Patent Application
20250005522
The following relates generally to the imaging arts, imaging device maintenance arts, service work order arts, artificial neural network (ANN) arts, and related arts.
Diagnostic medical imaging systems are complex devices having various components. If one or more of these components fail, the imaging system might deliver sub-optimal results (i.e., deteriorated image quality) or become completely unusable. The former case can lead to misdiagnosis or radiologists' productivity losses (e.g., more difficult result interpretation), while the latter case might result in reduced patient throughput. All scenarios can cause financial losses to the medical institution employing the device. From the patient's side, resulting misdiagnosis poses a potential health risk whilst system down-time may entail an increase in waiting times.
To ensure system down-time is kept to a minimum, timely diagnosis of component fault is key. Among several approaches used to diagnose medical system component failures, the typical one consists of scheduling planned visits of a service engineer, who in turn will perform a set of tests to evaluate the status of the machine. To avoid such resource demanding and cost-ineffective options, automatic approaches are also employed using remote log data storage transmitted from the system (i.e., on a dedicated server). In this case, the system assessment for certain component failures is done remotely to output existing failures (proactive models) or to predict future failures (i.e., predictive models).
The validation of either system diagnostic model type is sometimes done by assessing Service Work Order (SWO) reports, which indicates when a service action related to a particular system component was performed, and the failure mode addressed by the service action. The SWO reports can thus be mined to generate training data for training a diagnostic model. Unfortunately. SWO reports are often incomplete, inaccurate, or not sufficiently detailed to assess whether a part replacement was justified in relationship to a given failure mode. Moreover, the text field content of an SWO report (e.g., language of the description field) may be region dependent. The low quality of the contents of SWO reports makes it problematic to mine these reports for high quality training data to develop and validate models aiming at device failure detection.
Beyond the application of mining SWO reports for training diagnostic models, there are other reasons why an SWO report that is incomplete, inaccurate, or insufficiently detailed can be problematic. Service engineers conducting a current service call may consult a SWO report for a prior service call for information that may be pertinent to the current service call. Deficiencies in the prior SWO report can limit its usefulness for this purpose. Failure to record part replacement in an SWO report can also result in the part being replaced again prematurely.
The following discloses certain improvements to overcome these problems and others.
In one aspect, a non-transitory computer readable medium stores instructions executable by at least one electronic processor to perform a service work order (SWO) method. The method includes via a user interface (UI), receiving entry of a SWO report; applying at least one automated analysis to the SWO report to detect information missing from the SWO report and/or to generate a completeness score for the SWO report; via the UI, providing an indication of the information missing from the SWO report and/or the completeness score for the SWO report; and storing the SWO report in a SWO database.
In another aspect, a non-transitory computer readable medium stores instructions executable by at least one electronic processor to perform a service work order (SWO) method. The method includes via a UI, receiving entry of a SWO report; applying at least one automated analysis to the SWO report to generate a completeness score for the SWO report; via the UI, providing an indication of the completeness score for the SWO report; and storing the SWO report in a SWO database.
In another aspect, a non-transitory computer readable medium stores instructions executable by at least one electronic processor to perform a service work order (SWO) method. The method includes via a UI, receiving entry of a SWO report; applying at least one automated analysis to the SWO report to detect information missing from the SWO report; via the UI, providing an indication of the information missing from the SWO report: and storing the SWO report in a SWO database.
One advantage resides in providing a user interface (UI) for filling in an SWO report template and visualizing completeness or correctness information for various template fields.
Another advantage resides in enhancing the UI to provide similar SWO report-based recommendations to a field service engineer (FSE).
Another advantage resides in evaluating complex textual fields of an SWO report respective to completeness or other quality metrics.
Another advantage resides in generating recommendations for SWO report fields based on curated data.
Another advantage resides in extracting text patterns from an SWO report and clustering the text patterns around parts and failure modes of a medical device.
A given embodiment may provide none, one, two, more, or all of the foregoing advantages, and/or may provide other advantages as will become apparent to one of ordinary skill in the art upon reading and understanding the present disclosure.
The example embodiments are best understood from the following detailed description when read with the accompanying drawing figures. It is emphasized that the various features are not necessarily drawn to scale. In fact, the dimensions may be arbitrarily increased or decreased for clarity of discussion. Wherever applicable and practical, like reference numerals refer to like elements.
Data mining of service work order (SWO) reports to generate training data for training diagnostic models, or to identify ways to improve medical imaging device servicing practices and procedures, is desirable. However, these procedures have been hindered by poor quality of the SWO reports. Ideally, these reports would provide detailed information in which all fields of the SWO report template are filled in. Especially, free-form text fields would ideally be filled in with sufficient detail to capture all salient aspects of the service work. In practice, however, fields of the SWO reports are often left empty or incomplete, and in particular the information entered into the free-form text fields may be cursory or lack important details.
Recognizing these problems, the following proposes a data quality improvement system that operates in real-time as the service engineer (SE) fills out the SWO report. A curated database of high quality SWO reports is collected. This curated database is then used to train an artificial neural network (ANN) model or set of ANN models to assign completeness scores to SWO reports. In some embodiments. ANN models or other diagnostics (e.g., natural language processing or “NLP” analyses) are also developed to score specific free-form text entry fields and/or to identify specific types of incomplete information for the specific free-form text entry fields.
Because the ANN and/or NLP models are operating on text, they are fast and can be applied in (near) real-time. For example, when an SE submits an SWO report these models can be run against the report, and if it scores below some acceptable threshold (e.g. a threshold in a range of 80%-90% in some nonlimiting examples), then the SE can be informed of the deficiency and encouraged (or required) to update the SWO report accordingly. Similarly, if specific incomplete information in a specific free-form text entry field is identified by an ANN or NLP model, then a recommendation to add that information can be provided.
The curation of the curated database of high quality SWO reports can be done manually. or in an automated or semi-automated fashion. In some embodiments, once the system is running any SWO report with a score above some excellence threshold may be added to the curated database, and the ANN model may be retrained occasionally with the expanded database, or based on existing data or model quality monitors. In another variant, the SE can also grade the scoring and/or recommendations provided by the system and this feedback can also be used to update the ANN models. Other types of voting schemes for ranking SWO reports and adding high quality reports to the curated database are also contemplated.
The following also discloses the curated database of high quality SWO reports can also be beneficially leveraged during the subsequent data mining. For example, NLP can be used to decompose free-form text entry field content of the curated SWO reports to extract technology labels for the reports, such as failure modes, and clustering around failure modes or other technology aspects can be used to derive data pools of curated SWO reports related to specific failure modes that can then be analyzed by data mining techniques, such as training and validating diagnostic models under development.
With reference to
The service device 102 includes a display device 105 via which servicing-related information such as log data, prior SWO reports, a parts ordering interface. alerts generated by predictive failure models, an SWO entry user interface (UI), and/or so forth are displayed. The service device 102 also preferably allows the service engineer to interact with the servicing support system via at least one user input device 103 such a mouse, keyboard, or touchscreen. The service device further includes an electronic processor 101 and non-transitory storage medium 107 (internal components which are diagrammatically indicated in
In illustrative
With continuing reference to
To this end, the electronic processor 113 of the backend server 111 implements at least one artificial neural network (ANN) 134 (which can be trained) to detect information missing from an SWO report 136 (i.e., a “current SWO report 136 for a current servicing session) and/or to generate a completeness score 138 for an SWO report. The training of the ANN 134 includes using historical SWO reports having a completeness score 138 exceeding a predetermined completeness score threshold. Additionally or alternatively, NLP can be used for detecting missing information.
The non-transitory computer readable medium 127 stores instructions executable by the electronic processor 113 of the backend server 111 to provide a user interface (UI) 140 for display on the display device 105 of the service device 102.
The non-transitory storage medium 127 stores instructions executable by the electronic processor 113 of the backend server 111 to perform a SWO method 200 of generating and storing a SWO report following a servicing session of the medical device 120.
With continuing reference to
To begin the method 200, at an operation 202, the SWO report entry UI 140 is provided on the display device 105 of the SE electronic processing device 102, via which the FSE or RSE enters the current SWO report 136 for the current servicing session of the medical device 120. The current SWO report 136 is then received and stored at the backend server 111.
However, before saving or filing the SWO report, at least one automated analysis is applied to the SWO report 136 to detect information missing from the SWO report 136 and/or to generate a completeness score 138 for the SWO report 136. This can be performed in a variety of manners. In some embodiments, at an operation 204, the automated analysis includes detecting information missing from the SWO report 136. To do so, at least one natural language process (NLP) operation is applied to the SWO report 136 to detect information missing from the SWO report 136. In another embodiment, at an operation 206, the automated analysis includes generating a completeness score 138 for the SWO report 136.
In some embodiments, the NLP operation 204 and/or the completeness score operation 206 can be performed using the ANN 132. To do so, one or more historical SWO reports 136 are retrieved from the SWO database 111. A completeness score 138 according to a predetermined completeness score threshold is then generated for each retrieved historical SWO report 136. The scored historical SWO reports 136 that meet or exceed the predetermined completeness score threshold are then used to train the ANN 134 to perform the NLP operation 204 and/or the completeness score operation 206. In some embodiments, textual features can be extracted from the historical SWO reports 132 to provide labels for the historical SWO reports 132.
The scored historical SWO reports 136 that meet or exceed the predetermined completeness score threshold are also stored in the SWO database 111. In some embodiments, the scoring of the historical SWO reports 136 can include receiving a user input from the SE via the service device 102, in which the user input is indicative of the historical SWO report 134 should be stored in the SWO database 111.
The ANN 134 can be updated (i.e., retrained) with the addition of new historical SWO reports 132 (i.e., SWO reports 132 generated by other SEs during servicing sessions of other medical devices). To do so, additional historical SWO reports 132 are received at the SWO database 111. These additional historical SWO reports 132 are then scored with completeness scores 138 according to the predetermined completeness score threshold, and the additional historical SWO reports 132 with completeness scores 138 that meet or exceed the predetermined completeness score threshold are used to update the SWO database 111. The ANN 134 can then be retrained with these additional stored historical SWO reports 132.
At an operation 208, an indication 142 is provided. via the UI 140, of either the information missing from the current SWO report 136 and/or the completeness score 138 for the SWO report 134. In one example, the completeness score 138 for the current SWO report 136 can be provided on the UI 140. If the completeness score 138 for the current SWO report 136 does not exceed the predetermined completeness score threshold. then the indication 142 can include an indication for the SE to update the current SWO report 136. Depending upon the embodiment, this can take the form of a suggestion to update the report, or can be more strongly enforced, for example by not permitting filing or saving the SWO report until it is scored above the predetermined completeness score threshold. In another example, the indication 142 can include the detected missing information from the current SWO report 136. The indication 142 can further include suggest text to add to the current SWO report 136 allow the current SWO report 136 to exceed the predetermined completeness score threshold. Again, this can optionally be more strongly enforced, for example by not permitting filing or saving the SWO report until the missing information is added. The completed SWO report 136 can then be stored in the SWO database 111.
In some embodiments, the completed and verified SWO report 136 can be used to train a diagnostic model 144 (implemented in the SWO database 111) to determine a root cause of the medical imaging device 120 fault, and predict a maintenance operation to be performed on the medical imaging device 120. To do so, for example, the completed and verified SWO reports 136 stored in the SWO database 111 can be data-mined to determine a root cause of the medical imaging device 120 fault, and predict a maintenance operation to be performed on the medical imaging device 120. Advantageously, the completed and verified SWO reports 136 are used to improve the reliability of the analysis performed by the diagnostic model 144 in the mining operation.
The following describes the system 100 and the method 200 in more detail. The system 100 comprises two-side quality improvement system for SWO data used in system diagnostic model assessments. The first side considers ensuring better data quality at a source-side (i.e. during SWO creation by an SE) and the client-side (i.e. the time when key insights are extracted for model creation or validation by the backend server 111).
The source side part of the system 100 (i.e., the service device 102) aims at supporting the FSE during the process of compiling an SWO report 136 for a given problem. It comprises a standardized template for which the % of each field being complete and correct can be checked and recommendations can be provided based on entries of e.g. predefined options and other previous user entries.
The completeness and correctness for 1-entry fields (e.g. city, customer, serial numbers) is relatively trivial to judge, by checking whether an entry exists and then cross-referencing against proper field values (e.g. list of cities, known customers, correct serial number format using e.g. a regular expression). Additionally, the reference data can be used to support the filling out each of the specified 1-entry fields. The fields can also be pre-populated from the system log file data.
For more complex SWO fields like root cause description, additional info, etc. the quality assessment is done via a neural network model 132. This model 132 has been trained on curated SWO data from the field, which has been judged as complete and correct. It outputs a % score corresponding to the match with the curated data. A certain percentage (e.g. 80% or 90%) could be deemed as acceptable. Note that one can also consider generating reference data for these fields using system domain knowledge, in other words emulating ideal complex text fields for various components and failure modes.
For further quality improvement of the aforementioned fields, a recommender system can also be deployed to filling out suggestions. This system essentially searches for similar entries in the curated data and makes recommendations on filling out the text fields in question.
On the client side (i.e., at the backend server 111), one is interested in knowing to which component and failure mode the given SWO report 136 was related to and, most importantly, whether the associated action (check, part replacement, etc.) was justified. The client would search a given system component (e.g., x-ray tube) and get data on possible failure modes and associated text field patterns extracted from relevant historical SWO reports 134.
For the implementation of such functionality, employs Natural Language Processing (NLP) approaches to decompose SWO text fields (e.g., description) into separate text patterns. These text patterns are generated based on curated data used in the source-side part of the proposed quality improvement system. The curated data also provides information on associated failure modes. As a result, the decomposed text patterns are clustered based on failure mode.
The operation provides the user a pool of data that, on the one hand, has sufficient textual information in general, but also is clustered around failure modes based on various relevant text patterns. The data pool can be used to assess whether a given part replacement was justified by assessing the text patterns. Furthermore, the compiled database is readily usable for system diagnostic model development and verification.
In some embodiments, during filling out of the SWO, there is already a pre-population of information for meta data from the imaging device 120. A list of devices (e.g., RF coils, gradient systems etc.) is available from, for example, log data or configuration files from the site. In some examples, a unique identifier is needed for this information, such as a unique coil name (i.e., 12NC (order number)), a serial number, and so forth. The list can be a drop-down list from which a coil can be selected. Table 1 shows an example of an SWO report 136.
A non-transitory storage medium includes any medium for storing or transmitting information in a form readable by a machine (e.g., a computer). For instance, a machine-readable medium includes read only memory (“ROM”), solid state drive (SSD), flash memory, or other electronic storage medium; a hard disk drive, RAID array, or other magnetic disk storage media; an optical disk or other optical storage media; or so forth.
The methods illustrated throughout the specification, may be implemented as instructions stored on a non-transitory storage medium and read and executed by a computer or other electronic processor.
The disclosure has been described with reference to the preferred embodiments. Modifications and alterations may occur to others upon reading and understanding the preceding detailed description. It is intended that the exemplary embodiment be construed as including all such modifications and alterations insofar as they come within the scope of the appended claims or the equivalents thereof.
| Number | Date | Country | Kind |
|---|---|---|---|
| 22164296.0 | Mar 2022 | EP | regional |
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/EP2022/082528 | 11/21/2022 | WO |
| Number | Date | Country | |
|---|---|---|---|
| 63284083 | Nov 2021 | US |
