Diagnostic Tools

Abstract

Disclosed herein are methods for diagnosing a medical condition in a patient. The method includes acquiring a plurality of physiological values ​​from a patient and executing a first model configured to determine a risk value for at least one of a plurality of medical conditions based on the physiological values, wherein executing the first model acquires a first risk value for the at least one medical condition based on a first one of the acquired physiological values ​​and weights the first risk value based on a second of the acquired physiological values ​​to determine a total risk value for the at least one medical condition for the patient.

Description

[Technical field]

[0001] The present disclosure relates to systems and methods for diagnosing one or more medical conditions, which employ machine learning techniques to identify risk for such medical conditions and determine appropriate diagnostic pathways. [Background technology]

[0002] Certain medical conditions, such as cancer, have proven extremely difficult to diagnose by traditional methods. To confirm such a diagnosis, there may not be a single symptom that directly alerts a clinician to a specific diagnosis or directs them down a specific path, such as testing or investigation. Instead, these conditions are a collection of hundreds of different diseases, each with their own signs, symptoms, and risk factors, which may be vague in the early stages and overlap with less serious conditions. Therefore, identifying the condition or the appropriate diagnostic path at an early stage when symptoms first appear, for example at a general practitioner appointment, may be particularly challenging.

[0003] Furthermore, the information that would help a practitioner make a decision or reach a diagnosis may come from many different sources: for example, given the multifaceted manifestations of a medical condition, it may not be possible using traditional methods for a clinician to obtain all the information necessary to make such a decision immediately.

[0004] In recent years, there has been an increase in running both cloud-based technologies and application software to address a variety of technical problems. Furthermore, with increased computing power and access to a growing amount of suitable training data, machine learning techniques are finding increasing application to provide improved solutions to many technical problems. In particular, the use of neural networks has become increasingly popular as a way to provide elegant and effective solutions to many such technical problems.

[0005] SUMMARY OF THE DISCLOSURE Embodiments of the present application may address some or all of the above-mentioned problems, as well as other technical problems. Summary of the Invention [Problem to be solved by the invention]

[0006] The embodiments and examples of the present disclosure aim to address at least some of the above mentioned technical and / or related problems. The embodiments of the present invention are as set out in the independent claims, with optional features as set out in the dependent claims. The embodiments and examples may provide a more efficient way of determining a patient's risk of developing a number of different medical conditions based on the range of specific physiological values ​​they exhibit, and may provide a more efficient way of providing recommended courses of action to diagnose the conditions. They may also provide a more efficient way of identifying conditions for which a patient is at risk, and may provide a more efficient way of providing recommended treatment pathways based on the individual requirements of a particular health authority or clinician. [Means for solving the problem]

[0007] In one embodiment, a method for diagnosing a medical condition of a patient is provided, comprising: acquiring a plurality of physiological values ​​from the patient; and executing a first model configured to determine a risk value for at least one of a plurality of medical conditions based on the physiological values, wherein executing the first model comprises: acquiring a first risk value for the at least one medical condition based on a first value of the acquired physiological values; and weighting the first risk value based on a second value of the acquired physiological values ​​to determine a total risk value for the at least one medical condition for the patient.

[0008] In one embodiment, a method is provided for diagnosing a medical condition of a patient comprising acquiring a plurality of patient factors from the patient and executing a first model configured to determine a risk value for one of a plurality of medical conditions based on the patient factors, wherein executing the first model comprises weighting a baseline risk value for the at least one medical condition based on the acquired patient factors to determine a risk value for the at least one patient medical condition.

[0009] In one embodiment, a method is provided for diagnosing a medical condition in a patient comprising obtaining a plurality of physiological values ​​from the patient and executing a first model configured to determine a risk value for one of a plurality of medical conditions based on the physiological values, wherein executing the first model comprises weighting the physiological values ​​and determining a risk value for the at least one medical condition based on the weighted physiological values.

[0010] Each risk value may indicate the likelihood that the patient has the corresponding medical condition.

[0011] Running the first model may comprise sequentially weighting the baseline risk value by each of a plurality of physiological values.

[0012] Executing the first model may comprise determining a risk associated with each of the physiological values, and may further combine each of the associated risks to provide a risk value for the at least one medical condition. Weighting the physiological values ​​may apply a weighting factor to each of the physiological values. For example, the weighting factor may be constant or may define a mathematical function. Determining the risk value for the at least one medical condition may combine the weighted physiological values. Applying the weighting factor may comprise obtaining the weighting factor and / or calculating the weighting factor based on, for example, data indicative of an association between a physiological parameter corresponding to the physiological value and the plurality of medical conditions, for example, the data may be obtained from a data store.

[0013] Applying weighting may modify the calculated risk value by using a second set of values, which may be, for example, additional patient information, such as patient demographic data.

[0014] The mapping data may be provided, e.g., acquired and stored by a processor, and used as the first model. The mapping data may indicate, for each of a plurality of medical conditions, a subset of physiological parameters relevant for the calculation of a risk value for that medical condition. The mapping data may alternatively or additionally provide a subset of medical conditions for each of the physiological parameters, e.g., may comprise an indication for each of the physiological parameters whether it is relevant for an assessment of risk for each of the selected medical conditions. The first model may map each of the acquired physiological values ​​to a medical condition for which they have an associated risk based on the mapping data. Thus, for a given set of physiological values, only one subset of medical conditions may be selected, and only risk values ​​for these conditions may be calculated, modified, and / or weighted.

[0015] The data may be obtained by a first model comprising an indication of a functional relationship between a percentage of risk of having one or more medical conditions and a physiological parameter corresponding to a physiological value, e.g., a risk value for one or more of the plurality of medical conditions may vary as a mathematical function of the physiological parameter. For example, there may be a proportional, exponential, or other mathematical relationship between the parameter and the risk value of the medical condition. The weighting data may comprise this indication.

[0016] The method may further comprise comparing each of the total risk values ​​to a selected threshold and, if one or more of the total risk values ​​exceeds a corresponding threshold, selecting at least one diagnostic pathway for diagnosing a corresponding medical condition.

[0017] Selecting the diagnostic pathway may include executing a second model configured to select at least one of a plurality of possible diagnostic pathways based on a plurality of sets of pathway parameter values, each set may be associated with one of the diagnostic pathways.

[0018] Executing the second model may comprise selecting the associated path if one of the path parameter values ​​meets a threshold.

[0019] The method may further comprise obtaining a patient dataset comprising a set of physiological values ​​and a set of indicators each identifying the presence or absence of a diagnosis of one of a plurality of medical conditions, and modifying the first model based on the patient dataset.

[0020] The method may further comprise obtaining a diagnostic pathway dataset for one of a plurality of diagnostic pathways, the pathway dataset comprising a plurality of pathway parameter values ​​for the diagnostic pathway, and modifying the second model based on the pathway dataset.

[0021] Modifying the second model may involve weighting each of the pathway parameters associated with that diagnostic pathway in the second model by the obtained corresponding pathway parameter value.

[0022] At least one of the first model and the second model may comprise a Bayesian model.

[0023] In one embodiment, a method is provided for diagnosing a medical condition of a patient, comprising obtaining a plurality of physiological values ​​from a patient, mapping the physiological values ​​to a plurality of associated medical conditions, determining a total risk value associated with each of the medical conditions based on the mapped physiological values, comparing each of the total risk values ​​to a threshold, and outputting diagnostic pathway data for diagnosing the medical condition if the total risk value exceeds the threshold.

[0024] Determining a total risk value associated with each of the medical conditions may comprise determining, for each physiological value, an indication of risk associated with each of the medical conditions to which it is mapped.

[0025] The method may further comprise obtaining risk data indicative of a predetermined association between the plurality of physiological parameters and the plurality of medical conditions, and wherein the mapping of the at least one physiological value to the plurality of medical conditions and determining a total risk value for each of the medical conditions are based on the risk data.

[0026] Outputting the diagnostic pathway data may select at least one of a plurality of possible diagnostic pathways.

[0027] The method may further comprise obtaining a plurality of pathway parameter values ​​for each of the possible diagnostic pathways, and selecting at least one pathway is based on the pathway parameter values.

[0028] Selecting at least one of the plurality of possible diagnostic pathways may comprise selecting a pathway if any of the parameter values ​​of the pathway meets a corresponding threshold value.

[0029] Determining the total risk value for the condition may be iterating over past patient data, which may comprise a plurality of data sets each relating to a patient, each of the data sets comprising a set of physiological values ​​and a set of indices each identifying the presence or absence of a diagnosis for one of the indices.

[0030] The method may further comprise revising the total risk value for at least one index based on historical patient data.

[0031] The method may further comprise modifying pathway parameter values ​​based on historical patient data.

[0032] The physiological values ​​may relate to one or more of several different classes of patient factors and parameters, including, but not limited to, signs and symptoms exhibited by the patient, investigation or test results, demographic data and risk factors (e.g., whether the patient smokes, is overweight, etc.).

[0033] In another embodiment, there is provided a method for training an artificial neural network for diagnosing a medical condition, comprising: executing as the neural network a first model configured to provide a relationship between a set of physiological parameters and a risk value for at least one of a plurality of medical conditions; outputting an indication of total risk of the at least one medical condition from acquired physiological values ​​corresponding to the physiological parameters; acquiring a patient dataset comprising the set of physiological values ​​and a set of indicators each identifying a presence or absence of a diagnosis for one of the plurality of medical conditions; and modifying the first model based on the patient dataset.

[0034] The method may further comprise obtaining a plurality of said patient data sets and iteratively refining the first model based on each of the patient data sets.

[0035] The patient dataset may further comprise a pathway dataset for one of the plurality of diagnostic pathways, the pathway dataset may comprise a plurality of pathway parameter values ​​for the pathway, and the method may be to execute a second model as a neural network, the second model being configured to a) provide an association between each of the plurality of pathway parameters and each of the plurality of diagnostic pathways, and b) select one of the diagnostic pathways based on the modified second model and the acquired physiological values, and to modify the second model based on the pathway dataset.

[0036] The method may further comprise iteratively refining the second model based on each of the patient data sets.

[0037] In one embodiment, there is provided a computer program product comprising program instructions configured to program a processor to perform any of the methods described.

[0038] The embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings. [Brief description of the drawings]

[0039] [Figure 1] FIG. 1 outlines one example of a process for providing a clinician with risk and diagnostic pathways for one or more medical conditions. [Diagram 2] FIG. 2 illustrates an example process for providing a diagnostic pathway for one or more medical conditions based on a patient's physiological values. [Diagram 3] FIG. 3 shows an example of a process for providing risk and diagnostic information to a clinician. [Figure 4] FIG. 4 illustrates an example process for using machine learning to provide a diagnostic pathway for one or more medical conditions. [Diagram 5] FIG. 5 shows another example of a process for providing risk and diagnostic information to clinicians using machine learning. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0040] 1 shows an overview of a process 100 used to select a diagnostic pathway for diagnosing a medical condition. Such a process allows a clinician to efficiently receive information about a patient's risk of having one or more medical conditions and a recommended pathway for diagnosing these conditions based on data obtained from the patient and the relevant rules and guidelines of the particular medical service to which the clinician belongs. Each of the described steps can be performed, for example, by a control device in a server connected via a wide area network (WAN) to multiple devices that can input the relevant data.

[0041] In this example, as a first step, data about the patient is obtained (101), for example by a processor. For example, as described in more detail below, the data can comprise patient factors, such as physiological values ​​associated with the patient. The data may be entered into a computing device, such as a PC, smartphone, or tablet, by a clinician examining the patient and transmitted to an external server. These data are then used to determine (102) one or more risk values, for example the percentage of risk of a number of conditions that the patient has, based on the collected values. If it is determined that the patient is at risk for one or more conditions, for example if the patient has a calculated risk higher than a selected risk threshold, then one or more appropriate diagnostic pathways are selected (103) to diagnose the condition. The information, for example the conditions for which the patient is at risk, and the selected diagnostic pathways are then provided to the clinician's device, for example to be displayed. Each of these steps is described in more detail below with reference to subsequent figures.

[0042] FIG. 2 illustrates an example of a process 200 executed by a controller of a computing device to determine both a patient's risk of having one or more particular medical conditions and the appropriate diagnostic pathway to diagnose such conditions. In this example, first, physiological values ​​of the patient are obtained (201). The physiological values ​​may relate to one or more of several different classes of patient factors and parameters, and may comprise, but are not limited to, signs and symptoms exhibited by the patient, investigation or test results, risk factors (e.g., whether the patient smokes, is overweight, etc.), and / or demographic data. These values ​​may be entered into the computing device by a clinician at a user interface of the device, for example, using an application. By way of example, the values ​​may be the presence or absence of a sign or symptom exhibited at a clinical appointment, the results of a particular laboratory test (blood, urine, etc.) indicating the presence or absence of a particular substance, or a numerical value indicating the concentration of a particular substance, positive or negative indicators for weight, height, BMI and age, indicators of the patient's medical history, such as diagnoses of previously diagnosed medical conditions, and other demographic data.

[0043] Some or all of this physiological value data may be recorded by a clinician and entered into the device at the time of the patient's consultation or appointment. This data may also have been previously recorded or may have been stored in a data store. The physiological values ​​entered at the clinician's device may be transmitted over such a network to a remote server, and the calculations described herein may be performed at the server. If the physiological values ​​are stored in a data store, such a store may be local to the server or may be accessible over a network, such as a WAN, such that a controller at the server may obtain and / or extract the physiological value data from the data store. These values ​​are then used to determine a risk value for at least one of the plurality of medical conditions. The plurality of medical conditions comprises a set of predefined conditions, and a risk value is obtained for each condition in the set. As an example, the plurality of medical conditions may correspond to several different types of cancer.

[0044] It is understood that a risk value may refer to the likelihood that a patient has a medical condition. For example, if a risk value for a medical condition is calculated to be 6% for a given combination of physiological values, then 6 out of 100 patients exhibiting that particular combination of physiological values ​​would be predicted to have the condition.

[0045] These risk values ​​may be calculated by a first model based on known and / or predicted associations between the parameters corresponding to the obtained values ​​and a number of medical conditions. For example, the first model may be comprised of decision tree logic, where each of the obtained physiological values ​​may be cross-referenced throughout the decision tree logic to calculate a risk value for each medical condition. The decision tree logic used may be specific to a particular healthcare provider (e.g., a Clinical Commissioning Group (CCG)) to which the user, e.g., a clinician, belongs. In particular, data may be obtained that indicates an association between parameters such as the above example and the likelihood of having a particular medical condition. This data may comprise statistical correlations based on one or more sets of stored data such as clinical guidelines. For example, the set of guidelines may indicate, for each physiological value, a set of risk percentages that indicate the likelihood of diagnosis of a number of medical conditions. For example, if a physiological value indicates only the presence of a particular symptom, there may be an increased risk in one or more medical conditions associated with that value. Alternatively, where the physiological value is a numerical value, there may be a functional correlation between the percentage of risk of having one or more medical conditions and the parameter to which the numerical value is associated, e.g., the risk value for a condition may vary as a mathematical function of the physiological parameter. For example, there may be a proportional, exponential, or other mathematical relationship between the parameter and the risk value for the medical condition. As an example, where the physiological value is related to a numerical physiological parameter, such as the concentration of a substance detected in a blood test, as the concentration increases, the patient's risk of having a particular medical condition may also increase according to a particular functional relationship defined by the guideline data.

[0046] The first model can use this data (e.g., from a guideline, etc.) to calculate a risk value for one or more medical conditions based on the acquired physiological values. For example, the model can use the guideline data to determine, for each acquired physiological value, a risk value for one or more medical conditions associated with that respective physiological value. The first model can then further combine each of these risk values ​​such that, for a particular combination of acquired physiological values, an aggregate risk value for each of a plurality of medical conditions is determined.

[0047] Table 1 below is just one example of various combinations of physiological values ​​that represent symptoms in this case and can be used to determine risk percentages for several different types of cancer. [Table 1]

[0048] The first model combines the risk values ​​for each of the physiological values ​​based on the guideline data to provide a total risk value for each of the medical conditions. For example, for each medical condition, there may be a baseline risk value corresponding to the known or estimated prevalence of that condition in the general population. This data may be stored on a server and accessed by the control device for performing the calculations. For example, a look-up table may indicate baseline risk values ​​for each of a number of medical conditions. The first model may modify the baseline risk values ​​based on those physiological values ​​and in response to obtaining the above-mentioned physiological values. For example, each baseline risk value for each medical condition may be weighted by each of the obtained physiological values.

[0049] For each of a plurality of medical conditions, data indicative of a baseline risk may be obtained and subsequently modified based on the obtained physiological values. For example, the guideline data described above may comprise baseline risk data. Each of the obtained physiological values ​​may have a plurality of weighting factors associated therewith, each weighting factor corresponding to one of the medical conditions. In response to obtaining a particular physiological value, the first model may modify the baseline risk for each medical condition based on each of the weighting factors associated with that physiological value (e.g., the risk may be multiplied by the weighting factor). This process may then be repeated sequentially for each of the obtained physiological values, for example as shown in Table 1 above. It is understood that the sequential modification of the risk values ​​may not simply comprise a simple linear addition, and that the first model may obtain and apply additional weighting factors to modify the risk values ​​based on the overall combination of the obtained physiological values. The guideline data may comprise all of the weighting factors described above, or the weighting factors may be determined based on information obtained from the guideline data.

[0050] In some cases, data is obtained and combined by the first model from multiple data sources, e.g., multiple sets of guideline data, for use in calculating the risk value described above. For example, each set of guideline data may provide a statistical correlation between the physiological parameters and multiple medical conditions described above, and the first model may determine new correlations between the physiological parameters and the medical conditions by combining data from each data source (e.g., data from each guideline). For example, the model may weight the data provided by each data source to provide a relationship between the physiological values ​​and the medical conditions, e.g., using preselected weights selected by the clinician and / or the broader medical service due to the relative importance of the data sources. For example, if data from different sets of guidelines or different sources provide different numerical relationships between the physiological parameters and the risk value of one of the medical conditions, two or more data sets may be combined. For example, the values ​​or functions representing the correlations may be averaged, and the average may be weighted by the weighting factors applied to each data set. For example, if the dataset providing the correlation data is from a recommended source, this data may be weighted more heavily than data from less recommended sources. This weighted data may be stored on a server.

[0051] The weighting factor used by the first model to calculate the risk value of each medical condition is calculated based on an average of the weighting factors obtained from different sets of guideline data. For example, if two or more sets of guidelines provide a weighting factor for a particular medical condition and physiological value, the two coefficients may be averaged to obtain a new weighting factor for applying the risk value to the first model. This averaging may be a weighted average, e.g., the average may be weighted based on the source of the guideline data. For example, more trusted or recommended sources of guideline data may be weighted more heavily than those that are less recommended.

[0052] In some examples, the guideline data further comprises mapping data. The mapping data may indicate, for each of a plurality of medical conditions, those physiological parameters that are relevant for the calculation of a risk value for that medical condition. In some examples, the mapping data may alternatively or additionally comprise an indication for each physiological parameter whether it is relevant for the assessment of risk for each of the medical conditions, the medical conditions being selected. Thus, the first model may use the mapping data to map each of the obtained physiological values ​​to the medical conditions for which they have an associated risk. In this way, only a subset of the medical conditions may be selected in a given set of physiological values, and only the baseline risk for those conditions may be modified, for example based on the weighting factors described above.

[0053] The mapping data can be stored, for example on a server, as a look-up table, such that the controller can, based on the acquired physiological values, map the values ​​to the particular medical condition to which they relate, and can modify the baseline risk based on the stored statistical correlation between the parameters and the medical condition.

[0054] In other examples, a full set of risk values ​​may be determined for all physiological values, e.g., a risk value associated with all medical conditions (e.g., all medical conditions in a set) may be determined for each physiological value, with this calculation being based on the guideline data described above. In such examples, even if mapping data is not available or available, e.g., a physiological value is not relevant to determining the risk of a condition (e.g., the weighting factor for that condition is zero), a risk value for that condition is still calculated.

[0055] The order of processing the calculations may vary from case to case. For example, each physiological value may first be mapped to a number of disease states, and a risk value for each disease state may be calculated by modifying the baseline risk for that disease state. This may be repeated for each of the acquired physiological values, and each risk value may be modified in turn for each of the acquired physiological values, and a cumulative risk value for each disease state may be determined for a particular combination of acquired physiological values.

[0056] In other examples, a risk value may be determined for each medical condition in turn, for example, a first risk value for a first medical condition to which at least one physiological value is mapped may be determined based on a combination of physiological values, followed by a second medical condition, and so on.

[0057] In some examples, the calculated risk value may be weighted (203) with additional factors, such as demographic data, to modify the calculated risk value. Such data may be obtained from a data store on a server or external device, as described above. As an example, this weighting data may comprise geographic data of the patient, such that a patient exhibiting one set of symptoms in one location may have a different risk of diagnosis of a medical condition than another patient exhibiting the same symptoms in a different location, such as based on the environmental characteristics of the area. The weighting data may comprise, for example, age, sex, and ethnicity data. In other examples, this data may instead be provided as part of the physiological values ​​obtained as described above in step 201. In this case, these factors may simply be combined with other physiological values, such as symptoms, by a model in a single step to determine a risk value for each medical condition.

[0058] In another example, weighting data is used to modify the baseline risk values ​​for each of a number of medical conditions, prior to the introduction of physiological values.

[0059] A lookup table is generated for each patient, where all physiological values ​​obtained for that patient are listed along with weighting (e.g., demographic) data. Each entry in the table may then be associated with a corresponding risk value for each of a number of medical conditions, and a particular set of values ​​that are combined for each condition to provide a total risk value for each of the conditions.

[0060] Each of the calculated total risk values ​​for each medical condition is then compared to a threshold value (204). Such threshold value may be a predetermined value based on the preferences and / or requirements of the clinician or the broader medical service. Data representing the threshold value may be stored on a server or retrieved by the controller to apply the calculated risk value to each medical condition, e.g., as a filter. If the calculated risk value is greater than the threshold, then that medical condition is selected and carried forward for the remainder of the process, and all medical conditions with calculated risk values ​​less than the threshold are removed or discarded (205).

[0061] Different thresholds may be selected and applied depending on the healthcare provider to which the user accessing the service belongs. For example, a user, such as a clinician, may belong to a particular healthcare provider with an associated user identifier, e.g., a group identifier. In this way, when a clinician uses the service, e.g., by logging on to an application on their device and entering values ​​as described above, data obtained from the device by the server may comprise the identifier. In this way, the relevant thresholds for the corresponding healthcare provider can be selected and applied to the calculated risk value in the patient. A particular diagnostic pathway is determined for one or more medical conditions for which the determined risk value is greater than a threshold (206). Such pathways may be determined based on stored data, such as based on the above-mentioned guidelines or data sources indicating preferred diagnostic pathways for a particular medical condition at a particular healthcare provider to which the user belongs. As an example, the data store may comprise, for each medical condition, a list of one or more available pathways, e.g., in the form of a look-up table. In response to one of the medical conditions selected as described above, the controller may retrieve data from the table indicating a corresponding diagnostic pathway.

[0062] Alternatively or additionally, the pathways may be determined based on the acquired physiological values ​​and a set of pathway parameters, with each pathway having an associated set of parameter values ​​for each of the pathway parameters. These values ​​may be stored in a data store or device, such as a remote server. Alternatively, these values ​​may be calculated based on the stored data and the acquired physiological values. Table 2 below shows an example of this data in three different diagnostic pathways for one particular combination of physiological values. [Table 2]

[0063] Thresholds may be set for one or more of these parameters, e.g., maximum time to completion, maximum time to diagnosis, maximum time to treatment, maximum cost, minimum sensitivity and / or minimum specificity. Each pathway that meets all of the associated thresholds may be selected. Alternatively, one pathway may be selected based on a combination of each pathway parameter value and a physiological value. For example, a weighting may be applied to each pathway parameter, and the parameter values ​​in each pathway may be combined based on the weighting to provide a rating or score for each pathway, and the pathway with the best rating may be selected. Alternatively, one of the available pathways with the minimum or maximum value of one of the parameters may be selected. Alternatively, one or more pathways may be selected to diagnose the condition using a combination of the above-mentioned selection methods.

[0064] The selection of a diagnostic pathway may be based on the obtained user identifier data described above, such as data obtained by the server from the user's device when the user logs on to the service. For example, the user identifier may indicate the healthcare provider with which the user is affiliated. The server obtains the user identifier from the device and, in response, obtains data regarding the healthcare provider (e.g., from a data store on the server). For example, data may be obtained indicating diagnostic pathways available at that provider and / or data indicative of pathway parameter thresholds described above. A diagnostic pathway may then be selected based on this data.

[0065] In other examples, the user identifier data may itself comprise an indication of the diagnostic pathways available to the provider and may further comprise data indicative of the pathway parameter thresholds mentioned above, such that this data is obtained by the server directly from the user's device.

[0066] Once one or more pathways have been selected, an indication identifying them is output to a user, such as a clinician (107). For example, if the steps described above are performed on a server, the indication may be provided over a network to a clinician's computing device where it can be displayed to the clinician who can take appropriate action. Data sent to the clinician's device may include: Data identifying a medical condition for which the patient has a calculated risk greater than a threshold associated with the medical condition Detailed information about both the conditions and pathways, as well as data identifying the pathways selected to diagnose these conditions

[0067] FIG. 3 illustrates another example of a process 300 for providing patient risk and recommended pathway information to a clinician using an application. In this method, a clinician first logs into an application on a device such as a PC, smartphone or tablet computer (301). Once this is done, the clinician is presented with a user interface such as a form having fields where data about the patient can be entered (302), such as physiological values ​​comprising risk factors, symptoms, signs and findings as described above. This data is then taken by a first model and used to determine the risk of the patient having multiple medical conditions, such as different types of cancer, and to detect whether any of these risks meet a threshold for that medical condition, for example as described above. For example, this first model may comprise decision tree logic at the particular medical service (e.g. clinical referral group) to which the clinician or patient belongs, and the physiological value data is cross-referenced (303) across the decision tree logic to calculate a risk value. This determination may be made by one or other of the methods described. This cross-referencing and calculation may occur at a remote server or device, retrieving data entered from the clinician's device as described above. A recommended pathway may be determined for diagnosing one or more medical conditions, for example, when thresholds are met as described above, and / or based on data and / or guidelines provided by an associated medical service. Once the medical conditions for which the patient is at risk and the recommended pathway are identified, they are presented to the clinician in a user interface on their device (304). For example, data indicative of this information may be sent back to the device from the server.

[0068] FIG. 4 illustrates an example process 400 for performing machine learning to improve determining a risk value for a medical condition and improve route selection for diagnosis of the condition.

[0069] In this method, a first model is executed (401) to determine risk values ​​for a plurality of medical conditions for the patient based on the physiological values ​​and weighting data substantially as described above, and to determine which conditions exceed selected thresholds. However, whereas in the above-described method the statistical correlations provided by the first model are based on one or more sets of clinical guidelines, in this method the statistical correlations between the physiological values ​​and the medical conditions may instead or additionally be based on historical patient data collected by the presently described process 400. In particular, the first model is modified by the historical patient data, as described below.

[0070] Once a relevant medical condition has been identified by the first model, a second model is executed (402) to select a diagnostic pathway based, for example, on physiological values ​​and pathway parameter values ​​as described above. However, as with the first model, the parameter values ​​associated with each pathway may be updated or modified based on data obtained by process 400 instead of, or in addition to, being derived from predefined stored data.

[0071] Once a pathway has been selected, the patient's progress can be tracked and the values ​​of the pathway parameters obtained and the outcome of the diagnosis (positive or negative) can be recorded. In this way, a patient data set is obtained (403) and comprises the following indicators: Physiological data obtained about the patient (e.g., patient factors) Any weighting data (e.g. demographic data) Calculated risk scores for each of several medical conditions Selected diagnostic path - Recorded route parameter values ​​for the selected route The outcome of the diagnostic pathway (e.g., whether there was a positive or negative diagnosis)

[0072] The first model may then be modified 404 based on this data. For example, the associations between various physiological parameters and the risk of a medical condition may be updated based on physiological data obtained for the patient and the outcome of the diagnosis. For example, the first model may comprise a Bayesian model that implements Bayes' theorem to modify a medical condition risk value associated with a given physiological value or combination of physiological values. More specifically, Bayes' theorem states:

number

[0073] In this case, P(A|B) can correspond to the probability that a patient has one of the specific medical conditions given the presence of a particular physiological value or combination of values, P(A) can correspond to the probability that a patient has a particular medical condition independently (e.g., a baseline risk of a medical condition), P(B) can correspond to the general likelihood of a patient exhibiting a particular physiological value or combination of physiological values, and P(B|A) can correspond to the probability of a patient having a particular medical condition exhibiting a particular physiological value or combination of values. Thus, by recording the patients' outcome data and the initial physiological values ​​they exhibited, the probability can be updated, thereby updating the likelihood that a patient with a particular physiological value or combination of values ​​(e.g., P(A|B)) has one of the specific medical conditions. This process can be performed iteratively, such that the probability is updated each time new patient data is recorded.

[0074] Similarly, the second model may be modified based on this data (305). For example, certain pathway parameter values ​​may be updated based on corresponding values ​​recorded for pathways used for a particular patient. In this case, pathway parameter values ​​associated with each of the diagnostic pathways (described above) may be modified based on values ​​recorded for each diagnostic pathway when the patient used that pathway. The second model may also comprise a Bayesian model.

[0075] This process can be performed iteratively, with both the first and second models updated for each patient for which process 400 is performed.

[0076] FIG. 5 illustrates another process 500 for providing patient risk and recommended pathway information to clinicians using machine learning techniques. In this method, as a first step 501, the patient's physiological values ​​can be entered into an application as described above and provided to a remote server or device. Alternatively or additionally, such data can be retrieved by the server from a data store such as an electronic medical record. Once these values ​​are retrieved, an initial risk of one or more medical conditions is calculated based on a particular combination of the retrieved values, for example, using the methods described above. Demographic data can be entered by or retrieved from the clinician's device, or this data can be pre-recorded and retrieved from a data store.

[0077] This initial risk is then modified (503) by applying a first Bayesian model. In particular, historical patient data is obtained and iterated from one or more databases such that the statistical correlations used to calculate the risk value are updated based on actual physiological values ​​from the patient. The historical patient data comprises physiological values ​​of patients both with and without one or more of the conditions for which the risk is calculated. Such databases may be stored and retrieved from a data store local to the server or may be retrieved from other data sources over a network, e.g., a WAN. The initial risk value for each of the multiple conditions may be updated by iterating over this data. For example, each data point may be used in turn by the first model to modify parameters used to calculate the risk value, resulting in an updated risk value being calculated for each iterated data point. Once all historical patient data in the set has been iterated, a final cumulative risk value for the one or more conditions may be provided.

[0078] The threshold for detection (504) can be set based on the requirements of the relevant health authority, for example, if the health authority wishes to diagnose all patients with greater than a 3% risk of having a particular medical condition, the threshold can be set at 3%.

[0079] By way of example, an indication of this threshold may be stored on a server and accessed by the model once the risk value is calculated. The detection threshold may be modified, for example, based on the requirements of a particular health authority. For example, stored data indicating the threshold may be modified based on data received from a remote device.

[0080] The revised risk values ​​are then compared to thresholds (505), and those medical conditions, e.g., types of cancer, for which the calculated risk values ​​are greater than the thresholds are presented, e.g., data indicative of those medical conditions are sent to a clinician's device for display, and combinations of physiological values ​​for which the risk values ​​do not exceed the thresholds for any of the medical conditions are then excluded from further analysis (506).

[0081] The diagnostic pathways available for each condition may be established based on information from health authorities (507). For example, multiple tests or referrals may be available for a given condition. This data may also be stored on a server and accessed by the model once the risk value is calculated. This data may also be modified based on data received, for example, by a remote device. A second Bayesian model may then be applied based on a combination of physiological values, data indicative of available pathways, and historical patient data (508). In particular, the historical patient data may be stored in a database as described above, and may include, in addition to the information described above, for each patient, an indication of the diagnostic pathways the patient underwent, as well as pathway parameter values ​​associated with the diagnostic pathway. Such pathway data may include, for example, but is not limited to, one or more of the pathway cost, time to completion, time to diagnosis, time to treatment, sensitivity, and specificity, as shown in Table 2 above. The second model is configured to determine one or more appropriate pathways for the treatment of one or more medical conditions for which the patient is at risk based on such pathway parameters, and the model is iteratively updated based on historical patient data, for example by sequentially modifying parameter values ​​associated with each pathway based on each of the historical patient data points.

[0082] The updated model can be retroactively applied to other patients going through the diagnostic process to provide an updated set of parameters such as cost, time, and accuracy for each pathway based on historical patient data (509).

[0083] To determine the recommended routes, data indicative of thresholds for each route parameter may be set (510), e.g., based on requirements of a health authority. For example, each parameter threshold (e.g., minimum and / or maximum) may be set, e.g., as a dial, and any route that meets the requirements of each parameter may be selected as the recommended route. Once one or more recommended routes are selected, they are presented to the clinician (511), e.g., along with an indication of a medical condition to which the patient is at risk. For example, data indicative of the recommended route may be transmitted from a server on which the above-described calculations are performed to the clinician's device, e.g., over a WAN.

[0084] From the above description, it is clear that the embodiments shown in the drawings are merely exemplary, with features that may be generalized, removed, or replaced as described herein and in the claims. Referring generally to the drawings, it is clear that schematic functional block diagrams are used to illustrate the functionality of the systems and devices described herein. Furthermore, processing functions may be provided by devices supported by electronic devices. However, it is clear that the functionality need not be so divided and should not be interpreted as a specific structure of hardware other than that described below and claimed. The functionality of one or more elements shown may be further subdivided and / or distributed throughout the disclosed device. In some embodiments, the functionality of one or more elements shown may be integrated into a single functional unit.

[0085] As will be understood by those skilled in the art in the context of this disclosure, each embodiment described herein can be implemented in a variety of different ways. Any feature of any aspect of this disclosure can be combined with any of the other aspects of this disclosure. For example, method aspects can be combined with apparatus aspects, and features described with reference to the movement of certain elements of the apparatus may be provided in a method that does not use those particular types of apparatus. Furthermore, each feature of each embodiment is intended to be separable from the features with which it is described in combination, unless the other features are expressly stated as essential to its operation. Each of these separable features can, of course, be combined with any of the other features of the embodiment in which it is described, or with any of the other features or combinations of features of any of the other embodiments described herein. Furthermore, equivalents and variations not described above can also be used without departing from the invention.

[0086] Certain features of the methods described herein may be implemented in hardware, and one or more functions of the apparatus may be performed in method steps. It is also clear that in the context of this disclosure, the methods described herein do not have to be performed in the order in which they are described, nor in the order in which they are shown in the drawings. Thus, disclosed aspects described in terms of products or apparatus are also intended to be performed as methods, and vice versa. The methods described herein may be performed in computer programs, hardware, or any combination thereof.

[0087] In some examples, the controller may be provided by a general purpose processor configured to perform a method, for example, according to any one of the methods described herein. In some examples, the controller may comprise digital logic, such as a field programmable gate array, FPGA, application specific integrated circuit, ASIC, digital signal processor, DSP, or other suitable hardware. In some examples, one or more memory elements may store data and / or program instructions used to perform the operations described herein. An embodiment of the present disclosure provides a tangible non-transitory storage medium comprising program instructions to program and execute a processor to perform any one or more of the methods described herein and / or to provide a data processing device described herein. The controller may comprise an analog control circuit that provides at least a portion of this control functionality. An embodiment provides an analog control circuit configured to perform any one or more of the methods described herein.

[0088] In some examples, one or more memory elements may store data and / or program instructions used to perform the operations described herein.Embodiments of the present disclosure provide a tangible, non-transitory storage medium comprising program instructions to program and execute a processor to perform any one or more of the methods described herein and / or to provide a data processing apparatus described herein and / or.

[0089] Other examples and modifications of the present disclosure will be apparent to those of ordinary skill in the art in light of the present disclosure.

Claims

1. obtaining a plurality of physiological values ​​from the patient; executing a first model configured to determine a risk value for at least one of a plurality of medical conditions based on the physiological value; Executing the first model includes: obtaining a first risk value for at least one of the medical conditions based on a first one of the obtained physiological values; weighting the first risk value based on a second one of the acquired physiological values ​​to determine a total risk value for the at least one medical condition in the patient; comparing each of the total risk values ​​to selected thresholds and selecting at least one diagnostic pathway for diagnosing a corresponding medical condition if one or more of the total risk values ​​exceed a corresponding threshold; selecting the diagnostic pathway comprises executing a second model configured to select at least one of a plurality of possible diagnostic pathways based on a plurality of sets of pathway parameter values; Each set is associated with one of the diagnostic pathways. A method for diagnosing a medical condition in a patient.

2. 10. The method of claim 1 , executing the first model comprises weighting a risk value of at least one of the medical conditions by each of the plurality of physiological values ​​in turn. method.

3. The method according to claim 1 or 2, executing the second model includes selecting the diagnostic pathway associated with one of the pathway parameter values ​​if the one of the pathway parameter values ​​satisfies a threshold value; method.

4. The method according to any one of claims 1 to 3, acquiring a patient dataset comprising a set of physiological values ​​and a set of indices each identifying a presence or absence of a diagnosis for one of the plurality of medical conditions; and revising the first model based on the patient data set. method.

5. The method according to any one of claims 1 to 4, Obtaining a diagnostic pathway data set for one of the plurality of diagnostic pathways; and modifying the second model based on the diagnostic pathway dataset. the diagnostic pathway data set comprises a plurality of pathway parameter values ​​for the diagnostic pathway; modifying the second model comprises optionally weighting each of the pathway parameter values ​​associated with the diagnostic pathway in the second model by the corresponding pathway parameter value obtained. method.

6. Obtaining a plurality of physiological values ​​from the patient; Mapping the physiological values ​​to a plurality of relevant medical conditions; determining a total risk value associated with each of said medical conditions based on said mapped physiological values; comparing each of said total risk values ​​to a threshold; outputting diagnostic pathway data for diagnosing the medical condition if the total risk value exceeds the threshold; Outputting the diagnostic pathway data includes: obtaining a plurality of path parameter values ​​for each of the diagnostic paths; and selecting at least one of the plurality of diagnostic pathways based on the pathway parameter values. A method for diagnosing a medical condition in a patient.

7. 7. The method of claim 6, Determining the total risk value associated with each of the medical conditions comprises determining, for each physiological value, an indication of risk associated with each of the medical conditions to which it is mapped; and / or obtaining risk data indicative of a predetermined association between a plurality of physiological parameters and a plurality of said medical conditions; mapping the physiological values ​​to the plurality of medical conditions; and Determining the total risk value for each of the medical conditions is based on the risk data; and / or and selecting at least one of the plurality of diagnostic pathways optionally comprises selecting the diagnostic pathway if any of the pathway parameter values ​​of the diagnostic pathway meet a corresponding threshold value. method.

8. 8. The method according to claim 6 or 7, determining the total risk value for a medical condition comprises iterating through historical patient data; the historical patient data optionally comprises a plurality of data sets each associated with a patient; each of said data sets comprising a set of physiological values ​​and a set of indices each identifying the presence or absence of a diagnosis of one of said disease conditions; and optionally revising the total risk value for at least one of the medical conditions based on the historical patient data. method.

9. 9. The method of claim 8, modifying the pathway parameter values ​​based on the historical patient data. method.

10. As a neural network, executing a first model configured to provide a relationship between a set of physiological parameters and a condition risk value for at least one of a plurality of medical conditions and to output an indication of a total risk for at least one of the medical conditions from the acquired physiological values ​​corresponding to the physiological parameters; The neural network includes: a) providing an association between each of a plurality of pathway parameter values ​​and each of a plurality of diagnostic pathways; b) executing a second model configured to select one of the diagnostic pathways based on the acquired physiological values ​​and the pathway parameter values; A set of physiological values; a set of indices each identifying the presence or absence of a diagnosis for one of the plurality of medical conditions; a pathway data set of one of the plurality of diagnostic pathways, the pathway data set including a plurality of pathway parameter values ​​for the diagnostic pathway; obtaining a patient dataset comprising: revising the first model based on the patient data set; modifying the second model based on the route data set; A method for training an artificial neural network to diagnose a medical condition.

11. 11. The method of claim 10, obtaining a plurality of said patient data sets and iteratively refining said first model based on each of said patient data sets. method.

12. 12. The method of claim 11, and iteratively refining the second model based on each of the patient data sets. method.

13. A recording medium having recorded thereon program instructions arranged to program a processor to carry out a method according to any one of claims 1 to 12.

Images