Patent Application
20190006041
With the increasing trend toward building interoperable systems, data from disparate sources are aggregated, allowing for prevalent electronic use of previously underutilized clinical markers for disease, alone or in combination with more traditional clinical markers. Such previously underutilized diagnostic-related information can be extracted from radiology reports and combined with clinical markers of disease, such as fever, tachycardia, or low oxygen levels in blood, to evaluate a patient's health status. Radiology reports include results of a reading of an imaging exam for a patient, and may also include information regarding suggested follow-up and recommendations. Chest x-ray (CXR) reports, a type of radiology report, provide vital information for diagnosing clinical conditions, including for example, pneumonia, acute respiratory distress syndrome (ARDS), and cardiogenic pulmonary edema (CPE). Infiltrate information is a type of diagnostic-related information that may be extracted from a chest x-ray report. However, diagnostic-related information in a CXR report is frequently buried in the text of a report, which requires a skilled reader to understand and interpret. Furthermore, the report reader may introduce bias into the report interpretation, depending on his or her field of experience and level of expertise.
Electronic extraction of diagnostic-related information is difficult, as it is resource-intensive to code the diagnostic-related information in real-time and define a standard set of keywords for reviewing a CXR report. Moreover, annotation of CXR reports is often not disease specific, and keyword searches are often insufficiently sensitive if the searches do not include a sufficient number of search terms. Keyword searches are frequently insufficiently specific when the searches do not include natural language processing considerations like the progression of a condition, or the negation of identified diagnostic-related information and keywords.
A method for extracting information on infiltrates from an imaging report, and applying the report infiltrate information to clinical guidelines and clinical support applications, comprising: receiving the imaging report; parsing the report for infiltrate-related keywords; classifying the report based on the infiltrate-related keywords and at least one of directional words indicating a location of the infiltrates, negation words, and words canceling the negation words; presenting the classified report on a dashboard for clinical decision-making; and applying the classified report in a real-time clinical application.
A system for extracting information on infiltrates from an imaging report, and applying the report infiltrate information to clinical guidelines and clinical support applications, comprising: a non-transitory computer readable storage medium storing an executable program; a memory storing a keyword database, including infiltrate-related keywords, directional words, and negation words; and a processor executing the executable program to cause the processor to: receive the imaging report; parse the report for the infiltrate-related keywords; classify the report based on the infiltrate-related keywords and at least one of the directional words indicating a location of the infiltrates, the negation words, and words canceling the negation words; present the classified report on a dashboard for clinical decision-making; and apply the classified report in a real-time clinical application.
A non-transitory computer-readable storage medium including a set of instructions executable by a processor, the set of instructions, when executed by the processor, causing the processor to perform operations, comprising: receiving an imaging report; parsing the report for infiltrate-related keywords; classifying the report based on the infiltrate-related keywords and at least one of directional words indicating a location of the infiltrates, negation words, and words canceling the negation words; presenting the classified report on a dashboard for clinical decision-making; and applying the classified report in a real-time clinical application.
The exemplary embodiments may be further understood with reference to the following description and the appended drawings, wherein like elements are referred to with the same reference numerals. The exemplary embodiments relate to a system and methods for extracting infiltrate information from chest x-ray reports, classifying the reports, and applying infiltrate information from classified chest x-ray reports to a real-time clinical application for detecting and predicting acute illness. Although exemplary embodiments specifically describe chest x-rays, it will be understood by those of skill in the art that the system and method of the present disclosure may be used to extract infiltrate information from reports of any type of study or exam requiring a reading, classify the reports according to the infiltrate information, and employ classified reports within illness detection and prediction applications, within any of a variety of hospital settings.
As shown in
The memory 107 includes a database 108, which stores prior and current chest x-ray reports for a patient, and a keyword database, including infiltrate-related keywords, directional words, negation words, and words canceling the negation words, etc. For example, exemplary databases may include electronic medical records (EMR), electronic health records (EHR), cloud-based databases, or data warehouses, etc. The processor 104 may be implemented with engines and modules including, for example, a calculation engine 105 and a clinical decision support (CDS) manager 106.
The calculation engine 105 extracts infiltrate information from a chest x-ray report and classifies the report based on presence, pathology, and location of infiltrates on the chest x-ray. The calculation engine 105 uses the classified reports to render user dashboards, select clinical decision support (CDS) algorithms, run computerized clinical guidelines, run advanced CDS algorithms, apply CDS algorithm selection rules for scheduling CDS algorithms, and check for compliance with medical policies. In an exemplary embodiment, the calculation engine 105 places the classified infiltrate report on a user dashboard or chart, along with medical test information on user interface 102 and display 110, to aid a medical care provider in clinical decision making. In another exemplary embodiment, the classified chest x-ray infiltrate report may be presented to a medical care provider via electronic health medical records on user interface 102 and display 110, to aid a medical care provider in clinical decision making.
In a further exemplary embodiment, the calculation engine 105 applies the classified infiltrate reports in computerized clinical guidelines, and runs the computerized guidelines to test for a specific disease. In another exemplary embodiment, the calculation engine 105 applies the classified infiltrate report in advanced clinical decision support (CDS) applications, to test for a specific disease. For example, the advanced CDS application may be a statistical model testing for a particular disease or clinical condition.
In another exemplary embodiment, the clinical decision support (CDS) manager 106 applies the classified reports to determine a schedule for running CDS algorithms, based on the infiltrate classification of the reports. In a further exemplary embodiment, the calculation engine 105 applies the classified infiltrate report to provide notification of disease or clinical conditions in a health institution. In another exemplary embodiment, the calculation engine 105 applies the classified infiltrate report to detect the presence of disease or a clinical condition in a health institution, or check for compliance with medical policies, including for example, hospital policies, safety protocols, or quality protocols.
In step 201, the calculation engine 105 receives a chest x-ray report. In step 202, the calculation engine 105 extracts infiltrate information from the chest x-ray report. In step 204, the calculation engine 105 classifies the report based on the extracted infiltrate information. In step 206, the calculation engine 105 presents the classified infiltrate report to a medical care provider to aid in clinical decision making or determining compliance with hospital policy.
In step 208, the calculation engine 105 places the classified infiltrate report on a user dashboard or chart, along with medical test information, e.g. signs and symptoms of illness, patient demographics, patient vital signs, and biomarkers of disease or clinical conditions, etc. In step 209, the medical care provider makes a clinical decision based on the classified infiltrate report and the additional medical information on the dashboard. For example, while reviewing the user dashboard that presents the classified infiltrate report indicating the presence of unilateral infiltrates, elevated temperature (a fever), and presence of high microbe count of K. pneumoniae bacteria, a medical care provider may deduce the presence of pneumonia.
Returning to
In step 212, the calculation engine 105 applies the classified infiltrate report in advanced clinical decision support applications. In step 213, the classified infiltrate report is placed into a model that tests for a specific disease or clinical condition. In an exemplary embodiment, the advanced clinical decision support application may include an early detection model for acute respiratory distress syndrome (ARDS). For example, the model provides early detection of ARDS presence.
In step 214, the clinical decision support (CDS) manager 106 determines a schedule for running one or more CDS algorithms, based on the presence and classification of infiltrates on the CXR report. For example, the CDS manager 106 may optimize its performance and scheduling by automatically running only the relevant CDS algorithms. In step 215, the CDS manager 106 checks whether infiltrates are found in the report, using the CXR infiltrates algorithm. In step 219, when no infiltrates are present in the report, the CDS manager 106 does not run the CDS algorithm requiring CXR infiltrates as an input.
In step 217, the CDS manager 106 checks whether bilateral infiltrates are found in the report, using the CXR infiltrates algorithm. In step 216, when infiltrates are present in the report, or infiltrates are present but bilateral infiltrates are not present, the CDS manager 106 automatically runs only the relevant CDS algorithm, e.g. the algorithm that requires CXR infiltrates as an input. CDS algorithms that require CXR infiltrates as input include for example, algorithms for the detection of congestive heart failure (CHF), pneumonia, ARDS prediction, and tuberculosis.
In step 218, when bilateral infiltrates are present in the report, the CDS manager 106 automatically runs only the CDS algorithm that requires bilateral CXR infiltrates as an input. CDS algorithms that require bilateral CXR infiltrates as input including for example, ARDS guidelines and an algorithm for the ARDS-CPE exclusion model.
In step 220, the calculation engine 105 applies rules to check for clinical conditions in a health institution, or to check for compliance with medical policies, including, for example, hospital policies, or protocols for quality and safety reporting.
In step 302, the calculation engine 105 detects keywords in the report relating to infiltrates, including, for example, infiltrates, consolidation, pneumonia, inflammation, infection, opacities, edema, multilobular, flooding, and densities, etc.
In step 303, the calculation engine 105 checks for the presence of infiltrate-related keywords. In step 304, when no infiltrate-related keywords are present in any phrase of the report, the report is classified as “no infiltrates.” In step 305, when infiltrate-related keywords are present, the calculation engine 105 checks for the presence of directional words in the phrase containing the keyword. Exemplary directional words include, for example, bilateral, unilateral, bibasilar, both, right, and left, etc. In step 306, if there are no directional keywords in the phrase containing the keyword, the phrase is classified as “undecided infiltrates.” This “undecided infiltrates” classification indicates that infiltrates are present in the phrase, and specifies that the particular location or side of the infiltrates is not clear. In step 310, the phrases classified as “undecided infiltrates” are stored with all non-negated phrases in the report. In step 311, the calculation engine 105 assigns a class to the report. If there are no other phrases containing keywords in the report, beside the phrases classified “undecided infiltrates,” the calculation engine 105 classifies the report as “undecided infiltrates” in step 311. In another exemplary embodiment, rules are applied to further classify the phrase or report classified as “undecided infiltrates” as bilateral or unilateral.
In step 307, in a further exemplary embodiment, once the calculation engine 105 has detected directional words in the phrase with the keyword, the calculation engine checks for negation words in the phrase, including for example, no, not, negative, resolved, and without, etc. When no negation words are detected in the phrase with the keyword and directional words, in step 310, the keyword and direction from the directional words are not negated, e.g. the phrase has a keyword and directional words that are not ignored. Next, the non-negated keyword and direction for each phrase is stored, for all phrases with a non-negated keyword and direction in the report.
In step 308, once negation words are detected in the phrase, the calculation engine 105 checks for words that cancel the negation words in the phrase containing the keyword, including, for example, multiple negatives and adjectives, e.g. partially, slightly, not resolved, and no resolution, etc.
In step 309, when no words that cancel the negation words are detected in the phrase containing both the keyword and negation words, the keyword and direction are negated, e.g. ignored. For example, when a phrase contains a keyword “infiltrates,” the directional word “left,” negation words, and no words canceling the negation words, the keyword “infiltrates” and the direction of “left” are ignored.
In step 310, when words that cancel the negation words are detected in the phrase containing the keyword, directional words, and negation words, the negation words are canceled, and the keyword and direction from the directional words in the phrase are not negated, e.g. the keyword and directional words in the phrase are not ignored. Next, the non-negated keyword and direction for each phrase is stored, for all phrases with a non-negated keyword and direction in the report. In step 311, the calculation engine 105 assigns a class to the report, based on the keyword and directional words in the phrases of the report. Exemplary classifications include: “bilateral,” “unilateral,” or “undecided infiltrates,” etc.
In step 418, when the directional word “unilateral” is not present, the calculation engine 105 determines whether the directional word “left” is in the phrase. In step 419, when the directional word “left” is not present, the calculation engine 105 checks if the directional word “right” is in the phrase. In step 420, when the directional word “left” is present, the calculation engine 105 checks if the directional word “right” is also present. In step 421, when the directional words “left” and “right” are both not present, the calculation engine 105 classifies the report as “undecided infiltrates,” which indicates that infiltrates are present, but the location or side of the infiltrates is unclear. In step 422, when the directional word “left” is present and the word “right” is not present, or when the directional word “right” is present and the word “left” is not present, the calculation engine 105 classifies the report as “unilateral.” In step 423, when the directional words “left” and “right” are both present, the calculation engine 105 classifies the report as “bilateral.”
Table 1 below also presents exemplary rules determining the class assignment of the chest x-ray reports according to the presence of the directional words in the phrase.
In an exemplary embodiment, as depicted in steps 502-507, the calculation engine 105 applies the classified reports to run a computerized clinical guideline. For example, a computerized clinical guideline may be the guidelines for the detection of acute respiratory distress syndrome (ARDS). An exemplary computerized clinical guideline for detecting VAP is shown in steps 502-507. As another example, the classified infiltrate report could be used in the American-European Consensus Conference (AECC) and Berlin clinical guidelines for acute respiratory distress syndrome (ARDS), to test for the presence and severity of ARDS. The ARDS guidelines include for example, definitions of ARDS and are coded to the presence of bilateral infiltrates in the report, along with lab test results and ventilator settings. In another exemplary embodiment, the classified infiltrate reports could be used with clinical guidelines, e.g. detection of pneumonia, ventilator-associated pneumonia (VAP), cardiogenic pulmonary edema (CPE). In a further exemplary embodiment, the reports could be used for computerized clinical protocol-driven therapy recommendations.
An exemplary computerized clinical guideline for detecting VAP is shown in steps 502-507. In step 502, the calculation engine 105 checks whether infiltrates are present on two or more chest x-ray reports. In step 503, when infiltrates are not found in two or more reports, the ventilator-associated pneumonia (VAP) is not detected. In step 504, when infiltrates are present on two or more reports, the guideline for detecting a specific disease, for example, VAP, is applied. In step 505, the calculation engine 105 checks whether the parameters of the specific disease guideline are met. For example, in the VAP guideline, the calculation engine 105 checks for the presence of one of the following parameters: white blood cell count (WBC) of 1000 cells/mm3 less than 4, or greater than or equal to 12, age over 70 years and Glasgow Coma Scale (GCS) less than 13, or temperature greater than 38° C. In step 507, when the disease detection guideline parameters are satisfied, the disease is detected. For example, when any one of the above parameters is present, the calculation engine 105 checks for the presence of two of the following conditions: respiratory rate (RR) greater than 20, arterial oxygen partial pressure to fractional inspired oxygen ratio (PaO2/FiO2) less than or equal to 240, and ventilator required. When two of these conditions are present, VAP is detected. In step 506, when the disease detection guideline parameters are not satisfied, the disease is not present. For example, when either the first set of VAP detection parameters (WBC, age, GCS, temperature) or the second set of parameters (RR, PaO2/FiO2, ventilator requirement) is not satisfied, VAP is not detected.
In another exemplary embodiment, in step 509, the classified infiltrate reports may be used in advanced clinical decision support (CDS) applications, for example, in an early detection model for acute respiratory distress syndrome (ARDS). In this exemplary embodiment, in step 510, the classified infiltrate reports may be input into a model to test for a specific disease. For example, the CXR infiltrate classifications (no infiltrate, undecided, unilateral, bilateral) may each be coded with distinct numerical values, and the values for the classified reports may be input into a logistic regression or time series model with clinical information on patient vital signs, patient hospital status (admission, discharge, transfer) known as “ADT,” and Intensive Care Unit (ICU) patient admission source. For example, this logistic regression model provides early detection of ARDS presence.
In a further exemplary embodiment, the classified infiltrate reports may be used to schedule and run relevant clinical decision support (CDS) algorithms. In step 511, when the CXR infiltrates algorithm determines that infiltrates are present in the report, the CDS manager 106 checks for the presence of bilateral infiltrates in the report. In step 512, when the CXR infiltrates algorithm determines that infiltrates are present in the report or that infiltrates are present, but bilateral infiltrates are not present, the CDS manager 106 runs only CDS algorithms relevant to the infiltrates in the report, for example, CDS algorithms that require infiltrates as an input, e.g. CDS algorithms for detecting congestive heart failure (CHF), pneumonia, ARDS prediction, and tuberculosis.
In step 513, when the CXR infiltrates algorithm determines that bilateral infiltrates are present, the CDS manager 106 runs only CDS algorithms relevant to the infiltrates in the report, for example, CDS algorithms that require bilateral infiltrates as an input, e.g. CDS algorithms for ARDS guidelines, or an ARDS-CPE exclusion model. In an exemplary embodiment, in the Intensive Care Unit (ICU), medical care providers may need to determine whether edema in the lung is present due to cardiogenic or non-cardiogenic causes. For example, when bilateral infiltrates are found in the report, the CDS manager 106 runs the ARDS-CPE exclusion model, which eliminates a cardiogenic cause for the edema, when the model indicates ARDS (a non-cardiogenic cause) is suspected to be present.
In step 603, when infiltrates are detected in the report, the calculation engine 105 checks for signs of a specific disease, for example, pneumonia. In an exemplary embodiment, the detection of infiltrates and signs of a specific clinical condition or disease could be used to report the presence of the clinical condition or disease in a health institution. In step 604, the calculation engine 105 determines whether the clinical procedure for the specific disease was followed. For example, in the case of a procedure for pneumonia, the calculation engine 105 first checks for ventilation, and then checks whether oral antiseptic has been ordered. That is, the calculation engine checks if any of the clinical procedure steps were not followed. In step 605, when the calculation engine determines that the clinical procedure for the specific disease was not followed, the calculation engine identifies the procedure as non-compliant with the hospital indication policy. For example, when one or both steps of the clinical procedure for pneumonia were not followed, the calculation engine identifies the procedure as non-compliant with hospital policy.
In step 606, when the clinical procedure for the specified disease was followed, the calculation engine 105 identifies the procedure as compliant with the hospital indication policy. For example, when all required steps of the procedure for pneumonia were followed, the procedure is identified as compliant with hospital policy.
In step 607, when the infiltrates are detected in the chest x-ray report in step 601, the calculation engine 105 then determines whether the detection had proper indications under the hospital policy. For example, the detection of infiltrates on a report may be a trigger for the calculation engine 105 to check whether the patient meets criteria for the hospital indication for chest x-ray policy, e.g. the calculation engine 105 runs a set of rules to see whether the patient meets necessary criteria that would have required a doctor to order a chest x-ray. As an example of a proper indication under the hospital policy, under the hospital's indication policy for chest x-rays, suspected pneumonia may be a valid criterion for ordering chest x-rays.
When the infiltrates are detected in the chest x-ray report in step 601 and the detected infiltrates are confirmed as having proper indications under the hospital policy in step 607, in step 609, the calculation engine 105 identifies the procedure as compliant with the hospital indication policy, a good use of resources, and an effective indication policy. Returning to the above example, the identification of infiltrates in a report may be the hospital indication requiring a doctor to order a chest x-ray, under the hospital indication policy. Thus, for example, when infiltrates are detected in a chest x-ray report in step 601, in step 607, it is determined that the use of chest x-rays complies with the hospital indication procedure for pneumonia, for ordering chest x-rays when infiltrates are detected in a report. In this example, when the physician orders a chest x-ray in compliance with hospital policy (e.g., patient had suspected pneumonia), and the patient's chest x-ray has infiltrates, then the hospital policy was effective in recommending the chest x-ray order. For example, quality officers at the institution may use the method to continuously monitor the effectiveness of hospital policies over time, and amend the policies as necessary, to improve the clinical workflow and quality of care provided to patients at the institution.
In step 608, once the calculation engine 105 determines that the detection of infiltrates did not have proper indications for performing a specific procedure under the hospital policy in step 607, the calculation engine identifies the procedure as non-compliant with the hospital indication policy, where the procedure may signal ineffective hospital indication for ordering the procedure. For example, when the positive detection of infiltrates is not a valid criterion for the hospital indication policy of using chest x-rays, e.g., in checking for asthma, the use of the chest x-ray is identified as non-compliant with hospital indication policy, and ineffective hospital indication for ordering the x-ray for asthma. In this example, the hospital x-ray order indication policy may be ineffective because the asthma cases show detected infiltrates, but the asthma cases are not covered by the current chest x-ray order indication policy, which may signal a need to change the chest x-ray order indication policy. Further in this example, when the physician orders a chest x-ray out of compliance with hospital policy (e.g., patient had asthma, but did not have suspected pneumonia), and the patient's chest x-ray had infiltrates, then the hospital policy for ordering chest x-rays should be corrected, if this scenario of detected infiltrates that is not compliant with hospital chest x-ray ordering policy is found for many patients' chest x-rays in the institution.
In another exemplary embodiment, once the calculation engine 105 identifies the procedure as compliant with the hospital indication policy in step 607, the calculation engine checks for the presence of infiltrates in the report in step 601. In step 602, when no infiltrates are detected despite a proper indication under hospital policy, the calculation engine 105 identifies a potential wasted medical resource, signaling an ineffective indication policy for ordering the chest x-ray. For example, after the calculation engine 105 determines the chest x-ray report results are compliant with the hospital policy indication in step 607 for requiring a doctor to order a chest x-ray, in step 601, the absence of detected infiltrates in the report is identified. As an example, the calculation engine 105 may determine in step 602 that the ordered chest x-ray is a potential wasted medical resource, where the chest x-ray may not have been required, for a lower-severity lung condition where infiltrates are not detected.
In a further exemplary embodiment of the invention, the calculation engine 105 may extract information on alternate risk factors, such as cardiac anomalies, for example, from the chest x-ray reports, to input into other predictive algorithms.
Those of skill in the art will understand that the above-described exemplary embodiments may be implemented in any number of matters, including, as a software module, as a combination of hardware and software, etc. For example, the exemplary method 200, exemplary method 300, exemplary method 400, exemplary method 500, exemplary method 600, calculation engine 105, and clinical decision support (CDS) manager 106 may be programs containing lines of code that, when compiled, may be executed by a processor.
Those skilled in the art will understand that the engines 105-106 may be implemented by the processor 104 as, for example, lines of code that are executed by the processor 104, as firmware executed by the processor 104, as a function of the processor 104 being an application specific integrated circuit (ASIC), etc.
It will be apparent to those skilled in the art that various modifications may be made to the exemplary embodiments, without departing from the spirit or the scope of the invention. Thus, it is intended that the present invention cover modifications and variations of this invention provided they come within the scope of the appended claims and their equivalents.
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/EP2016/069696 | 8/19/2016 | WO | 00 |
| Number | Date | Country | |
|---|---|---|---|
| 62210023 | Aug 2015 | US |
