Lab Med Online.  2023 Jul;13(3):189-198. 10.47429/lmo.2023.13.3.189.

Liver Fibrosis Biomarker Validation Using Machine Learning Algorithms

• Kim MK
• Kwon HJ
• Shin K
• Park H
• Woo HY
• Park CH
• Kwon MJ

• KMID: 2552743
• DOI: http://doi.org/10.47429/lmo.2023.13.3.189

Abstract

Background
Liver fibrosis which causes several liver diseases, requires early screening and management. The gold standard for fibrosis assessment, liver biopsy, has recently been replaced by noninvasive scores. In this study, we validated liver fibrosis-associated biomarkers using machine learning techniques applied in medical research and evaluated their prediction models.
Methods
Noninvasive scores were assayed in 144 patients who underwent transient elastography (TE). The patients were divided into three groups (< 7 kPa, 7–10 kPa, ≥ 10 kPa) according to their TE results. Feature selection and modeling for predicting liver fibrosis were performed using random forest (RF) and support vector machine (SVM).
Results
Considering the mean decrease in impurity, permutation importance, and multicollinear analysis, the important features for differentiating between the three groups were Mac-2 binding protein glycosylation isomer (M2BPGi), platelet count, and aspartate aminotransferase (AST). Using these features, the RF and SVM models showed equivalent or better performance than noninvasive scores. The sensitivities of RF and SVM models for predicting ≥ 7 kPa TE results were higher than noninvasive scores (83.3% and 90.0% vs. < 80%, respectively). The sensitivity and specificity of RF and SVM models for ≥ 10 kPa TE result was 100%.
Conclusions
We used machine learning techniques to verify the usefulness of established serological biomarkers (M2BPGi, PLT, and AST) that predict liver fibrosis. Conclusively, machine learning models showed better performance than noninvasive scores.

Keyword

Liver fibrosis; Stiffness measurements; M2BPGi; Machine learning; Random forest; Support vector machine

Figure

• 

  Fig. 1 Noninvasive fibrosis score distribution according to groups by FibroScan. Boxes designate the interquartile range (25–75 percentile), and the middle line represents the median. The error bar represents minimum and maximum values. Group 1 shows no or minimal fibrosis (<7 kPa), group 2 shows moderate fibrosis (7–10 kPa), and group 3 shows severe fibrosis (≥10 kPa). NFS: nonalcoholic fatty liver disease fibrosis score; APRI: AST (Aspartate transaminase)–platelet ratio index; FIB-4: fibrosis 4.

• 

  Fig. 2 Feature selection for differentiating between the three groups. The mean decrease in impurity (left panel) and permutation importance (right panel) were calculated to assess the importance of features. Abbreviations: M2BPGi, Mac-2 binding protein glycosylation isomer; PLT, platelet count; AST, aspartate transaminase; PT, prothrombin time; WBC, whole blood cells; T_bil, total bilirubin; HTN, hypertension; Cr, creatinine; ALB, albumin; GGT, gamma-glutamyl transferase; BMI, body mass index; DM, diabetes mellitus; HL, hyperlipidemia.

• 

  Fig. 3 Five-fold cross-validation composite receiver operating characteristic (ROC) curves for predicting (A) moderate (≥7 kPa) and (B) severe fibrosis (≥10 kPa). Five-fold cross-validation separate ROC curves for predicting (C) moderate (≥7 kPa) and (D) severe fibrosis (≥10 kPa). The random forest and support vector machine are shown in the left and right panels, respectively. Abbreviation: AUC, area under the curve.

Reference

1. Campana L, Esser H, Huch M, Forbes S. 2021; Liver regeneration and inflammation: from fundamental science to clinical applications. Nat Rev Mol Cell Biol. 22:608–24. DOI: 10.1038/s41580-021-00373-7. PMID: 34079104.

2. Rockey DC, Bell PD, Hill JA. 2015; Fibrosis - a common pathway to organ injury and failure. N Engl J Med. 372:1138–49. DOI: 10.1056/NEJMra1300575. PMID: 25785971.

3. Pellicoro A, Ramachandran P, Iredale JP, Fallowfield JA. 2014; Liver fibrosis and repair: immune regulation of wound healing in a solid organ. Nat Rev Immunol. 14:181–94. DOI: 10.1038/nri3623. PMID: 24566915.

4. Kim BK, Fun J, Yuen MF, Kim SU. 2013; Clinical application of liver stiffness measurement using transient elastography in chronic liver disease from longitudinal perspectives. World J Gastroenterol. 19:1890–900. DOI: 10.3748/wjg.v19.i12.1890. PMID: 23569334. PMCID: PMC3613104.

5. Khalifa A, Rockey DC. The utility of liver biopsy in 2020. Curr Opin Gastroenterol. 2020; 36:184–91. DOI: 10.1097/MOG.0000000000000621. PMID: 32097176.

6. Fallatah HI, Akbar HO, Fallatah AM. 2016; Fibroscan compared to FIB-4, APRI, and AST/ALT ratio for assessment of liver fibrosis in Saudi patients with nonalcoholic fatty liver disease. Hepat Mon. 16:e38346. DOI: 10.5812/hepatmon.38346. PMID: 27642348. PMCID: PMC5018361.

7. Amernia B, Moosavy SH, Banookh F, Zoghi G. 2021; FIB-4, APRI, and AST/ALT ratio compared to FibroScan for the assessment of hepatic fibrosis in patients with non-alcoholic fatty liver disease in Bandar Abbas, Iran. BMC Gastroenterol. 21:453. DOI: 10.1186/s12876-021-02038-3. PMID: 34861841. PMCID: PMC8642865.

8. Moon HW, Park M, Hur M, Kim H, Choe WH, Yun YM. 2018; Usefulness of enhanced liver fibrosis, glycosylation isomer of Mac-2 binding protein, galectin-3, and soluble suppression of tumorigenicity 2 for assessing liver fibrosis in chronic liver diseases. Ann Lab Med. 38:331–7. DOI: 10.3343/alm.2018.38.4.331. PMID: 29611383. PMCID: PMC5895862.

9. Shirabe K, Bekki Y, Gantumur D, Araki K, Ishii N, Kuno A, et al. 2018; Mac-2 binding protein glycan isomer (M2BPGi) is a new serum biomarker for assessing liver fibrosis: more than a biomarker of liver fibrosis. J Gastroenterol. 53:819–26. DOI: 10.1007/s00535-017-1425-z. PMID: 29318378.

10. Saleh SA, Salama MM, Alhusseini MM, Mohamed GA. 2020; M2BPGi for assessing liver fibrosis in patients with hepatitis C treated with direct-acting antivirals. World J Gastroenterol. 26:2864–76. DOI: 10.3748/wjg.v26.i21.2864. PMID: 32550761. PMCID: PMC7284180.

11. Mak LY, Wong DK, Cheung KS, Hui RW, Liu F, Fung J, et al. 2022; Role of serum M2BPGi levels in predicting persistence of advanced fibrosis in chronic hepatitis B virus infection. Dig Dis Sci. 67:5127–36. DOI: 10.1007/s10620-022-07429-4. PMID: 35258755.

12. Choi R, Oh Y, Lee S, Lee SG. 2019; Evaluation of the serum Mac-2 binding protein glycosylation isomer test used for diagnosis and monitoring of liver fibrosis and the correlation of Mac-2 binding protein glycosylation isomer with hemoglobin A1c. Clin Lab. 65:197. DOI: 10.7754/Clin.Lab.2018.180719.

13. Tsochatzis EA, Gurusamy KS, Ntaoula S, Cholongitas E, Davidson BR, Burroughs AK. 2011; Elastography for the diagnosis of severity of fibrosis in chronic liver disease: a meta-analysis of diagnostic accuracy. J Hepatol. 54:650–9. DOI: 10.1016/j.jhep.2010.07.033. PMID: 21146892.

14. Mikolasevic I, Orlic L, Franjic N, Hauser G, Stimac D, Milic S. 2016; Transient elastography (FibroScan^®) with controlled attenuation parameter in the assessment of liver steatosis and fibrosis in patients with nonalcoholic fatty liver disease - Where do we stand? World J Gastroenterol. 22:7236–51. DOI: 10.3748/wjg.v22.i32.7236. PMID: 27621571. PMCID: PMC4997649.

15. Festi D, Schiumerini R, Marzi L, Di Biase AR, Mandolesi D, Montrone L, et al. 2013; Review article: the diagnosis of non-alcoholic fatty liver disease - availability and accuracy of non-invasive methods. Aliment Pharmacol Ther. 37:392–400. DOI: 10.1111/apt.12186. PMID: 23278163.

16. Oeda S, Tanaka K, Oshima A, Matsumoto Y, Sueoka E, Takahashi H. 2020; Diagnostic accuracy of FibroScan and factors affecting measurements. Diagnostics (Basel). 10:940. DOI: 10.3390/diagnostics10110940. PMID: 33198092. PMCID: PMC7696616.

17. Haq AU, Li JP, Saboor A, Khan J, Wali S, Ahmad S, et al. 2021; Detection of breast cancer through clinical data using supervised and unsupervised feature selection techniques. IEEE Access. 9:22090–105. DOI: 10.1109/ACCESS.2021.3055806.

18. Ahmad S, Khan S, Fahad M, Kumar Dutta A, Minh Dang L, Prasad Joshi G, et al. 2022; Deep learning enabled disease diagnosis for secure internet of medical things. Computers, Materials & Continua. 73:965–79. DOI: 10.32604/cmc.2022.025760.

19. Poonguzhali R, Ahmad S, Sivasankar PT, Babu SA, Joshi P, Joshi GP, et al. 2023; Automated brain tumor diagnosis using deep residual U-Net segmentation model. Computers, Materials & Continua. 74:2179–94. DOI: 10.32604/cmc.2023.032816.

20. Decharatanachart P, Chaiteerakij R, Tiyarattanachai T, Treeprasertsuk S. 2021; Application of artificial intelligence in chronic liver diseases: a systematic review and meta-analysis. BMC Gastroenterol. 21:10. DOI: 10.1186/s12876-020-01585-5. PMID: 33407169. PMCID: PMC7788739.

21. Spann A, Yasodhara A, Kang J, Watt K, Wang B, Goldenberg A, et al. 2020; Applying machine learning in liver disease and transplantation: a comprehensive review. Hepatology. 71:1093–105. DOI: 10.1002/hep.31103. PMID: 31907954.

22. Deo RC. 2015; Machine learning in medicine. Circulation. 132:1920–30. DOI: 10.1161/CIRCULATIONAHA.115.001593. PMID: 26572668. PMCID: PMC5831252.

23. Feng G, Zhang X, Zhang L, Liu WY, Geng S, Yuan HY, et al. Novel urinary protein panels for the non-invasive diagnosis of non-alcoholic fatty liver disease and fibrosis stages. Liver Int. 2023; 43:1234–6. DOI: 10.1111/liv.15565. PMID: 36924436.

24. Alkhouri N. 2017; Putting it all together: noninvasive diagnosis of fibrosis in nonalcoholic fatty liver disease in adults and children. Clin Liver Dis (Hoboken). 9:134–7. DOI: 10.1002/cld.636. PMID: 30992977. PMCID: PMC6467159.

25. Wei B, Feng S, Chen E, Li D, Wang T, Gou Y, et al. 2018; M2BPGi as a potential diagnostic tool of cirrhosis in Chinese patients with hepatitis B virus infection. J Clin Lab Anal. 32:e22261. DOI: 10.1002/jcla.22261. PMID: 28544156. PMCID: PMC6817288.

26. Jekarl DW, Choi H, Lee S, Kwon JH, Lee SW, Yu H, et al. 2018; Diagnosis of liver fibrosis with Wisteria floribunda agglutinin-positive Mac-2 binding protein (WFA-M2BP) among chronic hepatitis B patients. Ann Lab Med. 38:348–54. DOI: 10.3343/alm.2018.38.4.348. PMID: 29611385. PMCID: PMC5895864.

27. Cross TJ, Calvaruso V, Maimone S, Carey I, Chang TP, Pleguezuelo M, et al. 2010; Prospective comparison of Fibroscan, King's score and liver biopsy for the assessment of cirrhosis in chronic hepatitis C infection. J Viral Hepat. 17:546–54. DOI: 10.1111/j.1365-2893.2009.01210.x. PMID: 19874477.

28. Wong VW, Vergniol J, Wong GL, Foucher J, Chan HL, Le Bail B, et al. 2010; Diagnosis of fibrosis and cirrhosis using liver stiffness measurement in nonalcoholic fatty liver disease. Hepatology. 51:454–62. DOI: 10.1002/hep.23312. PMID: 20101745.

29. Breiman L, Friedman JH, editors. Classification and regression trees-[eBook]. New York: Routledge;2017.

30. Altmann A, Tolosi L, Sander O, Lengauer T. 2010; Permutation importance: a corrected feature importance measure. Bioinformatics. 26:1340–7. DOI: 10.1093/bioinformatics/btq134. PMID: 20385727.

31. Breiman L. 2001; Random forests. Mach Learn. 45:5–32. DOI: 10.1023/A:1010933404324.

32. Burger CJ. A tutorial on support vector machines for pattern recognition. Data mining and knowledge discovery. 1998; 2:121–67. DOI: 10.1023/A:1009715923555.

33. Hunter JD. 2007; Matplotlib: a 2D graphics environment. Computing in science & engineering. 9:90–5. DOI: 10.1109/MCSE.2007.55.

34. Pedregosa F, Varoquaux G, Gramfort A, Michel V, Thirion B, Grisel O, et al. 2011; Scikit-learn: Machine learning in Python. J Mach Learn Res. 12:2825–30.

35. Nishikawa H, Enomoto H, Iwata Y, Hasegawa K, Nakano C, Takata R, et al. 2016; Impact of serum Wisteria floribunda agglutinin positive Mac-2-binding protein and serum interferon-gamma-inducible protein-10 in primary biliary cirrhosis. Hepatol Res. 46:575–83. DOI: 10.1111/hepr.12595. PMID: 26418076.

36. Umemura T, Joshita S, Sekiguchi T, Usami Y, Shibata S, Kimura T, et al. 2015; Serum Wisteria floribunda agglutinin-positive Mac-2-binding protein level predicts liver fibrosis and prognosis in primary biliary cirrhosis. Am J Gastroenterol. 110:857–64. DOI: 10.1038/ajg.2015.118. PMID: 25916223.

37. Yamada N, Sanada Y, Tashiro M, Hirata Y, Okada N, Ihara Y, et al. 2017; Serum Mac-2 binding protein glycosylation isomer predicts grade F4 liver fibrosis in patients with biliary atresia. J Gastroenterol. 52:245–52. DOI: 10.1007/s00535-016-1235-8. PMID: 27349650.

38. Nishikawa H, Enomoto H, Iwata Y, Hasegawa K, Nakano C, Takata R, et al. 2016; Clinical significance of serum Wisteria floribunda agglutinin positive Mac-2-binding protein level and high-sensitivity C-reactive protein concentration in autoimmune hepatitis. Hepatol Res. 46:613–21. DOI: 10.1111/hepr.12596. PMID: 26406984.

39. Abe M, Miyake T, Kuno A, Imai Y, Sawai Y, Hino K, et al. 2015; Association between Wisteria floribunda agglutinin-positive Mac-2 binding protein and the fibrosis stage of non-alcoholic fatty liver disease. J Gastroenterol. 50:776–84. DOI: 10.1007/s00535-014-1007-2. PMID: 25326152.

40. Mizuno M, Shima T, Oya H, Mitsumoto Y, Mizuno C, Isoda S, et al. 2017; Classification of patients with non-alcoholic fatty liver disease using rapid immunoassay of serum type IV collagen compared with liver histology and other fibrosis markers. Hepatol Res. 47:216–25. DOI: 10.1111/hepr.12710. PMID: 26997642.

41. Nishikawa H, Enomoto H, Iwata Y, Kishino K, Shimono Y, Hasegawa K, et al. 2016; Clinical significance of serum Wisteria floribunda agglutinin positive Mac-2-binding protein level in non-alcoholic steatohepatitis. Hepatol Res. 46:1194–202. DOI: 10.1111/hepr.12662. PMID: 26836229.

42. Shigefuku R, Takahashi H, Nakano H, Watanabe T, Matsunaga K, Matsumoto N, et al. 2016; Correlations of hepatic hemodynamics, liver function, and fibrosis markers in nonalcoholic fatty liver disease: Comparison with chronic hepatitis related to hepatitis C virus. Int J Mol Sci. 17:1545. DOI: 10.3390/ijms17091545. PMID: 27649152. PMCID: PMC5037819.

43. Fujiyoshi M, Kuno A, Gotoh M, Fukai M, Yokoo H, Kamachi H, et al. 2015; Clinicopathological characteristics and diagnostic performance of Wisteria floribunda agglutinin positive Mac-2-binding protein as a preoperative serum marker of liver fibrosis in hepatocellular carcinoma. J Gastroenterol. 50:1134–44. DOI: 10.1007/s00535-015-1063-2. PMID: 25773774.

44. Toshima T, Shirabe K, Ikegami T, Yoshizumi T, Kuno A, Togayachi A, et al. 2015; A novel serum marker, glycosylated Wisteria floribunda agglutinin-positive Mac-2 binding protein (WFA(+)-M2BP), for assessing liver fibrosis. J Gastroenterol. 50:76–84. DOI: 10.1007/s00535-014-0946-y. PMID: 24603981.

45. Vapnik VN. 1999; An overview of statistical learning theory. IEEE Trans Neural Netw. 10:988–99. DOI: 10.1109/72.788640. PMID: 18252602.

46. LeCun Y, Bengio Y, Hinton G. 2015; Deep learning. Nature. 521:436–44.

47. Singh S, Fujii LL, Murad MH, Wang Z, Asrani SK, Ehman RL, et al. Liver stiffness is associated with risk of decompensation, liver cancer, and death in patients with chronic liver diseases: a systematic review and meta-analysis. Clin Gastroenterol Hepatol. 2013; 11:1573–84.