https://doi.org/10.1177/0161734618777262

Ultrasonic Imaging2018, Vol. 40(4) 232 –244

© The Author(s) 2018 Reprints and permissions:

sagepub.com/journalsPermissions.nav DOI: 10.1177/0161734618777262

journals.sagepub.com/home/uix

Article

Automated Identification and Localization of the Inferior Vena Cava Using Ultrasound: An Animal Study

Jiangang Chen1, Jiawei Li2,3, Xin Ding4, Cai Chang2,3, Xiaoting Wang4, and Dean Ta5

AbstractUltrasound measurement of the inferior vena cava (IVC) is widely implemented in the clinic. However, the process is time consuming and labor intensive, because the IVC diameter is continuously changing with respiration. In addition, artificial errors and intra-operator variations are always considerable, making the measurement inconsistent. Research efforts were recently devoted to developing semiautomated methods. But most required an initial identification of the IVC manually. As a first step toward fully automated IVC measurement, in this paper, we present an intelligent technique for automated IVC identification and localization. Forty-eight ultrasound data sets were collected from eight pigs, each of which included two frames in B-mode and color mode (C-mode) collected at the inspiration, and two cine loops in B-mode and C-mode. Static and dynamic automation algorithms were applied to the data sets for identifying and localizing the IVC. The results were evaluated by comparing with the manual measurement of experienced clinicians. The automated approaches successfully identified the IVC in 47 cases (success rate: 97.9%). The automated localization of the IVC is close to the manual counterpart, with the difference within one diameter. The automatically measured diameters are close to those measured manually, with most differences below 15%. It is revealed that the proposed method can automatically identify the IVC with high success rate and localize the IVC with high accuracy. But the study with high accuracy was conducted under good control and without considering difficult cases, which deserve future explorations. The method is a first step toward fully automated IVC measurement, which is suitable for point-of-care applications.

Keywordsultrasound, vessel identification, vessel localization, inferior vena cava, automated measurement

1Academy for Engineering and Technology, Fudan University, Shanghai, China2Department of Medical Ultrasound, Fudan University Shanghai Cancer Center, Shanghai, China3Department of Oncology, Shanghai Medical College, Fudan University, Shanghai, China4Department of Critical Care Medicine, Peking Union Medical College Hospital, Peking Union Medical College, Chinese Academy of Medical Sciences, Beijing, China5Department of Electronic Engineering, Fudan University, Shanghai, China

Corresponding Author:Dean Ta, Professor, Department of Electronic Engineering, Fudan University, RM519, Physics Building, No. 220 Handan Road, Yangpu district, Shanghai, P.R. China. Email: tda@fudan.edu.cn

777262 UIXXXX10.1177/0161734618777262Ultrasonic ImagingChen et al.research-article2018

Chen et al. 233

Background

Outperforming computed tomography (CT) and magnetic resonance imaging (MRI) in time effi-ciency and mobility, ultrasonography is increasingly employed as a point-of-care bedside diagnosis modality for clinicians.1,2 In particular, ultrasound is frequently used to quickly and conveniently check/measure a patient’s status in the emergency department (ED) and critical care unit (CCU).1,3 Therein, ultrasound measurement of the inferior vena cava (IVC) is widely accepted in ED and CCU.4-7 As reported by a number of studies, for example, on healthy pediatric patients,8 healthy blood donors,9 critically ill patients,10 and those with liver fibrosis or cirrhosis,11 variation in the ultrasound-measured diameter during respiration reflects the volume status. In the past decade, point-of-care ultrasound is increasingly used to measure the IVC in critically ill, ventilated patients for determination of fluid responsiveness and to guide fluid therapy.12-17

However, ultrasound measurements of the IVC face some challenges:

1. The measurement is made without standardization.18 There are longitudinal and cross-sectional imaging approaches in clinical practice, resulting in disputed diagnostic cutoffs.19

2. The measurement is labor intensive and time consuming, in that (a) the IVC moves with respiration and (b) ultrasound measurement requires skilled operations, leading to signifi-cant interoperator variability.20

3. Artificial errors are always considerable as a result of the special nature of the IVC measurement.19

To alleviate the current embarrassment, research efforts recently were devoted to developing computer-assisted methods for ultrasound IVC measurements. Reprehensively, Bellows et al.21 proposed an image-processing method to automatically track and measure both the diameter and respiratory variation of the IVC. However, the method required the operator to manually identify the IVC by marking the IVC’s centroid before the measurement. Mesin et al.22 developed a semi-automated method to track the movement of the IVC based on a moving M-line. Similarly, in the study, the operator was first asked to select two reference points to serve as a base for tracking the movement and rotation of the IVC, and then two more points to indicate an M-line. The movement of the M-line was then traced to obtain the maximum and minimum diameter, as well as the diameter variation of the IVC with respiration. However, serving as a base of the proposed methods, existing studies still require manual identification and localization of an IVC, making the IVC measurement less efficient and open to individual errors.

On the other hand, ultrasound is increasingly used by nonsonographic clinicians, in particular, those in pre-hospital settings and community hospitals, with portable and hand-held/mobile ultrasound scanners.23-26 In those situations, the clinicians lack ultrasound operation skills and have limited training opportunities, especially in developing countries. In addition, the clinician-patient ratio is significantly low, e.g., in China.

Due to the above drawbacks, smart algorithms for automated ultrasound scanning, e.g., fully automatic ultrasound IVC measurement, are needed. For this purpose, the first step is no doubt to identify and localize the IVC in an automated manner. In this study, we performed the first attempt to automatically identify and localize the IVC by testing ultrasound data collected from pigs.

Method

Eight pigs (female, age: 13-16 weeks) were scanned using a clinically available ultrasound machine CX-50 (Philips Healthcare, Bothell, Washington) equipped with a curved probe (C5-1,

234 Ultrasonic Imaging 40(4)

central frequency: 1-5 MHz). Two clinicians, one from the department of medical ultrasound of the Shanghai Cancer Center (Shanghai, China) and the other from the department of critical care medicine of Peking Union Medical College Hospital (Beijing, China), took part in the study. They were responsible for data collection and manual identification/localization of the IVC. Both clinicians have working experience of more than six years using ultrasound. The study was approved by the ethics committee of Fudan University, Shanghai, China.

Study Design

1. The animal was fixed on the operation platform in a supine position; 2. The animal was pre-anesthetized by the 846 mixture (0.8-1.0 mg/kg IM [intramuscular]).

Subsequently, anesthesia was maintained by 3% to 4% isoflurane. The animal was mechanically ventilated at a volume that was calculated by animal weight × 10 mL/kg at a breathing rate of 26 to 32 breaths/min;

3. During the whole study, the systolic and diastolic blood pressures, heart pulse rate, mean arterial pressure, and core temperature were monitored;

4. One clinician positioned the ultrasound transducer over the animal to identify the IVC in a longitudinal view (Figure 1). Two windows can be used, that is, the subcostal and the right upper quadrant (the latter uses the liver as a sonographic window);

5. After the IVC was identified, the ultrasound transducer was held in position by hand; 6. A frame at inspiration and a cine loop of 8 seconds (to cover more than one respiratory

cycle) in the B-mode were collected in the general abdominal imaging mode at a depth of 14 cm;

7. Step (6) was repeated but in color mode (C-mode). Note that the color window was preset to cover the center of the image, as shown in Figure 1(b);

8. Steps (4) to (7) were repeated two more times; 9. Steps (4) to (8) were repeated with the other sonographer; and10. The data in Digital Imaging and Communications in Medicine (DICOM) format were stored

on the ultrasound machine during the experiment, and then downloaded to a computer.

The above-depicted protocol resulted in a total of 48 ultrasound data sets, each of which included one frame in B-mode and one frame in C-mode collected at the inspiration, and one cine loop in B-mode and one cine loop in C-mode. Note that the frames and cine loops in each data set were collected in the same ultrasound scanning, that is, with the same probe position for the same pig by the same clinician. A laboratory-developed algorithm in Matlab (The MathWorks, Natick, Massachusetts, USA) was applied to test the collected ultrasound data for automated IVC identi-fication and localization.

Figure 1. IVC ultrasound images in (a) B-mode and (b) C-mode. IVC = inferior vena cava.

Chen et al. 235

Pre-process

The collected frames and those of the cine loops were first de-noised using a Gaussian filter with an 8 × 8 filter window. A region of interest (ROI) was preset to cover all useful information but reduce the data size significantly, as indicated in Figures 2(a) and 4(a) (white rectangle). Such an operation is done with the consideration that the algorithm may be employed in portable and hand-held ultrasound equipment where the computing resources are limited. Image enhancement based on contrast adjustment was then applied to the ROI (Figure 2c).

Candidate Region Selection

Several candidate regions are detected by either low intensity region detection in B-mode or color region detection in color mode, denoted as R nB [ ] or R nC [ ] , respectively. The former is conducted by first applying a threshold to the enhanced image (Figure 2c), resulting in a bina-rized image (Figure 2d). The candidate regions are extracted and indicated in Figure 2(e) (white circles). Note that the threshold is automatically determined by analysis of the histo-gram of the enhanced image (Figure 3b). The latter is processed by subtracting the blue part of a color image from its red counterpart, as shown in Figure 4(c). The candidate regions are presented in Figure 4(d) (white circles). The pixels inside R nB [ ] or R nC [ ] are denoted as

r n iB ,[ ] or r n iC ,[ ] , where r i n R nB B,[ ] ⊂ [ ] and r i n R nC C,[ ] ⊂ [ ] .

IVC Identification

Two different approaches are implemented to identify the IVC from the candidate regions.

1. Static approach

Several constraints are applied to all candidate regions detected in B-mode (Figure 2e) or C-mode (Figure 4d), that is, R nB [ ] or R nC [ ] , including

Figure 2. Presentation of automated approach based on B-mode. (a) IVC B-mode image (the white rectangle indicates the ROI), (b) the B-mode image of the ROI, (c) enhanced image of the ROI, (d) white and black image of the ROI, (e) the ROI with candidate regions, and (f) the ROI with identified IVC. IVC = inferior vena cava; ROI = region of interest.

236 Ultrasonic Imaging 40(4)

Figure 4. Presentation of automated approach based on C-mode. (a) IVC C-mode image (the white rectangle indicates the ROI), (b) the C-mode image of the ROI, (c) the difference between blue and red branches of the ROI in C-mode, (d) the ROI with candidate regions, and (e) the ROI with identified IVC. IVC = inferior vena cava; ROI = region of interest.

Figure 3. Histograms (left) and the corresponding images (right) of the B-mode (a) before and (b) after enhancement.

Chen et al. 237

a. Max R n Max P R nB B{ } { ( )}[ ] ∩ [ ]∫∫ for B-mode or

Max R n Max P R nC C{ } { ( )}[ ] ∩ [ ]∫∫ for C-mode

where R nB [ ]( )∫∫ , P R nB( )[ ] , R nC [ ]( )∫∫ and P R nC( )[ ] are the area and perim-

eter of the candidate regions detected in the B-mode and C-mode, respectively. Such constraints are determined due to the consideration that the IVC always demonstrates as the largest low intensity area in an ultrasound image.

b. L R n L R nX B y B[ ]( ) [ ]( ) for B-mode

or

L R n L R nX C y C[ ]( ) [ ]( ) for C-mode

where L R nX B [ ]( ) , L R ny B [ ]( ) , L R nX C [ ]( ) , and L R ny C [ ]( ) are the lengths of

the candidate regions in the x-dimension (horizontal) and y-dimension (vertical) in the B-mode and C-mode, respectively. This constraint is applied due to the observa-tion that the IVC is similar to a dark belt lying horizontally, whose length is much larger than its height, as shown in Figure 1. In this study, we set this constraint as

L R n L R nX B y B[ ]( ) > × [ ]( )3 for the B-mode and L R n L R nX C y C[ ]( ) > × [ ]( )3 for

the C-mode.2. Dynamic approach

max{ ( )}d R nt B [ ]∫∫ for B-mode

or

max{ ( )}d R nt C [ ]∫∫ for C-mode

where d R nt B( )[ ]∫∫ and d R nt C( )[ ]∫∫ denote the time variation of the areas of R nB [ ] and R nC [ ] , respectively. This constraint is selected with the assumption that the dimension of the IVC changes with respiration, demonstrating the largest time variation in area among the candidate regions.

3. Combination of static and dynamic methods

The above described static and dynamic methods are combined and applied collectively to identify and localize the IVC.

IVC Localization

Once the IVC is identified from the above procedure, a seed is automatically placed in the IVC

region, for example, at the pixel of r nB [ ] or r nC [ ] , which is the central of R nB [ ] or R nC [ ] ,

respectively. With r nB [ ] or r nC [ ] as a starting point, the IVC was localized in the following way:

1. With r nB [ ] or r nC [ ] as the center, a circle is created, and grows stepwise. The circle stops growing when it touches one of the boundaries of the IVC (denoted as touching

238 Ultrasonic Imaging 40(4)

boundary in the following). The touching boundary is determined where the standard deviation (SD) inside the circle encounters a sharp change (i.e., an increase of 50% in SD). Such an approach is employed since it is sensitive to location and local shape, although the vessel borders could also be detected by segmentation methods.27,28

2. The circle shifts vertically and away from the touching boundary determined in Step (1)

in a small step, for example, two pixels. Update r nB [ ] or r nC [ ] with the current circle center.

3. Repeat Steps (1) to (2) as long as to find both the upper and lower touching boundaries. As a result, the center of the circle is in the central line of the IVC.

4. The circle moves longitudinally along the long axis of the IVC toward the hepatic vein. Note that all the ultrasound scans follow the standard guideline, with the hepatic vein localized on the left side of the image.

5. The movement stops when the circle reaches the confluence of the hepatic vein, under the conditions that

i. The diameter of the circle increases significantly, e.g., an increase of more than 30% in diameter;

ii. The circle has more than two touching points with the vessel boundaries. This is under the consideration that the confluence of the hepatic vein is similar to a junc-tion of three roads. The circle may touch three boundaries at the confluence, as indicated in Figure 5 (red circle).

6. A marker ( Pa ) is automatically placed in the IVC central line 2 cm distal to the conflu-ence of the hepatic vein, as presented in Figure 5. The IVC diameter was then determined by the diameter of the circle at Pa , as denoted by ∅a in Figure 5.

The two clinicians were requested to localize the IVC manually. Blindly, each placed a marker ( Pm

i , i = 1,2) on the same ultrasound frame in which the automated algorithm places Pa . The marker was placed at the location following the same guideline as the earlier described, that is,

2 cm distal to the confluence of the hepatic vein. The midpoint Pm between Pm1 and Pm

2 are

Figure 5. The IVC ultrasound image demonstrating the locations of the confluence of the hepatic vein (center of the red circle) and where the IVC diameter is measured by manual (red star) and automated methods (center of the white circle). IVC = inferior vena cava.

Chen et al. 239

79.2%

66.7%

87.5%

72.9%

97.9%

0%

20%

40%

60%

80%

100%

Sta�c approach -B-mode

Sta�c approach -C-mode

Dynamic approach- B-mode

Dynamic approach- C-mode

Combine

Success rate of IVC iden�fica�on

Auto Manual

Figure 6. Success rates of different automated approaches and the manual method.

denoted as the final manual localization of the IVC, marked with a red star, as shown in Figure 5.

The diameter of the IVC at Pm was manually measured and denoted by ∅m .

Results

Figures 3(a) and 3(b) show the histogram of the original (Figure 2b) and enhanced (Figure 2c) ultrasound images. Table 1 lists the successful identification of the IVC by the different auto-mated approaches and manual measurement, which are also compared in Figure 6. The manual and automated localizations of the IVC were compared in terms of their distance, that is, P Pa m−

∅m

, where P Pa m− denotes the distance betweesn Pa and Pm , as presented in Figure 7.

The diameters measured at Pa and Pm are plotted in Figure 8(a). Figure 8(b) shows the normal-

ized difference between ∅a and ∅m , i.e., ∅

∅

−∅a m

m

. The automated and manual methods are

compared in terms of the Bland-Altman plot of the IVC diameter in Figure 8(c).

Table 1. Identification Number and Success Rate of Different Automated Approaches.

Identification No. Success rate

Auto Manual Auto Manual P-value

Static approach: B-mode 38 48 79.2% 100% 0.001Static approach: C-mode 32 48 66.7% 100% <0.0005Dynamic approach: B-mode 42 48 87.5% 100% 0.035Dynamic approach: C-mode 35 48 72.9% 100% <0.0005Combine 47 48 97.9% 100% 1.0

240 Ultrasonic Imaging 40(4)

Discussions

In this study, we tested an intelligent method for automated identification and localization of the IVC on animal subjects. The work of the study serves as a first step toward fully automated IVC measurement, that is, identify and localize the IVC in an automated way before the measurement.

In the study, the two sonographers successfully identified the IVC from all the data sets (suc-cess rate = 100%), while the automated approaches successfully identified as many as 47 from a total of 48 cases (success rate = 97.9%) (Table 1 and Figure 6). Therein, the static approach based on the C-mode gave the worst success rate (66.7%), since the Doppler signal is sensitive to the scanning angle of the probe, that is, the ultrasound machine does not always get a good blood signal during inspiration. On the other hand, the dynamic approach based on the C-mode gave a higher success rate than its static counterpart. This is because the dynamic approach does not rely on a single frame, which significantly increases the possibility of obtaining frames with good blood signals. It is noticed that the dynamic approach based on B-mode images presented the highest success rate among the individual approaches. Such an achievement may be attributed to the facts that (a) the ultrasound data were collected by sonographers/clinicians with rich experi-ence and under good controls and (b) image enhancement significantly improves the contrast of the image, as shown in Figure 2(c) and compared in terms of the histogram in Figure 3, which makes the automated selection of the threshold and extraction of candidate regions easy and robust. Importantly, the combination of both static and dynamic approaches based on both B-mode and C-mode leads to the highest success rate (97.9%). Another issue that may affect the success rate is the fact that the pigs were under ventilation during the experiment, which leads to large movement of the IVC. It is noted that all the proposed approaches fails to identify the IVC in one case due to poor image quality.

Different automated approaches met challenges in IVC identification in different aspects. Static approaches have the drawback that the analysis is based on a single frame obtained at inspiration, which does not always produce good signals, i.e., the Doppler signal. As to the approaches based on the B-mode images, a great challenge is the rib shadow, which creates significant interference when applying the automated algorithm to identify the IVC. The dynamic B-mode approach is less influenced by the rib shadow than the static counterpart because the rib shadows move, and change in position and size during respiration. As a result, the IVC is not always immersed in the shadow. The effect of the rib shadows should be compensated in future studies.

Figure 7. Normalized distance between the locations determined by manual and automated methods.

Chen et al. 241

Figure 8. (a) IVC diameters determined by manual measurement and automated approach, (b) normalized difference of the IVC diameters by manual measurement. IVC = inferior vena cava.

To further evaluate the algorithm, the manual and automated localizations of the IVC were compared and presented in Figure 7. It is observed that the automated localization of the IVC is close to the manual counterpart, with the distances less than one diameter. It is, thus, concluded that the automated algorithm can localize the IVC with high accuracy.

The automatically and manually measured diameters are compared in Figure 8. It is observed that the automatically measured diameters are close to those manually measured, with most differ-ences below 15% (Figure 8b). Those cases with larger differences may be attributed to the differ-ence in measurement nature of the manual and automated methods, in that, the manual measurement is based on the experience of the clinician to locate the IVC boundary, while the automated approach relies on the contrast of the gray value of the pixels at the boundaries and the threshold determined from the histogram (Figure 3). The data demonstrated in the Bland-Altman plot (Figure 8c) indicates that the automated approach is in good agreement with the manual method.

242 Ultrasonic Imaging 40(4)

Automated IVC measurement has been attempted in several studies recently published, in which the dynamic change of the IVC diameter with respiration was automatically traced and measured.21,22 However, in those studies, the IVC was identified and localized manually, making the methods semi-automated. Toward fully automated IVC measurement, that is, without the interference of a human, it is the first step that the IVC is identified and localized in an automated way. In this study, we tried the first attempt to automatically identify and localize the IVC, which may be the basis of the future work toward fully automated IVC measurement. The clinical application of the algorithm may be realized by implementing it to an ultrasound machine to achieve automated identification, localiza-tion, and movement tracking and measurement of the IVC by pressing only one button. In real clini-cal application, the algorithm may process ultrasound data in the sequence of static B-mode, dynamic B-mode, static Color-mode, and then dynamic Color-mode. The process stops when the IVC is identified and localized at any step. In addition, the switch between B-mode and Color-mode can be realized automatically. Thus, the application of the algorithm would not increase the clinician’s work load but make the operation more efficient. The computational cost of the proposed algorithm is not high. The whole process takes less than 0.1 seconds in a computer (CPU: 2.6GHz Core-2, RAM: 8GB, Operation system: Window 7 64bit). Aware of this and considering that analysis on DICOM data in the study does not require high computation capability, the proposed algorithm is applicable to portable and mobile ultrasound devices.

There are several limitations of the study. First, the ultrasound data were collected by experienced clinicians. The study was conducted under good control and without considering difficult cases. For example, sometimes the vein is not easily visible with artefacts like gas scattering the ultrasound rays. In the case of hypovolemic patients, the vein could be closed. Moreover, the confluence of hepatic veins is not always clearly visible. Second, the algorithms were tested in animal subjects. However, respiration cycles of animals could be different from those of human. In addition, humans may exhibit different anatomies in terms of depth, dimension, and the ultrasound image quality, e.g., contrast and noise. Third, the study was only conducted on healthy subjects under ventilations without consider-ation of any pathological conditions and under natural breathing. Fourth, considering that the IVC measurement is a three-dimensional problem,29,30 in this study, we did not consider the case if the IVC has irregular shapes (e.g., saber shape or irregularly varying diameter along the vessel).

Future study will include (a) improving the algorithm to alleviate the influence of affecting factors on the IVC identification and localization (e.g., rib shadows and gas), and apply to diffi-cult cases as aforementioned; (b) testing the automated algorithm on novice-collected data, in particular those from subjects under a variety of conditions, for example, human subjects of dif-ferent pathologies under natural breathing or ventilation; (c) exploring the three-dimensional problem of the IVC measurement and the effect of irregular shapes of the IVC on the measure-ment of the diameter; and (d) developing algorithms for IVC movement tracking and measure-ment, thereby providing a fully automated IVC measurement technique.

Conclusion

In this study, we tried the first attempt to automatically identify and localize the IVC on animal subjects, with an aim to assist clinicians in measuring the IVC in a fully automated way. With manual identification and localization of the IVC as the ground truth, it is concluded that the combination of the proposed automated approaches can identify and localize the IVC with high success rate and accuracy. In clinical practice, the operator should be familiar with the features of the approach and practice to adapt the acquisition to the needs of the algorithm. In future stud-ies, algorithms for automated movement tracking and measurement of the IVC will be developed to achieve a fully automated IVC measurement technique. In addition, human subjects of differ-ent pathologies under natural breathing and ventilation will be enrolled. The algorithm will be improved by accounting for the impact of affecting factors, e.g., rib shadows.

Chen et al. 243

Acknowledgments

The authors would like to thank the reviewers in advance for their comments and suggestions.

Ethical Approval and Consent to Participate

This study was approved by the ethics committee of Fudan University.

Availability of Data and Materials

The datasets during and/or analyzed during this study are available from the corresponding author on rea-sonable request.

Author Contribution

Jiangang Chen and Jiawei Li contributed equally.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publi-cation of this article: This work was supported by the Natural Science Foundation of China (11525416).

ORCID iD

Dean TA https://orcid.org/0000-0001-6651-4491

References

1. Shrestha GS. Point-of-care ultrasonography: a practical step in the path to precision in critical care. Crit Care. 2017;21(1):215.

2. Huang Q, Luo Y, Zhang Q. Breast ultrasound image segmentation: a survey. Int J Comput Assist Radiol Surg. 2017;12(3):493-507.

3. Wang X-t, Ding X, Zhang H-m, Chen H, Su L-x, Liu D-w. Lung ultrasound can be used to predict the potential of prone positioning and assess prognosis in patients with acute respiratory distress syndrome. Crit Care. 2016;20(1):385.

4. Finnerty NM, Panchal AR, Boulger C, Vira A, Bischof JJ, Amick C, et al. Inferior vena cava measure-ment with ultrasound: what is the best view and best mode? West J Emerg Med. 2017;18(3):496-501.

5. Au AK, Matthew Fields J. Ultrasound measurement of inferior vena cava collapse predicts propofol induced hypotension. Am J Emerg Med. 2017;35(3):508-9.

6. Airapetian N, Maizel J, Alyamani O, Mahjoub Y, Lorne E, Levrard M, et al. Does inferior vena cava respiratory variability predict fluid responsiveness in spontaneously breathing patients? Crit Care. 2015;19(1):400.

7. Charbonneau H, Riu B, Faron M, Mari A, Kurrek MM, Ruiz J, et al. Predicting preload respon-siveness using simultaneous recordings of inferior and superior vena cavae diameters. Crit Care. 2014;18(5):473.

8. Haines EJ, Chiricolo GC, Aralica K, Briggs WM, Van Amerongen R, Laudenbach A, et al. Derivation of a pediatric growth curve for inferior vena caval diameter in healthy pediatric patients: brief report of initial curve development. Crit Ultrasound J. 2012;4(1):12.

9. Lyon M, Blaivas M, Brannam L. Sonographic measurement of the inferior vena cava as a marker of blood loss. Am J Emerg Med. 2005;23(1):45-50.

10. Akkaya A, Yesilaras M, Aksay E, Sever M, Atilla OD. The interrater reliability of ultrasound imaging of the inferior vena cava performed by emergency residents. Am J Emerg Med. 2013;31(10):1509-11.

244 Ultrasonic Imaging 40(4)

11. Kitamura H, Kobayashi C. Impairment of change in diameter of the hepatic portion of the inferior vena cava: a sonographic sign of liver fibrosis or cirrhosis. J Ultrasound Med. 2005;24(3):355-59; quiz 60-61.

12. Anderson KL, Jenq KY, Fields JM, Panebianco NL, Dean AJ. Diagnosing heart failure among acutely dyspneic patients with cardiac, inferior vena cava, and lung ultrasonography. Am J Emerg Med. 2013;31(8):1208-14.

13. Machare-Delgado E, Decaro M, Marik PE. Inferior vena cava variation compared to pulse contour analysis as predictors of fluid responsiveness: a prospective cohort study. J Intensive Care Med. 2011;26(2):116-24.

14. Barbier C, Loubieres Y, Schmit C, Hayon J, Ricome JL, Jardin F, et al. Respiratory changes in infe-rior vena cava diameter are helpful in predicting fluid responsiveness in ventilated septic patients. J Intensive Care Med. 2004;30(9):1740-46.

15. Gomez Betancourt M, Moreno-Montoya J, Barragan Gonzalez AM, Ovalle JC, Bustos Martinez YF. Learning process and improvement of point-of-care ultrasound technique for subxiphoid visualization of the inferior vena cava. Crit Ultrasound J. 2016;8(1):4.

16. Huang Q, Zeng Z, Li X. 2.5-Dimensional Extended Field-of-View Ultrasound. IEEE Transactions on Medical Imaging. 2017;37(4):851-59.

17. Huang Q, Wu B, Lan J, Li X. Fully automatic three-dimensional ultrasound imaging based on conven-tional B-Scan. IEEE trans biomed circ sys. 2018;12(2):426-36.

18. Wallace DJ, Allison M, Stone MB. Inferior vena cava percentage collapse during respiration is affected by the sampling location: an ultrasound study in healthy volunteers. Acad Emerg Med. 2010;17(1):96-99.

19. Zhang Z, Xu X, Ye S, Xu L. Ultrasonographic measurement of the respiratory variation in the inferior vena cava diameter is predictive of fluid responsiveness in critically ill patients: systematic review and meta-analysis. Ultrasound Med Biol. 2014;40(5):845-53.

20. Akkaya A, Yesilaras M, Aksay E, Sever M, Atilla OD. The interrater reliability of ultrasound imaging of the inferior vena cava performed by emergency residents. Am J Emerg Med. 2013;31(10):1509-11.

21. Bellows S, Smith J, McGuire P, Smith A. Validation of a computerized technique for automatically tracking and measuring the inferior vena cava in ultrasound imagery. Stud Health Technol Inform. 2014;207:183-92.

22. Mesin L, Pasquero P, Albani S, Porta M, Roatta S. Semi-automated tracking and continuous monitor-ing of inferior vena cava diameter in simulated and experimental ultrasound imaging. Ultrasound Med Biol. 2015;41(3):845-57.

23. Stoica R, Heller EN, Bella JN. Point-of-care screening for left ventricular hypertrophy and concen-tric geometry using hand-held cardiac ultrasound in hypertensive patients. Am J Cardiovasc Dis. 2011;1(2):119-25.

24. Evangelista A, Galuppo V, Mendez J, Evangelista L, Arpal L, Rubio C, et al. Hand-held cardiac ultrasound screening performed by family doctors with remote expert support interpretation. Heart. 2016;102(5):376-82.

25. Kratz T, Exner M, Campo dell’Orto M, Timmesfeld N, Schuttler KF, Efe T, et al. A pocket-sized hand held ultrasound system for intraoperative transthoracic echocardiography by anaesthesiologists: a fea-sibility study. Technol Health Care. 2016;24(3):309-15.

26. Siso-Almirall A, Kostov B, Navarro Gonzalez M, Cararach Salami D, Perez Jimenez A, Gilabert Sole R, et al. Abdominal aortic aneurysm screening program using hand-held ultrasound in primary health-care. PLoS One. 2017;12(4):e0176877.

27. Luo Y, Liu L, Huang Q, Li X. A novel segmentation approach combining region-and edge-based infor-mation for ultrasound images. Biomed Res Int. 2017;2017:1-18.

28. Vercio LL, Orlando JI, del Fresno M, Larrabide I. Assessment of image features for vessel wall seg-mentation in intravascular ultrasound images. Int J Comput Assist Radiol Surg. 2016;11(8):1397-1407.

29. Blehar DJ, Resop D, Chin B, Dayno M, Gaspari R. Inferior vena cava displacement during respiropha-sic ultrasound imaging. Crit Ultrasound J. 2012;4(1):18.

30. Huang Q, Zeng Z. A review on real-time 3D ultrasound imaging technology. Biomed Res Int. 2017;2017:1-20.