Convolutional neural network detects and estimates severity of cirrhosis using ECG findings

June 17, 2022

Cirrhosis is among the leading causes of death worldwide, but early detection and outcome prediction in patients with this condition remain important areas of unmet need. Patients with early-stage cirrhosis can remain asymptomatic, and most patients are diagnosed later when complications have already developed.

Cirrhosis is associated with cardiac dysfunction and distinct electrocardiogram (ECG) abnormalities that correlate with liver disease severity. Recent studies led by Mayo Clinic Cardiovascular Medicine investigators Itzhak Zachi Attia, Ph.D., Peter A. Noseworthy, M.D., and Paul A. Friedman, M.D., have shown that when applied to the findings obtained from digitized 12-lead ECGs, deep learning-based artificial intelligence (AI) models using convolutional neural networks (CNNs) can enable the prediction of various cardiac and non-cardiac conditions.

Building on that knowledge, Mayo Clinic researchers conducted a study to test whether a CNN trained on a sample of patients with cirrhosis would be able to detect signs of cirrhosis and produce a numerical AI-Cirrhosis-ECG (ACE) score that correlates with disease severity. The results of this study were published in the American Journal of Gastroenterology in 2022. Mayo Clinic gastroenterologist Douglas (Doug) A. Simonetto, M.D., served as the article's corresponding author and Joseph C. Ahn, M.D., a Mayo Clinic gastroenterology fellow, served as first author.

Study methods

The Mayo Clinic researchers identified 5,212 adult patients with advanced cirrhosis who underwent liver transplantation at Mayo Clinic transplant centers between 1988 and 2019, and a control group with 20,728 age- and sex-matched individuals without liver disease. Data from these participants were used to train, validate and test a CNN. Within the control group, 70% were assigned to the training set, 10% were assigned to the validation set and 20% were assigned to the testing set. The study's primary outcome was the CNN model's performance in distinguishing patients with cirrhosis from controls using the ECG findings. Researchers also assessed the association between the ACE score and the severity of patients' liver disease.

Key findings

  • The ACE score performed well in classifying ECGs from patients with cirrhosis versus controls. The area under the curve (AUC) within the testing set was 0.908, with 84.9% sensitivity and 83.2% specificity, and this performance remained consistent after additional matching for medical comorbidities.
  • The ACE score was positively associated with markers of liver disease severity, including the model for end-stage liver disease-sodium (MELD-Na) score.
  • Longitudinal trends in the ACE scores before and after liver transplantation mirrored the progression and resolution of liver disease.

According to Drs. Simonetto and Ahn, this was the first study to apply deep learning-based AI technology to demonstrate the presence of a strong cirrhosis-related ECG signal and quantify the signal in a manner that correlates with the severity of liver disease.

When asked to summarize how these findings might guide clinical practice in the future, Drs. Ahn and Simonetto see a number of possibilities. "Given that ECGs are used worldwide and are one of the most commonly ordered medical tests, the use of this test for detection of cirrhosis may enable early detection and timely treatment in a broad population of patients," explains Dr. Ahn.

"The novel relationship between AI-enabled ECG analysis and cirrhosis holds promise as the basis for future low-cost tools and applications that facilitate early detection, timely treatment and prediction of clinical outcomes in patients with cirrhosis," says Dr. Simonetto.

Plans for future research building on these results are already underway. "Because the original CNN was trained using patients with advanced cirrhosis who were sick enough to need liver transplantation, we are in the process of developing another model that can detect cirrhosis at earlier stages," says Dr. Ahn.

Given that the initial study suggested a strong relationship between the magnitude of the ACE score and the severity of liver disease, researchers have also begun analyzing the associations between the ACE score and several liver-related outcomes, including hepatic decompensation, liver-related death or liver transplantation.

"For widespread implementation, external validation of our algorithm at sites outside of Mayo Clinic will also be essential," says Dr. Simonetto.

For more information

Ahn JC, et al. Development of the ACE score: An electrocardiogram-based deep learning model in cirrhosis. The American Journal of Gastroenterology. 2022;117:424.

Refer a patient to Mayo Clinic.