Noninvasive diagnosis of liver fibrosis: Utility of data mining of both ultrasound elastography and serological findings to construct a decision tree

Norihisa Yada, Masatoshi Kudo, Norifumi Kawada, Shuichi Sato, Yukio Osaki, Akihisa Ishikawa, Hisaaki Miyoshi, Michiie Sakamoto, Masayoshi Kage, Osamu Nakashima, Akiko Tonomura

Research output: Contribution to journalArticle

8 Citations (Scopus)

Abstract

Objective: Although liver biopsy is the gold standard for viral liver disease management, it is invasive and the sampling error rate is problematic. Real-time tissue elastography (RTE), a recently developed method of ultrasound elastography, can be used to assess liver fibrosis noninvasively but the overlap between fibrosis stages limits its ability to assess liver fibrosis adequately when used alone. Methods: A multicenter collaborative study involving 542 patients with chronic viral hepatitis and cirrhosis who were scheduled to undergo liver biopsy compared the image features obtained from RTE image analysis, the liver fibrosis index (LFI), and pathological diagnosis. RTE and a blood test were performed on the same day as the liver biopsy. Data mining was also performed to construct a decision tree, and its diagnostic performance for assessing liver fibrosis was evaluated. Results: The LFI was higher in patients with chronic hepatitis C (CHC) than in those with chronic hepatitis B (CHB). When a decision tree was constructed by data mining of RTE and serological findings, the diagnostic accuracy was very high for all fibrosis stages, with respective rates at F1, F2, F3, and F4 of 94.4, 54.1, 38.7, and 81.3% for patients with CHC and of 97.1, 50.0, 43.8, and 80.6% for patients with CHB. Conclusions: The variation in LFI values between the different etiologies appears to reflect the difference in the development style of liver fibrosis. The decision tree for assessing liver fibrosis constructed by data mining of both RTE and serological findings had a high diagnostic performance in assessing liver fibrosis and shows promising clinical utility.

Original languageEnglish
Pages (from-to)63-72
Number of pages10
JournalOncology (Switzerland)
Volume87
DOIs
Publication statusPublished - 2014 Apr 20

    Fingerprint

Keywords

  • Data mining
  • Liver fibrosis
  • Liver fibrosis index
  • Real-time tissue elastography
  • Strain elastography

ASJC Scopus subject areas

  • Oncology
  • Cancer Research

Cite this