ELECTRICA
Original Articles

Evaluating the Spectrum of AI: From Deep Learning to Traditional Models in Identifying Celiac Disease Marsh Levels

1.

Department of Biomedical Engineering, İstanbul-Cerrahpaşa University, Faculty of Engineering, İstanbul, Türkiye

2.

Department of Infomatics, İstanbul University, İstanbul, Türkiye

3.

Institute of Cyber Security for Society (iCSS) & School of Computing, University of Kent, Canterbury, UK

4.

Department of Gastroenterology, Health Sciences University, Van Education and Research Hospital, Van, Türkiye

ELECTRICA 2024; 24: 748-754
DOI: 10.5152/electrica.2024.24068
Read: 133 Downloads: 65 Published: 08 November 2024

Abstract
Celiac disease develops due to the consumption of gluten and presents symptoms similar to other disorders, causing a delay in diagnosis. If left untreated, celiac disease increases the likelihood of autoimmune conditions, neurological problems, and specific cancers such as lymphoma. This study aimed to create an improved and dependable classification system for predicting celiac Marsh levels which are crucial for diagnosing and treating celiac disease. Precise categorization of the severity levels of celiac disease can notably improve medical diagnosis and patient care. We employed various classification models, including a deep learning neural network using PyTorch and conventional classifiers like decision tree, random forest, gradient boosting, and naive Bayes, to identify celiac disease severity levels. Our dataset included 182 adults (132 females, 50 males) with clinical symptoms and blood test results, diagnosed across Marsh levels 0 to 4. Among these, 72 individuals were not adhering strictly to a gluten-free diet, and 106 were partially following dietary restrictions. PyTorch model achieved 80% accuracy in identifying Marsh levels, with precision, recall, and F1-score metrics of 0.81, 0.80, and 0.70, respectively. In contrast, the decision tree, random forest, and gradient boosting classifiers each achieved a perfect accuracy of 100%, with precision, recall, and F1-scores of 1.00. The naive Bayes classifier performed worse, with 55% accuracy and precision, recall, and F1-scores of 0.67, 0.44, and 0.53, respectively. Most models performed well in categorizing celiac disease severity using clinical features and blood tests. Our analysis highlights the most effective model for predicting Marsh levels, improving diagnostic precision and patient care. This study underscores the importance of data-driven methods in medical diagnoses to enhance decision-making and benefit patient outcomes.

Cite this article as: E. Keskin Bilgiç, İ. Zaim Gökbay and Y. Kayar, "Evaluating the spectrum of AI: from deep learning to traditional models in identifying celiac disease Marsh levels," Electrica, 24(3), 748-754, 2024.

Files
EISSN 2619-9831