Researchers at MIT and the medical professionals at the Massachusetts General Hospital (MGH) predict that in the near future, artificial intelligence (AI) will be able to assist anesthesiologists in the operating room.
The team – made up of neuroscientists, engineers, and physicians have demonstrated a machine learning (ML) algorithm that has been able to automate propofol dosage. The algorithm has been trained using deep reinforcement learning and in many trials runs; so far it has exceeded the performances of traditional software in sophisticated, physiology-based simulations of patients using an application of deep reinforcement learning.
The software has also matched a human anesthesiologists’ performance in the experiments where it was given data from a few procedures.
The software has two neural networks – actor and a critic – which have learned the necessary science of keeping patients unconscious while also critiquing the validity of their actions. This has allowed the computers’ ability to maintain a constant and safe dosage to keep patients unconscious while procedures are going on. This means that the anesthesiologists can be free to perform responsibilities in the operating room, such as ensuring patients remain immobile, experience no pain, remain stable, and receive adequate oxygen.
The ‘actor and ‘critic’ neural networks work simultaneously – the actor decides how much drugs to dose while the critic guides and questions the actor’s action to help it improve and maximizes “rewards” specified by the programmer. For the research, the algorithm has three different settings – one that penalized just overdose, one that questioned supplying any dose, and one that had no penalties.
The critic questioned the actor’s every dose and chastised him to limit the doses to the bare minimum necessary under the “dose penalty” approach, which was the most efficient approach. The “no penalties” approach saw the actor dose too much, while the “overdose penalty” saw the actor does too little.
Further, the algorithm was tested on real patient consciousness data, collected from operating rooms which revealed its challenges and strengths. “During most tests, the algorithm’s dosing choices closely matched those of the attending anesthesiologists after unconsciousness had been induced and before it was no longer necessary. The algorithm, however, adjusted dosing as frequently as every five seconds, while the anesthesiologists (who all had plenty of other things to do) typically did so only every 20-30 minutes,” observed co-lead author, Marcus Badgeley.
To the researchers own admission, the algorithm is not yet optimised for inducing unconsciousness therefore, the anesthesiologist has to maintain the process of inducing unconsciousness and helping patients regain consciousness back. Another challenge that co-led author Gabe Schamberg reveals is that the accuracy of patients’ data on unconsciousness has to be extremely accurate.
To overcome this challenge, MIT researchers and MGH are working towards improving data interpretation such as brain wave signals, to improve the quality of patient monitoring data under anaesthesia.
Source: indiaai.gov.in
