Alexander JC, and Joshi GP.
Journal Club Summary
This paper is focused on anaesthesiology, automation and artificial intelligence. The paper stated that previously there were difficulties incorporating automation and artificial intelligence into anaesthesia. The main reasons for the failure of AI incorporation stemmed from the complexity associated with anaesthetic practice and the lack of well defined ruled based algorithms.
Furthermore, the key differences between artificial intelligence and machine learning was explored. Artificial intelligence is divided into weak AI and general AI and machine learning involves machines evolving and adapting their algorithms to improve. Machine learning is the key focus of implementation of AI into clinical practice, with the vision of guideline incorporation to improve future clinical care.
The current uses of artificial intelligence in medicine was explored, with GE healthcare utilizing it currently with NVIDIA to analyze complex radiological images. An AI algorithm was used to detect liver and kidney lesions in order to lower radiation doses and improve examination times.
The first use of AI was in the early 1950s, when Bickford described the utilization of EEG to monitor anaesthetic depth.
The incorporation of AI into anaesthetic practice revolves around the triad of anaesthesia/hypnosis, analgesia and muscle relaxation. Utilizing a closed-loop feedback system for any of these parameters will allow the incorporation of AI or semi-automation into future anaesthetic practice.
An example of this in current clinical practice, is “McSleepy”, which was used in America to deliver a semi-autonomous anaesthetic. A computer, with a specified software programme, with the key algorithms connected to infusion pumps was used to deliver the anaesthesia. Another example of this was “Sedasys”, which was developed by Johnson and Johnson and had FDA approval, but was later discontinued due to lack of commercial potential.
The main advantages of AI are that it will result in lower doses of delivered anaesthetic agent, reduced recovery time, improved documentation and attain tighter control within a specific range of target variables.
The main problems with AI implementation will be that each algorithm will need careful planning of sometimes semi-complex tasks with strict monitoring and security control to prevent abuse or patient harm. Furthermore, AI will be unable to undertake complex tasks that require a lot of manual dexterity. Also, it will require current generation anaesthesia incorporation, by current practitioners to shape the future of AI incorporation.
In conclusion, artificial intelligence can be used to assist rather than replace anaesthetic care. AI forming a symbiotic relationship with the anesthetist will ensure improved efficiency and better clinical care.
Summary by Dr C Velho. Presented at Journal Club Meeting 08 November 2018.
Feature image from http://www.flickr.com/photos/gleonhard/33661760430