Development of a Smart Clinical Decision Support System for Screening of Celiac Disease using AI

Abstract

In the era of Artificial Intelligence (AI) and Intelligent Computing, the revolution in the healthcare sector is underway. Studies have focused on the development of decision-support processes for healthcare professionals to enhance disease screening and diagnostics. This study proposes a decision support system for diagnosing celiac disease (CD) using primary data, a complex autoimmune disorder affecting millions worldwide. CD diagnosis is complicated by socioeconomic factors, healthcare disparities, and limited access to advanced facilities and diagnostic technologies. Conventional methods are cost-prohibitive and lack of awareness contributes to underdiagnoses or misdiagnoses in developing countries. The study focused on improving detection rates of CD by utilizing AI-based approaches. The study aimed to ensure that no cases of CD go undetected and minimize the risk of misdiagnosing celiac cases as non-celiac. The experimentation phase employed 5 automated classifiers available in Google Colab Notebook: decision trees, Bayesian classifier, XGBoost algorithm, support vector machine, and logistic regression. The assessment parameters considered encompassed accuracy, sensitivity, specificity, and the area under the ROC curve (AUC). These models were selected for their proven ability to handle both continuous and categorical data, including categorical dependent variables, within a classification task. Additionally, considering the limitations of previous applications of AI-based diagnostic methods, comprehensive data preprocessing and feature engineering techniques have been introduced including the application of Recursive Feature Elimination (RFE). Among the array of AI models examined the XGBoost classifier showed the highest accuracy of 97.0%, a sensitivity of 0.98, a false-negative ratio of 1 and an AUC of 0.91. The outcomes of the study are helpful in terms of a step towards the development of a smart clinical decision support system (CDSS). Future directions include market validation of the proposed process and transformation into a smart application for ease of adoption for the end users. The study's methodology, which encompasses primary data collection, robust preprocessing, and meticulous feature engineering, not only enhances predictive accuracy but also establishes a pioneering CD data repository. This repository, brimming with comprehensive patient information, is poised to reshape CD research. In essence, the study's detailed results xv underscore the transformative potential of AI-driven diagnostic approaches in tackling the complexities of celiac disease.