Neural network prediction of difficult tracheal intubation risk by using the patient’s face image

Abstract

Background. The prognosis of the difficult tracheal intubation remains an essential problem. The effectiveness of using predictors does not allow to foreseen such situation accurately.
 The purpose of the study was to develop a predictive system and evaluate its effectiveness in difficult tracheal intubation based on facial image analysis combined with the most significant predictors of difficult intubation.
 Materials and methods. A database based on the registration of difficult intubation predictors was developed. It was based on the patients face images with marked reference points. It allowed to estimate the information signs associated with the difficult tracheal intubation. The degree of intubation severity was determined directly during the intubation process according to the proposed original scale of severity.
 Results. The classifier was synthesized by using the self-organization neural network method. The trained neural network was the basis of the classifier model implemented as a computer application. The sensitivity of the difficult tracheal intubation prognosis was 90.90%, specificity was 97.02%, the prognostic value of the positive result was 58.82%, the negative one was 99.56%.
 Conclusions. The proposed decision support system allows patients to be stratified into groups according to the degree of difficult tracheal intubation risk. In addition, the self-learning process of the system continues as the new data become available. This allows to improve its efficiency continuously.