CROSS REFERENCE TO RELATED APPLICATIONS [0001] This application claims priority to, and the benefit of, co-pending U.S. provisional application entitled “Artificial intelligence System and Method for Generating Medical Impressions from Text-Based Medical Reports” having serial no. 63/159,066, filed March 10, 2021, which is hereby incorporated by reference in its entirety.

TECHNICAL FIELD

[0002] This disclosure relates to a system and method for using artificial intelligence (AI) to automatically generate medical impressions from text-based reports of medical findings.

BACKGROUND

[0003] AI is used in many applications and industries to make predictions based on historical data and to automate various processes based on the predictions. Machine learning (ML) is a subset of AI that involves the ability of a machine to learn from historical data and make decisions without having to be explicitly programmed. A variety of ML algorithms are currently used for this purpose, such as supervised learning algorithms, unsupervised learning algorithms, deep learning algorithms and reinforcement learning algorithms.

[0004] In clinical practice, medical documentation is a tedious task. Most medical professionals generate text-based reports by typing or using voice-recognition software. It is common for medical professionals to document specific medical findings first and then summarize their reports later, often generating an assessment and plan or conclusion without or with recommendations. For example, a radiologist may view medical images and use voice recognition software to dictate their medical findings report. After generating the medical findings report, the radiologist will spend time and effort to generate a conclusion, or impression, without or with recommendations. The radiologist often has to re-read sections of their report and/or recall or review national standards and guidelines to generate the conclusion and recommendations. A similar approach is used by other physicians that review images or conduct procedures (e.g., colonoscopy, pathology review, surgery, biopsy). Furthermore, many medical professions evaluate patients to generate a note in the electronic medical record (e.g., history and physical or routine clinical progress note), often ending the report with an assessment and plan. [0005] For each of these scenarios, the clinician typically documents findings in text form and then manually uses this information to generate a summary that could include a conclusion, assessment, plan, and/or recommendation.

[0006] In the field of radiology, ML has been used to generate a medical impression or conclusion based on a report of medical findings. For example, a known ML tool processes a digital text-based report of medical findings to generate a medical impression or conclusion based on the findings. A medical professional first trains an ML model to learn how to generate medical impressions or summaries based on the particular medical professional’s reports of medical findings. Once the tool is trained, the medical professional can dictate a medical findings report and then the ML tool converts the speech into a digital text-based report and processes the text- based report using the medical professional’s ML model to generate the impression.

[0007] While such ML tools are useful, the medical professional who prepares the medical findings report is often not the same medical professional who signs off on the medical impression section of the report. For example, in some cases, trainees may dictate reports of medical findings without knowing which attending physician will ultimately approve and sign off on the final report that includes both the medical findings and the impression or summary. Since each attending physician has their own dictating style, a trainee dictating the medical findings may have used verbiage that is more commonly used by Attending Physician A if the trainee believes that the final signor on the report is likely to be Attending Physician A.

However, if Attending Physician B ends up finalizing the report, she may not like the verbiage used in the impression section of the report.

[0008] A need exists for an AI system and method that enable more accurate and efficient medical impressions to be automatically generated from reports of medical findings, regardless of who generates the medical findings report section.

SUMMARY

[0009] Aspects of the present disclosure are related to the use of artificial intelligence for generation of medical impressions from medical reports. In one aspect, among others, an artificial intelligence (AI) system for generating impression reports from digital text-based medical findings reports comprises at least one processor comprising: first machine learning (ML) logic configured to perform one or more ML algorithms that train at least N AI models associated with N medical professionals, respectively, using medical reports associated with the N medical professionals, respectively; second ML logic configured to use the trained AI models to generate impressions based on digital text-based medical findings reports received by the second ML logic, the second logic receiving a designation of at least one of the N medical professional and selecting the AI model associated with the designated medical professional to be used to generate an impression from a digital text-based medical findings report received by the second ML logic; and third ML logic configured to use the selected AI model to process the digital text-based medical findings report received by the second ML logic to generate the impression. The AI system can comprise processing circuitry including the at least one processor.

[0010] In one or more aspects, a medical professional who generates the digital text-based medical findings report that is processed by the third ML logic can be different from the designated medical professional. A medical professional who generates the digital text-based medical findings report that is processed by the third ML logic can be a trainee and the designated medical professional is a trained physician. A medical professional who generates the digital text-based medical findings report that is processed by the third ML logic can be someone other than a physician and the designated medical professional is a trained physician. The designation of the medical professional can be made by the medical professional who generated the digital text-based findings that were processed by the third logic. The designation of the medical professional can be made by someone other than the medical professional who generated the digital text-based findings that were processed by the third logic. In various aspects, the third ML logic can use the selected AI model to perform natural language processing (NLP) when processing the digital text-based medical findings report received by the second ML logic to generate the impression. The third ML logic can use the selected AI model to detect and interpret laboratory data in the digital text-based medical findings report received by the second ML logic to generate the impression.

[0011] In another aspect, an artificial intelligence (AI) method for generating impression reports from digital text-based medical findings reports comprises, in first machine learning (ML) logic, performing one or more ML algorithms that train at least N AI models associated with N medical professionals, respectively, using medical reports associated with the N medical professionals, respectively; in second ML logic, receiving a designation of at least one of the N medical professional and selecting the AI model associated with the designated medical professional to be used to generate an impression from a digital text-based medical findings report received by the second ML logic; and in third ML logic, using the selected AI model to process the digital text-based medical findings report received by the second ML logic to generate the impression. In one or more aspects, the third ML logic can use the selected AI model to perform natural language processing (NLP) when processing the digital text-based medical findings report received by the second ML logic to generate the impression. The third ML logic can use the selected AI model to detect and interpret laboratory data in the digital text-based medical findings report received by the second ML logic to generate the impression.

[0012] In various aspects, a medical professional who generates the digital text-based medical findings report that is processed by the third ML logic can be different from the designated medical professional. A medical professional who generates the digital text-based medical findings report that is processed by the third ML logic can be a trainee and the designated medical professional is a trained physician. A medical professional who generates the digital text-based medical findings report that is processed by the third ML logic can be someone other than a physician and the designated medical professional is a trained physician. The designation of the medical professional can be made by the medical professional who generated the digital text-based findings that were processed by the third logic. The designation of the medical professional can be made by someone other than the medical professional who generated the digital text-based findings that were processed by the third logic.

BRIEF DESCRIPTION OF THE DRAWINGS [0013] 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. [0014] Fig. 1 is a flow diagram illustrating the work flow of the AI system and method in accordance with a representative embodiment.

[0015] Fig. 2 is a flow diagram showing some of the possible scenarios that can occur and benefit from using the AI system and method.

[0016] Fig. 3 is a block diagram of the AI system in accordance with a representative embodiment.

DETAILED DESCRIPTION

[0017] Illustrative embodiments are disclosed herein of an AI system and method that enable more accurate and efficient medical impressions to be automatically generated from reports of medical findings. The AI system and method automatically generate a summary, or impression, of a medical report from digitized text-based findings entered by a clinician. A user who may or may not be the person who generated the report of medical findings designates a medical professional who will be finalizing the report. The AI system selects the AI model that is associated with the medical professional who has been designated and uses the AI model to generate the impression section of the report.

[0018] Some of the main benefits of the AI system and method are improved efficiency, accuracy, and standardization and reduced errors and omissions. The inventive principles and concepts apply to a wide variety of medical professions. The summary or impression can include a conclusion, assessment, plan, and/or a recommendation. Using AI to automate the report summary or impression improves efficiency, accuracy and standardization while also reducing errors and omissions.

[0019] For ease of discussion, the report summary, impression, conclusion, assessment, plan and/or recommendation will be referred to herein collectively as the impression. The medical findings and the impression may be combined into a final report with the medical findings contained in one section of the final report and the impression contained in another section of the final report. Alternatively, the impression can be part of a separate report.

[0020] The AI system can be a stand-alone system or it can be integrated with one or more other systems, such as, for example, a voice recognition system that is used to generate the text- based medical findings from speech input.

[0021] Millions of reports are available for training the AI system. The AI system can be trained in a generalized fashion for each use case (e.g., Radiology reporting for CT, pathology reporting for cytology, etc.). The generalized AI can then be sub-trained to match the specific impression style of individual clinicians by using their own prior reports for training. The generalized AI can also be trained to match selected best-practice standards of one or more national leaders in the field. These leaders can be, for example, large teams/committees that develop consensus guidelines.

[0022] Most trainees and junior clinical providers first learn to make important clinical findings, but struggle to generate high-level impressions. The AI system can be trained on an expert and applied by a novice, essentially upgrading the novice to expert level.

[0023] As indicated above, in some cases, trainees may dictate cases without knowing which attending physician will ultimately approve the final report. Since each attending physician has their own dictating style, a trainee may have used verbiage that may be more commonly used by Attending Physician A if the trainee believes that the final signor on the report is likely to be Attending Physician A. However, if Attending Physician B ends up finalizing the report, he or she may not like the verbiage used in the impressions section of the report. In accordance with a representative embodiment, although the medical findings section is dictated by a trainee, the AI system uses the AI model trained on the attending physician who will be signing off on the final report to generate the impression. This leads to efficiencies because the signor will not have to make extensive corrections to the impressions section of the report.

[0024] The AI system can also be used to grade a trainee’s report or to help train the trainee. If the trainee is given access to a senior Attending Physician’s dictation style model, then the trainee can complete the dictation to produce both the medical findings and impressions sections. Then, the AI system can use the AI model associated with the senior Attending Physician to generate the impression. The trainee can then compare the impression section the trainee prepared with the impressions section generated by the AI system to determine how far or close the trainee’s impression section is to the impressions section generated by the AI system using the Attending Physician’s model.

[0025] In accordance with a representative embodiment, the AI system uses natural language processing (NLP) to detect items in the text-based findings report that can be linked to national or international best practice standards, guidelines or recommendations. For example, a radiologist might mention an incidental lung nodule that measured 6x6 mm. The AI system can detect this in the findings section of the report and create an impressions section that links to established international guidelines (e.g., Fleischner Society Guidelines for Incidental Lung Nodules). The NLP can also be used to detect measurements and automatically generate a graph or table as a part of the final report or note. The graph or table can include data from that single report or capture and present longitudinal data (e.g., changes in the size of a lung nodules or aortic aneurysm).

[0026] In accordance with a representative embodiment, NLP is combined with other data to generate impressions. For example, in general medicine progress notes of History and Physical (H&P) notes, the physician may dictate their physical examination findings, but will also include laboratory data and data from other tests, and then generate an Assessment and Plan based on all of this data. Since a fair bit of the laboratory data is numerical data, this information can directly generate diagnoses. For example, if potassium levels are low, that leads to a diagnosis of hypokalemia. Similarly, blood pressure readings which may be included in such progress notes or H&Ps may provide a diagnosis directly. In some cases, these values are simply cut-and-pasted in text form from other computer screens; however, much of this data may already be tagged as laboratory data in the report. In accordance with a representative embodiment, the AI system can use hard-coded rules to recognize and interpret the tagged laboratory data and to generate the impression based in part on the laboratory data. This can significantly improve the models of the AI system since they do not rely only on NLP, but also on hard-corded rules.

[0027] In the following detailed description, a few illustrative, or representative, embodiments are described to demonstrate the inventive principles and concepts. For purposes of explanation and not limitation, representative embodiments disclosing specific details are set forth in order to provide a thorough understanding of an embodiment according to the present teachings. However, it will be apparent to one having ordinary skill in the art having the benefit of the present disclosure that other embodiments according to the present teachings that depart from the specific details disclosed herein remain within the scope of the appended claims. Moreover, descriptions of well-known apparatuses and methods may be omitted so as to not obscure the description of the representative embodiments. Such methods and apparatuses are clearly within the scope of the present teachings.

[0028] The terminology used herein is for purposes of describing particular embodiments only, and is not intended to be limiting. As used in the specification and appended claims, the terms “a,” “an,” and “the” include both singular and plural referents, unless the context clearly dictates otherwise. Thus, for example, “a device” includes one device and plural devices.

Relative terms may be used to describe the various elements’ relationships to one another, as illustrated in the accompanying drawings. These relative terms are intended to encompass different orientations of the device and/or elements in addition to the orientation depicted in the drawings. It will be understood that when an element is referred to as being “connected to” or “coupled to” or “electrically coupled to” another element, it can be directly connected or coupled, or intervening elements may be present.

[0029] It will be understood that when an element is referred to as being “connected to” or “coupled to” or “electrically coupled to” another element, it can be directly connected or coupled, or intervening elements may be present.

[0030] The term “memory” or “memory device”, as those terms are used herein, are intended to denote a non-transitory computer-readable storage medium that is capable of storing computer instructions, or computer code, for execution by one or more processors. References herein to “memory” or “memory device” should be interpreted as one or more memories or memory devices. The memory may, for example, be multiple memories within the same computer system. The memory may also be multiple memories distributed amongst multiple computer systems or computing devices.

[0031] A “processor,” as that term is used herein encompasses an electronic component that is able to execute a computer program or executable computer instructions. References herein to a computer comprising “a processor” should be interpreted as one or more processors or processing cores. The processor may, for instance, be a multi-core processor. A processor may also be, for example, a controller, such as a microcontroller, an application specific integrated circuit (ASIC), a digital signal processor (DSP), combinational logic configured to function as one or more state machines, etc. A processor may also refer to a collection of processors within a single computer system or distributed amongst multiple computer systems. An example of such a distributed computer system is a cloud computing system. The term “computer” should also be interpreted as possibly referring to a collection or network of computers or computing devices, each comprising a processor or processors. Instructions of a computer program can be performed by multiple processors that may be within the same computer or that may be distributed across multiple computers.

[0032] Exemplary, or representative, embodiments will now be described with reference to the figures, in which like reference numerals represent like components, elements or features. It should be noted that features, elements or components in the figures are not intended to be drawn to scale, emphasis being placed instead on demonstrating inventive principles and concepts. [0033] Fig. 1 is a flow diagram illustrating the flow of the AI system and method in accordance with a representative embodiment. Block 101 represents ML logic of the AI system receiving historical data in the form of previously-generated medical reports that will be used to generate respective AI models for each of Physicians 1 - N, where N is a positive integer that is greater than or equal to one. Typically, N will be equal to the number of attending physicians at a healthcare facility that could potentially sign off on the final report. The ML logic is then trained by using medical reports that have been signed off on by the N physicians to train N respective AI models, i.e., one AI model for each physician, as indicated by block 102.

[0034] In accordance with an embodiment, during training (model development), not all of the prior reports that were signed by the attending physician are used for training. Rather, only those that were entirely (both findings and impression) prepared (e.g., dictated) by the attending physician will be used for model development. This ensures that only the attending physician’s nuanced language is captured. Most current dictation systems (for example, Nuance Powerscribe) clearly show which reports were dictated by the attending physician alone and which ones were initially dictated by a trainee before being signed off by an attending physician. [0035] A second model can also be generated and trained for each of the N physicians to capture the attending preferred impression style when trainees dictate the reports. Thus, for this model, reports that were initially dictated by trainees will be used in the training phase.

[0036] Finally, a third model may use all of the reports for the model generation/training phase.

[0037] A variety of ML algorithms are available that are suitable for generating the AI models. There will typically be multiple reports associated with each physician that are used to train the AI models. The greater the number of reports that are available, the more accurate the AI system will be at predicting the impressions that it is to generate. The ML algorithm can be, for example, supervised learning algorithms, unsupervised learning algorithms, deep learning algorithms and reinforcement learning algorithms.

[0038] Once the AI system has been trained, it is used to generate the impressions. At block

111, the trained ML logic receives a medical findings report. As indicated above, the medical findings report is a digital text-based report that is typically generated by typing on a computer and/or using a dictation device with voice recognition software. The inventive principles and concepts are not limited with respect to the manner in which the digital text-based file is generated. As another example, the digital text-based report can also be generated from handwriting that has been entered on a digital pad using a stylus, recognized by logic of the pad and converted into the digital text-based file.

[0039] The medical findings report can be uploaded to the AI system, as indicated by block 111, and the user can designate the AI model that is to be used to generate the impression section of the report. For example, a nurse practitioner may generate the medical findings report and then upload it to the AI system along with a designation of the physician who is expected to sign off on the final report. Alternatively, the designation can be made at some other time after the medical findings report has been uploaded. Also, the designation does not have to be made by the person who generated the medical findings report.

[0040] At block 112, the trained ML logic selects, based on the designation, one of the AI models to be used to generate the impression. At block 113, the ML logic uses the selected AI model to generate the impression. The impression is then output from the AI system. The impression can be output from the AI system in the form of a final report that includes the medical findings section and the impressions section, ready to be signed and/or edited by the physician associated with the selected AI model.

[0041] The final report output at block 113 preferably is used to further train the AI model associated with the physician, as indicated by the arrow returning to block 101. Also, if the physician edits the final report, the edited final report can be used by the AI system at block 102 to retrain the respective AI model. The ML algorithm that performs the training at block 102 preferably weights more recently generated reports more heavily than older reports. This ensures that the most up-to-date standards or best practices are used to generate the impressions at block 113. This also ensures that changes over time in style or practice by the physicians are reflected in the respective AI models.

[0042] A variety of modifications can be made to the flow diagram shown in Fig. 1. For example, in some cases the person dictating a medical findings report may not know which physician will ultimately sign off on the final report that includes the impression section. In such cases, the designation may designate more than one or all of the potential signors of the final report, in which case impressions are generated at block 113 using the AI models associated with the designations. As another example, the designation can be made by the physician who will be signing off on the final report once that is known, which will result in a more efficient use of computational resources. Fig. 2 is a flow diagram showing some of the possible scenarios that can occur using the AI system and method. [0043] Fig. 3 is a block diagram of the AI system 200 in accordance with a representative embodiment. A processor 210 of the system 200 has ML logic that is configured to perform one or more ML algorithms 220, as described above with reference to Fig. 1. The ML algorithm(s) 220 can be implemented in hardware, software, firmware, or a combination thereof. A memory device 230 of the system 200 stores any computer instructions comprising the algorithm(s) 220. The memory device 230 may also contain the medical reports database that is used by the ML algorithm(s) 220 to train the AI models. In some embodiments, a medical reports database 250 that is external to the AI system 200 can be accessible by the processor 210 to perform the method described above with reference to Fig. 1. The system 100 may optionally include a display device 211, a printer 212 and an input device 213, such as a keyboard, a mouse and/or a voice recognition module. The components 210, 211, 212, 213, 230 and 250 are in communication with one another over a network or bus 215, which can be a wired or wireless link. These components can be co-located or they can be distributed over one or more networks. [0044] A number of software components can be stored in the memory 230 and executable by the processor 210. In this respect, the term "executable" means a program file that is in a form that can ultimately be run by the processor 210. Examples of executable programs may be, for example, a compiled program that can be translated into machine code in a format that can be loaded into a random access portion of the memory 230 and run by the processor 210, source code that may be expressed in proper format such as object code that is capable of being loaded into a random access portion of the memory 230 and executed by the processor 210, or source code that may be interpreted by another executable program to generate instructions in a random access portion of the memory 230 to be executed by the processor 210, etc.

[0045] An example of another software component that can be stored in memory device 230 for execution by the processor 210 is voice recognition software that is used to generate the digital text-based medical findings from speech entered on a dictation device, which may be input device 213. Also, AI can be used to generate the digital text-based medical findings, in which case the ML algorithm(s) 220 can include one or more ML algorithms that generate the digital text-based medical findings based on dictated medical findings. In any case, the ML algorithm(s) 220 include the algorithm(s) discussed above with reference to Fig. 1 that use the AI models to generate the impressions from the digital text-based medical findings.

[0046] Although a single processor 210 is shown in Fig. 2, the system 200 may comprise multiple processors that are distributed over a network. For example, if the processor 210 performs voice recognition to generate the digital text-based reports, a different processor (not shown) may perform the ML algorithm(s) that apply the trained AI models to generate the impressions. That processor could be, for example, part of a cloud computing center to which the digital text-based medical findings are uploaded. Likewise, the memory device 230 can be distributed memory. [0047] An executable program may be stored in any portion or component of the memory 230 including, for example, random access memory (RAM), read-only memory (ROM), hard drive, solid-state drive, USB flash drive, memory card, optical disc such as compact disc (CD) or digital versatile disc (DVD), floppy disk, magnetic tape, or other memory components. In addition, the RAM may comprise, for example, static random access memory (SRAM), dynamic random access memory (DRAM), or magnetic random access memory (MRAM) and other such devices. The ROM may comprise, for example, a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or other like memory device.

[0048] It should be noted that illustrative embodiments have been described herein for the purpose of demonstrating principles and concepts of the invention. As will be understood by persons of skill in the art in view of the description provided herein, many modifications may be made to the embodiments described herein without deviating from the scope of the invention.

For example, while the inventive principles and concepts have been described primarily with reference to a particular system configuration and method, the inventive principles and concepts are equally applicable to other configurations and methods. Also, many modifications may be made to the embodiments described herein without deviating from the inventive principles and concepts, and all such modifications are within the scope of the invention, as will be understood by those of skill in the art.