Methods and systems for patient monitoring

A method for a computing device, comprising:

• receiving a plurality of electronic signals from a plurality of sensors, each electronic signal of the plurality of electronic signals corresponding to physiological measurements of a patient;
• predicting, for each electronic signal with a corresponding predictor cloud of a plurality of predictor clouds, an expected value for the electronic signal, each predictor cloud of the plurality of predictor clouds corresponding to a respective, different, individual sensor of the plurality of sensors, each predictor cloud comprising a long short-term memory neural network and a recurrent neural network configured in series, such that all output of the long short-term memory neural network comprises all input to the recurrent neural network, and stored in non-transitory memory of the computing device;
• concatenating the expected value for each electronic signal into a combined prediction via concatenation in a string or chain, the combined prediction comprising expected values for the plurality of electronic signals;
• determining, with a multilayer perceptron, whether to generate an alarm based on the combined prediction;
• and
• generating the alarm responsive to output of the multilayer perceptron indicating to generate the alarm.

A system, comprising:

• a plurality of patient monitoring devices, each patient monitoring device of the plurality of patient monitoring devices configured to generate an electronic signal representing a measurement of a physiological parameter of a patient;
• and
• a processor communicatively coupled to the plurality of patient monitoring devices and an alarm device, the processor configured with instructions in non-transitory memory that, when executed, cause the processor to:
• input each electronic signal received from each of the plurality of patient monitoring devices into a corresponding predictor cloud of a plurality of predictor clouds stored in the non-transitory memory and generate an expected value, each predictor cloud corresponding to a different individual patient monitoring device of the plurality of patient monitoring devices, each predictor cloud comprising a long short-term memory neural network and a recurrent neural network configured in series such that all output of the long short-term memory neural network is input to the recurrent neural network;
• concatenate the expected value for each electronic signal into a combined prediction via concatenation in a string or chain, the combined prediction comprising expected values for the plurality of patient monitoring devices;
• input the combined prediction into a multilayer perceptron trained to determine if an alarm should be generated based on the combined prediction;
• evaluate the combined prediction with the multilayer perceptron to determine whether the alarm is false, wherein the alarm is false if the combined prediction is less than a threshold;
• and
• responsive to the alarm generated based on at least one of the electronic signals, suppress the alarm responsive to outputs of the plurality of predictor clouds indicating the alarm is false.