AI model based on imaging, blood tests may help diagnose biliary atresia
Study: Mobile app processes the data and estimates probability of disease
Written by |
A new artificial intelligence (AI)-based model that combines information from imaging scans and blood levels of a protein called matrix metalloproteinase-7 (MMP-7) may aid in the early detection and accurate diagnosis of children with biliary atresia, a study shows.
“Our multicenter study confirms that combining ultrasound imaging AI with [blood] MMP-7 creates a minimally invasive diagnostic model superior to either method alone, revolutionizing early BA detection,” researchers wrote.
Based on the results, researchers developed a mobile app to facilitate the clinical application of the diagnostic model. Clinicians upload an ultrasound image of the liver and surrounding structures, along with blood MMP-7 levels, and the app processes the data and estimates the probability that the child has biliary atresia.
The study, “Development and validation of a minimally invasive diagnostic model for biliary atresia using artificial intelligence,” was published in World Journal of Pediatrics.
Procedure using fluorescent dye is current gold standard for diagnosis
Biliary atresia is a liver disease that occurs when the tubes, or ducts, that carry the digestive fluid bile from the liver to the small intestine are absent or blocked. This leads to slowed bile flow, known as cholestasis, that can cause bile to accumulate to toxic levels in the liver (damaging the organ) and leak into the bloodstream.
Symptoms of biliary atresia include jaundice (yellow discoloration of the skin and whites of the eyes), dark urine, and pale stools.
Cholangiography, a procedure that uses a fluorescent dye to evaluate bile flow in the bile ducts, is the gold standard for diagnosing biliary atresia and typically involves open surgery.
Therefore, there is a need for accurate, noninvasive methods to diagnose biliary atresia and differentiate it from conditions that cause similar symptoms. Ultrasound scans have been “widely used for initial evaluation of suspected BA [biliary atresia],” the researchers wrote, with several features helping to discriminate the disease from others.
An increasing number of studies have also pointed to MMP-7, a protein elevated in biliary atresia patients, as a highly specific blood biomarker.
Infants in study had jaundice caused by blocked bile ducts
In this study, a team of Chinese researchers evaluated the performance of an AI-based model that combines ultrasound scans and MMP-7 levels to discriminate children with biliary atresia from those with other cholestatic diseases.
The study involved 348 infants with jaundice caused by blocked bile ducts. Data from 56 infants with biliary atresia (30%) and 131 without, who were recruited at a single center, were used to train the AI model. Additionally, data from 100 infants with biliary atresia and 61 without, recruited at six centers, were used to validate the model.
In both training and validation groups, children with biliary atresia were significantly younger (51 to 55 days versus 67 to 73 days), more commonly girls, and had significantly higher blood levels of MMP-7 and a liver damage marker than those without biliary atresia.
Additionally, they had more abnormalities in ultrasound scans, such as an abnormal gallbladder, the organ where bile is stored, absent bile ducts, or an enlarged liver artery.
The team calculated the area under the curve (AUC), a statistical measure that assesses how well a given test can distinguish between two groups. AUC values range from 0 to 1, where 0.5 indicates no discriminative power and 1 indicates perfect discrimination.
The ultrasound-based AI model alone showed AUCs of 0.945 and 0.909 in the training and validation groups, respectively, while the MMP-7-based model showed AUCs of 0.916 and 0.907, respectively.
This novel combination model not only provides high diagnostic accuracy but also addresses challenges inherent in traditional diagnostic methods. The improved specificity, although not large, is of clinical significance to BA patients to avoid unnecessary diagnostic surgery.
The combined AI model, using both ultrasound information and blood MMP-7 levels, significantly improved AUC to 0.985 in the training group and 0.949 in the validation group. This demonstrated a better performance than ultrasound or MMP-7 alone.
The method’s sensitivity, meaning its ability to correctly identify children with biliary atresia, ranged from 89.8% to 98.2%, while the specificity, or the ability to correctly rule out those without biliary atresia, ranged from 91.4% to 93.1%.
To facilitate the clinical application of the new AI-based diagnostic model, the researchers developed a mobile app. The program requires at least one ultrasound image, including the gallbladder or the portal vein, the main blood vessel supplying the liver, and blood MMP-7 levels.
Upon submitting the data, the app processes it, and the subsequent page displays a prediction of the likelihood that the infant has biliary atresia.
“This novel combination model not only provides high diagnostic accuracy but also addresses challenges inherent in traditional diagnostic methods,” the researchers wrote. “The improved specificity, although not large, is of clinical significance to BA patients to avoid unnecessary diagnostic surgery.”
