Neural Networks Trained to Apply Anesthetics

Apparently an example of direct process control. Risk involved?

Neural Network Trained to Control Anesthetic Doses, Keep Patients Under During Surgery

ZDNet.  BY Charlie Osborne

Researchers from the Massachusetts Institute of Technology (MIT) and Massachusetts General Hospital trained a neural network to administer anesthetic during surgery. The team trained the network on patient data, and on models that track levels of consciousness during an operation and recommended drug doses. Using a simulated environment and virtual patients, the researchers conducted trials to improve the model over time via a "cross-entropy" methodology. The network outperforms the proportional-integral-derivative (PID) controllers currently utilized to determine the right levels of anesthetic to use, and to administer it. MIT's Gabriel Schamberg said, "Deep neural networks allow us to make a model with many continuous input data, so our method generated more coherent control policies than prior table-based policies."