Elsevier

Computers in Biology and Medicine

Volume 89, 1 October 2017, Pages 18-23
Computers in Biology and Medicine

Machine-learning-based classification of real-time tissue elastography for hepatic fibrosis in patients with chronic hepatitis B

https://doi.org/10.1016/j.compbiomed.2017.07.012Get rights and content

Highlights

  • Real-time tissue elastography is noninvasive and accurate stage of hepatic fibrosis.

  • Four classical classifiers to diagnosis of hepatitis B disease fibrosis stage.

  • Machine-learning methods are better for evaluation and promising in application.

Abstract

Hepatic fibrosis is a common middle stage of the pathological processes of chronic liver diseases. Clinical intervention during the early stages of hepatic fibrosis can slow the development of liver cirrhosis and reduce the risk of developing liver cancer. Performing a liver biopsy, the gold standard for viral liver disease management, has drawbacks such as invasiveness and a relatively high sampling error rate. Real-time tissue elastography (RTE), one of the most recently developed technologies, might be promising imaging technology because it is both noninvasive and provides accurate assessments of hepatic fibrosis. However, determining the stage of liver fibrosis from RTE images in a clinic is a challenging task. In this study, in contrast to the previous liver fibrosis index (LFI) method, which predicts the stage of diagnosis using RTE images and multiple regression analysis, we employed four classical classifiers (i.e., Support Vector Machine, Naïve Bayes, Random Forest and K-Nearest Neighbor) to build a decision-support system to improve the hepatitis B stage diagnosis performance. Eleven RTE image features were obtained from 513 subjects who underwent liver biopsies in this multicenter collaborative research. The experimental results showed that the adopted classifiers significantly outperformed the LFI method and that the Random Forest(RF) classifier provided the highest average accuracy among the four machine algorithms. This result suggests that sophisticated machine-learning methods can be powerful tools for evaluating the stage of hepatic fibrosis and show promise for clinical applications.

Introduction

Hepatic fibrosis is the middle stage of hepatitis of various types, of liver cirrhosis and of any late period liver-related diseases. Patients with hepatitis C or B virus who chronically suffer from fibrosis are more vulnerable to hepatocellular carcinoma. Therefore, early detection and prompt intervention are important. It has been reported that more aggravated hepatic fibrosis increases the risk of hepatocellular carcinoma [1]; however, some interferon treatments can slow this process and reduce morbidity from hepatocellular carcinoma [2], which is why the Guide to the Prevention and Treatment for Chronic Hepatitis B considers it essential to evaluate the stage of hepatic fibrosis both for treatment and for monitoring treatment effects.

However, determining the stage of liver fibrosis is still arduous. Although a liver biopsy is widely considered the gold standard for assessing liver fibrosis, the procedure's invasiveness can cause complications such as hemorrhage and pneumothorax. Moreover, limited sample numbers and sizes usually lead to suboptimal accuracy [3], [4], [5]. Another option, serological examination, has some clinical implications; however, it has been reported that its sensitivity and specificity vary greatly. Moreover, serological examination cannot satisfy the requirements for clinical treatments by itself because it is unable to accurately stage liver fibrosis [6], [7]: it can distinguish between hepatic fibrosis and cirrhosis but cannot accurately determine the stage of hepatic fibrosis. In addition, this type of examination is not sufficiently specific because it can be affected by metabolism, extracellular-matrix diseases and late-stage cancers [8]. Therefore, the Asian-Pacific Association for the Study of the Liver recommends elastography instead of serological examination. Many scholars have sought other noninvasive methods to diagnose liver fibrosis, especially physical methods.

One existing noninvasive method is imaging diagnosis, which can completely evaluate the entire organ. Therefore, imaging examination has been widely adopted clinically. Imaging diagnosis can involve ultrasound, Computed Tomography (CT) or Magnetic Resonance Imaging (MRI); of these, CT and MRI are expensive. Ultrasound is the most feasible and inexpensive method. Unlike the others, it is suitable for use on patients with medical metal devices; consequently, ultrasound is widely accepted. Owing to the prevalence of ultrasound contrast and ultrasonic elastography, ultrasound can provide not only information about hepatic morphology and blood dynamics but also information about physical properties such as liver stiffness or elasticity modulus. In view of this capacity, ultrasound may emerge as one of the most important noninvasive methods for evaluating liver fibrosis.

RTE is also increasing in popularity and is considered to be among the most promising methods for staging hepatic fibrosis; however, it is unable to distinguish between intermediate degrees of liver fibrosis [9]. The calculations of both the elastic ratio and fibrosis index of RTE have shown intra-and inter-observer variability [10]. These methods have already been applied to the research of hepatitis C [11], non-alcoholic liver disease [12] and hepatitis B [13]. However, their classification accuracy reaches only approximately 70% for chronic hepatitis B patients, which cannot satisfy clinical demands.

A similar study conducted by Wu et al. [13] obtained predictions of significant fibrosis with the help of a multiple regression statistical analysis. However, multiple regression is a linearly dependent method, while the relationships between the image parameters and stages of hepatitis B are not linear.

As we know, machine-learning (ML) and pattern-recognition methods have been widely studied for early diagnosis of hepatitis diseases, such as in analyses of biochemical indices and clinical figures of hepatic fibrosis [14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24], [25]. However, because FibroScan has been adopted for assessing the stage of liver fibrosis in recent years [26], [27], [28], studies have seldom been conducted to evaluate the performance of machine learning in real-time ultrasonic tissue elastography. This paper focuses on the analysis of data in patients with chronic hepatitis B via RTE. The data indicate the rigidity of tissue indirectly by the displacement caused by mechanical heart impulses. The RTE software was developed by Hitachi (HI VISION Ascendus; Hitachi Aloka Medical, Tokyo, Japan) and offers 11 parameters to facilitate the quantitative evaluation of liver fibrosis [11], [29]. A total of 513 subjects suffering from chronic viral hepatitis and cirrhosis were studied and enrolled in a multicenter collaborative hospital study. Four classical pattern-recognition methods, which included Naïve Bayes, Random Forest (RF), K-Nearest Neighbor (KNN) and Support Vector Machine (SVM), were applied to build an aided decision-support system to estimate the stage of hepatic fibrosis.

The remainder of the paper is organized as follows. Section 2 presents a brief introduction to the four machine-learning methods used in this study. Section 3 focuses on the study methodology, in which RTE images were acquired to extract features and to perform further data preprocessing. Section 4 presents the experimental results. Finally, Section 5 draws conclusions, discusses the findings, and describes some plans for future work.

Section snippets

Machine-learning-based classification

The existing method, which is intended to determine the fibrosis stage via RTE, is to obtain the LFI as the fibrosis index [13] using a linear regression equation. However, this type of equation has limited functionality, and satisfactory results are often difficult to obtain. In this paper, we employ four classical machine learning methods to predict the stage of hepatic fibrosis based on features extracted from numerous RTE images. Our approach will improve the accuracy of staging hepatic

Acquiring RTE images

The RTE equipment used in this research was a color Doppler ultrasound diagnostic instrument, the Hitachi HI VISION Preirus (Hitachi Aloka Medical Corporation, Tokyo Japan) and a EUP-L52 linear probe (Hitachi Aloka Medical) with a frequency range of 3–7 MHz. This instrument is equipped with real-time elastography software that can analyze tissue dispersion quantitatively, attaining 11 elastography parameters. The equipment is also equipped with a strain histogram measurement data-processing

Data collection

The study protocol was approved by the independent ethics committees of the involved institutions, and the purpose of this research was explained to the patients. Written informed consent was obtained in advance. This prospective, multicenter, cross-sectional study was conducted at the following eight hospitals in China: Third Affiliated Hospital (Sun Yat-sen University), Guangdong General Hospital, Beijing Youan Hospital (Capital Medical University), First Affiliated Hospital (Guangxi Medical

Discussion and conclusions

When applying machine learning to hepatic fibrosis, previous research has focused primarily on the prediction of survival vs. mortality groups [14], [17], [18], [19], [20], [21], [23], [24], [25]; few investigations have been conducted on the four stages of liver fibrosis [15], [26], [27], [28]. Moreover, most of these have focused only on hepatitis C [15], [16], [28]. To our knowledge, no prior studies have applied machine-learning techniques to assess the stage of liver fibrosis in patients

Acknowledgments

The authors thank Zhang Qi for the data collection work, which was a valuable contribution to this study. This study was supported by the 973 Project (2013CB329401), the Natural Science Foundation of China (61573080, 91420105) and the Science and Technology Project of Sichuan Province (2015SZ0141).

References (34)

Cited by (80)

  • Artificial intelligence in gastroenterology and hepatology

    2023, Artificial Intelligence in Clinical Practice: How AI Technologies Impact Medical Research and Clinics
  • Artificial Intelligence in Hepatology- Ready for the Primetime

    2023, Journal of Clinical and Experimental Hepatology
  • Artificial intelligence based liver portal tract region identification and quantification with transplant biopsy whole-slide images

    2022, Computers in Biology and Medicine
    Citation Excerpt :

    Although numerous histopathological staging systems have been utilized for liver fibrosis evaluation in current clinical practice, including Knodell, Metavir, Ishak, and Scheuer systems, only manual reviews or semi-quantitative evaluations are conducted by these staging systems, resulting in large inter- and intra-observer variability [3–5]. To reduce such variations, evaluation methods based on artificial intelligence (AI) algorithms, such as random forests, K-nearest neighbors, and support vector machines, have been developed to provide objective diagnostic tools for liver fibrosis staging [6–8]. In contrast to these conventional machine learning methods, deep learning has emerged as a powerful tool for diverse biomedical image processing studies due to its great success across different image modalities [9].

View all citing articles on Scopus
View full text