Ultrasound tissue classification: a review
Summary (5 min read)
Introduction
- Different medical imaging modalities are available and widely used nowadays in clinical practice to create images of the human body, such as computed tomography (CT), magnetic resonance imaging (MR), positron emission tomography (PET), and ultrasound (US).
- Driven by the unmet clinical needs to distinguish different tissue types (e.g., healthy versus diseased) in US, tissue characterization and classification have received much attention in recent years [1], [2].
- Unfortunately in practice most extracted features have low discriminative power.
- This paper attempts to provide a review on ultrasound tissue classification, particularly focusing on recent advances in this area.
II. CLINICAL APPLICATIONS
- Medical ultrasound has a very broad range of clinical uses.
- Ultrasound tissue classification can be applied in many clinical fields, for instance, tissue classification plays an important role in ultrasoundbased cancer diagnosis, e.g., by classifying the tissue regions as benign or malignant.
- Here the authors briefly introduce the main clinical applications that have received the most attention in the recent literature.
A. Cardiology
- The primary aim of non-invasive cardiac imaging is to provide information on the diagnosis and severity of underlying cardiac conditions [7].
- Echocardiography (ultrasound imaging of the heart) is the most common cardiac imaging procedure performed in clinical practice, due to its portability, low cost, and patient acceptance.
2 JOURNAL OF LATEX CLASS FILES, VOL. 13, NO. 9, SEPTEMBER 2014
- Echocardiography has also been one of the driving application areas of medical ultrasound [8].
- Cardiac imaging techniques characterise the underlying tissue directly, by assessing a signal from the tissue itself, or indirectly, by inferring tissue characteristics from global or regional function [7].
- One challenge for ablation therapy is the monitoring of the temperature rise and the extend of the ablated region in the tissue.
- In the era of atherosclerosis, intravascular ultrasound (IVUS), a catheter-based imaging technique, has evolved as a valuable technique for diagnosis and intervention for coronary disease, by providing more precise measurement on intimal thickness and vulnerable plaques.
- Tissue classification in IVUS images can automatically predict vulnerable plaques as well as quantify the amount of the different tissues; many approaches have been proposed in literature [31]–[36], which will be discussed in detail in the following sections.
C. Breast Cancer
- Breast cancer is the most common cancer in women globally, and early detection is the key to reduce the death rate.
- To improve cancer detection in dense breasts, personalized breast cancer screening with ultrasound has been proposed for women with dense breasts and women with elevated risk factors for developing breast cancer.
- Furthermore, BUS is more convenient and safer for clinical use [39] and it can be used as an alternative screening device to mammography for women with harmful mutations in either BRCA1 (breast cancer gene type1) or BRCA2 since no radiation is involved.
- Because of the standard imaging procedure, it is possible to perform temporal analysis on prior and current exams.
- Since the ABUS images are standardly defined, many researcher paid lots of attention on tumor classification [45]–[49] and cancer detection [50]–[57] using various techniques.
D. Prostate Cancer
- Prostate cancer is the second most common cancer worldwide and the most common malignancy in men [62].
- It is due to the abnormal and uncontrolled cell mutation and replication in the prostate gland.
4 JOURNAL OF LATEX CLASS FILES, VOL. 13, NO. 9, SEPTEMBER 2014
- As with any cancer, early detection and treatment is vital for good survival rates of prostate cancer.
- By analyzing the characteristics of the tissue regions, tissue classification in TRUS provides a malignancy map to guide biopsy, which can reduce the number of unnecessary biopsy.
- Later different types of features were considered together with more sophisticated classifiers [67]– [69], for instance, textural features and spectral parameters extracted from RF data were employed in [67].
- It has been shown that the combination of CDUS and gray-scale ultrasound can detect a greater number of prostate cancers than gray-scale ultrasound alone [74].
- The classification of lesions in ultrasound liver image usually depends heavily on the characteristics of the lesions including internal echo, morphology, edge, echogenicity, and posterior echo enhancement.
III. FEATURE EXTRACTION
- The traditional approaches to ultrasound tissue classification start with extracting discriminative features from the ultrasound signal or image.
- Here the authors group them into three categories.
- By identifying spatial variations in pixel intensities and quantifying them into numerical features, texture analysis has been widely adopted and different kinds of texture features are available in the literature: Gabor filter responses, derivatives of Gaussian filters, wavelet transform, cooccurrence matrices, Local Binary Patterns, fractal spaces, Markov random fields, and so on.
- In [87], four types of texture features, GLCM, LBP, Gabor filters and the shading of the polar image, are extracted to form a feature vector of 68 dimensions for IVUS tissue classification.
- The second-order co-occurrence features are often supplemented by first order intensity distribution statistics, such as mean, variance, skew and kurtosis.
6 JOURNAL OF LATEX CLASS FILES, VOL. 13, NO. 9, SEPTEMBER 2014
- Thus, for the same type of tissue, different system parameters might lead to different texture patterns.
- For this purpose, in [87], the raw RF signals were exploited to reproduce ultrasound images with a unique and well controlled set of imaging parameters.
- In addition to the above texture features, the morphological features are also often considered for tissue classification in ultrasound.
- In contrast, the morphological parameters are less dependent on system parameters and acquisition characteristics, thus more consistent compared to texture features.
- Due to aggressiveness of the growth, malignant tumours tend to have an irregular shape while benign tumours tend to have a spherical or oval shape.
B. RF-signal-based approaches
- The US image formation process, i.e. from raw RF signals to US images, introduces a certain number of approximations (such as envelope detection and log compression), thus a certain amount of information is lost.
- Following the seminal work of Lizzi et al. [103], [130], spectral parameters of local RF signals are the most often used features, based on the hypothesis that different tissue types behave differently in the frequency domain.
- Spectral features are calculated based on the estimated local power spectrum of RF signals, which can be computed by using the Fourier transform or by the AutoRegressive (AR) model.
- Similarly, in [31], the plaque tissues are classified by comparing the full spectrum of a tissue sample to the ones in the training database.
- The Rayleigh distribution is also widely used to describe homogeneous areas in US images.
8 JOURNAL OF LATEX CLASS FILES, VOL. 13, NO. 9, SEPTEMBER 2014
- Beyond the feature categories discussed above, other features such as scatterer size and the speed of sound can also be considered.
- It is known that the ultrasound wave propagates with different speed in tissue with different density.
- Given the relationship between the acoustic impedance and the tissue density, the relative acoustic impedance can be used as a parameter for tissue characterization [35].
C. Feature Combination and Selection
- Individual features usually have limited discriminative power.
- Escalera et al. [32] considered texture features (GLCM, LBP, Gabor, and shadow), RF-based features (full spectrum, two global spectral features), and slope-based features [104] for tissue classification in IVUS.
- In [88], the texture and spectral features in combination with the RF time series features result in the best performance.
- In another study [42], the GLCM features are combined with the Nakagami parametric image for breast tumour detection.
- Feature selection approaches can be categorized into two classes: filter and wrapper.
IV. CLASSIFIER DESIGN
- After feature extraction, the next step is to design a classifier to automatically label tissue types based on the features.
- Support Vector Machine — SVM [141] is a discriminant method based on the Bayesian learning theory.
- In [110], after dimensionality reduction of full spectrum, SVM classification was performed for separating cancer from normal prostate tissue.
- In [31], the tissue classifier contains a bank of tissue detector arrays, e.g., 10 for each tissue type.
- In this respect their method share similarities with the random forests algorithms.
10 JOURNAL OF LATEX CLASS FILES, VOL. 13, NO. 9, SEPTEMBER 2014
- The k-NN classifier can capture complex boundary among different classes but it is computationally expensive.
- In [78], the k-NN, neural network, and Bayes classifier were utilized to classify liver tissues.
- In [108], with spectrum analysis of RF signals, four non-linear classifiers were trained for prostate tissue classification: multi-layer-perceptron neural networks, logitboost algorithms, SVM, and stacked restricted Boltzmann machines.
- By embedding ECOC in the potential functions for Discriminative Random Fields (DRF), Ciompi presented in [35] a multi-class classification technique called ECOC-DRF for IVUS tissue classification.
- Therefore, different classifiers can potentially offer complementary information, and the combination of various classifiers could improve the performance.
V. DEEP LEARNING APPROACHES
- Nowadays deep learning has become popular as a self-taught approach in which features are computed in an automatic manner instead of combining manual designed features.
- Both CNN and FCN have been applied extensively in ultrasound.
- In many medical image classification cases, the number of labeled data are limited for training.
- The remaining layers of the new network are initialized randomly and trained according to the new task [183].
- Recurrent Neural Networks (RNN) — Recurrent neural networks [195] are often used to process a sequence of data.
VI. FDA-CLEARED MACHINE LEARNING ALGORITHMS
- Outside the research regime, there are FDA cleared tissue classification algorithms or products are being accepted in the clinic to benefit patients in real practice.
- In breast imaging, lesion classification products by Koios Medical [212] for AI-based clinical decision support in 2D ultrasound have received clearance.
- In cardiac imaging, FDA recently approved the first AI-guided medical imaging tool by Caption Health [214] for the use in cardiac ultrasounds in which tissue-classification algorithms are behind.
- The tool can automatically capture video clips, and saves the best video clip acquired from a particular view for reviewing.
- In prostate imaging, Focal Healthcare provides a FDA-cleared solution [216] for MR guided ultrasound biopsy for identifying cancer tissues which help urologists perform fusion biopsy procedures more efficiently and accurately.
VII. DISCUSSIONS
- Compared to techniques in computed tomography (CT), magnetic resonance (MR) and X-rays, the techniques of tissue classification in ultrasound are less applied.
- These factors lead to different looks of ultrasound images while the robustness of the tissue classification will depend on the similarity of training data for developing the techniques and the testing data in clinical practice.
- To alleviate this problem, the best approach is to improve the standardization of the imaging via technology or guideline to make the procedure less-user dependent and image-looks consistent.
- For prostate cancer, the fusion of TRUS and MRI for biopsy guidance may offer improved results by combining the strengths of the two imaging modalities [218].
- In [111], ovarian tissue features extracted from photoacoustic spectra data, beam envelopes and co-registered ultrasound and photoacoustic.
12 JOURNAL OF LATEX CLASS FILES, VOL. 13, NO. 9, SEPTEMBER 2014
- Images are used to characterize cancerous vs. normal tissue using a SVM classifier.
- One of the critical issues in tissue classification research is the creation of a reliable data set with ground truth.
- Large databases are necessary for training powerful classifiers and validating the trained classifiers.
- Currently, the ground truth is mainly obtained by expert annotations or the histopathological analysis, which is a slow and complex process, leading to limited data with annotation.
- In [33], [96], an approach was presented to enhance the in vitro training data set by selectively including examples from in vivo data for plaque characterization.
VIII. CONCLUSIONS
- This paper has presented a survey on ultrasound tissue classification, particularly focusing on recent development in this area.
- SHELL shan et al.: ULTRASOUND TISSUE CLASSIFICATION: A REVIEW 13 REFERENCES [1].
- Chayakrit Krittanawong, Anusith Tunhasiriwet, HongJu Zhang, Zhen Wang, Mehmet Aydar, and Takeshi Kitai.
Did you find this useful? Give us your feedback
Citations
21 citations
Cites background from "Ultrasound tissue classification: a..."
...Ultrasound tissue classification has been identified as one of the most active fields of clinical study, powered by several major clinical applications (Shan et al., 2020)....
[...]
...Ultrasound image detection is a medical diagnosis method that utilizes dispersed or reflected ultrasound echo data to identify lesion regions in the human body, depending on the variation in acoustic impedance of various human tissues (Shan et al., 2020)....
[...]
8 citations
3 citations
2 citations
References
123,388 citations
[...]
79,257 citations
73,978 citations
49,590 citations
33,597 citations
Related Papers (5)
Frequently Asked Questions (14)
Q2. What are the common features extracted from RF signals?
The features extracted from RF signals are not subject to machine dependent processing, subsampling, interpolation, quantization and even operator-dependent settings [35].
Q3. What is the common use of ultrasound for cardiac imaging?
Echocardiography (ultrasound imaging of the heart) is the most common cardiac imaging procedure performed in clinical practice, due to its portability, low cost, and patient acceptance.
Q4. What is a well-known statistical tool for extracting texture information from images?
Grey Level Co-occurrence Matrix (GLCM) [122]: GLCM is a well-known statistical tool for extracting texture information from images.
Q5. What is the popular and successful deep learning architecture?
Convolutional Neural Networks (CNN) — Convolutional neural network is the most popular and successful deep learning architecture.
Q6. How many spectral parameters were computed from the power spectrum of the RF data?
To differentiate different cardiac tissue types, thirteen spectral parameters were computed from the power spectrum of the RF data in three different bandwidth ranges [11].
Q7. What is the way to extract spectral features from TRUS images?
Considering the RF signals may not be available for spectral analysis, the work [68] proposes to extract spectral features from TRUS images, where each ROI is first scanned to form 1-D signals and then spectral features are extracted.
Q8. What is the significance of spectral analysis of RF signals?
It is shown that spectral analysis of RF signals, which is related to the physical scatter properties, can potentially be used for monitoring the evolution of HIFU lesions.
Q9. Why is it difficult to find a boundary separating different tissue types in the feature space?
Due to the large variations in US images (signals), most features have low discriminative power and it is often difficult to find a boundary separating different tissue types in the feature space.
Q10. What are the common US imaging parameters used to improve the visualization of certain tissue?
In clinical practice, to improve the visualization of certain tissue, the physicians often change the US imaging parameters such as contrast, depth and gain.
Q11. What is the common method for spectral parameters of local RF signals?
Following the seminal work of Lizzi et al. [103], [130], spectral parameters of local RF signals are the most often used features, based on the hypothesis that different tissue types behave differently in the frequency domain.
Q12. What are the methods used to estimate the temperature increase during the ablation process?
Methods have also been developed to estimate the temperature increase during the ablation process by detecting the change of sound speed, attenuation coefficient, and backscattering.
Q13. What were the features considered for analyzing the ablated regions in the tissue?
In an earlier work [21], both the textural and spectral features were considered for analyzing the ablated regions in the tissue.
Q14. What is the way to improve the image quality of the procedure?
To alleviate this problem, the best approach is to improve the standardization of the imaging via technology or guideline to make the procedure less-user dependent and image-looks consistent.