LC-Pred: A Transformer-Based Interactive Interface for Liver Cirrhosis Prediction.

Objectives: Incorporating transformer, an innovative deep learning-based model with an authenticated and user-friendly client-server web application namely LC-Pred, this study highlights the early prediction of liver cirrhosis (LC), which will be beneficial for both the clinicians and the LC patients. LC-Pred delivers both single and bulk prediction abilities making it fast, sturdy, and secure to be used in healthcare sectors. Methods: This study uses a total of 1098 real-time patients' data having both LC and nonliver cirrhosis (NLC) cases to implement and compare traditional scoring systems and AI-based models, by using 20 clinical parameters with two demographic data such as patients' age and gender. Results: The tool consists of authentication, PDF report generation and spontaneous interface elevated for clinical workflow incorporation. Transformer model exhibits the highest accuracy among all the traditional and artificial intelligence (AI) models and is selected to be linked with LC-Pred for classification of cirrhosis. Transformer model accomplishes a vigorous performance with precision recall area under the curve (PR-AUC) of 0.907, receiver operating characteristics area under the curve (ROC-AUC) of 0.989, sensitivity/recall of 0.857 (for LC detection) and specificity of 0.947 (for NLC prediction), Brier score is of 0.027 and test accuracy of 0.977. Conclusion: The tool exhibits noteworthy upgradation in AI-assisted hepatology, over traditional scoring techniques like model for end-stage liver disease (MELD) and Child-Pugh, by stabilizing the technical intricacy with clinical efficacy. This note delineates the application framework, model training principles, evaluating results and the importance of implementing an aligned AI system to be utilized by the clinicians.