurement systems, instrument fault detection and diagnosis, measurement of electromagnetic compatibility, and measurements on RF and telecommunication systems. He is co-author of more than 100 scientific articles, most of which were published in relevant international journals. He is a member of the Italian Association for Electrical and Electronic Measurements (GMEE), the National Inter-University Consortium for Telecommunications (CNIT), the Italian Federation of Electrical, Electronic, Automation, Information and Telecommunication (AEIT).

CAPRIGLIONE et al.: MEASUREMENT SYSTEM TO REALIZE 3-D CAROTID OCCLUSION MEASUREMENT

Luigi Ferrigno (M’04) received the M.S. degree in electronic engineering from the University of Salerno, Salerno, Italy, and the Ph.D. degree in electrical engineering from the University of Napoli, Napoli, Italy. He is currently an Associate Professor of Electrical and Electronic Measurements and the Chief of the Metrological Laboratory LAT105 with the Department of Electrical and Information Engineering at the University of Cassino and Southern Lazio, Italy. His current research interests include the realization and characterization of wireless sensor networks, the realization of the measurement system for nondestructive testing, the characterization of electric system and components in nonsinusoidal conditions, and characterization of wired and RF digital transmission apparatuses.

Gianfranco Miele (S’06–M’08) was born in Cassino, Italy, on May 26, 1979. He received the M.S. degree (cum laude) in telecommunication engineering and the Ph.D. degree in electrical and information engineering from the University of Cassino, in 2004 and 2008, respectively. Since 2011, he has been an Adjunct Researcher of electrical and electronic measurements with the Dipartimento di Ingegneria Elettrica e dell’Informazione (DIEI), formerly the Dipartimento di Automazione, Elettromagnetismo, Ingegneria dell’Informazione e Matematica Industriale, University of Cassino and Southern Lazio, Cassino, Italy. In 2008, he was awarded the Carlo Offelli Prize for the best Ph.D. dissertation in electronic measurement subject titled “Design and implementation of an apparatus for reliable and repeatable power measurement in DVBT systems.” His current research interests include electrical and electronic measurements, and, in particular, design and implementation of innovative methods for performance assessment of RF telecommunication systems and communication networks, image-based measurement systems, measurement of electromagnetic compatibility, and DSP-based measurement systems. He is a member of the Italian Association for Electrical and Electronic Measurements (GMEE) and of the IEEE “Instrumentation and Measurement Society.”

757

Vincenzo Paciello (M’08) was born in Salerno, Italy, in 1977. He received the M.S. degree in electronic engineering and the Ph.D. degree in information engineering from the University of Salerno, Fisciano, Italy, in 2002 and 2006, respectively. Since 2008, he is an Assistant Professor of electrical and electronic measurements with the University of Salerno. He has been with the new Department of Industrial Engineering of the same university since January 2011. His current research interests include mechanical and electronic measurements, wireless sensor networks, instrument interfaces, and digital signal processing for advanced instrumentation.

Alfredo Paolillo (M’08) was born in Belvedere Marittimo, Italy, in 1972. He received the M.S. degree in electronic engineering and the Ph.D. degree in information engineering from the University of Salerno, Salerno, Italy, in 2000 and 2004, respectively. He is an Assistant Professor of electronic measurements with the Faculty of Engineering, University of Salerno, Italy. He has been with the Department of Industrial Engineering of the University of Salerno since January 2011. His main interests are in measurement systems based on vision and on numerical signal analysis.

Paolo Sommella (M’11) received the M.S. degree in electronic engineering and the Ph.D. degree in information engineering from the University of Salerno, Fisciano, Italy, in 2004 and 2008, respectively. He has been a Research Fellow with the University of Salerno since 2010. His current research interests are instrument fault detection and diagnosis, measurement in software engineering, and biomedical image processing.