Surgical computing system with intermediate model support
A surgical computing device may include a processor configured to implement two neural networks, a primary neural network trained with a procedure focus and support neural network trained with a patient focus. Data indicative of a surgical patient, a target procedure, and a proposed procedure plan may be input to the support neural network. The support neural network may generate a patient specific mapping from this data. The patient specific mapping and the data indicative of a surgical patient, a target procedure, and a proposed procedure plan may be input to the primary neural network. The primary neural network may output a modified procedure plan that is different from the proposed procedure plan. In an example, the support neural network is trained to isolate anatomical elements. And the primary neural network is trained to identify procedure plans associated with improved patient outcomes. Such a two-network approach may facilitate the use of diverse training data sets to better refine procedure plans for a variety of uses.
1 . A device comprising:
a processor configured to:
receive first data indicative of a surgical patient, a target procedure, and a proposed procedure plan;
generate a patient specific mapping from the first data via a first neural network trained independent of the target procedure, wherein the patient specific mapping indicates geometry data extracted by the first neural network;
process, via a second neural network, the first data and the patient specific mapping that indicates the geometry data, wherein the second neural network is trained with data associated with the target procedure;
obtain, from the second neural network, a modified procedure plan that is different from the proposed procedure plan; and
send the proposed procedure plan and the modified procedure plan to a surgical support system.
2 . The device of claim 1 , wherein the first neural network is further trained to isolate anatomical elements.
3 . The device of claim 1 , wherein the processor configured to process, via the second neural network, the first data and the patient specific mapping further comprises the processor being further configured to:
determine, via the second neural network, a correlation between the surgical patient and the target procedure based on the first data; and
generate the modified procedure plan from the correlation via the second neural network.
4 . The device of claim 3 , wherein the proposed procedure plan and the modified procedure plan are sent to the surgical support system to cause the proposed procedure plan and the modified procedure plan to be displayed to a user.
5 . The device of claim 1 , wherein the first data comprises imaging data, procedure data, and patient-specific health data.
6 . The device of claim 1 , wherein the patient specific mapping comprises a data element indicative of patient anatomy targeted by the target procedure.
7 . The device of claim 1 , wherein the modified procedure plan reduces a complication rate associated with the target procedure.
8 . The device of claim 1 , wherein the patient specific mapping further indicates a registration based on a plurality of anatomical landmarks.
9 . The device of claim 1 , wherein the modified procedure plan indicates a port location associated with the geometry data, and wherein the proposed procedure plan does not indicate the port location.
10 . The device of claim 1 , wherein the modified procedure plan indicates a surgical instrument associated with the geometry data, and wherein the proposed procedure plan does not indicate the surgical instrument.
11 . The device of claim 1 , wherein the modified procedure plan indicates a tumor margin associated with the geometry data, and wherein the proposed procedure plan does not indicate the tumor margin.
12 . The device of claim 1 , wherein the modified procedure plan indicates a mobilization approach associated with the geometry data, and wherein the proposed procedure plan does not indicate the mobilization approach.
13 . The device of claim 1 , wherein data used to the train the first neural network is different from that used to the train the second neural network.
14 . The device of claim 1 , wherein an output of the first neural network corresponds to a data element used in the training of the second neural network.
15 . A method comprising:
receiving first data indicative of a surgical patient, a target procedure, and a proposed procedure plan;
generating a patient specific mapping from the first data via a first neural network trained independent of the target procedure, wherein the patient specific mapping indicates geometry data extracted by the first neural network;
processing, via a second neural network, the first data and the patient specific mapping that indicates the geometry data, wherein the second neural network is trained with data associated with the target procedure;
obtain, from the second neural network, a modified procedure plan that is different from the proposed procedure plan; and
sending the proposed procedure plan and the modified procedure plan to a surgical support system.
16 . The method of claim 15 , wherein the first neural network and the second neural network are independently trained.
17 . The method of claim 15 , wherein the method further comprises:
determining, via the second neural network, a correlation between the surgical patient and the target procedure based on the first data; and
generating the modified procedure plan from the correlation via the second neural network.
18 . The method of claim 15 , wherein an output of the first neural network corresponds to a data element used in the training of the second neural network.
19 . A method comprising:
training a first neural network independent of a target procedure;
training a second neural network with data associated with the target procedure;
inputting first data indicative of a surgical patient, the target procedure, and a proposed procedure plan to the first neural network to generate a patient specific mapping, wherein the patient specific mapping indicates geometry data extracted by the first neural network;
inputting the first data and the patient specific mapping that indicates the geometry data to the second neural network;
obtaining, from the second neural network, a modified procedure plan that is different from the proposed procedure plan; and
sending the proposed procedure plan and the modified procedure plan to a surgical support system.
20 . The method of claim 19 , wherein the method further comprises:
determining, via the second neural network, a correlation between the surgical patient and the target procedure based on the first data; and
generating the modified procedure plan from the correlation via the second neural network.