Computer implemented determination method and system
Methods for providing a computer implemented medical diagnosis are provided. In one aspect, a method includes receiving an input from a user comprising at least one symptom of the user, and providing the at least one symptom as an input to a medical model. The method also includes deriving estimates of the probability of the user having a disease from the discriminative model, inputting the estimates to the inference engine, performing approximate inference on the probabilistic graphical model to obtain a prediction of the probability that the user has that disease, and outputting the probability of the user having the disease for display by a display device.
1. A method for providing a computer implemented medical diagnosis, the method comprising:
receiving an input from a user comprising at least one symptom;
providing the at least one symptom as an input to a medical model comprising:
a probabilistic graphical model, wherein said probabilistic graphical model is a generative model comprising probability distributions and relationships between symptoms and diseases;
a discriminative model comprising a neural network pre-trained to approximate the probabilistic graphical model, wherein the neural network is configured to approximate any of a plurality of posterior marginal distributions given an arbitrary set of evidence; and
an inference engine configured to perform Bayesian inference via statistical sampling on said probabilistic graphical model, the inference engine comprising a graphical processing unit and said statistical sampling being performed using parallel processing,
performing Bayesian inference via statistical sampling on said probabilistic graphical model with said inference engine to obtain a prediction of the probability that the user has a disease comprising:
performing statistical sampling on said probabilistic graphical model comprising sampling, from one or more posterior marginal distributions, values for each of a plurality of nodes of the probabilistic graphical model, wherein the one or more posterior marginal distributions are derived using the neural network pre-trained to approximate the probabilistic graphical model; and
outputting the probability of the user having the disease for display by a display device.
2. The method of claim 1 , wherein the statistical sampling is implemented via tensor operations.
3. The method of claim 1 , wherein the statistical sampling is importance sampling comprising an importance sampling proposal distribution.
4. The method of claim 3 , wherein the current estimate of the posterior probability that a user has a disease given the symptoms provided by the user, is mixed with the importance sampling proposal distribution.
5. The method of claim 4 , wherein probabilities are clipped.
6. The method of claim 1 , wherein during training of the neural network some of the data of the samples has been masked to allow the neural network to produce data which is robust to the user providing incomplete information about their symptoms.
7. The method of claim 1 , wherein the inference engine is adapted to perform importance sampling over conditional marginals.
8. The method of claim 1 , wherein the neural network comprising a plurality of sub-networks that can approximate the outputs of the probabilistic graphical model.
9. The method of claim 1 , wherein the neural network is a single neural network that can approximate the outputs of the probabilistic graphical model.
10. The method of claim 1 , wherein the probabilistic graphical model is a noisy-OR model.
11. The method of claim 1 , wherein determining the probability that the user has one or more diseases further comprises determining whether further information from the user would improve the diagnosis and requesting further information.
12. The method of claim 1 , wherein the medical model receives information concerning the symptoms of the user and risk factors of the user.
13. A non-transitory carrier medium comprising computer readable code configured to cause a computer to perform a method for providing a computer implemented medical diagnosis, the method comprising:
receiving an input from a user comprising at least one symptom;
providing the at least one symptom as an input to a medical model comprising:
a probabilistic graphical model, wherein said probabilistic graphical model is a generative model comprising probability distributions and relationships between symptoms and diseases;
a discriminative model comprising a neural network pre-trained to approximate the probabilistic graphical model, wherein the neural network is configured to approximate any of a plurality of posterior marginal distributions given an arbitrary set of evidence; and
an inference engine configured to perform Bayesian inference via statistical sampling on said probabilistic graphical model, the inference engine comprising a graphical processing unit and said statistical sampling being performed using parallel processing,
performing Bayesian inference via statistical sampling on said probabilistic graphical model with said inference engine to obtain a prediction of the probability that the user has a disease comprising:
performing statistical sampling on said probabilistic graphical model comprising sampling, from one or more posterior marginal distributions, values for each of a plurality of nodes of the probabilistic graphical model, wherein the one or more posterior marginal distributions are derived using the neural network pre-trained to approximate the probabilistic graphical model; and
outputting the probability of the user having the disease for display by a display device.
14. A system for providing a computer implemented medical diagnosis, the system configured to:
receive an input from a user comprising at least one symptom;
provide the at least one symptom as an input to a medical model comprising:
a probabilistic graphical model, wherein said probabilistic graphical model is a generative model comprising probability distributions and relationships between symptoms and diseases;
a discriminative model comprising a neural network pre-trained to approximate the probabilistic graphical model, wherein the neural network is configured to approximate any of a plurality of posterior marginal distributions given an arbitrary set of evidence; and
an inference engine configured to perform Bayesian inference via statistical sampling on said probabilistic graphical model, the inference engine comprising a graphical processing unit and said statistical sampling being performed using parallel processing,
perform approximate inference on the probabilistic graphical model with said inference engine to obtain a prediction of the probability that the user has a disease comprising:
performing statistical sampling on said probabilistic graphical model comprising sampling, from one or more posterior marginal distributions, values for each of a plurality of nodes of the probabilistic graphical model, wherein the one or more posterior marginal distributions are derived using the neural network pre-trained to approximate the probabilistic graphical model; and
output the probability of the user having the disease for display by a display device.
15. The method of claim 1 , wherein the probabilistic graphical model comprises at least 96 nodes.
16. The method of claim 1 , wherein the probabilistic graphical model comprises at least 768 nodes.
17. The method of claim 1 , wherein the probabilistic graphical model comprises a medical knowledge graph comprising at least 1000 nodes.