Systems and related methods for diagnostic decision support by AI-assisted and optimized care assistance tools

Abstract

A system for assisting in nursing a diagnosed medical patient. The system includes an edge device, a plurality of patient monitoring devices, and a database including medical information. The edge device is configured to obtain a care plan for a patient; obtaining a care suggestion by inputting the diagnosis and one or more medical information corresponding to the patient, monitoring data corresponding to the patient, and medical facility inventory information into a patient care model; comparing the care suggestion to the care plan to obtain a care insight, the care insight including suggested care not included in the care plan; information corresponding to the care insights is output, and the care plan is modified based on the care insights in response to user input that selects the care insights through the human-machine interface.

Description

Technical Field

[0001] Embodiments of the subject matter disclosed herein relate to caring for patients in a healthcare setting and, in particular, to providing assistance in caring for patients using an artificial intelligence (AI)-assisted guidance tool. Background Art

[0002] Healthcare providers (e.g., physicians, nurse practitioners, and physician assistants) utilize a thought process called differential diagnosis when diagnosing patients. Differential diagnosis involves determining multiple possible diagnoses for a patient based on available patient health data and the patient's symptoms. The physician then improves the differential diagnosis by performing additional diagnostic tests or evaluating additional patient health data until he / she is confident that he / she has identified the correct diagnosis. Gathering patient health data for differential diagnosis can involve the caregiver searching for medical information about the patient in multiple different hospital (or clinic) information systems (e.g., called "healthcare information technology (HCIT) systems"). This process can be time-consuming and error-prone, especially when the patient has an extensive medical history. Additionally, when performing differential diagnosis and deciding how to treat a patient, the caregiver may need to consult or call up available medical standards and guidelines from memory. This can also be time-consuming and / or cause the caregiver to miss possible diagnoses. Summary of the Invention

[0003] According to one aspect of the present disclosure, a medical system may include: one or more patient monitoring devices, each patient monitoring device being configured to generate patient monitoring data by monitoring a patient's physiological parameters; at least one database storing medical information corresponding to the patient; and an electronic device including a power supply; a communication interface configured to communicate with the one or more patient monitoring devices and the at least one database via a network; a human-machine interface configured to provide information to a user and obtain information from the user; a memory configured to store instructions; and at least one processor configured to execute the instructions to: obtain a care plan for the patient via the communication interface; obtain care recommendations by inputting a diagnosis and one or more medical information corresponding to the patient, monitoring data corresponding to the patient, and medical facility inventory information into a patient care model; compare the care recommendations with the care plan to obtain care insights, the care insights including recommended care not included in the care plan; output information corresponding to the care insights via the human-machine interface, and in response to user input selecting the care insights via the human-machine interface, modify the care plan based on the care insights.

[0004] According to another aspect of the present disclosure, a patient care assistance method may include: obtaining a care plan for a patient; obtaining care recommendations by inputting the diagnosis and one or more medical information corresponding to the patient, the monitoring data corresponding to the patient, and the medical facility inventory information into a patient care model; comparing the care recommendations with the care plan to obtain care insights, the care insights including recommended care not included in the care plan; outputting information corresponding to the care insights, and modifying the care plan based on the care insights in response to user input selecting the care insights through the human-machine interface.

[0005] It should be understood that the above brief description is provided to introduce in a simplified form selected concepts that are further described in the detailed description. This does not mean identifying the key features or essential features of the claimed subject matter, the scope of which is uniquely defined by the claims that follow the detailed description. Furthermore, the claimed subject matter is not limited to embodiments that solve any disadvantages mentioned above or in any part of this disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] The present invention will be better understood by reading the following description of non-limiting embodiments with reference to the accompanying drawings, in which:

[0007] Figure 1 An exemplary healthcare provider assistance system according to one embodiment is schematically shown.

[0008] Figure 2 is an input / output diagram of a DAM assistant according to one embodiment.

[0009] Figure 3A is a graph of a differential diagnosis list and corresponding predicted probabilities over a 12-hour period according to one embodiment.

[0010] Figure 3B is a graph of a differential diagnosis list together with clinical assessments, monitored parameters, and performed tests over a 12-hour period according to an example.

[0011] Figure 3C is a graph of the change in predicted probability of diagnoses on a differential diagnosis list over a 12-hour period according to one embodiment.

[0012] Figure 4A is a data table generated by a DAM assistant at the time point of 0 hours corresponding to an example according to one embodiment Figures 3A to 3C is a data table generated by a DAM assistant at the time point of 4 hours corresponding to an example according to one embodiment

[0013] Figure 4B is a data table generated by a DAM assistant at the time point of 4 hours corresponding to an example according to one embodiment Figures 3A to 3C is a data table generated by a DAM assistant at the time point of 4 hours corresponding to an example according to one embodiment

[0014] Figures 5A to 5F Shows the output of the chatbot interface of the DAM assistant according to one embodiment.

[0015] Figure 6 Is a diagram of an electronic device on which the DAM assistant is deployed, partially deployed, operated, or accessed according to one embodiment.

[0016] Figure 7 Is a diagram of a network environment on which the DAM assistant is deployed, partially deployed, operated, or accessed according to one embodiment.

[0017] Figure 8 Is a flowchart of a method 800 for diagnosing and monitoring a patient according to one embodiment.

[0018] Figure 9 Is a flowchart of a process 900 for training an AI model to generate a differential diagnosis list and corresponding prediction probabilities according to one embodiment.

[0019] Figure 10 Is an input / output diagram of a patient care assistant according to one embodiment.

[0020] Figures 11A to 11D Shows an example of the output of a care assistant according to one embodiment.

[0021] Figures 12A to 12G Shows a chatbot interface for a care assistant according to one embodiment.

[0022] Figure 13A and Figure 13B Is a flowchart showing a care assistance method 1300 according to one embodiment. Detailed Description

[0023] The following description relates to various embodiments of a healthcare provider assistance system that compiles patient health data from multiple sources, processes the data to generate meaningful information, and provides guidance to healthcare providers. The disclosed system can be implemented using a Diagnostic and Monitoring (DAM) assistant. The DAM assistant can provide a dynamic differential diagnosis list that includes one or more diagnoses, and for each diagnosis, a prediction probability of diagnostic accuracy. The DAM assistant can also recommend additional parameters to monitor, diagnostic tests to perform, imaging studies to perform, and standards of care to initiate / implement. When data (such as data corresponding to clinical assessments, physiological parameters, diagnostic test results, and imaging studies) is provided to the DAM assistant, the dynamic differential diagnosis list can be continuously updated.

[0024] Figure 1It is a diagram of the environment in which a DAM assistant operates and / or a care assistant operates according to an embodiment. As Figure 1 shown, the environment in which the DAM assistant and / or the care assistant operates may include a hospital environment, a cloud computing environment, and a virtual care provider environment.

[0025] The hospital environment may include a hospital network 107 connected to a hospital electronic medical record (EMR) database 109, a pharmacy system 110, a clinical information system 111, a laboratory information system 112, a picture archiving system / vendor neutral archive (PACS / VNA) 113, a central monitoring unit 114, an edge computing device 116, a secondary alert notification system 117, and a patient care unit 120. The EMR database 109 may provide data corresponding to the demographics and medical history of a patient. The pharmacy system 110 may provide data corresponding to the medications associated with a patient. The laboratory information system 112 may provide data indicating the results of diagnostic tests associated with a patient. The PACS / VNA 113 may provide data associated with the imaging studies associated with a patient. The department clinical information system 111 may provide data indicating the clinical observations and scores associated with a patient. These systems may be used as data sources for the local edge computing device 116, which may include an HL7 integration engine.

[0026] Patient physiological monitors and other point-of-care devices (such as ventilators and infusion pumps) may be present in the ward 118, which is part of one or more patient care units 120, which are in turn connected to a local data aggregator or monitoring gateway 119. According to one embodiment, the DAM assistant 115 may be deployed on the edge computing device 116 and may notify the users of the central monitoring unit 114 and nurses / doctors via the secondary alert notification system 117. According to one embodiment, the DAM assistant 123 may be deployed as an accompanying service of the remote patient monitoring system 124 on the cloud and notify the clinical user 127 of the virtual care remote display 126 or the mobile device 128. The remote patient monitoring system 124 may be many things, such as a virtual care solution. The hospital environment may include additional systems that integrate and display information from different device vendors, electronic health records, laboratory systems, and treatment records, etc.

[0027] The patient care unit 120 may include one or more patient monitoring devices that monitor the physiological parameters of a patient. Non-limiting embodiments of patient monitoring devices include non-invasive blood pressure monitors, oxygen saturation monitors, electrocardiographs, electroencephalographs, temperature monitors, heart rate monitors, respiratory rate monitors, hemoglobin monitors, end-tidal carbon dioxide monitors, cardiac rhythm monitors, cardiac output monitors, non-invasive blood pressure monitors, and heart rate variability monitors.

[0028] The EMR database 109 can be an external database that the EMR module can access via a secure hospital interface, or the EMR database 109 can be a local database (e.g., housed on a hospital device). The EMR database 109 can be a database stored in a mass storage device that is configured to communicate over a secure channel (e.g., HTTPS and TLS) and store data in encrypted form. Additionally, the EMR software is configured to control access to the patient's electronic medical records such that only authorized healthcare providers can edit and access the electronic medical records. The patient's EMR can include patient demographics, family history, past medical history, existing medical conditions, current medications, allergies, surgical history, past medical screenings and procedures, past hospitalizations and visits, etc.

[0029] Figure 2 is an input / output diagram of the DAM assistant according to one embodiment. As Figure 2 shown, the DAM assistant can input medical history, clinical assessment information, device data feeds, test results, and provider predictions for one or more patients, as well as other forms of relevant data. The patient's medical history can include previously diagnosed conditions such as hypertension and / or obesity, current medications, and relevant family history. The clinical assessment can be based on observations made by healthcare providers at the time of admission. For example, a healthcare provider may observe that the patient is swollen, disoriented, sweating, etc. The device data feeds can include data feeds provided by medical devices that monitor the physiological parameters of one or more patients. For example, the device data feeds can include the patient's blood oxygen level (SPO2) data and heart rate data. The laboratory / imaging results can include the results of diagnostic laboratories such as blood tests and the results of imaging studies such as computed tomography. The provider predictions can include the predicted probability of the provider's diagnosis being accurate.

[0030] Using the inputs discussed above, the DAM assistant can generate a list of differential diagnoses along with the predicted probability for each diagnosis on the list. The predicted probability can indicate the predicted probability that the diagnosis is accurate. The DAM assistant can utilize an artificial intelligence-based diagnosis (AID) model that is trained to generate a list of differential diagnoses and associated predicted probabilities using an available training data set. For example, the training data can include inputs of medical history, clinical assessment information, device data feeds, and test results labeled with a list of differential diagnoses and predicted probabilities. As described below, the AID model can also be trained based on feedback from healthcare providers who use the DAM assistant.

[0031] Once the DAM assistant generates a list of differential diagnoses and associated predicted probabilities, the DAM assistant can provide the healthcare provider with the opportunity to enter modified predicted probabilities based on the healthcare provider's expertise and knowledge of the patient to obtain an assisted probability (user prediction). The AID model can then be trained based on the assisted probability, thereby incorporating the feedback of the care provider in future diagnoses. In this way, the future output of the AID model may be influenced by the input provided by the care provider, which can improve the accuracy of the AID model. The user can also choose to overwrite the predicted probability with the assisted probability and / or add additional diagnoses not shown in the DAM along with the corresponding predicted probabilities. Overwriting the predicted probability with the assisted probability may include entering the assisted probability into the model, and the model generates all associated information based on the newly entered probability. When the predicted probability is overwritten, the DAM assistant can continue to operate the AID model based on the originally generated predicted probability in parallel, so that the user can refer back to the unassisted predictions, recommendations, and weights.

[0032] For one or more diagnoses on the differential diagnosis list, the DAM assistant can provide recommendations to increase the probability of accurate diagnosis by the healthcare provider. The recommendations may include: diagnostic tests to be performed, imaging studies to be performed, additional physiological parameters to be monitored, and additional clinical assessments to be performed. The DAM assistant can also provide a weight for each recommendation. For example, the diagnosis of acute gastroenteritis may have a predicted probability of 45% without considering the patient's heart rate, and when the heart rate data indicates acute gastroenteritis, the predicted probability can increase to 55% (+10%), or when the heart rate data does not indicate acute gastroenteritis, the predicted probability can decrease to 40% (-5%). In this way, the DAM assistant can determine that heart rate monitoring has a weight of 20 (based on 5% plus 15%). Another example, based on the blood test values, the predicted probability of acute gastroenteritis can change from 45% to 75% (+30%) or 25% (-20%). That is, based on the blood test values indicating acute gastroenteritis, the predicted probability can increase to 75%, or based on the blood test values indicating no acute gastroenteritis, the predicted probability can decrease to 25%. In this way, the DAM assistant can determine that the blood test has a weight of 50 (based on 20% plus 30%).

[0033] The DAM assistant can determine the cost of implementing the recommendations and the relative cost-effectiveness. For example, if test A costs $10 and has a weight of 10, and test B costs $10 and has a weight of 5, then the cost-effectiveness of test A is 1, while the cost-effectiveness of test B is 0.5. This information can be provided by the DAM assistant to the healthcare provider to assist in selecting an economical care plan.

[0034] The differential diagnosis list and corresponding prediction probabilities can be autonomously updated by the DAM assistant based on data being input into the DAM assistant. For example, the DAM assistance can output an initial differential diagnosis list along with corresponding recommendations based on initial / admission diagnosis data, clinical condition data, and care provider observation data, as this may be the only data available at that time. Based on this initial output, the care provider can order the recommended diagnostic tests. Once the results from the recommended diagnostic tests are input into the DAM assistant, the DAM assistant can change the prediction probabilities of one or more diagnoses on the differential diagnosis list based on the relevant information obtained from the test results. In response to the diagnostic test results, the DAM assistant can also adjust the recommended tests, studies, and additional parameters to be monitored, as well as their associated weights. According to one aspect of the present disclosure, diagnoses can be added to and / or removed from the differential diagnosis list based on the diagnostic test results. The results of imaging studies or information from monitored parameters can also affect the differential diagnosis list, corresponding prediction probabilities, and recommendations in a similar manner as discussed above with respect to the diagnostic test results.

[0035] In a healthcare environment, a care provider may need to consider a vast amount of information when caring for one or more patients. As a result, it may be difficult for the care provider to retain, understand, and process this information in an appropriate manner to make informed decisions. Such a chaotic environment has the potential to lead to errors, thereby reducing the quality of care provided to the patient. The DAM assistant can be used to compile and process all of this information to provide the care provider with concise and useful information that can be easily retained, understood, and used to make informed decisions. By using the DAM assistant, the generated differential diagnosis list will assist the care provider in diagnosing the patient. It will also help the care provider determine which parameters to monitor, which diagnostic tests to perform, and which imaging procedures to carry out. In this way, the DAM assistant can improve the quality of care by ensuring that the care provider is aware of all relevant diagnoses as well as additional parameters to be monitored, tests to be performed, and imaging studies to be carried out to help accurately diagnose the patient. Additionally, by training the AID model of the assistant based on user input provided to the DAM assistant, the accuracy of the DAM assistant can be improved over time, thereby further enhancing the quality of care provided to the patient.

[0036] Figures 3A to 3C Referring to an example according to one embodiment, wherein the DAM assistant helps a care provider diagnose a patient. Figure 3A is a graph of the differential diagnosis list and corresponding prediction probabilities over a 12-hour period. Figure 3B is a graph of the differential diagnosis list along with monitored parameters and performed tests over a 12-hour period. In Figure 3B , (+) indicates that the information increased the prediction probability and (-) indicates that the information decreased the prediction probability. Figure 3CIt is a graph of the change in the predicted probability of a diagnosis on the differential diagnosis list within a 12-hour period.

[0037] In Figures 3A to 3C example, the patient is a 74-year-old diabetic female who was admitted to the ER and was found to have the following signs and symptoms during clinical evaluation: disorientation, sweating, facial swelling, dry tongue and extremities, shortness of breath, weakness and lethargy, loss of appetite, tremors, feeling cold, headache, and diarrhea. The patient had fever, body aches, and runny nose 10 days ago, which lasted for 3 days. The patient is taking the following medications: hydrochlorothiazide, lithium, and risperidone. The DAM assistant can input "existing condition (diabetes), medications taken (hydrochlorothiazide, lithium, and risperidone), age (74), and fever, body aches, and runny nose 10 days ago, which lasted for 3 days" as the medical history information. This data can be imported from a database such as an electronic medical record (EMR) and / or entered by the care provider. The DAM assistant can also input "disorientation, sweating, facial swelling, dry tongue and extremities, shortness of breath, weakness and lethargy, loss of appetite, tremors, feeling cold, headache, and diarrhea" as the clinical evaluation information.

[0038] Based on the above inputs, which represent the data entered into the DAM model at 0 hours (0 hours since admission) as shown in Figures 3A to 3C , the DAM assistant can generate the following candidate differential diagnosis list: acute gastroenteritis, hypertension, congestive heart failure, diabetes, and hypothyroidism. The DAM assistant can use the AID model to generate the differential diagnosis list. The AID model can utilize a neural network trained using any of the techniques of supervised, semi-supervised, unsupervised, and reinforcement learning, such as a convolutional neural network. The AID model can be trained to generate a differential diagnosis list and corresponding predicted probabilities based on the input of medical history data, clinical evaluation data, physiological parameter data, and / or test result data. For example, the AID model can be trained based on a labeled dataset in which the medical history data, clinical evaluation data, physiological parameter data, and / or test result data are labeled with a differential diagnosis list including one or more diagnoses. The AID model can also be trained based on care provider predictions, such as when the care provider modifies the predicted probabilities generated by the AID model.

[0039] According to one embodiment, the AID model can utilize multiple neural networks trained based on different training data. For example, the first neural network can be trained to generate a list of differential diagnoses and corresponding prediction probabilities. The second neural network can be trained to suggest additional parameters to be monitored and tests to be performed, as well as associated weights. According to one embodiment, the AID model can include a neural network in which a first part of the layers is trained to generate a list of differential diagnoses, and the first and second parts of the layers are trained to generate a list of differential diagnoses and prediction probabilities. According to one embodiment, the AID model can include a neural network in which a first part of the layers is trained to generate suggested parameters, and the first and second parts of the layers are trained to generate suggested parameters and associated weights.

[0040] As Figure 3A shown, the DAM assistant generates accurate prediction probabilities for each diagnosis regarding differential diagnoses. That is, by inputting the initial input item into the AID model, the DAM assistant can predict that the diagnosis of acute gastroenteritis is 90% accurate, hypertension is 25% accurate, congestive heart failure is 45% accurate, diabetes is 25% accurate, and hypothyroidism is 35% accurate based on the data input at 0 hours. Since they are co-morbidities, the sum of all prediction probabilities may be greater than 100%. That is, each prediction probability can be between 0% and 100% because the prediction probability refers to the probability of the diagnosis it corresponds to.

[0041] As Figure 3B shown, at 4 hours (4 hours since admission), physiological parameters such as blood pressure, heart rate, and cardiac output (CO) are monitored. As Figure 3B shown, at 4 hours, the new data indicates a high heart rate (110 beats per minute), low blood pressure (90 / 70), and normal CO. The DAM assistant can obtain this new data at 4 hours based on user input and / or by monitoring the data stream of physiological sensors. Using this physiological data, the AID model can update the prediction probabilities, as Figure 3A shown. That is, based on the fast heart rate and low blood pressure, the DAM assistant determines that the prediction probability of acute gastroenteritis increases by 5%; based on the low blood pressure data, the prediction probability of hypertension decreases by 15% to 0%; and based on the normal CO, the prediction probability of congestive heart failure decreases by 15%. Even if the new data input at 4 hours may not be directly related to the diagnoses of diabetes and hypothyroidism, the AID model can adjust the prediction probabilities of these diagnoses based on the correlations learned through training and the changes in the prediction probabilities of other diagnoses, as Figure 3A shown.

[0042] At 8 hours, as Figure 3BAs shown, the results of echocardiogram imaging studies and blood test values are received and input into the AID model of the DAM assistant. The imaging studies and blood test values can be input by the user or obtained from a database, such as a database where laboratory technicians input laboratory results. Based on the echocardiogram study results indicating that the patient's heart is functioning normally, the AID model changes the predicted probability of congestive heart failure to 0%, and based on the blood test values indicating normal blood sugar levels, the AID model changes the predicted probability of diabetes to 0%. Based on the predicted probabilities of hypertension, congestive heart failure, and diabetes being zero, and other data (including low hormone levels over 8 hours) indicating that the patient has acute gastroenteritis and hypothyroidism, the AID model changes the predicted probabilities of acute gastroenteritis and hypothyroidism to 100%.

[0043] Figure 3C is a visual representation of the predicted probability of each diagnosis within a 12-hour period starting at the time of patient admission. As Figure 3C shown, based on the data input over time, the predicted probabilities are refined to remove potentially inaccurate diagnoses. For illustrative purposes, new data input at 4 hours and 8 hours are used. The AID model of the DAM assistant can receive data at any interval. For example, the model can input a data feed from sensors monitoring the patient's physiological parameters and modify the predicted probability based on detecting an anomaly or no anomaly in the data feed. Similarly, the AID model can modify the predicted probability in response to test results being input, regardless of a regular interval as used in this example.

[0044] According to one embodiment, the DAM assistant can recommend the standards of care to be initiated / implemented. For example, the DAM assistant can recommend starting a sepsis protocol for the patient. According to another embodiment, the DAM assistant can generate and output a scoring system to evaluate the deterioration of a patient suspected of having sepsis.

[0045] Figure 4A and Figure 4B are data tables generated by the DAM assistant at different time points according to one embodiment. Figure 4A The specific example shown in Figures 3A to 3C corresponds to the 0-hour of the example in Figure 4B The specific example shown in Figures 3A to 3C corresponds to the 4-hour of the example in

[0046] As Figure 4A shown, the DAM assistant can generate a list of differential diagnoses for the patient and the corresponding predicted probabilities. The DAM assistant can provide this list of differential diagnoses and the corresponding predicted probabilities to the care provider assigned to the patient. The DAM assistant can also receive and input the care provider's predicted probabilities. For example, as Figure 4AAs shown, based on the expertise of the care provider and the interaction with the patient, the care provider can predict that the probability of the patient having acute gastroenteritis is 75%. Therefore, the prediction of the care provider is 15% lower than the 90% prediction of the DAM assistant. According to one embodiment, the difference between the predicted probability of the DAM assistant and the predicted probability of the care provider can be used to train the AID model of the DAM assistant.

[0047] As Figure 4A shown, the DAM assistant can recommend diagnostic tests, imaging studies, and additional parameters for one or more diagnoses. Additionally, the DAM assistant can determine an increase in the estimated predicted probability of a diagnosis based on a recommendation indicating a correct diagnosis and a decrease in the estimated predicted probability of a diagnosis based on a recommendation indicating an incorrect diagnosis. Weights can be assigned to each recommendation by adding the predicted increase and the predicted decrease. For example, as Figure 4A shown, monitoring the patient's ECG is predicted to increase the predicted probability of congestive heart failure by 20% based on a positive diagnosis indicated by the ECG and decrease the probability by 20% based on a negative diagnosis indicated by the ECG. Therefore, ECG monitoring has a weight of 40 (20% + 20%). This weight indicates the estimated correlation between the recommendation and the diagnosis. For example, a patient with high blood pressure will have hypertension, while a patient with normal or low blood pressure will not have hypertension. Therefore, as Figure 4A shown, increasing blood pressure (BP) monitoring has a weight of 100 because only this one recommendation is needed to determine whether the diagnosis is correct or incorrect. Another example is that a liver enzyme test does not clearly indicate hypothyroidism. Therefore, a liver enzyme blood test has a weight of only 30.

[0048] According to one embodiment, the predicted probability increases and predicted probability decreases for diagnostic tests, imaging studies, and additional parameters can be generated by the AID model, such as by the neural network of the AID model. For example, the model can be trained based on labeled training data including examples of scenarios that include the results of diagnostic tests, imaging studies, and / or additional parameters. The data can be labeled to indicate a predicted probability increase or a predicted probability decrease for one or more of the received results. The model can also be trained based on the care provider's prediction. For example, once the results are input into the DAM model, the care provider can input the predicted probability of a correct diagnosis, which is different from the predicted probability provided by the DAM model. Then, when determining future predicted probability increases and decreases associated with diagnostic tests, imaging studies, and additional parameters, the model can be trained based on this difference.

[0049] The DAM assistant can also determine relative cost - effectiveness to further assist the care provider in selecting diagnostic tests, imaging studies, and / or additional parameters. As Figure 4AAs shown, each diagnostic test, imaging study, and / or additional parameter can be assigned a cost to be given. The cost information can be provided by a database including up-to-date hospital-specific pricing. Using these costs and weights, the relative cost-effectiveness can be determined by dividing the weight by the cost. For example, as Figure 4A shown, for acute gastroenteritis, since monitoring blood pressure has a weight of 23 and a cost of $10, increasing blood pressure monitoring has a relative cost-effectiveness of 2.3.

[0050] According to one embodiment, the DAM assistant can also input additional laboratory tests, imaging studies, and parameters that are not recommended, as well as corresponding increases and decreases in predicted probabilities. For example, when a care provider believes that a test that will assist in diagnosing the patient has not been recommended by the DAM assistant, the care provider can input the additional test and the associated increases and decreases in predicted probabilities. The added information can then be displayed to the care giver throughout the diagnostic process. The added information can also be used to train the AID model for future similar scenarios.

[0051] As described above, in the Figures 3A to 3C example, at 4 hours, the predicted probabilities of the diagnoses on the differential diagnosis list change in response to monitoring cardiac output, blood pressure, and heart rate. Figure 4B is an example of data provided by the DAM assistant at 4 hours in response to new cardiac output, blood pressure, and heart rate data being input into the DAM assistant.

[0052] As Figure 4B shown, since cardiac output, blood pressure, and heart rate are being monitored, monitoring of these parameters is no longer recommended. Additionally, since new data is being considered, the increased / decreased probabilities associated with the recommendations are different from those at 0 hours. For example, for congestive heart failure, once cardiac output has been considered, ECG data is less likely to affect the predicted probability. Therefore, the increased predicted probability and the decreased predicted probability of increasing ECG monitoring change from 20% and 20% respectively to 10% and 10%.

[0053] According to one aspect of the present disclosure, the DAM assistant can also recommend parameters to be removed from monitoring. To determine these parameters, the DAM assistant can track the parameters being monitored and determine the values of these parameters. Based on the value of the parameter being below a threshold, the DAM assistant can recommend ending the monitoring of the parameter to prevent waste and unnecessary expenses. For example, if the DAM assistant determines that the SPO2 data has a minimum value in assisting with diagnosis, the DAM assistant can recommend ending the monitoring of SPO2.

[0054] According to one embodiment, the AID model can determine the values and costs of the parameters being monitored. Based on this information, the DAM assistant can determine the relative cost value of each parameter being monitored. The relative cost value can be provided to the care provider to assist in determining which parameters to end monitoring.

[0055] According to one embodiment, the final result of a diagnosis can be used to train the model. For example, regarding Figures 3A to 3C the example shown, the DAM assistant can determine that the final diagnosis is accurate gastroenteritis and hypothyroidism. The DAM assistant can ask the user to confirm that the determined final diagnosis is accurate to ensure the training data is correct. Then the final diagnosis information can be used to train the AID model for similar scenarios.

[0056] Figures 5A to 5E A chatbot interface of the DAM assistant according to one aspect of the present disclosure is shown. According to a non-limiting embodiment, the DAM assistant can interact with the user via a computing device having a user interface or a human-machine interface (such as a personal computer, a tablet computer, or a handheld device such as a cellular phone). The user can receive data from the computing device and can also input data into the computing device. Figures 5A to 5E Different outputs of the chatbot provided on the computing device for interacting with the user are shown.

[0057] Figure 5A An initial message provided to the care provider (user) receiving the patient is shown. The initial message can provide the user with relevant information for appropriately diagnosing and monitoring the patient, such as information corresponding to a clinical assessment and the patient's medical history. The chatbot can also ask the user if they want help diagnosing the patient. For example, as Figure 5A shown, the chatbot can state the following: "Patient X was admitted due to the following symptoms: disorientation, sweating, facial swelling, dry tongue and limbs, shortness of breath, weakness and lethargy, loss of appetite, tremors, feeling cold, headache, and diarrhea. Patient X has a history of fever, body aches, and runny nose, which started 10 days ago and ended 3 days ago. Patient X is taking the following medications: hydrochlorothiazide, lithium, and risperidone. Do you want help diagnosing this patient? [Yes / No]"

[0058] In response to the user selecting "Yes" to indicate a need for assisted diagnosis, the chatbot can output a list of differential diagnoses and the corresponding prediction probabilities generated by the DAM assistant, as Figure 5B shown. The user can select one or more of the prediction probabilities to enter a care provider prediction different from the generated prediction probabilities. The user can also select a diagnosis to view additional information corresponding to that diagnosis.

[0059] For example, in response to the input of one or more care provider predictions, the chatbot may ask the user if they wish to overwrite the predicted probabilities with the care provider predictions, as Figure 5C shown. If the user chooses not to overwrite, the chatbot may display the generated predicted probabilities together with the care provider predictions during the diagnostic process. If the user chooses to overwrite, the care provider predictions may be input into the AID model, and new recommendations and weights may be generated based on the newly input probabilities. When the predicted probabilities are overwritten, the AID model may also continue to generate data based on the initial predicted probabilities such that the data remains accessible to the user.

[0060] For example, in response to the user selecting Figure 5B acute gastroenteritis in Figure 5D shown, the chatbot may output the received parameters, diagnostic tests, and imaging study information, as well as the recommended parameters, diagnostic tests, and imaging study information, as Figure 5D shown. The example of Figure 5D corresponds to 0 hours in FIGS. 3 and 4. As Figure 4A shown, since there are no monitored parameters and no diagnostic test / imaging study results are received at 0 hours, no information is shown as having been received. Since the DAM assistant has recommended additional parameters to monitor, as Figure 4A shown, the chatbot recommends monitoring blood pressure, heart rate, and SPO2. For each recommendation to assist the user in selecting parameters, the chatbot may also provide the user with an increase and decrease in the predicted probability of accurate diagnosis, as well as the relative cost-effectiveness.

[0061] Figure 5E shows the output provided by the chatbot in response to the user selecting Figure 5B congestive heart failure in Figure 5D shown. As Figure 5D shown, the chatbot provides the user with recommendations for additional parameters to monitor and imaging studies to perform, as well as the corresponding increase / decrease in the predicted probability, and the relative cost-effectiveness. By making this information readily available, the user can make quick and informed decisions, thereby improving the quality of care for the patient.

[0062] Figure 5F shows the output provided by the chatbot in response to the user selecting acute gastroenteritis at 4 hours in the example shown in FIGS. 3 and 4. As Figure 5F shown, since blood pressure and heart rate are being monitored, the chatbot may output data corresponding to these parameters. The data may be displayed in real-time or near real-time and may be continuously updated based on live data feeds received from sensors by the DAM assistant.

[0063] According to one embodiment, the DAM assistant may also display risk score data of a patient. For example, the risk score may be obtained from an EMR database and displayed by a chatbot interface to assist a user in diagnosing the patient.

[0064] According to one embodiment, the DAM assistant may be deployed, partially deployed, operated, or accessed on an electronic device 600 in a network environment 700 as Figure 7 shown. Figure 6 For illustration only, and without departing from the scope of the present disclosure, other embodiments of the electronic device and the environment may be used. Figure 6 and Figure 7 For illustration only, and without departing from the scope of the present disclosure, other embodiments of the electronic device and the environment may be used.

[0065] As Figure 6 shown, the electronic device 600 may include at least one of the following: a bus 610, a processor 620 (or processors), a memory 630, an interface 640, and a display 650.

[0066] The bus 610 may include circuitry for connecting components 620, 630, 640, and 650 to each other. The bus 610 may serve as a communication system for transferring data between components or between electronic devices.

[0067] The processor 620 may include one or more of the following: a central processing unit (CPU), a graphics processing unit (GPU), an accelerated processing unit (APU), an integrated many-core (MIC), a field programmable gate array (FPGA), or a digital signal processing (DSP). The processor 620 may control at least one of the other components of the electronic device 600, and / or perform operations related to communication or data processing. The processor 620 may execute one or more programs stored in the memory 630.

[0068] The memory 630 may include volatile and / or non-volatile memory. The memory 630 may store information related to at least one other component of the electronic device 600 and used for driving and controlling the electronic device 600, such as one or more commands, data, programs (one or more instructions), or application programs, etc. For example, the commands or data may formulate an operating system (OS). The information stored in the memory 630 may be executed by the processor 620.

[0069] The application programs may include one or more embodiments as discussed above. These functions may be executed by a single application program or multiple application programs each performing one or more of these functions.

[0070] The display 650 may include, for example, a liquid crystal display (LCD), a light emitting diode (LED) display, an organic light emitting diode (OLED) display, a quantum dot light emitting diode (QLED) display, a microelectromechanical systems (MEMS) display, or an electronic paper display. The display 650 may also be a depth sensing display, such as a multi-focus display. The display 650 is capable of presenting various contents (such as text, images, videos, icons, or symbols), for example.

[0071] The interface 640 may include an input / output (I / O) interface 641, a communication interface 642, and / or one or more sensors 643. The I / O interface 641 serves as an interface that may transfer commands or data, for example, between a user or other external devices and other components of the electronic device 600. The I / O interface 641 may provide a human-machine user interface for providing information to the user and receiving information from the user.

[0072] The sensor 643 may measure a physical quantity or detect an activation state of the electronic device 600, and may convert the measured or detected information into an electrical signal. For example, the sensor 643 may include one or more cameras or other imaging sensors for capturing images of a scene. The sensor 643 may also include: a microphone, a keyboard, a mouse, a touch screen, one or more buttons for touch input, a gyroscope or gyro sensor, a barometric pressure sensor, a magnetic sensor or magnetometer, an acceleration sensor or accelerometer, a grip sensor, a proximity sensor, a color sensor (such as a red, green, blue (RGB) sensor), a biophysical sensor, a temperature sensor, a humidity sensor, an illuminance sensor, an ultraviolet (UV) sensor, an electromyogram (EMG) sensor, an electroencephalogram (EEG) sensor, an electrocardiogram (EGG) sensor, an infrared (IR) sensor, an ultrasonic sensor, an iris sensor, or a fingerprint sensor. The sensor 643 may also include an inertial measurement unit. In addition, the sensor 643 may include a control circuit for controlling at least one of the sensors included herein. Any of these sensors 643 may be located within the electronic device 600 or coupled to the electronic device. The sensor 643 may be used to detect touch inputs, gesture inputs, hover inputs using an electronic pen or a body part of the user, etc.

[0073] The communication interface 642 may, for example, be capable of establishing communication between the electronic device 600 and external electronic devices (such as Figure 7 the first electronic device 602, the second electronic device 604, or the server 606 as shown). As Figure 7 shown, the communication interface 642 may be connected to the network 710 through a wireless or wired communication architecture to communicate with external electronic devices. The communication interface 642 may be a wired or wireless transceiver or any other component for transmitting and receiving signals.

[0074] Figure 7An exemplary network configuration 700 according to one embodiment is shown. Figure 6 The electronic device 600 can be connected to a first external electronic device 602, a second external electronic device 604, or a server 606 via a network 710. The electronic device 600 can be a wearable device, a wearable device that can be installed with an electronic device (such as a FIMD), etc. When the electronic device 600 is installed in an electronic device 602 (such as a FIMD), the electronic device 600 can communicate with the electronic device 602 through a communication interface 642. The electronic device 600 can be directly connected to the electronic device 602 to communicate with the electronic device 602 without involving a separate network. The electronic device 600 can also be an augmented reality wearable device including one or more cameras, such as glasses.

[0075] The first external electronic device 602, the second external electronic device 604, and the server 506 can each be a device of the same or different type as the electronic device 600. According to some embodiments, the server 606 can include a group of one or more servers. In addition, according to some embodiments, all or some of the operations performed on the electronic device 600 can be performed on another or multiple other electronic devices (such as the electronic devices 602 and 604 or the server 606). Additionally, according to some embodiments, when the electronic device 600 should automatically or upon request perform a certain function or service, the electronic device 600 can request another device (such as the electronic devices 602 and 604 or the server 506) to perform at least some functions associated therewith, rather than performing the function or service independently or additionally. Another electronic device (such as the electronic devices 602 and 604 or the server 606) may be capable of performing the requested function or additional functions and transmitting the execution result to the electronic device 600. The electronic device 600 can provide the requested function or service by processing the received result as it is or additionally. For this purpose, for example, cloud computing, distributed computing, or client - server computing technologies can be used. Although Figure 6 and Figure 7 shows that the electronic device 600 includes a communication interface 642 for communicating with the external electronic devices 602 and 604 or the server 606 via the network 710, but according to some embodiments, the electronic device 600 can operate independently without a separate communication function.

[0076] The server 606 can include components 610, 620, 630, 640, and 650 that are the same or similar to those of the electronic device 600 (or its suitable subset). The server 506 can support driving the electronic device 600 by performing at least one of the operations (or functions) implemented on the electronic device 600. For example, the server 606 can include a processing module or a processor that can support the processor 620 of the electronic device 600.

[0077] Wireless communication may be capable of using, for example, at least one of the following as a cellular communication protocol: Long Term Evolution (LTE), Long Term Evolution-Advanced (LTE-A), Fifth Generation Wireless System (5G), millimeter wave or 60 GHz wireless communication, Wireless USB, Code Division Multiple Access (CDMA), Wideband Code Division Multiple Access (WCDMA), Universal Mobile Telecommunications System (UMTS), Wireless Broadband (WiBro), or Global System for Mobile Communications (GSM). Wired connections may include, for example, at least one of the following: Universal Serial Bus (USB), High-Definition Multimedia Interface (HDMI), Recommended Standard 232 (RS-232), or Plain Old Telephone Service (POTS). Network 710 includes at least one communication network, such as a computer network (e.g., Local Area Network (LAN) or Wide Area Network (WAN)), the Internet, or a telephone network.

[0078] Although Figure 7 an example of a network configuration 700 including an electronic device 600, two external electronic devices 602 and 604, and a server 606 is shown, various changes may be made to Figure 7 it. For example, network configuration 700 may include any number of each component in any suitable arrangement. Generally speaking, computing and communication systems have a wide variety of configurations, and Figure 7 the scope of the present disclosure is not limited to any particular configuration. Additionally, although Figure 7 an operating environment in which various features disclosed in this patent document may be used is shown, these features may be used in any other suitable system.

[0079] The event detection method may be written as a computer-executable program or instructions that can be stored in a medium.

[0080] The medium may continuously store the computer-executable program or instructions, or temporarily store the computer-executable program or instructions for execution or download. Additionally, the medium may be any of various recording media or storage media that combine a single piece or multiple pieces of hardware, and the medium is not limited to a medium directly connected to the electronic device 600, but may be distributed over a network. Examples of the medium include magnetic media (such as hard disks, floppy disks, and magnetic tapes), optical recording media (such as CD-ROMs and DVDs), magneto-optical media (such as floppy optical disks, as well as ROM, RAM), and flash memory, which are configured to store program instructions. Other examples of the medium include recording media and storage media managed by an application store that distributes applications or by a website, server, etc. that provides or distributes various other types of software.

[0081] The DAM assistant may be provided in the form of downloadable software. A computer program product may include a product in the form of a software program that is electronically distributed by a manufacturer or an electronic marketplace (e.g., a downloadable application). For electronic distribution, at least a portion of the software program may be stored in a storage medium or may be generated temporarily. In such a case, the storage medium may be the storage medium of a server or server 606.

[0082] Figure 8 is a flowchart of a method 800 for diagnosing and monitoring a patient according to one embodiment. The method may be implemented on an electronic device 600, in a network environment 700, and / or in Figure 1 an environment. Method 800 may be implemented using the DAM assistant.

[0083] At operation 801, medical information may be obtained. The medical information may include a clinical assessment based on observations made by a healthcare provider at the time of the patient's admission. The medical information may also include the patient's medical history and prescription drug use. For example, the patient's medical history may include previously diagnosed conditions such as hypertension and / or obesity, current medication treatments, and relevant family history; and the clinical assessment observations may include that the patient exhibits swelling, disorientation, and sweating.

[0084] At operation 802, a differential diagnosis list may be generated based on the medical information obtained at operation 801. The differential diagnosis list may be generated by a model such as an AI model, a rule-based model, or a combination of both. The model may take the medical information as input and output the differential diagnosis list. Exemplary differential diagnosis lists may include acute gastroenteritis, hypertension, congestive heart failure, diabetes, and hypothyroidism.

[0085] At operation 802, the model may generate a prediction probability for each diagnosis on the differential diagnosis list. The prediction probability may indicate the estimated accuracy of the corresponding diagnosis. For example, based on the input medical information entered at operation 801, the model may predict that the patient has a 90% chance of having acute gastroenteritis.

[0086] At operation 802, recommendations may be generated for one or more diagnoses. The recommendations may include additional parameters to be monitored, diagnostic tests to be performed, and imaging studies to be performed. For example, based on the medical information entered at operation 801, the model may recommend monitoring blood pressure, heart rate, and oxygen saturation levels to further confirm the diagnosis of acute gastroenteritis.

[0087] At operation 802, a weight is generated for each recommendation. The recommendation weight may indicate the change in the predicted probability of a diagnosis based on the information corresponding to the recommendation considered in that diagnosis. The recommendation weight can be determined by increasing the predicted probability based on the recommendation indicating that the diagnosis is accurate and decreasing the predicted probability based on the recommendation indicating that the diagnosis is inaccurate, and the increased and decreased predicted probabilities can be combined to determine the weight of the recommendation.

[0088] A cost - benefit can also be generated for each recommendation. The cost - benefit can be determined by dividing the weight of the recommendation by the cost of implementing the recommendation. For example, if the cost of the first recommendation is $100 and it has a weight of 5, the cost - benefit will be 0.05, and if the cost of the second recommendation is $10 and it has a weight of 20, the cost - benefit will be 2. The care provider can then use this information to determine an economical diagnostic plan.

[0089] At operation 803, once the differential diagnosis list and the corresponding predicted probabilities are provided to the care provider, the care provider can optionally enter a care - provider prediction. The care - provider prediction can be entered when the care provider believes that the predicted probability generated by the model is inaccurate. For example, the care provider may believe that the predicted probability of 90% for acute gastroenteritis is too high, and thus may enter a care - provider prediction of 50%. This prediction can be used to train the model for generating future predicted probabilities. The care - provider prediction can also be used to adjust the current predicted probability.

[0090] The care provider can also enter one or more additional diagnoses not listed in the generated differential diagnosis list and the predicted probability of that diagnosis. These added diagnoses can then be added to the diagnosis list that is continuously updated in operation 804. The addition of the diagnoses can also be used to train the AID model for future predictions.

[0091] At operation 804, if a care - provider prediction has been entered, the user can choose to override the predicted probability with the care - provider prediction. For example, the chatbot can ask the user whether he / she wants to override the predicted probability with the care - provider prediction. The override made by the care provider can be recorded along with a reasonable reason. This information can be used to train the AID model.

[0092] At operation 805, based on the user selection, a care provider prediction overrides the predicted probability. The care provider prediction can be input into the AID model as the new predicted probability, and the model can then generate new recommendations and weights based on the newly input probability. For example, the model can input a 50% care provider prediction and adjust the predicted probability of acute gastroenteritis to 50%. Since the predicted probability of acute gastroenteritis is now much lower, the model can determine that the predicted probabilities of other diagnoses are higher. Additionally, since the predicted probability of acute gastroenteritis has decreased by 30%, the recommendation will likely have a greater impact on the predicted probability. Therefore, the model can also adjust the weight of the recommendation in response to the care provider prediction.

[0093] At operation 806, once the diagnosis list is set, the model can continuously update the differential diagnosis list, predicted probabilities, recommendations, and / or recommendation weights. According to one embodiment, the model can input a real-time or near real-time data feed of the parameters being monitored. The model can analyze the data to determine if it indicates that the differential diagnosis list, predicted probabilities, recommendations, and / or recommendation weights should be adjusted. In response to determining that an adjustment is to be made, the items can be updated autonomously without any user intervention. For example, blood pressure can be monitored in response to a recommendation to monitor blood pressure. In response to the monitored blood pressure being low, the model can decrease the predicted probability of hypertension. According to another embodiment, the user can monitor a parameter and input information into the model based on that monitoring, such as low blood pressure. The model can then use the information to adjust the predicted probability and, when applicable, notify the care team.

[0094] The model can also be continuously updated based on diagnostic tests and imaging study results. For example, once a diagnostic test is performed, the diagnostic test results can be input into the diagnostic information system. The diagnostic information system can then provide the information to the model. For example, a blood test can indicate that the patient has high thyroid hormones. Based on the information being input into the model, the model can update the differential diagnosis list and / or update the predicted probability of hypothyroidism to a lower value. Similarly, once an imaging study is performed, its results can be input into a program or database that provides the results to the model. The model can then update the differential diagnosis list and / or predicted probability in response to receiving the results. For example, whenever the DAM assistant notices a change in the patient's condition, the assistant can notify the provider and care team of the change (e.g., diagnosis, parameter recommendation). The notification can be configurable based on thresholds.

[0095] According to one embodiment, when the DAM assistant is receiving new data, a list of recommended parameters for which monitoring should be ended can also be continuously generated and updated. For example, based on the DAM assistant determining that monitoring the ECG is of little or no value in diagnosing acute gastroenteritis, the DAM assistant can recommend ending ECG monitoring. The DAM assistant can use the AID model to determine the values of the monitored parameters. For example, the amount by which the AID model uses the ECG data to contribute to the diagnosis can be determined. This amount can then be compared to a pre-determined threshold or cost for that monitoring.

[0096] The care provider can optionally input a care provider prediction in response to the updated list. The care provider prediction can be used by the model as discussed above with respect to operation 803.

[0097] According to an embodiment, operations 804 and 805 can be repeated until the patient is fully diagnosed. According to another embodiment, once a diagnosis is obtained, the DAM assistant using the above-described model or a separate model can continue to assist the care provider in monitoring the patient. For example, the DAM assistant can continue to monitor the patient's physiological parameters and / or recommend a monitoring plan. For example, once a final diagnosis has been made for the patient, the DAM assistant can recommend a monitoring interval for the parameter being monitored. The DAM assistant can also track the monitoring time to ensure that the parameters are within the recommended intervals. Any changes in the patient while being monitored by the system can result in new diagnosis or care plan recommendations.

[0098] According to one embodiment, once the patient has been discharged, the DAM assistant can continue to assist the care provider in monitoring the patient. The DAM assistant can recommend a post-discharge monitoring plan including the physiological parameters to be monitored and the monitoring intervals. The DAM assistant can input data from home monitoring equipment to further assist in monitoring the patient. For example, the DAM assistant can track the monitoring intervals and the abnormal monitoring data. The DAM assistant can also provide recommendations based on the abnormal monitoring data.

[0099] Figure 9 is a flowchart of a process 900 for training an AI model to generate a list of differential diagnoses and corresponding prediction probabilities according to one embodiment. Figure 9 The process 900 can be used to train a model of the DAM assistant, such as the AID model, a neural network of the AID model, or a part of the neural network of the AID model.

[0100] At operation 901, a trained AI model is obtained. The AI model can be trained to generate a list of differential diagnoses and corresponding prediction probabilities based on inputs of medical history, clinical assessment, physiological parameters, diagnostic test results, and / or imaging study results. The AI model can be trained based on a labeled dataset, in which the medical history data, clinical assessment data, physiological parameter data, and / or test result data are labeled with a list of differential diagnoses including one or more diagnoses. The AI model can utilize a neural network trained using any of supervised, semi-supervised, unsupervised, and reinforcement learning techniques, such as a convolutional neural network.

[0101] At operation 902, the AI model can be trained to generate a list of differential diagnoses and corresponding prediction probabilities based on inputs of medical history, clinical assessment, physiological parameters, diagnostic test results, and / or imaging study results. For example, the AI model can generate a list of differential diagnoses described in Figures 3A to 3C the 0 hours based on medical history information and clinical assessment information.

[0102] At operation 903, the user can input the prediction probabilities that the diagnoses on the list of differential diagnoses are accurate. For example, the user can input user predictions for one or more diagnoses based on the user's knowledge of the patient's condition.

[0103] At operation 904, the difference between the prediction probabilities and the user predictions can be determined. For example, if the AI model predicts that the probability of a patient having acute gastroenteritis is 90% and the user's prediction for acute gastroenteritis is 80%, the difference will be 10%.

[0104] At operation 905, the learnable parameters of the AI model can be adjusted based on the determined difference. For example, the learnable parameters of the AI model can be adjusted using known training techniques where the user predictions are used as ground truth.

[0105] When more user predictions are obtained by using the DAM assistant, the training process 500 can be infinitely repeated to further improve the accuracy of the AI model used therein.

[0106] According to one aspect of the present disclosure, once a diagnosis is determined, a patient care assistant can be implemented to assist the care team in monitoring and treating the diagnosed patient. The patient care assistant can be integrated with the DAM assistant or implemented as a standalone assistant.

[0107] Once a patient is diagnosed, dynamically controlling the care of the patient can improve the quality of patient care while also reducing waste (e.g., unnecessary use of medical devices, unnecessary tests, unnecessary costs, unnecessary time spent at a facility). Dynamically controlling the care of the patient can include analyzing and / or adjusting the care plan based on receiving new information corresponding to the patient (e.g., test results, monitoring data).

[0108] According to one aspect, a patient care model can generate a care plan for a patient based on patient information available at the time of diagnosis. For example, the patient care model can input the patient's diagnosis, medical history, clinical assessment information, device data feeds, test results (e.g., laboratory tests, imaging tests, etc.), and other forms of relevant data, and based on this input data, the patient care model can generate and output an initial care plan.

[0109] According to another aspect, a patient care assistant can input the initial care plan. According to one aspect, the initial care plan can be a standardized care plan obtained from a database. For example, a hospital may have standardized care plans that can be selected based on the diagnosis and additional information corresponding to the patient. That is, a standardized care plan can be provided for patients over 55 years old diagnosed with bronchitis, while a separate standardized care plan can be provided for patients under 55 years old diagnosed with bronchitis. According to another aspect, the initial care plan can be created by a member of the care team. For example, a physician monitoring the patient can provide an initial care plan based on the diagnosis and their knowledge of the patient. According to another aspect, a standardized care plan can be provided to a member of the care team, and that member of the care team can modify the standardized care plan to obtain the initial care plan input into the patient care model.

[0110] The initial care plan can be input into the care assistant to provide a starting point for the care of the patient. The initial care plan can be assigned to a member of the care team so that consistent care can be provided to the patient from all members of the care team.

[0111] When new information corresponding to the patient is received, the care assistant can adjust the care plan based on this new information. For example, in response to receiving new information indicating that the patient's parameters have stabilized, the care assistant can remove or recommend removing additional monitoring of the stable parameters from the treatment plan. According to another example, in response to receiving new information indicating that the patient's parameters have stabilized, the care assistant can remove or recommend removing future tests related to the patient's parameters from the treatment plan. According to another example, in response to receiving new information indicating that the patient's parameters are outside a preset range, the care assistant can add or recommend adding additional monitoring or tests to the treatment plan. By continuously adjusting or recommending adjustments to the care plan based on new information, human errors that may lead to non-compliance with best practices can be avoided.

[0112] Figure 10 is an input / output diagram of a patient care assistant according to one aspect. As Figure 10As shown, the care assistant can input medical history, clinical assessment information, device data feeds, test results, user input, and facility inventory / availability for one or more patients, as well as other forms of relevant data. The medical history of a patient can include previously diagnosed conditions such as hypertension and / or obesity, current medications, and relevant family history. Clinical assessments can be based on observations made by healthcare providers. For example, a healthcare provider may observe that a patient has swelling, disorientation, sweating, etc. Device data feeds can include data feeds provided by medical devices that monitor physiological parameters of one or more patients. For example, device data feeds can include a patient's blood oxygen level (SPO2) data and heart rate data, as well as other parameters. Laboratory / imaging results can include the results of diagnostic laboratories such as blood tests and the results of imaging studies such as computed tomography. User input can include the addition, modification, and removal of care recommendations.

[0113] The decision-making of the care assistant can be provided by a patient care model. The patient care model can include AI-based models and non-AI-based models (e.g., statistical models, rule-based algorithms). According to one aspect, the patient care model can utilize multiple submodels.

[0114] The monitoring submodel of the patient care model can include one or more AI models such as neural networks that are trained to recommend a patient monitoring plan for one or more patients based on the patient's diagnosis, medical history, clinical assessment information, device data feeds, test results, imaging results, the output of the patient care model, and other forms of relevant data. At the start of treatment, the monitoring submodel can input the initial patient monitoring plan of the initial care plan and any patient information (e.g., diagnosis, parameter data, test results, etc.) to obtain monitoring recommendations. Monitoring recommendations can include the parameters to be monitored, the time range / intervals for monitoring these parameters, the tests to be performed, the timing of the tests, the imaging to be performed, the units to be placed, and other known patient monitoring recommendations.

[0115] According to one aspect, the neural network of the monitoring sub-model can be trained to generate suggestions using supervised learning. The supervised learning process can use annotated training data, where patient information is annotated with patient monitoring plan information. That is, the training data can include the patient's diagnosis, medical history, clinical assessment information, device data feeds, test results, and other forms of relevant patient data. The ground truth can be obtained from the annotation of the patient monitoring plan information. Based on the patient information input and the corresponding ground truth of the patient monitoring plan information, the neural network can be trained to generate a suggested patient monitoring plan. For example, a large dataset of real-world patient information and the corresponding patient monitoring plans can be used to train the neural network of the monitoring sub-model. According to one aspect, the sub-model can also be trained based on usage feedback. For example, the monitoring sub-model can be trained based on information corresponding to whether a care team member accepts, modifies, or discards a suggestion or a part of them.

[0116] The treatment sub-model can include one or more AI models, such as neural networks, which are trained to suggest treatment plans for one or more patients based on the patient's diagnosis, medical history, clinical assessment information, device data feeds, test results, and the output of the patient care model, as well as other forms of relevant data. At the start of treatment, the treatment sub-model can input the initial treatment plan of the initial care plan, as well as any patient data (e.g., diagnosis, parameter data, etc.) to obtain treatment suggestions. The treatment suggestions can include administering drugs or other therapeutic substances to the patient, administering treatments provided by a device or a care team member, etc.

[0117] According to one aspect, the neural network of the treatment sub-model can be trained to provide suggestions using supervised learning. The supervised learning can use annotated training data, where patient information is annotated with treatment plan information. That is, the training data can include the patient's diagnosis, medical history, clinical assessment information, device data feeds, and test results, as well as other forms of relevant patient data. The ground truth can be obtained from the annotated treatment plan information. Based on the patient data input and the corresponding ground truth based on the treatment plan information, the neural network can be trained to generate a suggested treatment plan. For example, a large dataset of real-world patient information and the corresponding patient monitoring plans can be used to train the neural network of the treatment sub-model. According to one aspect, the sub-model can also be trained based on usage feedback. For example, the treatment sub-model can be trained based on information corresponding to whether a care team member accepts, modifies, or discards a suggestion or a part of them.

[0118] The equipment sub-model may include one or more AI models, such as neural networks, which are trained to recommend treatment plans for one or more patients based on the patient's diagnosis, medical history, clinical assessment information, device data feeds, test results, the output of the patient care model, and facility inventory / availability (e.g., medical device inventory / availability), as well as other forms of relevant data. At the start of treatment, the equipment sub-model may input an initial equipment list for the initial care plan, along with any patient and facility data (e.g., diagnosis, parameter data, etc.) to obtain equipment recommendations. The equipment recommendations may include the types of devices for performing monitoring and treatment, as well as the types of devices for caring for the patient.

[0119] According to one aspect, the neural network of the equipment sub-model may be trained to provide recommendations using supervised learning. Supervised learning may use annotated training data, where patient information is annotated with treatment plan information. That is, the training data may include the patient's diagnosis, medical history, clinical assessment information, device data feeds, test results, and other forms of relevant patient data. The ground truth may be obtained from the annotated treatment plan information. Based on the patient data input and the corresponding ground truth based on the treatment plan information, the neural network may be trained to generate a recommended treatment plan. For example, a large dataset of real-world patient information and the corresponding patient monitoring plans may be used to train the neural network of the treatment sub-model. According to one aspect, the sub-model may also be trained based on usage feedback. For example, the equipment sub-model may be trained based on information corresponding to whether a care team member accepts, modifies, or discards a recommendation or a part of it.

[0120] The patient care model may include additional sub-models that are trained to assist care team members in caring for the diagnosed patient. For example, the sub-model may be trained to assist with resource allocation and inventory management, as well as to assess the risk of readmission.

[0121] Figures 11A to 11D An example of the output of a care assistant according to one aspect is shown. In this example, the care assistant inputs diabetic ketoacidosis, congestive heart failure, and potential sepsis as patient diagnoses, as well as the following patient information: 1. Echo impression: low ejection fraction < 45%, indicating congestive heart failure. 2. ECG: Electrocardiogram abnormal changes: sinus tachycardia, tall R waves with widened QRS, ST segment changes: flat T - T wave inversion. 3. Blood glucose: Random blood glucose test: 220 mg / dl. 4. Blood electrolytes: Na, K, Cl disorders: low potassium, high sodium and high chloride. 5. ABG profile: > Metabolic acidosis: PH - 7.1, high anion gap, low serum bicarbonate, and 6. Blood test: showing high TLC, indicating infection, serum lactate > 3 mmol / L.

[0122] Based on the above inputs, the patient care model may output Figure 11AThe initial care plan shown. As Figure 11A shown, the care plan may include a monitoring plan, a treatment plan, and a list of equipment. According to one aspect, the information used by the care assistant to obtain the monitoring plan may be generated by the monitoring sub-model, the information used by the care assistant to obtain the treatment plan may be generated by the treatment sub-model, and the information used by the care assistant to obtain the equipment may be obtained from the equipment sub-model. According to another aspect, the initial care plan may be generated and input by the care team.

[0123] As shown on the first day of the monitoring plan, the parameters to be monitored and the timing corresponding to that monitoring are listed. For example, vital signs such as ECG and invasive blood pressure (IVP) will be continuously monitored, while arterial blood gas (ABG) is monitored every two hours. As shown on the first day of the treatment plan, the treatments and their amounts and any necessary timings are listed. As shown on Day 1 of the equipment list, the recommended equipment for patient care is listed. The equipment may include devices for the recommended monitoring (e.g., patient monitors), treatment devices, and other patient care equipment (e.g., catheters).

[0124] In addition, the care assistant may recommend a location for the patient. In Figures 11A to 11D the example, on Day 1, due to the patient's condition and the monitoring to be performed, the care assistant recommends that the patient stay in the intensive care unit (ICU). When the patient's condition improves, the care assistant may recommend transferring the patient to another location, such as a transitional ward or the patient's home.

[0125] Figures 12A to 12G Shows a chatbot interface for a care assistant according to one aspect. Figures 12A to 12E Shows the chatbot outputting the initial care plan.

[0126] As Figure 12A shown, the chatbot may provide the patient diagnosis and additional patient information. By selecting "Yes", a care team member can obtain the initial recommended care plan from the care assistant. As Figure 12B shown, the recommended monitoring plan of the recommended care plan is displayed to the care provider. The care provider can modify the list, delete items from it, and add items to it through the chatbot. Any added items, deleted items, reasonable items, case end reviews, and modified items can be saved for training the patient care model. Once the monitoring list is acceptable to the care team member, the member can proceed to the treatment plan, as Figure 12C shown. Similar to the monitoring plan, the care provider can modify the treatment plan, delete items from it, and add items to it through the chatbot. Any added items, deleted items, and modified items can be saved for training the patient care model. Once the treatment plan is acceptable to the care team member, the member can proceed to the equipment, asFigure 12D As shown. Similar to the monitoring plan and treatment plan, care providers can modify the equipment list, delete items from it, and add items to it through the chatbot. Any added, deleted, and modified items can be saved for training the patient care model. Once a care team member finds that all the monitoring, treatment, and equipment plans are acceptable, that member can finalize the care plan.

[0127] The finalized care plan can be output by the care assistant to the facility where the patient is located. For example, the care plan can be output to multiple members of the care team so that the monitoring and treatment in the plan can be initiated. According to one aspect, the finalized list can be output to a database so that it can be used for further training of the patient care model. According to another aspect, the list can be output to an EMR database.

[0128] In Figures 11A to 11D the example shown, a care plan is generated daily based on new patient information received since the last recommendation was generated. For example, at 5:00 am every day, a new care plan can be generated for the care team members who conduct the morning rounds. The new patient information can include new monitoring information (e.g., device data feeds), new test results, and new observations entered by care team members. According to another aspect, a new care plan can be generated whenever new information is received. For example, the care assistant can generate a new care plan based on receiving new laboratory results because these results can indicate the need for new treatment and / or monitoring, or the current treatment and / or monitoring can be stopped.

[0129] As Figure 11A shown, on the second day, the care assistant may not recommend changing any of the monitoring plan, treatment plan, or equipment. In some cases, the newly generated care recommendation will be the same as the current care plan because the new information will not cause any changes. In these cases, since there is no difference between the recommended care plan and the current care plan, the care assistant may not recommend any changes.

[0130] Figure 11B Shows the recommendations output on the third and fourth days of care by the care assistant according to one aspect.

[0131] In Figures 11A to 11DIn the example, after generating the care advice for the second day and before generating the care advice for the third day, the care assistant receives information that the patient's blood sugar has stabilized and the patient's ECG has become normal. Based on this received information, the care assistant recommends i) removing the continuous blood glucose monitoring (CBGM), and ii) replacing the 12-lead ECG with a 6-lead ECG. Since the blood sugar has stabilized, there is no longer a need to monitor the blood sugar. Therefore, the care assistant recommends removing the blood sugar monitoring to prevent the wasteful use of facility resources (e.g., CBGM devices). Similarly, since the patient's ECG has become normal and the details from the 12-lead ECG are no longer necessary for proper care, the care assistant recommends switching to the more economical 6-lead ECG to save costs. In addition to the changes in monitoring, based on the stable blood sugar and normal ECG, the care assistant recommends reducing the dose of Monocef from 2g to 1g because the patient's condition has improved.

[0132] The advice, also known as care insights, can be obtained by comparing the proposed care plan with the current care plan. Any differences can be output by the care assistant as advice or care insights.

[0133] Figures 12E to 12G Shows the changes recommended by the care assistant output by the chatbot according to one aspect. As Figure 12E shown, the recommended changes to the monitoring plan are displayed to the care provider. The care provider can select the recommended changes, modify the items therein, and add items to it through the chatbot. Any added items, modified items, case end reviews, rationale items, and selections can be saved for training the patient care model. Once the monitoring list is acceptable to the care team member, the member can proceed to the treatment plan, as Figure 12F shown. Similar to the monitoring plan, the care provider can select the recommended changes to the treatment plan, modify the items therein, and add items to it through the chatbot. Any added items, modified items, and selections can be saved for training the patient care model. Once the treatment plan is acceptable to the care team member, the member can proceed to the equipment, as Figure 12G shown. Similar to the monitoring plan and the treatment plan, the care provider can select the recommended changes to the equipment list, modify the items therein, and add items to it through the chatbot. Any added items, modified items, and selections can be saved for training the patient care model. Once the care team member finds that all the monitoring, treatment, and equipment changes are acceptable, the member can finalize the changes to the care plan.

[0134] The final determined changes to the care plan can be output by the care assistant to the facility where the patient is located. For example, the changes can be output to multiple members of the care team, so that monitoring and treatment on the list can be initiated. According to one aspect, the changes can be output to a database so that they can be used for further training of the care assistant. According to another aspect, the changes can be output to an EMR database.

[0135] As Figure 11B shown, on the 4th day of care, based on receiving information indicating that the patient's symptoms are improving (e.g., improved vital signs, normal test results), the care assistant can recommend increasing the interval between blood tests and reducing the furosemide dose. Since these recommended changes do not involve changing any equipment, the care assistant may not recommend any equipment changes.

[0136] Figure 11C Shows the recommended monitoring, treatment, and equipment changes for the 5th - 7th days of care. As shown on the 7th day, the care assistant recommends transferring the patient to a transitional ward. The care assistant can recommend this transfer because it has received information indicating that the patient's condition has improved. Figure 11D Shows the change recommendations for the 8th - 14th days. As shown on the 14th day, due to the information indicating that the patient's condition has improved received by the care assistant, the care assistant recommends ending monitoring, treatment, and equipment.

[0137] As regarding Figures 11A to 11D discussed, the care assistant can provide recommendations to care team members based on newly received patient information. The recommendations of the care assistant can change based on the received new information and be input into the patient care model to generate recommended changes. For example, when the same initial information as in the Figures 11A to 11D example is input into the care assistant, the care assistant can output the same care plan. However, based on different patient responses to treatment, the care assistant can provide different recommendations.

[0138] In Figures 11A to 11D the example, on the 3rd day, the care assistant recommends removing the CBGM and reducing the ECG leads based on receiving new patient information indicating that the patient's blood sugar has stabilized and the condition is improving (e.g., information obtained since the previous recommendations were generated). In an alternative example where the patient's blood sugar is unstable, the care assistant may not recommend ending the CBGM.

[0139] In Figures 11A to 11D the example, on the 9th day, the care assistant recommends discharging the patient from the medical care facility because the patient's vital signs have become normal. In an alternative example where the patient's vital signs normalize on the 5th day, the care assistant can recommend discharging the patient on the 6th day.

[0140] As described above, any suggestions output by the care assistant can be modified or deleted by a care team member via the chatbot. The care team member can also add a monitoring plan, treatment, and / or equipment to the care plan via the chatbot. These deletions, modifications, and additions can be used by the care assistant when making future suggestions. For example, when a care team member adds a treatment to the care plan, the care assistant can suggest an adjustment to the treatment based on the newly received patient information. According to another example, when a care team member adds a treatment to the care plan, the care assistant can suggest removing another treatment that is redundant or interacts poorly with the added treatment.

[0141] Figure 13A and Figure 13B is a flowchart showing a care assistance method 1300 according to one aspect. The method 1300 can be implemented on an electronic device 600, in a network environment 700, and / or in an Figure 1 environment. The method 1300 can be implemented using a care assistant. The care assistant can interact with a user via a chatbot or other known user interface as Figures 12A to 12G shown.

[0142] At operation 1301, a diagnosis of a patient can be obtained. According to one aspect, the DAM assistant described above can be used to obtain the diagnosis. A care team member can use the DAM assistant to narrow down the list of possible diagnoses to obtain a final diagnosis list. According to another aspect, the diagnosis can be input by the user via a chatbot interface or a standardized form interface.

[0143] At operation 1303, a care plan can be obtained based on the diagnosis. According to one aspect, the care plan can be a standardized care plan obtained from a database. According to another aspect, the care plan can be created by a member of the care team. According to another aspect, a standardized care plan can be provided to a member of the care team, and the care team member can modify the standardized care plan to obtain the care plan input into the patient care model. According to another aspect, the care plan can be obtained by inputting the diagnosis and a trained AI model corresponding to other medical information of the patient (such as patient parameter data from medical devices, patient history data from the EMR, and laboratory / test result data). The AI model can be trained based on real-world data annotated based on a care plan serving as ground truth. The care plan can include a monitoring plan, a treatment plan, and a list of equipment.

[0144] At operation 1305, a patient care model can be used to obtain care recommendations for a patient. The patient care model can generate care recommendations based on diagnostic inputs, the patient's medical information, monitoring data obtained from the patient, and / or the facility's inventory information. For example, the patient care model can input patient parameter data, including real-time data feeds from sensors measuring the patient, medical history from electronic medical records, observations made by care team members, laboratory / test result information, available equipment at the facility, available beds at the facility, and other relevant information for obtaining a patient care plan.

[0145] The care recommendations can include a patient monitoring plan. The patient monitoring plan can include one or more parameters to be monitored, and timing information for the monitoring. The patient monitoring plan can be obtained from the monitoring sub-model of the patient care model.

[0146] The care recommendations can include a treatment plan. The treatment plan can include the type of medical treatment, the amount of the treatment type, and timing information for the treatment. The treatment plan can be obtained from the treatment sub-model of the patient care model.

[0147] The care recommendations can include an equipment list. The equipment list can include devices for monitoring the patient, devices for treating the patient, and / or patient care equipment. The treatment plan can be obtained from the equipment sub-model of the patient care model. According to one aspect, the equipment list sub-model can use the facility's inventory information to determine which equipment to recommend. For example, if 6-lead ECG monitoring is recommended for a patient, but the facility does not have a 6-lead ECG available, the equipment list sub-model can recommend a 12-lead ECG because the 12-lead will provide the same data as the 6-lead plus additional data that can be discarded.

[0148] At operation 1307, the care recommendations can be compared with the care plan to determine if any part of the care recommendations is different from the care plan. If the care plan is the same as the care recommendations, the patient can be adequately cared for and no adjustment to the care plan is needed, so the method proceeds to operation 1319, which is to receive new information. If the care recommendations are different from the care plan, the patient may not be adequately cared for, so an adjustment to the care plan may be needed.

[0149] If there is a difference between the care recommendation and the care plan at operation 1307, the method may proceed to operation 1309. At operation 1309, care insights may be obtained based on the difference between the care recommendation and the care plan. For example, the care plan may include 24-hour 12-lead ECG monitoring, 24-hour blood glucose monitoring, and blood electrolyte testing every 12 hours, while the care recommendation may include 36-hour 12-lead ECG monitoring, 24-hour blood glucose monitoring, blood electrolyte testing every 8 hours, and 36-hour SPO2 monitoring. In this case, the care insights may include 36-hour 12-lead ECG monitoring, blood electrolyte testing every 8 hours, and 36-hour SPO2 monitoring.

[0150] At operation 1311, the care insights may be output. The insights may be output to the user to notify the user of the proposed changes to the care plan and to allow the user to adjust the care plan based on the recommendations. For example, the care insights may be output to the user via a user interface such as Figures 12A to 12G shown by a chatbot. According to one aspect, the care insights may be distributed to one or more devices via a network.

[0151] At operation 1313, the user may select or modify the care insights, or add new care insights to add to the care plan. Based on the user's selection, modification, or addition of care insights, the care plan may be modified to include the selected, modified, or added insights at operation 1315. When considering the above example, a care team member may select 36-hour 12-lead ECG monitoring from the care insights. This selection may cause the 24-hour 12-lead ECG monitoring in the care plan to be replaced with 36-hour 12-lead ECG monitoring. In addition, the care team member may modify the SPO2 monitoring from 36h to 24h. Then the modified care is added to the care plan. If the care team member determines that 36-hour body temperature monitoring is necessary care for the patient, the care team member may add 36-hour body temperature monitoring as a new care insight and then add it to the care plan.

[0152] Based on the user not selecting, modifying, or adding care insights, the care plan may not be modified, and the method may proceed to Figure 13B operation 1319 in

[0153] At operation 1317, information that instructs a user to select, modify, or add care insights can be used to train a patient care model. For example, information indicating that a user input has selected a recommendation can be used to train the patient care model to recommend that recommendation more aggressively in future similar situations. Information indicating that a user input has modified a recommendation can be used to train the patient care model to recommend the modified recommendation in future similar situations. Information indicating that a user input has not selected a recommendation can be used to train the patient care model to recommend that recommendation less or not at all in future similar situations. Training can be performed using methods known in the art, such as supervised learning with annotated data.

[0154] At operation 1319, additional patient information can be obtained. The additional patient information (also referred to as new information) can include information received after a previous recommendation was generated. The additional information can include the patient's medical information, monitoring data obtained from the patient, and inventory information of the facility. For example, a patient monitor data stream and laboratory results received since the generation of the previous care recommendation can be additional information that will be used in generating an updated care recommendation.

[0155] At operation 1321, the additional information and the information used in generating the previous care plan can be used to obtain an updated care recommendation. The updated care recommendation can be generated by the patient care model in a manner similar to that described in operation 1305. In some cases, the additional information may not cause the updated care recommendation to be different from the previous care recommendation. However, in cases where the additional information indicates that the patient's status has changed, the additional information can cause the patient care model to generate an updated care recommendation that is different from the previously generated care plan.

[0156] The generation of the updated care recommendation can be triggered by an event. According to one aspect, an updated care recommendation can be generated whenever additional information is received. For example, the patient care model can generate an updated care recommendation in response to receiving new laboratory results. According to other aspects, the updated care recommendation can be generated at a preset interval. For example, an updated care recommendation can be generated at 5:00 am every day, thus providing the care team with a recommendation based on the latest information. According to another aspect, the updated care recommendation can be generated based on a user input. For example, when a care team member makes rounds in a healthcare facility, the care team member can instruct a care assistant to generate an updated care recommendation.

[0157] At operation 1323, the updated care recommendation can be compared with the care plan in a manner similar to operation 1307 to determine whether any part of the updated care recommendation is different from the care plan. Operations 1325, 1327, 1329, and 1331 can be performed in a manner similar to operations 1309, 1311, 1313, and 1315, respectively.

[0158] If no selection, revision, or addition is made at operation 1329, method 1300 may return to operation 1319, i.e., obtain new information for generating future recommendations. Similarly, if there is a selection, modification, or addition at operation 1329, method 1300 may return to operation 1319 once the care plan has been updated at operation 1331.

[0159] Care assistance method 1300 may assist a care team in caring for a patient. Care team members are often overworked and work in a busy environment, which may result in inadvertent errors that lead to suboptimal care. By interacting with the care team, care assistant implementing care assistance method 1300 can prevent inadvertent errors that may lead to suboptimal care. Additionally, care team members may not have the necessary knowledge to provide optimal care. By interacting with the care assistant while caring for the patient, the quality of care can be improved by providing the care team with information they may not have.

[0160] According to one aspect of the present disclosure, a medical system may include: one or more patient monitoring devices, each patient monitoring device being configured to generate patient monitoring data by monitoring a patient's physiological parameters; at least one database storing medical information corresponding to the patient; an electronic device including a power supply; a communication interface configured to communicate with the one or more patient monitoring devices and the at least one database via a network; a human-machine interface configured to provide information to a user and obtain information from the user; a memory configured to store instructions; and at least one processor configured to execute the instructions to: obtain, via the communication interface, medical information corresponding to the patient from the at least one database; obtain, via the communication interface, real-time or near-real-time patient monitoring data corresponding to the patient from the one or more patient monitoring devices; generate, based on at least one of the medical information and the patient monitoring data, a differential diagnosis list for the patient by an artificial intelligence (AI) model, the differential diagnosis list including one or more diagnoses and a predicted probability corresponding to each diagnosis, each predicted probability indicating the estimated accuracy of the corresponding diagnosis; provide, via the human-machine interface, the differential diagnosis list and the predicted probabilities to the user; and obtain, from the user and via the human-machine interface, user predictions for one or more diagnoses of the differential diagnosis list. The AI model includes a first neural network trained to generate predicted probabilities based on previously input user predictions.

[0161] According to one aspect of the present disclosure, the one or more patient monitoring devices include one or more of a blood pressure monitor, a blood oxygen monitor, an electrocardiogram, an electroencephalogram, a body temperature monitor, a heart rate monitor, a respiratory rate monitor, a hemoglobin monitor, an end-tidal carbon dioxide monitor, a heart rhythm monitor, a cardiac output monitor, and a heart rate variability monitor.

[0162] According to one aspect of the present disclosure, the at least one processor may be further configured to: continuously input patient monitoring data corresponding to a patient in real-time or near real-time into an AI model; and autonomously update the prediction probability in response to the patient monitoring data indicating that one or more of the differential diagnosis list and the prediction probability should be adjusted.

[0163] According to one aspect of the present disclosure, the first neural network may be trained to obtain a prediction probability by: comparing one or more prediction probabilities output by the first neural network with one or more user predictions; and adjusting the parameters of the first neural network based on the differences obtained from the comparison.

[0164] According to one aspect of the present disclosure, the at least one processor may be further configured to: generate one or more suggestions for a first diagnosis in the differential diagnosis list; and generate a suggestion weight for each of the one or more suggestions, the suggestion weight indicating the change in the prediction probability of the diagnosis based on information corresponding to the suggestions considered in the corresponding diagnosis.

[0165] According to one aspect of the present disclosure, the one or more suggestions may include one of additional parameters to be monitored, diagnostic tests to be performed, and imaging studies to be performed.

[0166] According to one aspect of the present disclosure, the at least one processor may be further configured to: for a first suggestion of a first diagnosis, increase the prediction probability based on the first suggestion indicating that the first diagnosis is accurate; decrease the prediction probability based on the first suggestion indicating that the first diagnosis is inaccurate; and generate a weight for the first suggestion by combining the increased prediction probability and the decreased prediction probability.

[0167] According to one aspect of the present disclosure, the at least one processor may be further configured to: autonomously adjust the prediction probability in response to information obtained from the implementation of a suggestion being input into the AI model.

[0168] According to one aspect of the present disclosure, the AI model may include two neural networks, namely: a first neural network, which is trained to generate a differential diagnosis list; and a second neural network, which is trained to generate the one or more suggestions.

[0169] According to one aspect of the present disclosure, the at least one processor may be further configured to: determine the relative cost-effectiveness of each suggestion based on the cost of the suggestion and the weight of the suggestion.

[0170] According to one aspect of the present disclosure, an electronic device may include: a power supply; a communication interface configured to communicate with one or more patient monitoring devices and at least one database via a network; a human-machine interface configured to provide information to a user and obtain information from the user; a memory configured to store instructions; and at least one processor configured to execute the instructions to: obtain medical information corresponding to the patient from the at least one database via the communication interface; obtain real-time or near-real-time patient monitoring data corresponding to the patient from the one or more patient monitoring devices via the communication interface; generate, by an artificial intelligence (AI) model, a differential diagnosis list for the patient based on at least one of the medical information and the patient monitoring data, the differential diagnosis list including one or more diagnoses and a predicted probability corresponding to each diagnosis, each predicted probability indicating the estimated accuracy of the corresponding diagnosis; provide the differential diagnosis list and the predicted probabilities to the user via the human-machine interface; and obtain user predictions for one or more of the diagnoses from the user and via the human-machine interface. The AI model may include a first neural network trained to generate predicted probabilities based on previously input user predictions.

[0171] According to one aspect of the present disclosure, an electronic device may include: a power supply; a communication interface configured to communicate with one or more patient monitoring devices and at least one database via a network; a human-machine interface configured to provide information to a user and obtain information from the user; a memory configured to store instructions; and at least one processor configured to execute the instructions to: obtain medical information corresponding to the patient from the at least one database via the communication interface; obtain real-time or near-real-time patient monitoring data corresponding to the patient from the one or more patient monitoring devices via the communication interface; generate, by an artificial intelligence (AI) model: a differential diagnosis for the patient based on at least one of the medical information and the patient monitoring data, the differential diagnosis list including one or more diagnoses; a predicted probability corresponding to each diagnosis, each predicted probability indicating the estimated accuracy of the corresponding diagnosis; one or more recommendations for a first diagnosis in the differential diagnosis list; and a recommendation weight for each of the one or more recommendations, the recommendation weight indicating the change in the predicted probability of the diagnosis based on information corresponding to the recommendations considered in the corresponding diagnosis; and provide the differential diagnosis list, the predicted probabilities, the one or more recommendations, and the one or more recommendation weights to the user via the human-machine interface. The one or more recommendations may include one of additional parameters to be monitored, diagnostic tests to be performed, and imaging studies to be performed.

[0172] According to another aspect of the present disclosure, the at least one processor may be configured to obtain user predictions for the diagnoses in the differential diagnosis list from the user and via the human-machine interface.

[0173] According to another aspect of the present disclosure, the at least one processor may be configured to input a user prediction into an AI model to generate one or more of new suggestions and new suggestion weights.

[0174] According to another aspect of the present disclosure, the at least one processor may be configured to obtain a value of a parameter being monitored, the value indicating an amount contributed by the parameter to a corresponding diagnosis.

[0175] According to another aspect of the present disclosure, the at least one processor may be further configured to: suggest ending the monitoring of the parameter being monitored based on the value being lower than a predefined threshold.

[0176] According to another aspect of the present disclosure, the at least one processor may be configured to: continuously input patient monitoring data corresponding to a patient in real-time or near real-time into an AI model; and autonomously update a prediction probability based on the patient monitoring data.

[0177] According to another aspect of the present disclosure, a first neural network may be trained to obtain a prediction probability by: comparing one or more prediction probabilities output by the first neural network with one or more user predictions; and adjusting parameters of the first neural network based on a difference obtained from the comparison.

[0178] According to another aspect of the present disclosure, the at least one processor is further configured to: for a first suggestion for a first diagnosis, generate an increase in the prediction probability based on the first suggestion indicating that the first diagnosis is accurate; generate a decrease in the prediction probability based on the first suggestion indicating that the first diagnosis is inaccurate; and generate a weight for the first suggestion by combining the increase in the prediction probability and the decrease in the prediction probability.

[0179] According to another aspect of the present disclosure, the at least one processor may be configured to: autonomously adjust the prediction probability in response to information obtained from the implementation of a suggestion being input into the AI model.

[0180] According to one aspect, a medical system may include: one or more patient monitoring devices, each patient monitoring device being configured to generate patient monitoring data by monitoring a patient's physiological parameters; at least one database storing medical information corresponding to the patient; and an electronic device, the electronic device including a power supply; a communication interface configured to communicate with the one or more patient monitoring devices and the at least one database via a network; a human-machine interface configured to provide information to a user and obtain information from the user; a memory configured to store instructions; and at least one processor configured to execute the instructions to: obtain a care plan for the patient via the communication interface; obtain care recommendations by inputting the patient's diagnosis and one or more medical information, the patient's monitoring data, and medical facility inventory information into a patient care model; compare the care recommendations with the care plan to obtain care insights, the care insights including recommended care not included in the care plan; output information corresponding to the care insights via the human-machine interface, and in response to user input selecting the care insights via the human-machine interface, modify the care plan based on the care insights.

[0181] According to one aspect, a system may include: a power supply; a communication interface configured to communicate with one or more patient monitoring devices and at least one database via a network; a human-machine interface configured to provide information to a user and obtain information from the user; a memory configured to store instructions; and at least one processor configured to execute the instructions to: obtain a care plan for a patient via the communication interface; obtain care recommendations by inputting the patient's diagnosis and one or more medical information, the patient's monitoring data, and medical facility inventory information into a patient care model; compare the care recommendations with the care plan to obtain care insights, the care insights including recommended care not included in the care plan; output information corresponding to the care insights via the human-machine interface, and in response to user input selecting the care insights via the human-machine interface, modify the care plan based on the care insights.

[0182] According to one aspect, the at least one processor may be configured to execute the instructions to: obtain care recommendations by obtaining a patient monitoring plan from a patient care model.

[0183] According to one aspect, the patient monitoring plan may include parameters to be monitored and timing information for monitoring the parameters.

[0184] According to one aspect, the at least one processor may be configured to execute the instructions to obtain a patient monitoring plan from a monitoring sub-model of the patient care model, the monitoring sub-model being trained based on previous user input.

[0185] According to one aspect, the at least one processor may be configured to execute the instructions to: obtain care advice by obtaining a treatment plan from a patient care model.

[0186] According to one aspect, the treatment plan may include a type of medical treatment, an amount of the treatment type, and timing information of the treatment.

[0187] According to one aspect, the at least one processor may be configured to execute the instructions to obtain a treatment plan from a treatment sub-model of the patient care model, the treatment sub-model being trained based on previous user input.

[0188] According to one aspect, the at least one processor may be configured to execute the instructions to: obtain care advice by obtaining a list of equipment from a patient care model.

[0189] According to one aspect, the list of equipment may include at least one of a device for monitoring a patient, a device for treating a patient, and patient care equipment.

[0190] According to one aspect, the at least one processor is configured to execute the instructions to obtain a list of equipment from an equipment sub-model of the patient care model, the equipment sub-model being trained based on previous user input.

[0191] According to one aspect, the at least one processor may be configured to execute the instructions to, based on receiving additional medical information and / or additional patient monitoring data: obtain updated care advice by inputting the additional medical information and / or the additional patient monitoring data into the patient care model; compare the updated care advice with a care plan to obtain updated care insights, the updated care insights including recommended care not included in the care plan; output information corresponding to the updated care insights, and in response to user input selecting the updated care insights of the updated care advice, modify the care plan based on the updated care insights.

[0192] According to one aspect, the at least one processor may be configured to execute the instructions at a preset time interval to: obtain updated care advice by inputting the additional medical information and / or the additional patient monitoring data into the patient care model, the additional medical information and / or the additional patient monitoring data being created after obtaining the care advice; compare the updated care advice with a care plan to obtain updated care insights, the updated care insights including recommended care not included in the care plan; output information corresponding to the updated care insights, and in response to user input selecting the updated care insights of the updated care advice, modify the care plan based on the updated care insights.

[0193] According to one aspect, the at least one processor may be configured to execute the instructions to: generate, by a differential diagnosis model, a differential diagnosis list for a patient, the differential diagnosis list being obtained by inputting at least one of diagnostic medical information corresponding to the patient and diagnostic monitoring data corresponding to the patient into the differential diagnosis model, the differential diagnosis list including one or more diagnoses; and obtain a diagnosis from the differential diagnosis model based on a user selection.

[0194] According to one aspect, user input for selecting a care insight includes user input for modifying the care insight.

[0195] According to one aspect, a patient care assistance method may include: obtaining a care plan for a patient; obtaining care recommendations by inputting the diagnosis and one or more medical information corresponding to the patient, the monitoring data corresponding to the patient, and medical facility inventory information into a patient care model; comparing the care recommendations with the care plan to obtain a care insight, the care insight including recommended care not included in the care plan; outputting information corresponding to the care insight, and modifying the care plan based on the care insight in response to user input selecting the care insight through the human-machine interface.

[0196] According to one aspect, obtaining care recommendations includes obtaining a patient monitoring plan, a treatment plan, and an equipment plan from a patient care model.

[0197] According to one aspect, the method may further include obtaining user input for each of the patient monitoring plan, the treatment plan, and the equipment plan.

[0198] According to one aspect, the method may further include one or more of the following: i) in response to receiving additional medical information and / or additional patient monitoring data and ii) at a preset time interval: obtaining updated care recommendations by inputting the additional medical information and / or the additional patient monitoring data into the patient care model; comparing the updated care recommendations with the care plan to obtain an updated care insight, the updated care insight including recommended care not included in the care plan; outputting information corresponding to the updated care insight, and modifying the care plan based on the updated care insight in response to user input selecting the updated care insight of the updated care recommendations.

[0199] According to one aspect, the monitoring data may include data from a plurality of real-time or near-real-time data feeds generated by a plurality of sources.

[0200] According to one aspect, the method may include obtaining medical information, including obtaining medical data from multiple sources and standardizing the medical data for input into a patient care model. For example, different medical devices may provide data in different formats. Thus, these different formats may be standardized for input into the care model.

[0201] According to one aspect, the method may further include obtaining real-time or near real-time monitoring data from multiple sources and standardizing the real-time or near real-time monitoring data for input into a patient care model.

[0202] As used herein, an element or step recited in the singular and preceded by the word "a" or "an" should be understood as not excluding a plurality of the elements or steps, unless expressly stated to the contrary. Further, a reference to "one embodiment" of the present invention is not intended to be construed as excluding the existence of additional embodiments that also incorporate the recited features. Additionally, unless expressly stated to the contrary, an embodiment that "comprises," "includes," or "has" an element or elements with a particular characteristic may include additional such elements that do not have that characteristic. The terms "comprising" and "in which" are used as the plain-language equivalents of the corresponding terms "including" and "wherein." Further, the terms "first," "second," and "third," etc. are used merely as labels and are not intended to impose numerical requirements or a particular positional order on their objects.

[0203] This written description uses examples to disclose the invention, including the best mode, and also enables one of ordinary skill in the relevant art to practice the invention, including making and using any device or system and performing any included method. The scope of the invention that can be patented is defined by the claims and may include other examples that occur to one of ordinary skill in the art. Such other examples are intended to fall within the scope of the claims if they have structural elements that are not different from the literal language of the claims, or if they include equivalent structural elements with insubstantial differences from the literal language of the claims.

Claims

1. A medical system, comprising: one or more patient monitoring devices, each patient monitoring device configured to generate patient monitoring data by monitoring a physiological parameter of a patient; at least one database storing medical information corresponding to the patient; and An electronic device, comprising: power supply; a communication interface configured to communicate with the one or more patient monitoring devices and the at least one database via a network; a human-machine interface configured to provide information to a user and obtain information from the user; a memory configured to store instructions; and at least one processor configured to execute the instructions to: obtaining a care plan for the patient through the communication interface; obtaining care recommendations by inputting a diagnosis and one or more medical information corresponding to the patient, monitoring data corresponding to the patient, and medical facility inventory information into a patient care model; comparing the care recommendations to the care plan to obtain care insights, the care insights including recommended care not included in the care plan or recommended removal of care included in the care plan; outputting information corresponding to the nursing insight through the human-machine interface, and In response to user input selecting the care insight through the human-machine interface, the care plan is modified based on the care insight.

2. A system, comprising: power supply; a communication interface configured to communicate with one or more patient monitoring devices and at least one database via a network; a human-machine interface configured to provide information to a user and obtain information from the user; a memory configured to store instructions; and at least one processor configured to execute the instructions to: obtaining a care plan for the patient through the communication interface; obtaining care recommendations by inputting a diagnosis and one or more medical information corresponding to the patient, monitoring data corresponding to the patient, and medical facility inventory information into a patient care model; comparing the care recommendations to the care plan to obtain care insights, the care insights including recommended care not included in the care plan or recommended removal of care included in the care plan; outputting information corresponding to the nursing insight through the human-machine interface, and In response to user input selecting the care insight through the human-machine interface, The care plan is modified based on the care insights.

3. The system of claim 2, wherein the at least one processor is configured to execute the instructions to: obtain the care recommendations by obtaining a patient monitoring plan from the patient care model.

4. The system of claim 3, wherein the patient monitoring plan comprises parameters to be monitored and timing information for monitoring the parameters.

5. The system of claim 3, wherein the at least one processor is configured to execute the instructions to obtain the patient monitoring plan from a monitoring sub-model of the patient care model, the monitoring sub-model being trained based on previous user input.

6. The system of claim 2, wherein the at least one processor is configured to execute the instructions to: obtain the care recommendations by obtaining a treatment plan from the patient care model.

7. The system of claim 6, wherein the treatment plan includes a type of medical treatment, an amount of the type of treatment, and timing information for the treatment.

8. The system of claim 6, wherein the at least one processor is configured to execute the instructions to obtain the treatment plan from a treatment sub-model of the patient care model, the treatment sub-model being trained based on previous user input.

9. The system of claim 2, wherein the at least one processor is configured to execute the instructions to obtain the care recommendations by obtaining an equipment list from the patient care model.

10. The system of claim 9, wherein the equipment list includes at least one of equipment for monitoring the patient, equipment for treating the patient, and patient care equipment.

11. The system of claim 9, wherein the at least one processor is configured to execute the instructions to obtain the equipment list from an equipment sub-model of the patient care model, the equipment sub-model being trained based on previous user input.

12. The system of claim 2, wherein the at least one processor is configured to execute the instructions to: based on receiving additional medical information and / or additional patient monitoring data: obtaining updated care recommendations by inputting the additional medical information and / or the additional patient monitoring data into the patient care model; comparing the updated care recommendations to the care plan to obtain updated care insights, the updated care insights including recommended care not included in the care plan; outputting information corresponding to the updated nursing insights, and In response to user input selecting the updated care insight of the updated care recommendation, the care plan is modified based on the updated care insight.

13. The system of claim 2, wherein the at least one processor is configured to execute the instructions to: at predetermined time intervals: obtaining updated care recommendations by inputting additional medical information and / or additional patient monitoring data into the patient care model, the additional medical information and / or additional patient monitoring data being created after obtaining the care recommendations; comparing the updated care recommendations to the care plan to obtain updated care insights, the updated care insights including recommended care not included in the care plan; Information corresponding to the updated care insights is output, and in response to user input of the updated care insights selecting the updated care recommendations, the care plan is modified based on the updated care insights.

14. The system of claim 2, wherein the at least one processor is configured to execute the instructions to: generating a differential diagnosis list for the patient by a differential diagnosis model, the differential diagnosis list being obtained by inputting at least one of diagnostic medical information corresponding to the patient and diagnostic monitoring data corresponding to the patient into the differential diagnosis model, the differential diagnosis list including one or more diagnoses; and The diagnosis is obtained from the differential diagnosis model based on a user selection.

15. The system of claim 2, wherein the user input selecting the care insight comprises user input modifying the care insight.

16. A patient care assistance method, comprising: obtain a plan of care specific to the patient; obtaining care recommendations by inputting a diagnosis and one or more medical information corresponding to the patient, monitoring data corresponding to the patient, and medical facility inventory information into a patient care model; comparing the care recommendations to the care plan to obtain care insights, the care insights including recommended care not included in the care plan or recommended removal of care included in the care plan; outputting information corresponding to the care insights, and In response to user input selecting the care insight through the human-machine interface, the care plan is modified based on the care insight.

17. The method of claim 16, wherein obtaining the care recommendations comprises obtaining a patient monitoring plan, a treatment plan, and an equipment plan from the patient care model.

18. The method of claim 17, further comprising obtaining user input for each of the patient monitoring plan, the treatment plan, and the equipment plan.

19. The method of claim 16, further comprising one or more of: i) in response to receiving additional medical information and / or additional patient monitoring data and ii) at predetermined time intervals: obtaining updated care recommendations by inputting the additional medical information and / or the additional patient monitoring data into the patient care model; comparing the updated care recommendations to the care plan to obtain updated care insights, the updated care insights including recommended care not included in the care plan; Information corresponding to the updated care insights is output, and in response to user input of the updated care insights selecting the updated care recommendations, the care plan is modified based on the updated care insights.

20. The method of claim 16, wherein the monitoring data comprises data from a plurality of real-time or near real-time data feeds generated by a plurality of sources.

Images