Development and validation of a minimally invasive diagnostic model for biliary atresia using artificial intelligence.
Authors
Affiliations (12)
Affiliations (12)
- Department of Pediatric Surgery, Shanghai Key Laboratory of Birth Defect, and Key Laboratory of Neonatal Disease, Children's Hospital of Fudan University, Ministry of Health, 399 Wan Yuan Rd, Shanghai 201102, China.
- Department of Ultrasound, Children's Hospital of Fudan University, Shanghai 201102, China.
- Department of Pediatrics, University of Groningen, University Medical Center Groningen, Groningen, The Netherlands.
- Department of Population Health Sciences, Weill Cornell Medicine, New York, 10065, USA.
- Department of Pediatric Gastroenterology, School of Medicine, Chengdu Women's and Children's Central Hospital, University of Electronic Science and Technology of China, Chengdu 610091, China.
- Department of Pediatric Surgery, Shengjing Hospital of Chinese Medical University, Shenyang 110004, China.
- Department of Pediatric Surgery, Xiamen Children's Hospital, Xiamen 361006, China.
- Department of Pediatric Surgery, Affiliated Hospital of Zunyi Medical University, Zunyi 563000, China.
- Department of Pediatric Surgery, Xuzhou Children's Hospital, Xuzhou 221002, China.
- Department of Clinical Epidemiology, Clinical Trial Unit (CTU), Children's Hospital of Fudan University, 399 Wan Yuan Road, Shanghai 201102, China. [email protected].
- Department of Pediatric Surgery, Shanghai Key Laboratory of Birth Defect, and Key Laboratory of Neonatal Disease, Children's Hospital of Fudan University, Ministry of Health, 399 Wan Yuan Rd, Shanghai 201102, China. [email protected].
- Department of Pediatric Surgery, Shanghai Key Laboratory of Birth Defect, and Key Laboratory of Neonatal Disease, Children's Hospital of Fudan University, Ministry of Health, 399 Wan Yuan Rd, Shanghai 201102, China. [email protected].
Abstract
Ultrasound and serum matrix metalloproteinase-7 (MMP-7) hold great value in distinguishing biliary atresia (BA) from other cholestatic diseases. This study aims to assess the accuracy of an artificial intelligence (AI) based diagnostic model of ultrasound combined with serum MMP-7 in discriminative diagnosis of BA. This is a multicenter diagnostic study involving six medical centers in China. Patients with obstructive jaundice were enrolled. A set of morphological operators were employed to extract features of the ultrasound images to construct an AI algorithm. Logistic regression model was established with validation. Two cohorts with a total of 348 children with obstructive jaundice were recruited from January 2020 to April 2023. A retrospective cohort of 187 infants served as a training cohort; this included 56 BA and 131 non-BA patients. Serum MMP-7 testing model yielded an area under the receiver-operating characteristic curve (AUROC) of 0.916 [95% confidence interval (CI) = 0.876-0.956], sensitivity of 94.6% (95% CI = 85.1%-98.9%), specificity of 88.6% (95% CI = 81.8%-93.5%), and accuracy of 90.4% (95% CI = 85.2%-94.2%). Values for ultrasound testing model were 0.945 (95% CI = 0.902-0.987), 98.2% (95% CI = 90.5%-99.9%), 91.6% (95% CI = 85.5%-95.7%), and 93.6% (95% CI = 89.1%-96.6%), respectively. The combined AI model obtained an AUROC of 0.985 (95% CI = 0.971-0.999), sensitivity of 98.2% (95% CI = 90.5%-99.9%), specificity of 93.1% (95% CI = 84.4%-96.4%), and accuracy of 94.7% (95% CI = 90.4%-97.4%), respectively. Performance was confirmed using a multicenter prospective validation cohort of 161 patients that included 100 BA cases. An AI model combining ultrasound and serum MMP-7 demonstrated robust high sensitivity and specificity in the differential diagnosis of BA.