Machine learning for interconnected surgical theater architecture
Each of a plurality of edge computing devices are configured to receive data streams generated by at least one sensor forming part of a respective medical device (e.g., a sensor-equipped surgical tool, etc.) which, in turn, characterizes use of the respective medical device in relation to a particular patient. Each of the edge computing devices can execute at least one machine learning model which generates or from which model attributes are derived. The generated model attributes are anonymized using an anonymization technique such as k-anonymity. The anonymized generated model attributes are homomorphically encrypted and transmitted to a central server. Encrypted model attribute updates to at least one of the machine learning models are later received from the central server which results in the machine learning models executing on one or more of the edge computing devices to be updated based on the received encrypted model attribute updates.
1. A method comprising:
generating, by each of a plurality of edge computing devices, model attributes, each of the edge computing devices configured to receive one or more data streams generated by at least one sensor forming part of a respective medical device, the at least one sensor characterizing use of the respective medical device in relation to a particular patient, each of the edge computing devices executing at least one machine learning model;
anonymizing the generated model attributes using k-anonymity;
homomorphically encrypting the anonymized generated model attributes;
transmitting the anonymizing generated model attributes to a central server;
receiving encrypted model attribute updates to at least one of the machine learning models from the central server based on the transmitting; and
updating one or more of the machine learning models executing on one or more of the edge computing devices based on the received encrypted model attribute updates.
2. The method of claim 1 , wherein the model attributes comprise model hyperparameters.
3. The method of claim 2 , wherein the model hyperparameters comprise one or more of: neural network hidden layer size, number of convolution layers, batch normalization specifications, or activation layer type.
4. The method of claim 1 , wherein the model attributes comprise model information.
5. The method of claim 4 , wherein the machine learning model information can comprise one or more of: model name, data input and output characteristics, data dimensionality, training data size, testing data size, number of training iterations, learning rate, or optimization method.
6. The method of claim 1 , wherein the updating comprises: retraining the one or more of the machine learning models executing on one or more of the edge computing devices using the received encrypted updates.
7. The method of claim 6 , wherein the received updates comprise: encrypted model hyperparameters.
8. A method comprising:
receiving, by a cloud server from each of a plurality of edge computing devices, anonymized and encrypted model attributes, each of the edge computing devices configured to receive one or more data streams generated by at least one sensor forming part of a respective medical device, the at least one sensor characterizing use of the respective medical device in relation to a particular patient, each of the edge computing devices executing at least one machine learning model;
decrypting the received model attributes;
updating, by the cloud server, model attributes for one or more cloud-based machine learning models corresponding to one or more machine learning models being executed by the edge computing devices;
encrypting the updated model attributes; and
transmitting the encrypted updated model attributes to the edge computing devices executing at least one machine learning model having updated model attributes, the corresponding edge computing device decrypting the updated model attributes and updating the corresponding machine learning models based on the transmitted updated model attributes.
9. The method of claim 8 , wherein the model attributes comprise model hyperparameters.
10. The method of claim 9 , wherein the model hyperparameters comprise one or more of: neural network hidden layer size, number of convolution layers, batch normalization specifications, or activation layer type.
11. The method of claim 8 , wherein the model attributes comprise model information.
12. The method of claim 11 , wherein the machine learning model information can comprise one or more of: model name, data input and output characteristics, data dimensionality, training data size, testing data size, number of training iterations, learning rate, or optimization method.
13. The method of claim 8 , wherein the updating comprises: retraining the one or more of the machine learning models executing on one or more of the edge computing devices using the received encrypted updates.
14. The method of claim 13 , wherein the received updates comprise: encrypted model hyperparameters.
15. A system comprising:
a plurality of edge computing devices each configured to receive one or more data streams generated by at least one sensor forming part of a respective medical device, the at least one sensor characterizing use of the respective medical device in relation to a particular patient, each of the edge computing devices executing at least one machine learning model; and
a cloud-based computing system for training and updating the respective at least one machine model based on data received from the plurality of edge computing devices which has been anonymized and encrypted using homomorphic encryption prior to it being transmitted over a network by the edge computing devices, the cloud-based system sending updates over the network to the machine learning models.
16. The system of claim 15 further comprising:
a plurality of Internet of the Operating Theaters (IoT-OR) gateways providing communication interfaces between the edge computing devices and the cloud-based computing system to allow for the exchange of model attributes used for updating the respective machine learning models.
17. The system of claim 16 , wherein the model attributes comprise model hyperparameters.
18. The system of claim 17 , wherein the model hyperparameters comprise one or more of: neural network hidden layer size, number of convolution layers, batch normalization specifications, or activation layer type.
19. The system of claim 16 , wherein the IoT-OR gateways relay data from the edge computing devices to the cloud-based computing system.
20. The system of claim 16 , wherein the IoT-OR gateways process data from the edge computing devices prior to transmission to the cloud-based computing system.