Neural network for recommendation of shoulder surgery type
A computing system generates a plurality of training datasets from past shoulder surgery cases and uses the plurality of training datasets to train a neural network. Each output layer neuron in a plurality of output layer neurons of the neural network corresponds to a different class in one or more shoulder pathology classification systems. The computing system may obtain a current input vector that corresponds to a current patient. Additionally, the computing system may apply the neural network to the current input vector to generate a current output vector. The computing system may determine, based on the current output vector, a recommended type of shoulder surgery for the current patient.
1. A computer-implemented method comprising:
generating, by a computing system, a plurality of training datasets, wherein:
a neural network (NN) has an input layer, an output layer, and one or more hidden layers between the input layer and the output layer,
the input layer includes a plurality of input layer neurons, each input layer neuron in the plurality of input layer neurons corresponding to a different input element in a plurality of input elements,
the output layer includes a plurality of output layer neurons, each output layer neuron in the plurality of output layer neurons corresponding to a different output element in a plurality of output elements, the plurality of output elements including a plurality of surgery type output elements, each surgery type output element in the plurality of surgery type output elements corresponding to a different type of shoulder surgery in a plurality of types of shoulder surgery, wherein the plurality of types of shoulder surgery includes a standard total shoulder arthroplasty and a reverse shoulder arthroplasty;
each respective training dataset corresponds to a different training data patient in a plurality of training data patients and comprises a respective training input vector and a respective target output vector,
for each respective training dataset, the training input vector of the respective training dataset comprises a value for each element of the plurality of input elements,
for each respective training dataset, the target output vector of the respective training dataset comprises a value for each element of the plurality of output elements;
using, by the computing system, the plurality of training datasets to train the NN;
obtaining, by the computing system, a current input vector that corresponds to a current patient;
applying, by the computing system, the NN to the current input vector to generate a current output vector; and
determining, by the computing system, based on the current output vector, a recommended type of shoulder surgery for the current patient, wherein the recommended type of shoulder surgery is one of the plurality of types of shoulder surgery including the standard total shoulder arthroplasty and the reverse shoulder arthroplasty.
2. The computer-implemented method of claim 1 , wherein:
a confidence value function outputs confidence values, wherein all possible confidence values output by the confidence value function are within a predefined range,
applying the NN to the current input vector to generate the current output vector comprises, for each respective output layer neuron in the plurality of output layer neurons that corresponds to a surgery type output element in the plurality of surgery type output elements:
calculating, by the computing system, an output value of the respective output layer neuron; and
applying, by the computing system, the confidence value function with the output value of the respective output layer neuron as input to the confidence value function, wherein the confidence value function outputs a confidence value for the respective output layer neuron, and
for each respective output layer neuron in the plurality of output layer neurons that corresponds to a surgery type output element in the plurality of surgery type output elements:
the output element corresponding to the respective output layer neuron specifies the confidence value for the respective output layer neuron, and
the confidence value for the respective output layer neuron is a measure of confidence that the type of shoulder surgery corresponding to the surgery type output element that corresponds to the respective output layer neuron is optimal for the current patient.
3. The computer-implemented method of claim 2 , wherein the confidence value function is one of: a hyperbolic tangent function or a sigmoid function.
4. The computer-implemented method of claim 1 , wherein the plurality of types of shoulder surgery further includes a stemless standard total shoulder arthroplasty, a stemmed standard total shoulder arthroplasty, a stemless reverse shoulder arthroplasty, a stemmed reverse shoulder arthroplasty, an augmented glenoid standard total shoulder arthroplasty, and an augmented glenoid reverse shoulder arthroplasty.
5. The computer-implemented method of claim 1 , wherein:
the method further comprises:
obtaining, by the computing system, images of the current patient;
segmenting, by the computing system, the images to identify internal structures of the current patient; and
determining, by the computing system, based on relative positions of the identified internal structures of the current patient, a plurality of measurements, and
the plurality of input elements includes an input element for each measurement in the plurality of measurements.
6. The computer-implemented method of claim 1 , wherein the plurality of input elements includes measurements of morphological characteristics of the patient.
7. The computer-implemented method of claim 1 , wherein the plurality of input elements includes inputs elements specifying one or more of:
a glenoid version, a glenoid inclination, a glenoid orientation, a glenoid direction, a glenoid best fit sphere radius, a glenoid best fit sphere root mean squares error, a reverse shoulder angle, a critical shoulder angle, acromion humeral space, a glenoid humeral space, glenoid bone density scores for a humerus and a glenoid, humeral version, a humeral neck shaft angle, a humeral head best fit sphere radius and a root mean squares error of the humeral head best fit sphere radius, humeral subluxation, humeral orientation, humeral direction, a fatty infiltration measure of a rotator cuff, an atrophy measure of the rotator cuff, a patient age, patient gender, patient activities, or a patient body mass index.
8. The computer-implemented method of claim 1 , wherein the NN is a deep neural network.
9. A computing system comprising:
a storage system configured to store parameters of a deep neural network (NN) that has an input layer, an output layer, and one or more hidden layers between the input layer and the output layer; and
one or more processors configured to:
generate a plurality of training datasets, wherein:
the input layer includes a plurality of input layer neurons, each input layer neuron in the plurality of input layer neurons corresponding to a different input element in a plurality of input elements,
the output layer includes a plurality of output layer neurons, each output layer neuron in the plurality of output layer neurons corresponding to a different output element in a plurality of output elements, the plurality of output elements including a plurality of surgery type output elements, each surgery type output element in the plurality of surgery type output elements corresponding to a different type of shoulder surgery in a plurality of types of shoulder surgery,
wherein the plurality of types of shoulder surgery includes a standard total shoulder arthroplasty and a reverse shoulder arthroplasty;
each respective training dataset corresponds to a different training data patient in a plurality of training data patients and comprises a respective training input vector and a respective target output vector,
for each respective training dataset, the training input vector of the respective training dataset comprises a value for each element of the plurality of input elements,
for each respective training dataset, the target output vector of the respective training dataset comprises a value for each element of the plurality of output elements;
use the plurality of training datasets to train the NN;
obtain a current input vector that corresponds to a current patient;
apply the NN to the current input vector to generate a current output vector; and
determine, based on the current output vector, a recommended type of shoulder surgery for the current patient, wherein the recommended type of shoulder surgery is one of the plurality of types of shoulder surgery including the standard total shoulder arthroplasty and the reverse shoulder arthroplasty.
10. The computing system of claim 9 , wherein:
a confidence value function outputs confidence values, wherein all possible confidence values output by the confidence value function are within a predefined range,
applying the NN to the current input vector to generate the current output vector comprises, for each respective output layer neuron in the plurality of output layer neurons that corresponds to a surgery type output element in the plurality of surgery type output elements:
calculating, by the computing system, an output value of the respective output layer neuron; and
applying, by the computing system, the confidence value function with the output value of the respective output layer neuron as input to the confidence value function, wherein the confidence value function outputs a confidence value for the respective output layer neuron, and
for each respective output layer neuron in the plurality of output layer neurons that corresponds to a surgery type output element in the plurality of surgery type output elements:
the output element corresponding to the respective output layer neuron specifies the confidence value for the respective output layer neuron, and
the confidence value for the respective output layer neuron is a measure of confidence that the type of shoulder surgery corresponding to the surgery type output element that corresponds to the respective output layer neuron is optimal for the current patient.
11. The computing system of claim 10 , wherein the confidence value function is one of: a hyperbolic tangent function or a sigmoid function.
12. The computing system of claim 9 , wherein the plurality of types of shoulder surgery further includes a stemless standard total shoulder arthroplasty, a stemmed standard total shoulder arthroplasty, a stemless reverse shoulder arthroplasty, a stemmed reverse shoulder arthroplasty, an augmented glenoid standard total shoulder arthroplasty, and an augmented glenoid reverse shoulder arthroplasty.
13. The computing system of claim 9 , wherein:
the one or more processors are further configured to:
obtain images of the current patient;
segment the images to identify internal structures of the current patient; and
determine, based on relative positions of the identified internal structures of the current patient, a plurality of measurements, and
the plurality of input elements includes an input element for each measurement in the plurality of measurements.
14. The computing system of claim 9 , wherein the plurality of input elements includes measurements of morphological characteristics of the patient.
15. The computing system of claim 9 , wherein the plurality of input elements includes inputs elements specifying one or more of:
a glenoid version, a glenoid inclination, a glenoid orientation, a glenoid direction, a glenoid best fit sphere radius, a glenoid best fit sphere root mean squares error, a reverse shoulder angle, a critical shoulder angle, acromion humeral space, a glenoid humeral space, glenoid bone density scores for a humerus and a glenoid, humeral version, a humeral neck shaft angle, a humeral head best fit sphere radius and a root mean squares error of the humeral head best fit sphere radius, humeral subluxation, humeral orientation, humeral direction, a fatty infiltration measure of a rotator cuff, an atrophy measure of the rotator cuff, a patient age, patient gender, patient activities, or a patient body mass index.
16. The computing system of claim 9 , wherein the computing system comprises a mixed reality (MR) visualization device that includes the one or more processors.
17. A non-transitory computer-readable storage medium having instructions stored thereon that, when executed, configure a computing system to:
generate a plurality of training datasets, wherein:
a neural network (NN) has an input layer, an output layer, and one or more hidden layers between the input layer and the output layer,
the input layer includes a plurality of input layer neurons, each input layer neuron in the plurality of input layer neurons corresponding to a different input element in a plurality of input elements,
the output layer includes a plurality of output layer neurons, each output layer neuron in the plurality of output layer neurons corresponding to a different output element in a plurality of output elements, the plurality of output elements including a plurality of surgery type output elements, each surgery type output element in the plurality of surgery type output elements corresponding to a different type of shoulder surgery in a plurality of types of shoulder surgery, wherein the plurality of types of shoulder surgery includes a standard total shoulder arthroplasty and a reverse shoulder arthroplasty;
each respective training dataset corresponds to a different training data patient in a plurality of training data patients and comprises a respective training input vector and a respective target output vector,
for each respective training dataset, the training input vector of the respective training dataset comprises a value for each element of the plurality of input elements,
for each respective training dataset, the target output vector of the respective training dataset comprises a value for each element of the plurality of output elements;
use the plurality of training datasets to train the NN;
obtain a current input vector that corresponds to a current patient;
apply the NN to the current input vector to generate a current output vector; and
determine, based on the current output vector, a recommended type of shoulder surgery for the current patient, wherein the recommended type of shoulder surgery is one of the plurality of types of shoulder surgery including the standard total shoulder arthroplasty and the reverse shoulder arthroplasty.