SYSTEMS AND METHODS FOR CREATING AN EXPERT-TRAINED DATA MODEL
Presented are systems and methods for using expert knowledge to generate, train, and use a medical data model that uses medical data from a number of sources to generate likelihoods that a given set of symptoms is caused or related to one or more illnesses. Various embodiments accomplish this by parsing medical and non-medical data into keywords and target words to learn, e.g., based on a characteristic of the parsed words, an association between keywords and target words. Based on the learned associations, likelihood scores are then generated that represent, for example, a relationship between a set of symptoms and an illness, a treatment, and an outcome.
1 . A method for training a medical data model, the method comprising:
receiving medical data comprising sentences from one or more sources;
parsing the sentences to generate parsed words that comprise keywords and target words;
based on a characteristic of the parsed words, learning an association between a keyword and a first target word;
generating a score indicative of the association;
based on the score, generating a likelihood score that is representative of a relationship between, at least, the first target word and a second target word; and
outputting a result that is representative of the likelihood score.
2 . The method according to claim 1 , wherein the keyword comprises a modifier, and at least one of the first target word and the second target word comprises at least one of a symptom, an illness, a treatment, and an outcome.
3 . The method according to claim 2 , wherein the modifier comprises at least one of a qualifier and a quantifier.
4 . The method according to claim 1 , wherein outputting the result further comprises, based on the likelihood score, eliminating one or more potential illnesses.
5 . The method according to claim 1 , wherein the relationship between the first target word and the second target word is established in response to the second target word being selected from a list of potential illnesses.
6 . The method according to claim 1 , wherein the relationship between the first target word and the second target word is established in response to the first target word being selected from a list of potential symptoms.
7 . The method according to claim 6 , further comprising:
for the second target word that represents an illness, displaying on a monitor a first image that represents an area of a body;
in response to the area being selected, displaying the list of potential symptoms, the potential symptoms being related to the area of the body;
in response to a symptom being selected, displaying an indicator and a measure related to the indicator; and
and associating the symptom with the second target word to generate an association that indicates that the indicator is a factor in diagnosing the illness.
8 . The method according to claim 7 , further comprising inputting the association into a data model, the association enabling an identification of the illness based on the symptom.
9 . The method according to claim 7 , wherein displaying the first image comprises displaying a second image that comprises greater detail about the first area than the first image.
10 . A method for using a medical data model to make medical predictions, the method comprising:
inputting one or more keywords and a first set of target words into a model, the model having been trained to associate the one or more keywords and at least the first set of target words to make a prediction related to a second set of target words;
obtaining the prediction from the model; and
using the prediction to output a result.
11 . The method according to claim 10 , wherein the one or more keywords comprise a modifier, and at least one of the first set of target words and the second set of target words comprises at least one of a symptom, an illness, a treatment, and an outcome.
12 . The method according to claim 10 , wherein the result comprises an identification of an illness based on a set of symptoms.
13 . The method according to claim 12 , wherein making the prediction comprises, based on a likelihood score, eliminating one or more potential illnesses.
14 . The method according to claim 10 , wherein the one or more keywords comprise an indicator.
15 . The method according to claim 14 , wherein the indicator comprises at least one of a pain descriptor and a negative indicator.
16 . The method according to claim 14 , wherein the indicator comprises information related to at least one of an immunization, an allergy, a travel risk, an alcohol use, an occupational risk, a diet, a pet risk, a food source, a physical condition, and a neurological condition.
17 . The method according to claim 14 , wherein the indicator comprises past patient medical data.
18 . The method according to claim 10 , wherein the one or more keywords comprise a measure.
19 . The method according to claim 18 , wherein the measure comprises timing information.
20 . The method according to claim 18 , wherein at least one of the result and the measure comprise at least one of weight data, a range, an option, a frequency, a percentage, and a likelihood.