A computer-aided diagnostic system to discriminate SPIO-enhanced magnetic resonance hepatocellular carcinoma by a neural network classifier
Introduction
Hepatocellular carcinoma (HCC) is one of the most common cancers worldwide and a major cause of death in patients with cirrhosis. Most types of liver cirrhosis predispose to HCC. Effective local treatment requires the early detection and accurate diagnosis of HCC, however the detection and accurate characterization of HCC in the presence of advanced cirrhosis remains a clinical problem [1]. The examination of liver tissue pathology is performed with various medical imaging modalities such as ultrasonography (US), computed tomography (CT), or magnetic resonance (MR) imaging. MR imaging is the sensitive and robust imaging technique for the diagnosis of liver lesions (e.g. HCC). Gd-enhanced MR imaging has significantly improved the diagnostic accuracy of hypervascular HCC, but it is still problematic in evaluating hypovascular HCC [2]. MR imaging with a liver-specifc agent superparamagnetic iron oxides (SPIO) has been shown to improve the detectability of liver tumors and help in accurate characterization of hypovascular HCC, and which become a compensatory for Gd-enhanced imaging. However, radiologists correctly diagnose HCC from advanced cirrhosis background on liver SPIO-enhanced imaging being difficult because there are some shortcomings in liver SPIO-enhanced imaging. Firstly, structural and functional inhomogeneity in cirrhosis could make radiologists doing false-positive lesion decision after SPIO administration. Secondly, the decrease in signal intensity of cirrhotic liver with SPIO is limited compared to that in normal liver, and some well-differentiated HCCs may contain various numbers of kupffer cells which exhibit signal decrease after SPIO administration, these factors all worsen lesion-liver contrast and the rate of detection of HCC. So interpretation of these images is highly dependent on the ability and experience of the observer, in some cases even experienced radiologists resort to confirmation of diagnosis by invasive hepatic biopsy, and inexperienced radiologists or physicians often mistakenly interpret SPIO-enhanced MR images of the liver [3], [4]. The necessity of a more accurate diagnosis arises and to this end computer-aided diagnostic (CAD) systems can be employed. Medical doctors can use CAD as an alternative second opinion to make the diagnosis precisely and objectively and reduce the number of required invasive procedures [5], [6].
The evolution in medical image processing and artificial intelligence techniques has given researchers the opportunity to investigate the potential of CAD systems for the classification of liver tissue [7]. Various approaches, most of them using US B-scan and CT images, have been proposed based on different image parameters, such as texture features, estimated from first- and second-order gray-level statistics, and fractal dimension estimators, in combination with various classifiers [8], [9]. Recently, a few research efforts have been undertaken in order to classify liver tissue on MR imaging. Zhang et al. [10] used texture feature improving the classification of cirrhotic liver. McNeal et al. [11] investigated a method for measuring the volumes of human livers in vivo from MR imaging and subsequently displaying these livers in three dimensions. These methods are mostly limited to non-contrast MR imaging. To the best of our knowledge, there are no published studies on the application of texture based artificial neural network (ANN) classifier to discriminate HCC nodules from cirrhosis on SPIO-enhanced MR imaging.
The aim of this preliminarily study is assessing an ANN classifier based on tuxture for classification HCC nodules in the presence of diffused liver cirrhosis on SPIO-enhanced imaging in rat model which make preparation for using the classifier for clinical patients study .We utilize SPIO-enhanced images of rat HCC model induced by Diethylnitrosamine (DENA) to assess the proposed classifier since it is difficult collecting sufficient SPIO-enhanced images during short time in clinical and also the routine use of SPIO-enhanced imaging is still not warranted in clinical practice due to high cost and patient discomfort, and studies haves been shown that DENA-induced rat liver pathology contains a large variety of lesions that closely mimic the progress stage of cirrhosis to HCC in human and the DENA-induced rat liver pathology is a comprehensive and realistic model for the study of the mechanisms of carcinogenesis [4], [12].
Section snippets
Materials and methods
The proposed CAD system consists of two basic modules: the feature extraction and the classifier modules, as shown in Fig. 1. Regions of interest (ROIs) are first segmented manually from SPIO-enhanced images by an experienced radiologist, then fed to the feature extraction module and passed to the classifier module (Fig. 1). All datasets are stored in a PC (Pentium Dual-Core2.50 GHz with 1.99 GB RAM). MATLAB 6.5 is used to implement the ANN classifier.
Results and discussion
Results of texture features calculated for HCC and cirrhosis are shown below (Table 1). Statistical test shows that there is significant difference in each texture feature between HCC and cirrhosis. Of the 161 ROIs obtained from the rat model with liver lesions, the ANN classifies 101 ROIs of training set into two types of liver lesions with an accuracy of 98.02% (99/101) and 60 ROIs of testing set with an accuracy of 91.67% (55/60) shown (Table 2).
Texture, an important feature of image, has
Conclusion
In this paper, the implementation of a CAD system consisting of a feature extraction and a classification module has been presented aiming to discriminate cirrhosis and HCC on MR imaging. The feature extraction module calculates the average gray level of six texture parameters derived from GLCM in 161 ROIs cut by an experienced radiologist from rat SPIO-enhanced imaging. The classifier module has been implemented with a BP ANN and the high performance has achieved by the ANN classifier in the
Acknowledgement
This work is supported in part by the National Science Foundation of China under the grant of 30570475.
DongMei Guo A doctoral student of Dalian University of Technology, Department of Biomedical Engineering. She is also a radiologist of Second Affiliated Hospital, Dalian Medical University.
References (14)
- et al.
Application of superparamagnetic iron oxide to imaging of hepatocellular carcinoma
Eur J Radiol
(2006) - et al.
Imaging of hepatocellular carcinoma in the cirrhotic liver: challenges and controversies
Radiology
(2008) - et al.
Hepatocellular carcinomain in the cirrhotic liver: gadolinium-enhanced 3D T1-weighted MR imaging as a stand-alone sequence for diagnosis
Radiology
(2006) - et al.
Development of hepatomas in hyperplastic nodules induced in the rat liver: detection with superparamagnetic iron oxide-enhanced magnetic resonance imaging
Acad Radiol
(1996) - et al.
Computer-aided diagnosis of solid breast nodules: use of an artificial neural network based on multiple sonographic features
IEEE Trans Med Imag
(2004) - et al.
Classification of dynamic contrast-enhanced. Magnetic resonance breast lesions by support vector machines
IEEE Trans Med Imag
(2008) - et al.
A computer-aided diagnostic system to characterize CT focal liver lesions: design and optimization of a neural network classifier
IEEE Trans Inform Techn Biomed
(2003)
Cited by (14)
Symtosis: A liver ultrasound tissue characterization and risk stratification in optimized deep learning paradigm
2018, Computer Methods and Programs in BiomedicineCitation Excerpt :Lamb et al., [3] used dual energy CT (DECT) based technique for classification Computed Tomography (CT) liver images. Guo et al. [4] used neural networks for classification of Magnetic Resonance Imaging (MRI) liver images. Suri et al. [5] designed a computer aided liver data classifier for Ultrasound in 2012.
Quantitative contrast-enhanced harmonic EUS in differential diagnosis of focal pancreatic masses (with videos)
2015, Gastrointestinal EndoscopyCirrhosis classification based on MRI with duplicative-feature support vector machine (DFSVM)
2013, Biomedical Signal Processing and ControlCitation Excerpt :It is also studied that duplication cannot be simply reduced to a weighted feature. And as for classification, ANN is one of a powerful computational tool for learning the relationship between the input, such as clinical parameters and radiological findings, and the output, such as interstitial diseases [4,7,10]. Although NN has many successful examples in practice, its generalization ability is an important bottle-neck.
An automatic computer-aided diagnosis system for liver tumours on computed tomography images
2013, Computers and Electrical EngineeringCitation Excerpt :Hence, it is necessary to use texture features which describe the relative homogeneity or non-uniformity of the images. Thus, only six of the most commonly used texture descriptors for live tumour classification [16] including angular second moment, contrast, correlation, inverse difference moment, entropy and variance are used to extract texture features of cropped ROI. The Discrete Wavelet Transform (DWT) can be implemented with filter banks [17].
Medical image analysis with artificial neural networks
2010, Computerized Medical Imaging and GraphicsCitation Excerpt :ANN has also been widely applied in diagnosis of diseases in MR images. In Guo et al. [65], a computer-aided diagnostic system was proposed to classify rat liver lesions from MR imaging. Six parameters of texture characteristics and variance of 161 ROIs were calculated and assessed by gray-level co-occurrence matrices, then fed into a back propagation neural network classifier to classify the liver tissue into two classes namely cirrhosis and HCC.
Intelligent Magnetic Resonance Imaging analysis and processing technology for magnetic nanoparticles medical applications
2023, ACM International Conference Proceeding Series
DongMei Guo A doctoral student of Dalian University of Technology, Department of Biomedical Engineering. She is also a radiologist of Second Affiliated Hospital, Dalian Medical University.
TianShuang Qiu received the B.S. degree from Tianjin University, Tianjin, China, in 1983, the M.S. degree from Dalian University of Technology, Dalian, China, in 1993, and the PhD degree from Southeastern University, Nanjing, China, in 1996, all in electrical engineering. He conducted his post-doctoral research in the Department of Electrical Engineering at Northern Illinois University, DeKalb, UAS. He is currently a professor in the Department of Electronic Engineering, Dalian University of Technology.