Adaptive image inversion of contrast 3D echocardiography for enabling automated analysis
Introduction
Cardiovascular disease (CVD) is the leading cause of deaths worldwide [1]. To reduce the number of CVD mortality rate, it is necessary to have an efficient diagnostic method for timely detection and monitoring of disease status. For this purpose, cardiac imaging is frequently used by the cardiologist for the diagnosis of CVD and the study of cardiac structure as well as its function by automatic or semi-automatic analysis. Moreover, the segmentation of left ventricle (LV) endocardium or epicardium boundaries is an active area of research.
Among the cardiac imaging modalities, the echocardiography is generally accepted as a quick, non-invasive, cost effective and practical method for examining and exploring the cardiac structure as well as its function at rest and during stress. Real-time three-dimensional echocardiography (3DE) is a relatively recent addition to the echocardiography imaging scene [2] with success in providing more accurate and useful results. Despite the fact of being widely used in a large proportion of patients, the 3DE occasionally fails to produce diagnostically useful images [3]. The image quality is known to be affected by ultrasound physics and influenced by factors such as fat, rib spacing, ultrasound reflection angle, and lung breathing [3], [4]. Therefore, the echocardiography suffers from intensity dropout, missing features and speckle noise. These artefacts may further lead to inaccurate assessment and anomalous interpretation of the cardiac structure. Fig. 1 presents some examples of 2D cross-sections from 3DE images with intensity dropout, where the arrows point to the missing anatomical structures.
To tackle these limitations, contrast enhanced 3d echocardiography (C3DE) scans are acquired as an alternative choice with relatively improved image quality for patients with poor acoustic window. In contrast echocardiography, the contrast agent drugs containing hyper-echoic micro-bubbles are injected intravenously. The ultrasound characteristics of these contrast agents are distinctly different from the blood and the cardiac tissue. The contrast agents intravenously mixed in the blood produce strong backscatter to the ultrasound waves, thus causing an increase in the echocardiographic signal and resulting in better image quality [6].
The concentration of micro-bubbles in the LV blood pool is much higher than in vessels of the cardiac muscles, thus resulting in inverted image appearance (see Fig. 2) compared with routine 3DE image and often provides a clear LV endocardium boundary [7]. The visual appearance of endocardial delineation is often improved in contrast echocardiography, thus enhancing the cardiologist׳s diagnostic ability. Cardiologists often choose the contrast images over non-contrast images, when corresponding non-contrast image quality is poor due to poor acoustic window.
Although the image quality in C3DE is perceived to be improved for visual analysis, it seems to deteriorate for the purpose of automatic or semi-automatic analysis due to high speckle noise and intensity inhomogeneity within the LV cavity. In the C3DE images, due to this inverted appearance compared to the 3DE, the LV segmentation methods proposed for 3DE images fail on the C3DE images. Moreover, the commercial software (QLAB [8] and TomTec [9]) do not presently support automatic or semi-automatic C3DE segmentation. Therefore, the LV endocardial feature extraction and segmentation from C3DE images remain challenging problems. In the previous literature, no considerable work has been done to address these issues. Hence, there is a genuine need for the exploration of C3DE pre-processing technique to enable automated image analysis.
The C3DE image analysis appears to face the following difficulties, due to image quality, which hinder the subsequent automatic or semi-automatic image analysis:
- a)
Contrast intensity inhomogeneity within the LV cavity, as a result of reflection and scattering of the ultrasound beam by many uncorrelated free moving contrast microbubbles (see Fig. 3(a)).
- b)
There is a relatively low contrast or slow intensity change between brighter LV cavity and the myocardium border (see Fig. 3(b)). This low contrast between myocardium border and the LV cavity can obstruct the traditional LV edge driven feature extraction.
This work proposes an adaptive automatic pre-processing technique for image inversion in which the new inverted representation of a C3DE image is constructed to overcome the abovementioned difficulties. This inverted representation facilitates the further automatic or semi-automatic analysis.
This paper is organized as follows: Section 2 describes the relevant research work. Section 3 briefly elaborates the proposed work. Further, 4 Results, 5 Discussion present and discuss the results, respectively. Towards the end, the paper ends with concluding remarks in Section 6.
Section snippets
Related work
To the best of our knowledge, there has been no previous work in the literature on C3DE pre-processing to assist in subsequent image analysis. However, considerable work has been done in ultrasound and echocardiography image enhancement. Zwirn and Akselrod [10] proposed a histogram based technique for the noise reduction in the echocardiographic images. They used an adaptive method of brightness transfer function to increase the spread of the grey level histogram in order to improve image
Methods
The proposed work consists of estimating automatic threshold by a data analysis process through histogram analysis. Subsequently, this threshold is utilized to construct the inverse image. Below, we describe: (a) the simple scheme of inverse image construction, and (ii) the method for estimation of threshold through histogram analysis.
Data
We obtained C3DE scan sequences from 16 subjects which were recorded at the Alberta University Hospital, Canada. The recordings were acquired with a volumetric matrix-array X5 probe using a Philips iE33 ultrasound system. The image spatial dimensions for each scan were either 160×144×208 or 192×176×208 depending on the settings of the ultrasound system. It is noted that the C3DE is not yet used in routine clinical practice and it is used primarily for research studies, thus the data size may
Automatic threshold estimation
The performance of ATE of threshold through five validation schemes is presented in Table 3 to Table 7 respectively, which demonstrates the close match with manual threshold with a reasonable mean error. Through visual analysis of inverted image by changing the threshold manually, we observed that the variation in the threshold value by up to 10 points (intensity level) is tolerable (please see Fig. 6 illustrating this observation). In ATE, first two validation schemes are considered to
Conclusions
Even though the image quality in C3DE is perceived to be improved for visual analysis, but it seems to deteriorate for the purpose of automatic or semi-automatic analysis due to two main issues: (i) the contrast inhomogeneity within the LV cavity, and (ii) the relatively low contrast between the LV cavity and the myocardium border. Thus, the LV endocardial segmentation from C3DE images remains a challenging problem with little published work in the literature. To deal with this challenge, we
Conflict of interest statements
None declared.
Acknowledgement
We are grateful to Professor Harald Becher from Alberta University Hospital, Edmonton, Canada for the provision of C3DE data.
References (19)
- et al.
American Society of Echocardiography consensus statement on the clinical applications of ultrasonic contrast agents in echocardiography
J. Am. Soc. Echocardiogr.
(2008) - M. Nichols, N. Townsend, R. Luengo-Fernandez, J.Leal, A. Gray, P. Scarborough, M. Rayner, European Cardiovascular...
- et al.
3D echocardiography: a review of the current status and future directions
J. Am. Soc. Echocardiogr.
(2007) - et al.
Contrast echocardiography: evidence-based recommendations by European Association of Echocardiography
Eur. J. Echocardiogr.
(2009) - K.M. Rajpoot, Multi-view 3D Echocardiographic Image Analysis Ph.D. Thesis, Institute of Biomedical Engineering,...
- H. Becher P.N. Burns, Handbook of Contrast Echocardiography: Left Ventricular Function and Myocardial Perfusion:,...
- M. Ma, M. Stralen, J.H. Reiber, J.G. Bosch, B.P. Lelieveldt, Left ventricle segmentation from contrast enhanced fast...
- Philips QLAB: Cardiac 3D Quantification Advanced (3DQ Advanced) Available:...
- Tomtec Imaging System. Available: 〈http://www.tomtec.de/〉 (Last access date:...
Cited by (1)
Ultrasonic Enhancing Agents for the Cardiothoracic Anesthesiologist: A Focused Review of the 2018 American Society of Echocardiography Guidelines Update
2019, Journal of Cardiothoracic and Vascular AnesthesiaCitation Excerpt :Current commercial 3D software does not have the capability to automatically analyze opacified LV space and hence may fail to generate meaningful results. Shaheen et al. designed a new software algorithm to invert the brightness of 3D images.15 This method allows LV cavity enhanced by UEAs to appear dark, similar to that acquired by traditional 3D echocardiography, and enables semiautomatic LV quantification.