Computer-aided diagnosis of liver tumors on computed tomography images

https://doi.org/10.1016/j.cmpb.2017.04.008Get rights and content

Highlights

  • A computer-aided diagnosis (CAD) system was proposed to diagnose liver cancer using the features of tumors obtained from multiphase CT images.

  • Three kinds of features were extracted for each tumor, including 3-D texture feature (GLCM), 3-D shape (compactness and elliptic model), and kinetic curve characteristics (maximum enhancement, time to peak, uptake rate, and washout rate).

  • This study compared the performances of each combination of different feature sets. Combining all of the features achieved the highest accuracy (58/71, 81.69%), sensitivity (18/22, 81.82%), and specificity (40/49, 81.63%). The Az value of combining all features was 0.8713.

  • Combining texture, shape, and kinetic curve features may be able to differentiate benign from malignant tumors in the liver using our proposed CAD system.

Abstract

Background and objective

Liver cancer is the tenth most common cancer in the USA, and its incidence has been increasing for several decades. Early detection, diagnosis, and treatment of the disease are very important. Computed tomography (CT) is one of the most common and robust imaging techniques for the detection of liver cancer. CT scanners can provide multiple-phase sequential scans of the whole liver. In this study, we proposed a computer-aided diagnosis (CAD) system to diagnose liver cancer using the features of tumors obtained from multiphase CT images.

Methods

A total of 71 histologically-proven liver tumors including 49 benign and 22 malignant lesions were evaluated with the proposed CAD system to evaluate its performance. Tumors were identified by the user and then segmented using a region growing algorithm. After tumor segmentation, three kinds of features were obtained for each tumor, including texture, shape, and kinetic curve. The texture was quantified using 3 dimensional (3-D) texture data of the tumor based on the grey level co-occurrence matrix (GLCM). Compactness, margin, and an elliptic model were used to describe the 3-D shape of the tumor. The kinetic curve was established from each phase of tumor and represented as variations in density between each phase. Backward elimination was used to select the best combination of features, and binary logistic regression analysis was used to classify the tumors with leave-one-out cross validation.

Results

The accuracy and sensitivity for the texture were 71.82% and 68.18%, respectively, which were better than for the shape and kinetic curve under closed specificity. Combining all of the features achieved the highest accuracy (58/71, 81.69%), sensitivity (18/22, 81.82%), and specificity (40/49, 81.63%). The Az value of combining all features was 0.8713.

Conclusions

Combining texture, shape, and kinetic curve features may be able to differentiate benign from malignant tumors in the liver using our proposed CAD system.

Introduction

Liver cancer is the tenth most common cancer and the fifth and ninth most common causes of cancer death in men and women, respectively, in the USA [1]. Many cancers have a high survival rate if detected and treated early [2]. Computed tomography (CT) is one of the most common and robust imaging techniques for the detection of liver tumors [3]. CT provides images of the whole liver through multiphase sequential scans after the injection of contrast media [4].

Several studies have discussed CAD systems for liver tumors using CT images [5], [6], [7]. Chen et al. [5] reported a CAD system that used boundary detection to automatically identify the contours of the liver, and a neural network (NN) classification system using a spatial grey-level co-occurrence matrix (SGLCM) [8], [9] to distinguish a normal liver and two types of liver tumors, hepatoma and hemangioma. Gletsos et al. [7] used average gray levels and 48 texture characteristics with three hierarchically placed feedforward NNs to classify tumors into healthy, cyst, hemangioma, and hepatocellular carcinoma groups, however segmentation of the tumors was done manually using regions of interest. In the study by Zhang et al. [6], the authors used an automatic edge detection and subtraction process [10] to detect hepatocellular carcinoma and a 3-layer feedforward NN to differentiate the degree of cirrhosis in the liver, which was derived by integrating two shape and seven grey-level co-occurrence matrix (GLCM) texture features [8].

This study developed a computer-aided diagnosis (CAD) system with semi-automatic region growing segmentation technique. Three kinds of features were extracted for analyzing between benign and malignant, including features of texture [11], [12], [13], [14], shape [15], [16], [17], [18], and kinetic curve [19], [20], [21]. Texture was quantified using GLCM [8] including the relative position and density of the tumor. Shape was classified using an elliptic model [17], [18], and a kinetic curve was generated to represent changes in density between each phase on the CT images. To simplify the kinetic curve, we used the four features proposed by Chen et al. [20], [22] including maximum enhancement, time to peak, uptake rate, and washout rate.

Section snippets

CT scans

This retrospective study was approved by the institutional review board at our hospital and the requirement for informed consent was waived. The data used in this study were acquired from two CT scanners (Brilliance iCT 256, Philips Healthcare, Dutch; SOMATOM Definition AS+, Siemens Healthcare, German) with 100, 120, or 130 kV, automatic mA control, and without extra noise reduction processes. The parameters of the CT images were as follows: slice thickness 0.7 mm–1.5 mm, and image size

Methods

In this study, the CAD system was comprised of five parts: tumor segmentation, tumor registration, feature extraction, and tumor classification. The tumor region was segmented by using a region growing method [23] on the CT phase which is the most conspicuous focal liver lesion selected by one author (C.C.C. with 10 years of experience in the diagnosis of abdominal imaging). Subsequently, a tumor registration method was performed manually to align the tumor between each phase. Three kinds of

Results

The comparison of texture, shape, and kinetic curve features between benign and malignant tumors is listed in Table 2. For texture features, six features (G1, G4, G8, G9, G12, and G14) showed significant differences (p < 0.05) between the benign and malignant tumor. Compared with the benign tumors, the malignant tumors are relative larger in size and more spiculated (significant higher values of R and Rs). For the kinetic curve features, none of the four features were statistically

Discussion

Proliferation is the main property of tumors, especially malignant ones [33]. Further, angiogenesis is important since the proliferation of cancer cells which generates capillaries from the existing vessel for supplying nutrients [34], [35]. In this study, the intra-tumor texture generated from the immature and tortuous new capillaries [35] was quantified by GLCM texture analysis. To quantify the tumor shape, an ellipsoid fitting model was established and a series of morphological properties

Funding

The authors thank the Ministry of Science and Technology (MOST104-2221-E-002-062-MY3) of the Republic of China for the financial support.

References (36)

  • C. Mathers et al.

    The Global Burden of Disease: 2004 Update

    (2008)
  • T. Murakami et al.

    Ultrasonography, computed tomography and magnetic resonance imaging of hepatocellular carcinoma: toward improved treatment decisions

    Oncology

    (2011)
  • K. Mitsuzaki et al.

    Multiple-phase helical CT of the liver for detecting small hepatomas in patients with liver cirrhosis: contrast-injection protocol and optimal timing

    Am. J. Roentgenology

    (1996)
  • E.L. Chen et al.

    An automatic diagnostic system for CT liver image classification

    IEEE Trans. Biomed. Eng.

    (1998)
  • X. Zhang et al.

    CAD on liver using CT and MRI

    Med. Imaging Inf.

    (2008)
  • M. Gletsos et al.

    A computer-aided diagnostic system to characterize CT focal liver lesions: design and optimization of a neural network classifier

    IEEE Trans. Inf. Technol. Biomed.

    (2003)
  • R.M. Haralick et al.

    Textural features for image classification

    IEEE Trans. Syst. Man Cybern.

    (1973)
  • C.M. Wu et al.

    Texture features for classification of ultrasonic liver images

    IEEE Trans. Med. Imaging

    (1992)
  • Cited by (88)

    • Recent advancement in cancer diagnosis using machine learning and deep learning techniques: A comprehensive review

      2022, Computers in Biology and Medicine
      Citation Excerpt :

      GLCM forms a matrix representing probabilities of pixel-pair correlation and may be used to extract different statistical measures showing information about texture properties of an image i.e., correlation, contrast, dissimilarity, homogeneity, energy etc. Although GLCM is efficient feature descriptor to extract texture features from image, however it performs significantly well, only when its use case involve requirement of quantization of texture in an image [50,62,73,84,101,110,127,137,158]. Emergence of ML/DL-based techniques have enriched cancer diagnosis area with its overwhelmed efficiency in computation of complex problem with lower error-rate than humans.

    View all citing articles on Scopus
    View full text