Fibro-Scope V1.0.1: an artificial intelligence/neural network system for staging of nonalcoholic steatohepatitis

• Published in: Hepatology international 2023-06, Vol.17 (3), p.573-583
• Main Authors: Yamaguchi, Kanji, Shima, Toshihide, Mitsumoto, Yasuhide, Seko, Yuya, Umemura, Atsushi, Itoh, Yoshito, Nakajima, Atsushi, Kaneko, Shuichi, Harada, Kenichi, Watkins, Timothy, Okanoue, Takeshi
• Format: Article
• Language: English
• Subjects: Alanine; Alanine transaminase; Artificial Intelligence; Cholesterol; Cirrhosis; Clinical trials; Collagen (type IV); Colorectal Surgery; Diagnostic systems; Fibrosis; Hepatology; Humans; Liver - pathology; Liver Cirrhosis - diagnosis; Liver Cirrhosis - pathology; Medical diagnosis; Medicine; Medicine & Public Health; Neural networks; Neural Networks, Computer; Non-alcoholic Fatty Liver Disease - diagnosis; Non-alcoholic Fatty Liver Disease - pathology; Original Article; Patients; Sensitivity analysis; Surgery; Training; Triglycerides; Validation studies; γ-Glutamyltransferase
• ISSN: 1936-0533; 1936-0541
• DOI: 10.1007/s12072-022-10454-0

Background Fibro-Scope is an artificial intelligence/neural network system to determine the fibrosis stage in nonalcoholic steatohepatitis (NASH) using 12 parameters of the patient: age, sex, height, weight, waist circumference (WC), platelet count, and the levels of aspartate and alanine aminotransferase, gamma-glutamyltransferase, cholesterol, triglycerides, and type IV collagen 7S. However, measurement of WC is unstable and often missing from patient databases. Herein, we created Fibro-Scope V1.0.1 that has the same detection power as its predecessor, without the need to consider WC. Methods To build a new AI diagnostic system available for the global needs, data from 764 patients with NASH and bridging fibrosis (STELLAR-3) or compensated cirrhosis (STELLAR-4) that participated in two phase III trials were added to the Japanese data. Finally, the data of a total of 898 patients in the training and of 300 patients in the validation studies were analyzed, respectively. Results The discrimination of F0–2 from F3,4 through Fibro-Scope V1.0.1 was characterized by a 99.8% sensitivity, a 99.6% specificity, a 99.8% positive predictive value, and a 99.6% negative predictive value in a training study with gray zone analysis; similar effectiveness was also revealed in the analysis without a gray zone. In the validation studies with and without gray zone analysis, high sensitivity and specificity were also identified. Fibro-Scope V1.0.1 exerted a diagnostic accuracy for F3,4 advanced fibrosis that was comparable to that of the original Fibro-Scope and delivered high (> 92%) sensitivity and specificity. Conclusion Fibro-Scope V1.0.1 can accurately diagnose F3,4 fibrosis without the need of WC.