Multiscale superpixel method for segmentation of breast ultrasound

doi:10.1016/j.compbiomed.2020.103879

Computers in Biology and Medicine

Volume 125, October 2020, 103879

https://doi.org/10.1016/j.compbiomed.2020.103879 Get rights and content

Abstract

Background

In medical diagnostics, breast ultrasound is an inexpensive and flexible imaging modality. The segmentation of breast ultrasounds to identify tumour regions is a challenging and complex task. The major problems of effective tumour identification are speckle noise, artefacts and low contrast. The gold standard for segmentation is manual processing; however, manual segmentation is a cumbersome task. To address this problem, the automatic multiscale superpixel method for the segmentation of breast ultrasounds is proposed.

Methods

The original breast ultrasound image was transformed into multiscaled images, and then, the multiscaled images were preprocessed. Next, a boundary efficient superpixel decomposition of the multiscaled images was created. Finally, the tumour region was generated by the boundary graph cut segmentation method. The proposed method was evaluated with 120 images from the Thammassat University Hospital database. The dataset consists of 30 malignant, 30 benign tumors, 60 fibroadenoma, and 60 cyst images. Popular metrics, such as the accuracy, sensitivity, specificity, Dice index, Jaccard index and Hausdorff distance, were used for the evaluation.

Results

The results indicate that the proposed method achieves segmentation accuracy of 97.3% for benign tumors, 94.2% for malignant, 96.4% for cysts and 96.7% for fibroadenomas. The results validate that the proposed model outperforms selected state-of-the-art segmentation methods.

Conclusions

The proposed method outperforms selected state-of-the-art segmentation methods with an average segmentation accuracy of 94%.

Introduction

The World Health Organization estimated that over 9.8 million people died from cancer related diseases in 2018. Statistically, 1 in 6 global deaths is cancer related. An estimated 627,000 women died from breast cancer in 2018, which is 15% of the cancer deaths among women [1]. Early detection, screening and treatments are urgent steps to avoid complications in breast cancer. Imaging techniques like Magnetic Resonance Imaging (MRI), mammography, ultrasounds, and breast Computer Tomography (CT) are used for medical diagnosis. From the list of available imaging techniques, ultrasounds are commonly used because they are non-invasive, cheap and easy to access. However, the major problems with ultrasounds are their poor quality, acoustic shadowing and speckle noise. Speckle noise is a multiplicative noise that appears in a granular form. Furthermore, speckle noise is associated with echoes that originate from obstruction patterns. Speckle noise destroys the resolution, brightness and luminance of ultrasound images [2]. (see Fig. 12)

The gold standard for the segmentation of Breast Ultrasounds (BUSs) is manual segmentation. Manual segmentation is a procedure in which an experienced radiologist manually outlines and identifies tumours. Unfortunately, this procedure is cumbersome and time consuming, and hence the Computer-Aided Diagnostic (CAD) system has been introduced. CAD is a computerized tool to analyse medical images with a limited performance apparatus. Kandemir et al. [3] discussed the usage of CAD for ultrasound and medical images. Specifically, CAD is not a tool to replace clinicians; rather, it is a tool to act as a second interpreter for diagnosing medical images. In general, CAD is categorized into: semi-automatic and automatic. A semi-automatic CAD outlines tumours using user interaction tools; alternatively, automatic CAD does not require user interaction [4]. This paper utilized the automatic process to segment breast ultrasounds. In a nutshell, the proposed method creates multiscaled images from the original BUS image, and then the multiscaled images are preprocessed and a superpixel decomposition is created

The remainder of this paper is as follows. A literature review is given in Section 2. The materials and proposed method are reported in Section 3. The results of the numerical experiment are reported in Section 4. Finally, a discussion is given in Section 5 and the paper is concluded in Section 6.

Section snippets

Related work

Several methods have been proposed for the segmentation of BUSs. The commonly used methods are the watershed, graph cut, and active contour methods [5]. A detailed review of BUS segmentation is in Refs. [6,7]. In this section, a brief insight into various superpixel methods and related segmentation is given. A superpixel is a set of connected pixels with a similar colour, intensity or other low-level properties [8]. The major advantages of using superpixels are: reduced computational costs and

Data acquisition

The database used in this study contains 180 BUS images, i.e. 30 malignant, 30 benign tumors, 60 fibroadenoma, and 60 cyst images. Malignant, benign and fibroadenoma images were selected from patients with a histopathological diagnose. The malignant tumours have irregular boundaries; fibroadenoma images are characterized by low contrast; and cyst images include large irregular shadows, artefacts and random noise (see Fig. 1). The images were obtained by a Philips iU22 ultrasound machine from

Results

All experiments for this paper were implemented on Matlab R2018a with a system configuration of an Intel(R) 3.60 GHz CPU, 16 GB of RAM and a Windows 10 Operating System. In this paper, four cases of BUS segmentation have been reported. Case 1 is the result of the segmentation of other benign BUSs, Case 2 is the result of the segmentation of malignant BUSs, case 3 is the result of segmentation of cyst BUSs, and case 4 is the results of the segmentation of fibroadenoma BUSs.

Discussion

The proposed method involves the segmentation and evaluation of breast ultrasounds (case 1 to case 4). The results of the visual analysis indicate that the proposed method was close to the ground truth. The mean and standard deviation have been recorded for the proposed method and the competing methods. The values of different measures as seen from the different tables indicate that the proposed method is effective for the segmentation of breast ultrasounds (see Table 1).

Visual comparisons

Conclusion

In this paper, a novel multiscale superpixel method for segmentation of BUS images based on the boundary efficient graph cut method was presented. A real dataset of breast ultrasounds obtained from Thammasat University Hospital was used for the evaluation. Specifically, a multiscale image from the original BUS image was created, and then the BAS was used to remove speckle noise. Subsequently, a distance transform superpixel decomposition of the multiscale images was generated before finally

Declaration of competing interest

All authors in this paper have no potential conflict of interests.

Acknowledgements

This research is supported by the Thailand Research Fund grant RSA6280098 and the Center of Excellence in Biomedical Engineering of Thammasat University. We wish to thank the anonymous referees for their valuable remarks.

References (42)

M. Kandemir et al.
Computer-aided diagnosis from weak supervision: a benchmarking study
Comput. Med. Imag. Graph.
(2015)
V. Giannini et al.
A fully automatic computer aided diagnosis system for peripheral zone prostate cancer detection using multi-parametric magnetic resonance imaging
Comput. Med. Imag. Graph.
(2015)
K.M. Meiburger et al.
Automated localization and segmentation techniques for B-mode ultrasound images: a review
Comput. Biol. Med.
(2018)
L. Han et al.
Semi-supervised segmentation of lesion from breast ultrasound images with attentional generative adversarial network
Comput. Methods Progr. Biomed.
(2020)
X. Xi et al.
Breast tumor segmentation with prior knowledge learning
Neurocomputing
(2017)
Q. Huang et al.
Segmentation of breast ultrasound image with semantic classification of superpixels
Med. Image Anal.
(2020)
C. Keatmanee et al.
Initialization of active contours for segmentation of breast cancer via fusion of ultrasound, Doppler, and elasticity images
Ultrasonics
(2019)
A. Rodtook et al.
Automatic initialization of active contours and level set method in ultrasound images of breast abnormalities
Pattern Recogn.
(2018)
M. Wang et al.
Content-sensitive superpixel segmentation via self-organization-map neural network
J. Vis. Commun. Image Represent.
(2019)
J. Shi et al.
Joint sparse coding based spatial pyramid matching for classification of color medical image
Comput. Med. Imag. Graph.
(2015)

V. Kamalaveni et al.

Image denoising using variation of perona-malik model with different edge stopping functions

Procedia Comput. Sci.

(2015)

Z. Peng et al.

Interactive image segmentation using geodesic appearance overlap graph cut

Signal Process. Image Commun.

(2019)

V. Ng et al.

Determining the asymmetry of skin lesion with fuzzy borders

Comput. Biol. Med.

(2005)

B. Foster et al.

A review on segmentation of positron emission tomography images

Comput. Biol. Med.

(2014)

S. Zhou et al.

Active contour model based on local and global intensity information for medical image segmentation

Neurocomputing

(2016)

S. Bag et al.

An efficient recommendation generation using relevant Jaccard similarity

Inf. Sci.

(2019)

A. Singh et al.

A novel dice similarity measure for IFSs and its applications in pattern and face recognition

Expert Syst. Appl.

(2020)

N. Karunanayake et al.

Particle method for segmentation of breast tumors in ultrasound images

Mathematics and Computers in Simulation

(2020)

C. Sha et al.

A robust 2D Otsu's thresholding method in image segmentation

J. Vis. Commun. Image Represent.

(2016)

J. Guo et al.

An enhanced modulation signal bispectrum analysis for bearing fault detection based on non-Gaussian noise suppression

Measurement

(2020)

Global Health Observatory

Unit on Cancer

(2018)

Cited by (20)

Reviewing 3D convolutional neural network approaches for medical image segmentation
2024, Heliyon
Convolutional neural networks (CNNs) assume pivotal roles in aiding clinicians in diagnosis and treatment decisions. The rapid evolution of imaging technology has established three-dimensional (3D) CNNs as a formidable framework for delineating organs and anomalies in medical images. The prominence of 3D CNN frameworks is steadily growing within medical image segmentation and classification. Thus, our proposition entails a comprehensive review, encapsulating diverse 3D CNN algorithms for the segmentation of medical image anomalies and organs.
This study systematically presents an exhaustive review of recent 3D CNN methodologies. Rigorous screening of abstracts and titles were carried out to establish their relevance. Research papers disseminated across academic repositories were meticulously chosen, analyzed, and appraised against specific criteria. Insights into the realm of anomalies and organ segmentation were derived, encompassing details such as network architecture and achieved accuracies.
This paper offers an all-encompassing analysis, unveiling the prevailing trends in 3D CNN segmentation. In-depth elucidations encompass essential insights, constraints, observations, and avenues for future exploration. A discerning examination indicates the preponderance of the encoder-decoder network in segmentation tasks. The encoder-decoder framework affords a coherent methodology for the segmentation of medical images.
The findings of this study are poised to find application in clinical diagnosis and therapeutic interventions. Despite inherent limitations, CNN algorithms showcase commendable accuracy levels, solidifying their potential in medical image segmentation and classification endeavors.
Learning active contour models based on self-attention for breast ultrasound image segmentation
2024, Biomedical Signal Processing and Control
Computer-aided diagnosis (CAD) systems based on ultrasound have been developed and widely promoted in breast cancer screening. Due to the characteristics of low contrast and speckle noises, breast ultrasound image segmentation, one of the crucial steps of CAD systems, has always been challenging. Recently, the emerging Transformer-based medical segmentation methods, which have a better ability to model long dependencies than convolutional neural networks (CNNs), have shown significant value for medical image segmentation. However, due to the limited data with the high-quality label, Transformer performs weakly on breast ultrasound image segmentation without pretraining. Thus, we propose the Attention-Gate Medical Transformer (AGMT) for small breast ultrasound datasets, which introduces the attention-gate (AG) module to suppress background information and the average radial derivative increment ( $Δ A R D$ ) loss function to enhance shape information. We evaluate the AGMT on both a private dataset A and a public dataset B. On dataset A, the AGMT outperforms MT on the metrics of true positive ratio, jaccard index (JI) and dice similarity coefficient (DSC) by 6.4%, 2.3% and 1.9%, respectively. Meanwhile, when compared with UNet, the AGMT improves JI and DSC by 5.3% and 4.9%, respectively. The results show performance has significantly improved compared with mainstream models. In addition, we also conduct ablation experiments on the AG module and $Δ A R D$ , which prove their effectiveness.
A systematic review of retinal fundus image segmentation and classification methods using convolutional neural networks
2023, Healthcare Analytics
Retinal fundus images play a crucial role in the early detection of eye problems, aiding in timely diagnosis and treatment to prevent vision loss or blindness. With advancements in technology, Convolutional Neural Network (CNN) algorithms have emerged as effective tools for recognition, delineation, and classification tasks. This study proposes a comprehensive review of CNN algorithms used for retinal fundus image segmentation and classification. Our review follows a systematic approach, exploring diverse repositories to identify studies employing CNN to segment and classify retinal fundus images. Utilizing CNNs in the segmentation and classification of retinal fundus images can enhance the precision of segmentation outcomes and alleviate the sole dependence on human experts. This approach enables more accurate segmentation results, reducing the burden on human experts. A total of sixty-two studies are included in our review, analyzing aspects such as database usage and the advantages and disadvantages of the methods employed. The review provides valuable insights, limitations, observations, and future directions in the field. Despite certain limitations, the findings indicate that CNN algorithms consistently achieve high accuracies. The comprehensive examination of the included studies sheds light on the potential of CNN in retinal fundus image analysis.
ECAU-Net: Efficient channel attention U-Net for fetal ultrasound cerebellum segmentation
2022, Biomedical Signal Processing and Control
Citation Excerpt :
Based on the above two points, it is possible to assist the diagnosis with artificial intelligence technologies represented by deep learning. In the medical field, convolutional neural network (CNN) has also achieved good results in the detection and segmentation of musculoskeletal ultrasound [5], breast [6], liver tumors [7,8], glands [9–11], fetus [4,12,13], and other human tissues. Whereas medical images are different from natural images in some cases: 1) The application of medical image datasets is relatively narrow, and the research data should be collected specifically for different tasks; 2) The collection of medical image data should follow strict clinical standards and get permission from the medical committee; 3) Medical images need to be labeled by the experienced experts.
Ultrasound (US) examination of the fetal central nervous system is one of the significant tasks in the mid-term pregnancy inspection, in which the fetal cerebellum, as the key structure, attracts more attention from physicians. Therefore, accurate identification and segmentation of the cerebellum are particularly crucial for clinical diagnosis. In recent years, the development of deep learning has greatly improved the accuracy of computer automatic segmentation. This paper proposes an efficient channel attention U-Net algorithm for cerebellum segmentation (termed as ECAU-Net). The method employs the U-Net, which applies the encoder to extract feature representations and the decoder to locate segmentation results, as the backbone for segmentation. Combined with ECA modules, the one-dimensional convolutional layers with shared parameters are applied to replace the full connection layers in the conventional channel attention modules, which greatly reduces the number of model parameters without affecting performance. Moreover, we establish a US database, termed as JSUAH-Cerebellum, of fetal cerebellum images in the second trimester. During the training process, Gaussian blur and random flipping are utilized to boost the scale of the dataset. Comprehensive experiments were carried out with the Dice loss function on JSUAH-Cerebellum. In addition, we compared the proposed method with 12 other typical methods. The experimental results show that the mean Jaccard Score (JS) and Dice Similarity Coefficient (DSC) of the proposed algorithm are 86.01% and 91.35%, respectively, which are 2.86% and 3.73% higher than the baseline. Furthermore, for the worst-quality samples, the JS and DSC reach 0.55 and 0.68, respectively.
A method for segmentation of tumors in breast ultrasound images using the variant enhanced deep learning
2021, Biocybernetics and Biomedical Engineering
Citation Excerpt :
The graph-based method uses graph analysis for segmentation. Graph methods are simple, flexible with soft constraints [13,14]. DM methods segment images using a curved surface technique.
Breast cancer is a deadly disease responsible for statistical yearly global death. Identification of cancer tumors is quite tasking, as a result, concerted efforts are thus devoted. Clinicians have used ultrasounds as a diagnostic tool for breast cancer, though, poor image quality is a major limitation when segmenting breast ultrasound. To address this problem, we present a semantic segmentation method for breast ultrasound (BUS) images.
The BUS images were resized and then enhanced with the contrast limited adaptive histogram equalization method. Subsequently, the variant enhanced block was used to encode the preprocessed image. Finally, the concatenated convolutions produced the segmentation mask.
The proposed method was evaluated with two datasets. The datasets contain 264 and 830 BUS images respectively. Dice measure (DM), Jaccard measure, and Hausdroff distance were used to evaluate the methods. Results indicate that the proposed method achieves high DM with 89.73% for malignant and 89.62% for benign BUSs. Moreover, the results obtained validate the capacity of the proposed method to achieve higher DM in comparison with reported methods.
The proposed algorithm provides a deep learning segmentation procedure that can segment tumors in BUS images effectively and efficiently.
Multiscale hybrid algorithm for pre-processing of ultrasound images
2021, Biomedical Signal Processing and Control
One of the diagnostic tools for breast cancer screening is the ultrasound. Ultrasounds are characterized by low cost and non-invasive patterns. A major drawback of ultrasound is the multiplicative speckle noise. Speckle noise limits the effectiveness of images thereby reducing the efficiency of the test. This paper proposes a new algorithm to reduce noise.
The method is based on the combination of the multiscale approach, the wiener filter, and the new fast bilateral filter. A multiscale image was first created. Subsequently, the wiener filter was used for initial filtering and to reduce the mean square error of the multiscaled images. Finally, the new fast bilateral filter was used to remove speckle noise.
The algorithm was tested on 50 synthetic images degraded by speckle noise with varying intensity and 250 breast ultrasound (BUS) images from two datasets. The results were compared with selected state-of-the-art filters. The proposed approach shows better performance in terms of standard noise evaluation measures. The noise reduction in the US images of breast cancer (BUS) has been verified by selected conventional segmentation procedures. Results indicate that the proposed method reduced speckle better than other methods. Specifically, the proposed method produced a structural similarity index measure value of 95% for both benign and malignant tumors. In addition a peak signal-to-noise ratio of 30.41db and 30.75db for benign and malignant tumors were obtained. The proposed filter also achieves high accuracy in terms of the segmentation measures.
The proposed speckle noise reduction algorithm achieves better accuracies than other competing filters. The algorithm will act as a tool for speckle reduction in breast ultrasound images.

View all citing articles on Scopus

View full text

Multiscale superpixel method for segmentation of breast ultrasound

Abstract

Background

Methods

Results

Conclusions

Introduction

Section snippets

Related work

Data acquisition

Results

Discussion

Conclusion

Declaration of competing interest

Acknowledgements

Comput. Med. Imag. Graph.

Comput. Med. Imag. Graph.

Comput. Biol. Med.

Comput. Methods Progr. Biomed.

Neurocomputing

Med. Image Anal.

Ultrasonics

Pattern Recogn.

J. Vis. Commun. Image Represent.

Comput. Med. Imag. Graph.

Procedia Comput. Sci.

Signal Process. Image Commun.

Comput. Biol. Med.

Comput. Biol. Med.

Neurocomputing

Inf. Sci.

Expert Syst. Appl.

Mathematics and Computers in Simulation

J. Vis. Commun. Image Represent.

Measurement

Unit on Cancer