NCT06961240

THE ROLE OF ARTIFICIAL INTELLIGENCE TRAINED WITH PRE-MEASURED NUMERICAL DATA IN PREDICTION OF DIFFICULT INTUBATION

• Status: Completed
• Phase: —
• Study type: Observational
• Enrollment: 250 (actual)
• Sponsor: Kutahya Health Sciences University · Academic / Other
• Sex: All
• Age: 18 Years
• Healthy volunteers: Not accepted

Summary

This study aims to develop an artificial intelligence (AI)-based model to predict difficult intubation in patients undergoing general anesthesia. Since patients are apneic during intubation without spontaneous breathing efforts, minimizing apnea duration is critical. Traditional methods for predicting difficult intubation rely on physical markers such as sternomental distance, thyromental distance, mouth opening, neck extension, Mallampati score, neck circumference, and upper lip bite test. However, performing these assessments quickly and objectively in every patient is challenging. Therefore, utilizing computer-assisted imaging systems and AI techniques may facilitate clinical practice. In this study, 250 patients presenting to the anesthesia outpatient clinic, who provide informed consent, will be evaluated. Demographic data (age, gender, height, weight, body mass index) will be recorded. Measurements including mouth opening, thyromental distance, sternomental distance, and neck circumference will be performed. Additionally, Mallampati score, neck extension ability, and upper lip bite test results will be noted. Portrait photographs capturing shoulder and upper body anatomy from multiple angles will be taken. During the operation, the Cormack-Lehane score observed by anesthesiologists with at least three years of experience during intubation will also be recorded. The collected data will consist of both tabular (structured) data and visual data. Data preprocessing will involve cleaning missing and outlier values, encoding categorical variables, and normalizing continuous variables. Key anatomical points (e.g., chin tip, thyroid notch, sternum) will be identified using landmark detection algorithms on the images. Of the dataset, 200 patients will be used for model training and 50 patients for testing. Machine learning methods (Random Forest, Support Vector Machines, Gradient Boosting) and deep learning methods (Artificial Neural Networks, Convolutional Neural Networks) will be employed. Tabular and image data will first be modeled separately and then combined using ensemble methods. Model performance will be evaluated with metrics including accuracy, sensitivity, specificity, F1 score, and AUC-ROC. The models will be developed using Python programming language with libraries such as TensorFlow, Scikit-learn, and NumPy, supported by GPU-based computing. This study is unique in its aim to compare classical physical examination-based predictions with AI-based predictions, enhancing the accuracy of difficult intubation forecasts. Strengthening clinical decision-making processes and improving patient safety are among the primary goals. Inclusion Criteria: Patients aged 18 years and older Patients undergoing general anesthesia with endotracheal intubation Patients providing informed consent Exclusion Criteria: Patients under 18 years of age Pregnant patients Emergency surgery cases Patients with a history of facial surgeries that alter appearance Patients with prior head and neck surgeries Patients not receiving general anesthesia The results of this study aim to contribute to the development of a reliable, generalizable AI model for early prediction of difficult airways in clinical settings.

Conditions

• Artificial Intelligence (AI)
• Difficult Intubation

Interventions

Type	Name	Description
DIAGNOSTIC_TEST	AI-Based Difficult Airway Prediction Model	The "AI-Based Difficult Airway Prediction Model" is an artificial intelligence system designed to predict difficult intubation in patients undergoing general anesthesia. It combines clinical data (age, BMI, Mallampati score, mouth opening, thyromental distance, sternomental distance, neck circumference) and anatomical image data. Machine learning (Random Forests, SVM, Gradient Boosting) and deep learning (ANN, CNN) algorithms are used to classify airway difficulty. The model's predictions are compared with clinical assessments by anesthesiologists using Cormack-Lehane grading. The goal is to improve prediction accuracy, enhance airway management, and support clinical decision-making.

Timeline

• Start date: 2025-05-02
• Primary completion: 2025-10-30
• Completion: 2025-11-11
• First posted: 2025-05-07
• Last updated: 2025-11-19

Locations

1 site across 1 country: Turkey (Türkiye)