Artificial intelligence may have potential role in airway management

The state of AI in airway control, the most recent advancements in this subject, and its potential therapeutic applications are all covered in a recently published narrative review. A subfield of computer science called artificial intelligence (AI) is concerned with using algorithms to replicate and enhance human knowledge. It has expanded dramatically in recent years, mostly as a result of advancements in computing power and statistics. Artificial Intelligence comprises machine learning (ML), deep learning (DL), expert systems, language processing, voice recognition, and picture identification. While DL makes use of neural networks such as recurrent neural networks (RNN), artificial neural networks (ANN), and convolutional neural networks (CNN), machine learning (ML) also incorporates algorithm training. Images are detected and anomalous patterns are identified using DL, and until the desired result is obtained, ANNs continue the process after adapting to it. While RNNs are utilised for sequence prediction, robot control, voice recognition, natural language processing, and brain-computer interface, CNNs are employed for computer vision and classification applications.

AI in anaesthesia seeks to enhance the quality of anaesthetic treatment across the whole perioperative care cycle, maximise resource efficiency, and protect patient safety. Inadequate airway care may nevertheless result in potentially fatal consequences even in the presence of explicit standards and methods. It is possible to outperform current approaches in airway control by becoming proficient in AI for prediction and planning.

AI may aid with better knowledge, prevention, and control of issues associated to the airways, which makes it especially helpful in challenging airway evaluation and management. Artificial intelligence (AI) has been investigated to predict difficult intubation by incorporating subjective elements including facial appearance, voice traits, habitus, and other little-known aspects. Current approaches for evaluating difficult airways have weak to moderate discriminative ability.

Researchers have created a number of methods, such as video-assisted laryngoscopes (VLs) and facial and voice feature analysis, to anticipate problematic intubations. The CNN model by Hayasaka et al., the semi-supervised deep learning technique by Tavolara et al., the non-invasive automated face-analysis system by Cuendet et al., and the gradient boosting approach by Zhou and colleagues are a few of the most promising AI models.

The study of speech characteristics has shown potential in forecasting challenging airways. Five phonetic properties of vowels and formants were used to create an evaluation model that accurately led the prediction of problematic airways. To validate the prospective function of speech technology in forecasting problematic airways, however, multicentric research involving varied populations are required.

Video laryngoscopes (VLs) have been used to reduce the amount of time needed for tracheal intubation and to enhance the image of the glottic aperture. Nonetheless, there are two possible areas where the VL approach might be enhanced: Depth perception is impaired by two-dimensional (2D) screen visualisation, which may sometimes result in unsuccessful or delayed intubations. Furthermore, those executing endotracheal intubation must be able to use many instruments at once.

In an effort to improve upper airway structure visualisation with real-time input, researchers have attempted to integrate AI-based algorithms into the VLs that are now in use. A recent research used VL pictures obtained during emergency intubations to create an AI model for segmenting oral cavity features. Excellent accuracy and high specificity were shown by the CNN-based Configured Mask R system for the vocal cords, epiglottis, and cricoid cartilage.

It is essential to do more research on VL devices and the related deep-learning advancements. Future studies should concentrate on enhancing the design of airway structures by three-dimensional visualisation and a real-time feedback system that uses an AI interface. While there aren't any studies using AI to support and direct flexible fiberoptic intubations at the moment, this seems like an intriguing field to research.

The first robotic-aided endotracheal intubation was carried out on a mannequin. Robotic technology has been used in airway management. RRAIS, IntuBot, and the Kepler intubation robot system (KIS) are among further innovations. To improve out-of-hospital intubations, Wang X and colleagues created the compact, adaptable, and portable robotic-assisted remote intubation system (RRAIS). The Robotic Endoscope Automated Via Laryngeal Imaging for Tracheal Intubation (REALITI) system, which combines a video endoscope with a joystick and has both manual and automatic control modes for directing the endoscope tip into the glottis, was the subject of a proof-of-concept study by Biro and colleagues.

AI has also been used to the control of medication administration. Closed-loop anaesthesia delivery systems employ a bispectral index to regulate drug distribution in real-time. In comparison to manually operated systems, these systems function by continually monitoring the patient's physiological reactions, autonomously regulating the level of anaesthesia, and more precisely optimising medication doses. By incorporating additional control factors from huge patient datasets, AI-driven closed-loop control algorithms may improve the accuracy of anaesthetic administration.

By combining artificial intelligence (AI) with ultrasound- or CT-guided navigation devices, it may be possible to provide accurate endotracheal tube placement and real-time assistance with airway anatomy during endotracheal intubation. Real-time clinical data that integrates automation and AI algorithms might help identify airway obstacles and perhaps prevent unanticipated issues like oesophageal intubation or airway-related injuries. Because robotic systems may provide immediate feedback on processes carried out, they may make it possible to conduct real-time video conferences and tele-mentoring at faraway sites. To assess the effectiveness of AI-based solutions in airway control, more multicentric rigorous clinical trials and comparison studies are needed.

Even while AI-based research on airway management has made significant strides, there is still no practical use for it because of high costs, possible bias, and ethical and legal concerns. The following elements must be taken into account in order to overcome these obstacles: fully informed permission to apply data; data privacy and security; bias-free algorithms; and an AI interface that is easy to use.

Reference –

Naik, Naveen B.; Mathew, Preethy J.; Kundra, Pankaj1. Scope of artificial intelligence in airway management. Indian Journal of Anaesthesia 68(1):p 105-110, January 2024. | DOI: 10.4103/ija.ija_1228_23