Medical data processing device, medical data processing method, and program

Abstract

To assist healthcare professionals in verifying the sources of their citations. [Solution] The medical data processing device includes a processing circuit. The processing circuit receives a first medical text, extracts at least one item from the first medical text, receives a second medical text, determines whether there is a match between the extracted at least one item and the content of the second medical text, and if there is a match, determines the portion of the second medical text that matches the extracted at least one item.

Description

Technical Field

[0001] The embodiments disclosed in this specification generally relate to methods and apparatuses for processing text, for example, for learning and utilizing a model to find text matches from two or more information sources.

Background Art

[0002] Although having a very disruptive impact on many industries, multiple LLMs such as Generative Pre-trained Transformer (GPT) and Bard are publicly available. Many of these models are available via API access, while some others are available by downloading and running locally.

[0003] These models generally learn in an unsupervised manner on text corpora obtained from the Internet and can solve complex language expression tasks such as text summarization, answering test questions, and writing papers. The models can use both structured text and unstructured text. Also, the models can provide structured outputs in multiple formats such as, for example, the ".json" format.

[0004] Current LLMs already have a wide range of capabilities and will continue to improve in the future. In the future, it is highly likely that LLMs can be used to associate and reconcile data of completely different modalities and provide the reconciled data to users via an intermediate language form.

[0005] [[ID= twenty-three]] There are challenges that must be overcome in implementing LLMs in a Precision Clinical Decision Support (P-CDS) system.

[0006] LLMs can cause hallucination, producing highly plausible responses in a way that is difficult for users to detect. Especially in clinical applications, it is crucial to mitigate the risk of such outputs reaching users. [Prior art documents] [Patent Documents]

[0007] [Patent Document 1] Japanese Patent Publication No. 2024-177129 [Overview of the Initiative] [Problems that the invention aims to solve]

[0008] One of the problems that the embodiments disclosed in this specification and drawings aim to solve is to assist healthcare professionals in verifying sources of reference. However, the problems that the embodiments disclosed in this specification and drawings aim to solve are not limited to the above problem. Problems corresponding to the effects of each configuration shown in the embodiments described later can also be positioned as other problems. [Means for solving the problem]

[0009] The medical data processing device according to the embodiment includes a processing circuit. The processing circuit receives a first medical text, extracts at least one item from the first medical text, receives a second medical text, determines whether there is a match between the extracted at least one item and the content of the second medical text, and if there is a match, determines a portion of the second medical text that matches the extracted at least one item. [Brief explanation of the drawing]

[0010] Next, several embodiments shown in the following drawings will be described as non-limiting examples. [Figure 1] Figure 1 is a schematic diagram of the apparatus according to the embodiment. [Figure 2]Figure 2 is a schematic diagram of the text data processing method according to the embodiment. [Figure 3] Figure 3 is a diagram of a graphical user interface according to this embodiment. [Figure 4] Figure 4 is a schematic diagram of a text data processing method and an associated graphical user interface according to an embodiment. [Figure 5] Figure 5 shows a graphical user interface according to an embodiment. [Figure 6] Figure 6 is a schematic diagram of the method according to the embodiment. [Figure 7] Figure 7 shows a graphical user interface according to an embodiment. [Figure 8] Figure 8 shows a graphical user interface according to an embodiment. [Figure 9] Figure 9 shows a graphical user interface according to this embodiment. [Modes for carrying out the invention]

[0011] A medical data processing device is provided that includes a processing circuit based on one embodiment. The processing circuit receives a first medical text, extracts at least one item from the first medical text, receives a second medical text, determines whether there is a match between the extracted at least one item and the content of the second medical text, and if there is a match, determines the portion of the second medical text that matches the extracted at least one item.

[0012] A method for searching for matches in medical data text is provided based on one embodiment. The method is: Receiving a first medical text and extracting at least one item from the first medical text, Receiving a second medical text and determining whether there is a match between the extracted at least one item and the content of the second medical text, when there is a match, determining a part of the second medical text that matches the at least one item extracted.

[0013] Based on one embodiment, a non - temporary computer program product storing computer - readable instructions is provided. The instructions receive a first medical text and extract at least one item from the first medical text, receive a second medical text, determine whether there is a match between the at least one item extracted and the content of the second medical text, and when there is a match, cause to determine a part of the second medical text that matches the at least one item extracted.

[0014] FIG. 1 schematically shows a data processing apparatus 20 according to an embodiment. In this embodiment, the data processing apparatus 20 processes text data. In other embodiments, the data processing apparatus 20 may process other data as appropriate.

[0015] The data processing apparatus 20 includes an arithmetic unit 22. In this case, the arithmetic unit 22 is a personal computer (PC) or a workstation. The arithmetic unit 22 is connected to a display screen 26 and other display devices, and one or more input devices 28 such as a computer keyboard and a mouse. The display screen 26 may provide a user interface that outputs a display of the determination result of the presence or absence of a match to the user.

[0016] The arithmetic unit 22 acquires a data set from the data storage unit 30. The data set is appropriately acquired or generated by an arbitrary device or from an arbitrary source. Also, the data storage unit 30 may store at least one of the first medical text and the second medical text.

[0017] In one embodiment, at least a portion of the data includes, for example, medical record data obtained using scanner 24 or can be determined from medical record data.

[0018] Instead of or in addition to data storage unit 30, computing device 22 may receive data from one or more other data storage units (not shown). For example, computing device 22 may receive medical image data from one or more other data storage units (not shown) located remotely or from other information systems.

[0019] Also, computing device 22 provides processing resources for automatically or semi-automatically processing data. Computing device 22 includes processing circuit 32. Processing circuit 32 includes application programming interface (API) / communication circuit 34, data processing circuit 36, and interface circuit 38. Data processing circuit 36 performs multiple processes. The processes include providing data to and receiving data from the API circuit as part of the processes. Interface circuit 38 obtains user input and other inputs and / or outputs the results of data processing via the user interface.

[0020] In this embodiment, API / communication circuit 34, data processing circuit 36, and interface circuit 38 are each implemented within computing device 22 according to a computer program including computer-readable instructions for performing the method of the embodiment. However, in another embodiment, these circuits may be implemented as one or more application specific integrated circuits (ASICs) or field programmable gate arrays (FPGAs).

[0021] Furthermore, the arithmetic unit 22 includes the PC's hard drive and other components, such as RAM, ROM, a data bus, an operating system including various device drivers, and hardware devices including a graphics card. For clarity, Figure 1 omits the illustration of these components.

[0022] The data processing device 20 in Figure 1 performs the method illustrated and described below.

[0023] Figure 2 shows a schematic of a text data processing method 200 according to an embodiment, which is executed under the control of the processing circuit 32 of Figure 1. Refer to Figure 2. A first input text 40 is given to Model 42. The first input text 40 is structured text, unstructured text, or a combination of structured and unstructured text. Unlike structured text data, which has a unique structure that makes it easy to extract relevant information, free text, which is unstructured data, typically requires one or more of the following for text processing: high processing power and a large amount of context. Electronic health records contain a large amount of unstructured data in different formats. Free text constitutes the majority of such data.

[0024] The model 42 that processes the first input text 40 is a trained machine learning model. In the embodiment of Figure 2, model 42 comprises a Large Language Model (LLM). Model 42 is stored on a server located away from the apparatus of Figure 1. The processing circuit 32 sends and receives data to and from model 42 via the API / communication circuit 34. Under the control of the processing circuit 32, the API / communication circuit 34 sends appropriate prompts and other instructions and / or data to model 42 and receives output from model 42. The API / communication circuit 34 can communicate with model 42 via a network connection, such as the Internet or other direct or indirect connections. In another embodiment, model 42 is stored locally in the apparatus of Figure 1, rather than being stored in a remote location. The API / communication circuit 34 operates to communicate with at least one of the data storage unit and the external trained model. The external trained model is a trained model owned by a device located outside the data processing device 20. Furthermore, the externally trained model is capable of operating based on instructions from the processing circuit 32 or other communications, and performs at least one of the following: extracting at least one item from the first medical text, and determining whether there is a match between the extracted at least one item and the content of the second medical text. The API / communication circuit 34 then receives the result of at least one of the extraction and determination from the externally trained model.

[0025] In various embodiments, Model 42 comprises a transformer for processing text sequences and other types of deep learning architectures. Model 42 may comprise a generative pretrained transformer (GPT) or a chatbot such as a chatGPT transformer (ChatGPT). In other embodiments, other LLMs may be used as appropriate, such as GPT-2, GPT-3.5, GPT-4, PaLM, LLaMa, BLOOM, Ernie, T5, Claude or Claude2, or at least one of their derivatives or evolutions.

[0026] The first input text 40 may be referred to as the first user prompt or user query. The first user prompt modifies the output of the model 42. The first user prompt also contains text and is structured in a conversational format. The first user prompt requests the model 42 to perform one or more tasks related to processing the first input text 40. The first input text 40 includes, for example, at least one of the following: medical notes about a patient or other subject, the results of a diagnosis or other procedure, the results of a test or scan, or text related to such results.

[0027] Model 42 processes the first input text 40 and extracts at least one item. In this embodiment, the extracted item is referred to as a header. Model 42 generates a list of one or more headers 44 and a list of one or more first supporting quotes 46 from the first input text 40. Other collected data may also be used. The first supporting quotes 46 include one or more subsets of the first input text 40 selected by Model 42. The first supporting quotes 46 may also be text sentences from the first input text 40 selected by Model 42. The first supporting quotes 46 are selected based on a first user prompt or query that constitutes part of the first input text 40. The headers 44 include a subset of the first input text 40 selected by Model 42. The headers 44 are also selected based on a first user prompt that constitutes part of the first input text 40. One or more headers 44 may further include subsets or shortened versions of the first supporting quotes 46.

[0028] A header 44 containing a subset or abbreviated version of the first supporting citation 46 is selected by Model 42 based on the first user prompt. The first supporting citation 46 and the header 44 are related based on the concept of similarity or relationship between them. In the context of the input prompt, the text of the first supporting citation 46 supports the text of the header 44. The relationship between the headers 44 corresponding to the first supporting citation 46 is the criterion for selection by Model 42 and is defined by the first user prompt or query.

[0029] The header 44 is provided to the model 42 along with the second input text 48 and processed. The second input text 48 includes structured text, unstructured text, or a combination of structured and unstructured text. The second input text 48 includes a second user prompt or query that modifies the output of the model 42. The model 42 processes the second input text 48 to generate a list of one or more second supporting quotes 52. The second supporting quotes 52 include one or more subsets of the second input text 48 selected by the model 42. The second supporting quotes 52 are text sentences from the second input text 48 selected by the model 42. The second supporting quotes 52 are selected based on a second user prompt or query that constitutes part of the second input text 48.

[0030] In this embodiment, a second user prompt adjusts the output of Model 42 to find matches between Header 44 and Second Input Text 48. By processing the Second Input Text 48 and Header 44, Model 42 obtains a match status 50 and selects second supporting quotes 52 obtained from the Second Input Text 48 that match one or more Header 44s. The list of match statuses 50 includes, for example, a binary representation indicating whether or not there is a match between Header 44 and Second Input Text 48. The second supporting quotes 52 include one or more subsets of Second Input Text 48 that are selected by Model 42 and correspond to Header 44. The relationships between Header 44s corresponding to the second supporting quotes 52 are criteria used by Model 42 for selection and are defined by a second user prompt or query.

[0031] The output data of the text data processing method 200 is combined text data containing a header 44, a first supporting citation 46, a matching status 50, and a second supporting citation 52. The text data obtained from the output of the text data processing method 200 consists of a subset of the first text related to a matching subset of the second medical text. The header 44 and the second supporting citation 52 associated with the header 44 are considered to be related. In this embodiment, this data / method can automatically combine medical data from different sources or from different sections of the same source that match according to user-based criteria defined by user prompts. Applying automatic matching of medical data to a user's medical records is beneficial because it can accelerate diagnosis by finding patterns in large amounts of medical text data that are difficult to observe manually.

[0032] The text data processing method 200 of this embodiment has two input stages and two output stages for the model 42. In other embodiments, the text data processing method 200 may include inputting new data and processing the new data with past data three, four, or more times.

[0033] Figure 3 shows an image of the graphical user interface 300 according to an embodiment. Figure 3 is an image of the first input text 40 with added visual display, and shows a portion of the output data of the text data processing method 200 in the embodiment of Figure 2. The text shown in Figure 3 includes medical records, but in other embodiments, text of other modalities may be used.

[0034] Figure 3 shows a cursor 54 juxtaposed with a first token 56 or quote in the first input text 40. In this particular embodiment, the token is represented by a sentence. In other embodiments, the token may be represented by a word, phrase, paragraph, or clause. The sentence or token juxtaposed with the cursor 54 is highlighted because Model 42 has determined that the matching conditions described later are met. A pop-up text box is generated as intermediate text 58, showing a header 44 and a second supporting quote 52 associated with the header. Figure 3 does not show the second input text 48 used to search for a match with the header 44. Considering the rest of the text, the second supporting quote 52 that matches the first token 56 or quote is highlighted. Tokens or quotes that do not match the first token 56 are highlighted in a different way. The highlighting in Figure 3 is indicated by different types of shading. In various embodiments, features such as tooltips, popovers, or mouseovers may be used to represent, for example, the relevant parts of the first or second medical text. In another embodiment, in addition to or instead of color highlighting, other methods may be used to draw attention to text or portions of text, such as quotations. Examples include shading, the use of display elements such as text resizing, pointers or other graphical display elements, or other display elements.

[0035] A processing circuit, or a trained model based on the instructions of the processing circuit, performs the desired match search process. For example, determining whether a match exists involves determining whether the cognitive or semantic content of at least one extracted item is identical or consistent with at least a portion of the content of the second medical text. For example, the trained model may include a large-scale language model (LLM) or other language models.

[0036] Alternatively or additionally, determining whether there is a match may include at least one of the following: determining at least one criterion from at least one extracted item; or determining whether the content of the second medical text conforms to that at least one criterion. Or, determining whether there is a match between the at least one extracted item and the content of the second medical text may include determining the question represented by or included in that at least one item; and determining whether the answer to the question is affirmative or negative based on the second medical text.

[0037] As shown in Figure 3, the original text is linked to the second medical text, directly superimposed on relevant citations, and may be visually marked regarding the degree of agreement using a graphical user interface.

[0038] One feature of text data processing method 200 is to hide from the user connections that do not contain relevant citations (or citations generated by LLM hallucination). Additionally, transparency of the input span type is identified.

[0039] Figure 4 shows an overview of the text data processing method 400 according to the embodiment and an image of the associated graphical user interface.

[0040] Figure 4 shows an overview of the text data processing method 400 for medical text data according to the embodiment. Refer to Figure 4. Clinical input text 60 is provided to Model 62. Clinical input text 60 is structured text or unstructured free text. Clinical input text 60 includes clinical trial criteria. Clinical input text 60 may also include medical data such as medical records and patient records for one or more users.

[0041] Furthermore, the clinical input text 60 also includes a first user prompt. The first user prompt adjusts the output of the model 62. The first user prompt contains text and is structured in a conversational format. The first user prompt requests the model 62 to perform one or more tasks related to processing the clinical input text 60.

[0042] Model 62 processes the clinical input text 60 and generates a list of one or more clinical trial criteria 64 and a list of one or more clinical trial supporting citations 66 relating to the clinical trial criteria 64. Other collected data may be used. The clinical trial supporting citations 66 include one or more subsets of the clinical input text 60 selected by Model 62. The one or more clinical trial supporting citations 66 are selected based on a first user prompt that constitutes part of the clinical input text 60. The clinical trial criteria 64 may also include one subset of the clinical input text 60 selected by Model 62.

[0043] Clinical trial criteria 64 are selected based on a first user prompt or query that constitutes part of the clinical input text 60. One or more clinical trial criteria 64 may further include a subset or abbreviated version of clinical trial supporting citations 66. Header containing a subset or abbreviated version of clinical trial supporting citations 66 is selected by Model 62 based on the first user prompt. Clinical trial supporting citations 66 and clinical trial criteria 64 are related based on the concept of similarity or relationship between them. In the context of the ground truth each represents, the text of clinical trial supporting citations 66 supports the text of clinical trial criteria 64. The relationship between clinical trial criteria 64 corresponding to clinical trial supporting citations 66 is the criterion for selection by Model 62 and is defined by the first user prompt or query.

[0044] Clinical trial criteria 64 are provided to and processed by Model 62 along with patient records 68. Patient records 68 include structured or unstructured free text. Patient records 68 also include a second user prompt to adjust the output of Model 62.

[0045] In this embodiment, a second user prompt or query adjusts the output of Model 62 to find matches between clinical trial criteria 64 and patient records 68. Model 62 processes the patient records 68 and clinical trial criteria 64 and obtains match status 70 and patient record supporting citations 72 for each of the one or more clinical trial criteria 64 in list or other collected text data format. The list of match statuses 70 includes a binary representation indicating whether or not there is a match between the clinical trial criteria 64 and patient records 68. The patient record supporting citations 72 include one or more subsets of patient records 68 selected by Model 62 and corresponding to the clinical trial criteria 64. The relationships between the clinical trial criteria 64 corresponding to the patient record supporting citations 72 are the criteria used by Model 62 for selection and are defined by the second user prompt or query.

[0046] The output data of the text data processing method 400 is combined text data included in the clinical trial criteria 64, clinical trial supporting citations 66, agreement status 70, and patient record supporting citations 72. The text data obtained from the output of the text data processing method 400 consists of a subset of the first text associated with a matching subset of the second medical text. Applying this automated matching of medical data to a user's medical records is beneficial because it can accelerate diagnosis by finding patterns within large amounts of medical text data that are difficult to observe manually.

[0047] The text data processing method 400 of this embodiment has two input stages and two output stages for the model 62. In other embodiments, the text data processing method 400 may include inputting new data and processing the new data and past data multiple times.

[0048] Figure 4 also shows a subsection of the graphical user interface 402 related to the text data processing method 400. In the graphical user interface 402, various visual markers, described later, are superimposed on the clinical input text 60. In this embodiment, the clinical input text 60 is shown separately for each citation. For illustrative purposes, we assume that the text data processing method 400 is completed according to Figure 4 and that all data obtained from the method is available. A supporting citation is selected by placing the cursor 74 next to a particular supporting citation. Intermediate text 76 is also superimposed on the position of the cursor 74 on the clinical input text 60. The intermediate text 76 contains clinical trial criteria 64 related to the supporting citation selected by the cursor 74. In this particular embodiment, the clinical trial criteria 64 is "NSCLC diagnosis confirmed:", and the supporting citation is "NSCLC positive confirmed". As mentioned above, the clinical trial criteria are a subset or shortened version of the text of the supporting citation. The intermediate text 76 also contains patient record supporting citations 72 selected from the patient record 68. As described above, the patient record supporting citation 72 is selected based on the clinical input text 60, the text of the patient record 68, and user prompts related to both. In the graphical user interface 402, supporting citations that are not related to the clinical input text 60 and the patient record 68 are highlighted. In this embodiment, the patient record supporting citation 72 is "Diagnosis at admission: Non-small cell lung cancer".

[0049] Model 62 summarizes the text of the supporting citation “NSCLC positive confirmed” to “NSCLC diagnosis confirmed.” Further processing of patient record 68 determines the agreement or association deemed to be met by the GPT to be binary, and the supporting citation is highlighted. The supporting citation is accompanied by patient record supporting citation 72 “...diagnosis at admission, non-small cell lung cancer….” Model 62 correctly associates the abbreviation “NSCLC” with non-small cell lung cancer.

[0050] This allows for the concealment of connections without relevant citations (or citations generated by LLM hallucination) from the user. It also identifies the type of input span.

[0051] The cursor 74 is moved to the position of a different supporting citation within the clinical input text 60 to select that citation and update the intermediate text 76. In this case, the new intermediate text 76 includes the clinical trial criteria 64 and the patient record supporting citation 72 related to the selected supporting citation. If Model 62 determines that there is a match between the clinical trial criteria 64 and the patient record 68 related to the selected supporting citation, that supporting citation is highlighted; supporting citations that do not meet the match criteria are highlighted in a different way. The highlighting in Figure 4 uses different types of shading or cross-hatching.

[0052] Figure 5 shows an image of a graphical user interface 500 that can be used for parsing the data generated based on the text data processing method 400 and the text data processing method 200.

[0053] Here, we specifically detail Cohort 1 and Cohort 2 based on MET gene mutation status. Since there is no information about MET status in the patient records, this criterion is marked by corresponding shading. In Figure 5, the cursor is juxtaposed with the supporting citation "Cohort 1: MET gene mutation-positive patients who have not received prior treatment" or "Cohort 2: MET gene mutation-positive patients who have received prior treatment." Intermediate text 78 includes the relevant clinical trial criterion 64, "Cohort 1 or Cohort 2." Intermediate text 78 also includes the status "No information about MET gene mutation." It can be seen that Model 42 could not find text matching clinical trial criterion 64 within the patient record 68. Therefore, the supporting citation is highlighted in a different way. If the second medical text does not contain information about MET gene mutation, in this example, the second citation is not presented; instead, "No information about MET gene mutation" is presented and associated with the first supporting citation. The highlighting in Figure 5 uses different types of shading.

[0054] Figure 6 shows a text processing method 600 that uses a model to find text matches from two sources. In other embodiments, three or more sources may be used. A first text named Eligibility Criteria 80 and a first user query 82 are provided to the model 90. In this embodiment, the model 90 is a Generative Pre-trained Transformer (GPT). In other embodiments, the model 90 may be another LLM. The first user query 82 is "Summarize the following eligibility criteria into concise titles of no more than 5 words each. For each title, extract directly from the source text." The model 90 outputs a first query response 84 containing "Title" and "Excerpt". The title is similar to the header 44 in text data processing method 200 and the clinical trial criteria 64 in text data processing method 400, and the excerpt is similar to the first supporting citation 46 in text data processing method 200 and the clinical trial supporting citation 66 in text data processing method 400. Furthermore, patient record 86 and a second user query 88 are given to model 90, and a second query response 92 is output. The second user query 88 is "For each title, answer 'yes' or 'no' whether the following patient record meets the requirements. Extract directly from the record." As a result, model 90 generates titles and excerpts from the text of patient record 86, along with the 'yes' or 'no' answers to the second user query 88. The titles correspond to the headers in Figure 2. Also, for the second input text, titles or headers are generated in text data processing method 200 (Figure 2) and text data processing method 400 (Figure 4). For example, in a clinical trial matching search example, the LLM is required to summarize the criteria into headers / titles, each with supporting citations.

[0055] The output of the text processing method 600 is a matching criteria result 94, which shows a graphical user interface that integrates the results of model 90. The graphical user interface displays a list of criteria representing the answers to the second user query 88 and matching queries, using color coding or shading. The color coding or shading is selectable, and GPT is requested to provide structured output. The matching status (yes, no, unknown) may be mapped to the corresponding color or shade, for example, depending on the desired color coding or shading scheme.

[0056] In this embodiment, the following steps are provided: Step 1 (Clinical Trial Text): The LLM is asked to summarize the eligibility criteria in the header and citations. Step 2 (Patient Text): The header is matched to the patient record using evidence (citations) from the patient record. This ensures that citations from the clinical trial text match with relevant citations from the patient record.

[0057] Figure 7 shows the output provided to the user via a user interface according to another embodiment. In this example, the first medical text includes a combination of eligibility criteria documents and documents from Guideline 1 and Guideline 2, as shown in Figure 7. The second medical text includes the patient record documents shown in Figure 7. That is, one of the first and second medical texts includes clinical trial eligibility criteria or clinical practice guidelines, and the other of the first and second medical texts includes patient records containing multiple medical record documents. In this example, the items extracted from the first medical text and highlighted using the highlighting in Figure 7 are the text "Cytological or histological NSCLC diagnosis confirmed: ALK rearrangement negative" from eligibility criteria, the text "ALK-positive advanced non-small cell lung cancer (NSCLC)" from Guideline 1, and the texts "lorlatinib," "ALK-positive advanced non-small cell lung cancer (NSCLC)," and "crizotinib" from Guideline 2. In Figure 7, the corresponding section of the matching second medical text is highlighted by shading and includes the text: "An ALK mutation was found. The oncology team decided to initiate treatment with the ALK inhibitor crizotinib in addition to the ongoing chemotherapy regimen." In this example, the shading or highlighting represents agreement with a clinical trial or clinical guideline, rather than the nature of the agreement (for example, instead of the aforementioned yes, no, or unknown). In this example, agreement of eligibility criteria takes precedence.

[0058] Figure 8 shows the output provided to the user via a user interface according to another embodiment. This process is similar to that in Figure 2, except that in Figure 2, the first input text is a document containing general practitioner (GP) records. In contrast, in the example of Figure 8, the first medical text contains a simulated medical paper. In the figure, the first medical text is shown in a box shape (with some text blank), but in reality, it is understood to contain text such as "high blood glucose" and "colorectal cancer (CRC)". The second medical text contains blood measurement values ​​for patient John Smith at the time of his GP appointment on September 11, 2025. Various items are extracted from the first medical text and highlighted within the first medical text (Figure 8 includes the highlighting, but does not show all the words and sentences of the text written below the highlighting in order to avoid reproducing the entire paper in this document). The processing circuit determined that the high blood glucose criterion, an item extracted from the first medical text, and the blood glucose measurement result of 130 mg / dL in the second medical text were consistent. A quote corresponding to the extracted item ("high blood glucose") is selected from the first medical text, and a quote representing the matching section in the second medical text is selected from the second medical text. The quote corresponding to the extracted item from the first medical text and the matching section in the second medical text are highlighted on the user interface in Figure 8. Here, colors, shading, etc., can be used to distinguish related diseases / symptoms. For example, when highlighting text in a document, a first color, shading, or color can be used to highlight CRC, a second color, shading, or color can be used to highlight high blood glucose, and a third color, shading, or color can be used to highlight diabetes. Rather than indicating the nature of the match with color, shading, or color (e.g., finding a CRC diagnosis in the patient record), this example chooses to simply show the matches and select a color, shading, or color based on the content of the match.

[0059] Figure 9 shows the output provided to the user via a user interface according to another embodiment. Figure 9 is similar to Figure 8, but the user interface displays different information based on the cursor position. The first medical text includes an article from a simulated medical paper. The second medical text includes a memo from a GP appointment on June 10, 2025, for the same patient, John Smith, as in the embodiment of Figure 8. Various items are extracted from the first medical text, and each is highlighted. The processing circuit determines that the item "colon cancer" extracted from the first medical text matches the part of the second medical text, "screening results indicate suspected CRC." A citation corresponding to the items extracted from the first medical text ("colon cancer", "CRC") is selected, and a citation representing the matching portion of the second medical text ("screening results indicate suspected CRC") is selected. The corresponding portions of the quotations extracted from the first medical text and the second medical text are highlighted on the user interface in Figure 9 using color, shading, or other methods.

[0060] A system comprising a processing circuit is provided based on various embodiments. The processing circuit automatically searches for a match between two medical texts (or two sets of multiple medical texts) by the following operation: Along with supporting citations for each item, the source medical text is summarized into each item. Along with supporting citations from a second medical text, multiple items are linked to items within the target input medical text. The user interface displays the source text, overlaid with the identified quote from the target text.

[0061] A Large-Scale Language Model (LLM) extracts summaries and citations. The LLM is fine-tuned and / or includes examples of input-output pairs. The degree of agreement between source and target text is visually indicated on the user interface by color coding or shading of the text. The graphical user interface may display citations from the target text on the source text using tooltips, popovers, or mouseover functions. The input text can be unstructured, structured, or a mixture of structured and unstructured. The correlation / match search may be performed between clinical trial eligibility criteria as a source and target patient records. The patient records may include multiple medical record documents. The correlation / match search may be performed between clinical practice guidelines as a source and target patient records. These patient records may include multiple medical record documents. Alternatively, the correlation / match search may be performed between medical articles as a source and target patient records. That is, one of the first and second medical texts includes a medical article as a source, and the other of the first and second medical texts includes patient records containing multiple medical record documents.

[0062] While various embodiments have been described for displaying supporting citations associated with items via a user interface, in various embodiments, the supporting citations and highlighted items may be available in other desired ways, such as in the planning of procedures, including scan plans and drug prescriptions. In one embodiment, the user interface may be included in or accessible from, for example, a scanner, scan management software, a prescription management system, etc. Medical text may include scan protocol text, scan instructions, or prescription notes and workflows.

[0063] A medical data processing device comprising a processing circuit is provided based on various embodiments. The processing circuit receives a first medical text, extracts at least one item from the first medical text, receives a second medical text, determines whether there is a match between the extracted at least one item and the content of the second medical text, and if there is a match, determines the portion of the second medical text that matches the extracted at least one item.

[0064] The medical data processing device further comprises a user interface for displaying at least a portion of the first medical text, the display of which may include displaying and / or highlighting the at least one extracted item.

[0065] Furthermore, the user interface may display an image of a portion of the second medical text that matches the at least one extracted item, and the user interface may associate the image of the portion of the second medical text with the matching at least one extracted item.

[0066] Associating an image of a portion of the second medical text with the matching at least one extracted item on the user interface may include overlaying, linking, or displaying the portion of the second medical text with the matching at least one extracted item.

[0067] Some images in the second medical text may include quotations from the second medical text.

[0068] Some images from the second medical text may be displayed using tooltips, popovers, or mouseover functions.

[0069] The user interface may output a display indicating whether or not there is a match between the at least one extracted item and the content of the second medical text.

[0070] The display may include, depending on whether there is a match, at least one of the following: text highlighting, color coding, shading, or one or more display elements.

[0071] Determining whether there is a match between the at least one extracted item and the content of the second medical text may include determining whether the cognitive or semantic content of the at least one extracted item is identical to or consistent with at least a portion of the content of the second medical text.

[0072] Determining whether there is a match between the at least one item extracted and the content of the second medical text may include at least one of the following: a) Determine at least one criterion from the at least one item extracted, and determine whether the content of the second medical text conforms to the at least one criterion. b) Determine the question represented by or included in at least one of the items, and determine whether the answer to the question is affirmative or negative based on the second medical text.

[0073] Extracting at least one item from the first medical text may include at least one of the following: a) Select at least a portion of the above-mentioned medical text, b) Summarizing the contents of the first medical text and generating at least one item that represents the summarized contents.

[0074] The processing circuit may use a trained model to extract at least one of the following: extract at least one item from the first medical text; or determine whether there is a match between the extracted at least one item and the content of the second medical text.

[0075] The aforementioned trained model may include a large-scale language model (LLM) or other language models.

[0076] The pre-trained models may include at least one of the following: GPT-2, GPT-3.5, GPT-4, PaLM, LLaMa, BLOOM, Ernie, T5, Claude or Claude2, or their derivatives or evolutions.

[0077] One or both of the first medical text and the second medical text may be structured text, unstructured text, or a mixture of structured and unstructured text.

[0078] One of the first medical texts and the second medical text may include clinical trial eligibility criteria or clinical practice guidelines. The other of the first medical text and the second medical text may include patient records, which may include multiple medical record documents.

[0079] One of the first medical text and the second medical text may include medical articles as a source. The other of the first medical text and the second medical text may include patient records that include multiple medical record documents.

[0080] The medical data processing device may further include the following: A data storage unit that stores at least one of the first medical text and the second medical text, A display device that provides a user interface to the user that displays the result of whether or not there is a match, and A communication circuit capable of communicating with the data storage unit and at least one of the external trained models, wherein the external trained model is capable of operating based on instructions from the processing circuit or other communications, and performs at least one of the following: extracting at least one item from the first medical text and determining whether there is a match between the extracted at least one item and the content of the second medical text. A communication circuit that receives at least one of the results of extraction or judgment from the aforementioned trained model.

[0081] A method for finding matches in medical data text is provided based on various embodiments. The method is Receiving a first medical text and extracting at least one item from the first medical text, Receiving a second medical text and determining whether there is a match between the extracted at least one item and the content of the second medical text, If a match exists, the method comprises determining the portion of the second medical text that matches the at least one extracted item.

[0082] Based on various embodiments, a non-temporary computer program product is provided that stores computer-readable instructions. These instructions are Receiving a first medical text and extracting at least one item from the first medical text, Receiving a second medical text and determining whether there is a match between the extracted at least one item and the content of the second medical text, If a match exists, the system will perform the task of determining the portion of the second medical text that matches the at least one of the extracted items.

[0083] Furthermore, a user interface / query processing system is provided that allows patient records to be matched to specific clinical trials in a transparent manner.

[0084] While specific circuits have been described herein, in other embodiments, one or more functions of these circuits may be provided by a single processing resource or other component, or functions provided by a single circuit may be provided by a combination of two or more processing resources or other components. A description of a single circuit may include multiple components that realize the function of that circuit, regardless of whether these components are spatially separated. A description of multiple circuits may include a single component that realizes the functions of these circuits.

[0085] While several embodiments have been described, these embodiments are presented as examples only and are not intended to limit the scope of the invention. The novel methods, apparatus, and systems can be implemented in a variety of other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their variations are included in the scope and spirit of the invention, as well as in the claims and their equivalents. [Explanation of symbols]

[0086] 20 Data Processing Devices 22 Arithmetic unit 24 Scanners 26 Display screen 28 Input devices 30 Data storage unit 32 Processing Circuits 34. Application Programming Interface (API) / Communication Circuits 36 Data Processing Circuits 38 Interface Circuit

Claims

1. Upon receiving a first medical text, at least one item is extracted from the first medical text. Upon receiving the second medical text, it is determined whether there is a match between the extracted at least one item and the content of the second medical text. If a match is found, a processing circuit determines the portion of the second medical text that matches the extracted at least one item. A medical data processing device equipped with [a specific feature].

2. The processing circuit performs at least one of the following: extracting at least one item from the first medical text using a trained model; and determining whether or not there is a match between the extracted at least one item and the content of the second medical text. The medical data processing device according to claim 1.

3. The aforementioned trained model includes a Large-Scale Language Model (LLM) or other language models. The medical data processing device according to claim 2.

4. The pre-trained model includes at least one of the following: GPT-2, GPT-3.5, GPT-4, PaLM, LLaMa, BLOOM, Ernie, T5, Claude, or Claude2, or their derivatives or developments. The medical data processing device according to claim 3.

5. The system further includes a user interface for displaying at least a portion of the first medical text, The display includes displaying and / or highlighting at least one of the extracted items. The medical data processing device according to claim 1.

6. The user interface displays an image of a portion of the second medical text that matches the extracted at least one item, and associates the image of the portion of the second medical text with the matching extracted at least one item on the user interface. The medical data processing device according to claim 5.

7. Associating an image of a portion of the second medical text with the matching extracted at least one item on the user interface includes overlaying, linking, and displaying the portion of the second medical text and the matching extracted at least one item in proximity. The medical data processing device according to claim 6.

8. Some images in the second medical text include quotations from the second medical text. The medical data processing device according to claim 6.

9. Some images in the second medical text are displayed using tooltips, popovers, or mouseover functions. The medical data processing device according to claim 6.

10. The user interface outputs a display indicating whether or not there is a match between the at least one extracted item and the content of the second medical text. The medical data processing device according to claim 5.

11. The aforementioned display includes, depending on whether there is a match, at least one of the following: text highlighting, color coding, shading, or one or more other displays. The medical data processing device according to claim 10.

12. Determining whether there is a match between the at least one extracted item and the content of the second medical text includes determining whether the cognitive or semantic content of the at least one extracted item is identical to or consistent with at least a portion of the content of the second medical text. The medical data processing device according to claim 1.

13. Determining whether there is a match between the at least one item extracted and the content of the second medical text is: Determine at least one criterion from the at least one item extracted, and determine whether the content of the second medical text conforms to the at least one criterion. Determine a question represented by or included in at least one of the aforementioned items, and determine whether the answer to that question is affirmative or negative based on a second medical text. Including at least one of the following, The medical data processing device according to claim 1.

14. Extracting at least one item from the first medical text is: Selecting at least a portion of the first medical text, Summarizing the contents of the first medical text and generating at least one item that represents the summarized contents, Including at least one of the following, The medical data processing device according to claim 1.

15. One or both of the first medical text and the second medical text are structured text, unstructured text, or a mixture of structured and unstructured text. The medical data processing device according to claim 1.

16. One of the first medical text and the second medical text includes clinical trial eligibility criteria or clinical practice guidelines. The other of the first medical text and the second medical text includes patient records, which include multiple medical record documents. The medical data processing device according to claim 1.

17. One of the first medical text and the second medical text includes medical papers as sources. The other of the first medical text and the second medical text includes patient records, which include multiple medical record documents. The medical data processing device according to claim 1.

18. A data storage unit that stores at least one of the first medical text and the second medical text, A display device that provides a user interface to output to the user the result of determining whether or not there is a match, A communication circuit capable of communicating with at least one of the data storage unit and an externally trained model, wherein the externally trained model is capable of operating based on instructions from the processing circuit or other communications, and performs at least one of the following: extracting at least one item from the first medical text and determining whether there is a match between the extracted at least one item and the content of the second medical text, and the communication circuit receives the result of at least one of the extraction and determination from the externally trained model. The medical data processing device according to claim 1.

19. A method for finding matches in medical data text, the method is: Receiving a first medical text and extracting at least one item from the first medical text, Receiving a second medical text and determining whether there is a match between the extracted at least one item and the content of the second medical text, If a match is found, determine the portion of the second medical text that matches the extracted at least one item. A medical data processing method including [the specified term].

20. A non-temporary computer program product that stores computer-readable instructions, wherein the instructions are Receiving a first medical text and extracting at least one item from the first medical text, Receiving a second medical text and determining whether there is a match between the extracted at least one item and the content of the second medical text, If a match exists, the system will perform the following: determine the portion of the second medical text that matches the at least one extracted item. A program that instructs a computer to perform a process.

Images