Computer-implemented method and system for automatically providing disease information

Abstract

The invention relates to a computer-implemented method and system for automatically providing disease information relating to verification of a suspected diagnosis, comprising the following steps: providing at least one disease database, which contains a plurality of disease patterns relating to one or more diseases, and at least one patient database, the patient database comprises a plurality of electronic patient files with patient information; extracting a disease pattern for diagnosing a disease of the patient from at least one provided disease database based on the suspected diagnosis; matching the extracted sick pattern with patient information of an electronic patient profile of the patient loaded from a patient database by using a mathematical model for automatically determining at least one piece of disease information for suspected diagnosis; and outputting the determined disease information via the first user interface.

Description

Technical Field

[0001] The present invention relates to a computer-implemented method and system for automatically providing disease information related to validating a suspected diagnosis. Background Art

[0002] In the healthcare industry, different information systems are used to store patient-specific documents. These databases include, for example, databases of HIS (Hospital Information System), RIS (Radiology Information System), OIS (Oncology Information System), LIS (Laboratory Information System), etc. Most of these databases are text-based and are therefore not designed to provide content in a machine-readable form. Content can only be searched patient by patient for a specific date or a specific time, making it feasible to only roughly classify the content. Although for some patients there is already a long medical history and a large amount of associated documents and information, usually only a small portion of the available documents is important for the specific disease condition to be examined. Traditional information systems do not provide automatic support to users, such as treating specialists or corresponding assistants, when searching for documents that may be important or worth reading for the case being examined.

[0003] In order to be able to make a final assessment or evaluation of a clinical case, it is often not sufficient to only look at the medical images of the patient in question. For example, in some cases it is not possible or sufficient to distinguish a tumor or inflammation based on the images alone. In other cases, the patient in question may have a specific anatomical structure resulting from a previous clinical operation performed on the patient. Therefore, in order to complete a final evaluation of the current clinical case and thus clarify this situation, the radiologist as the attending specialist must use other information about the examined patient in addition to the image information.

[0004] For this reason, for example, an inspection system can be used, which utilizes a suitable search engine to search for a patient in the following manner, and the patient suffers from similar diseases or illnesses (for example, in interstitial lung disease, the ILD field). To this end, existing medical image records can be compared with the cases with known diagnosis stored in the database. Then similar cases with a detailed description of the disease are displayed to the radiologist. However, in these traditional inspection systems, only analyze and match medical image records, and do not analyze and match corresponding, known patient documents or other patient information. These information are usually not easy to access, and are for example distributed in a plurality of documents in an unstructured manner. This means that the doctor must either manually search for these information in a costly manner, or may not do these at all, because each patient cannot provide a search usually in the remaining scarce time in clinical daily life. If a plurality of differential diagnoses are available, when different information must be obtained for this reason, then the time consumption can be correspondingly multiplied. Summary of the invention

[0005] Against this background, the invention is based on the object of providing disease information for different types of diseases in relation to a suspected diagnosis established for an examined patient as quickly as possible and moreover reliably.

[0006] According to the invention, the object is achieved by a computer-implemented method having the features of the invention and / or by a computer-implemented system having the features of the invention.

[0007] Based on this, it is proposed that:

[0008] -A computer-implemented method for automatically providing disease information about verifying a suspected diagnosis, the method comprising the following steps: providing at least one disease database and at least one patient database, the disease database containing a plurality of disease patterns (Erkrankungsmusters) about one or more diseases, and the patient database containing a plurality of electronic patient files with patient information; extracting disease patterns for diagnosing a patient's disease from at least one provided disease database based on the suspected diagnosis; matching the extracted disease patterns with patient information of the patient's electronic patient file loaded from the patient database by using a mathematical model for automatically determining at least one disease information for the suspected diagnosis; and outputting the determined disease information via a first user interface.

[0009] -A computer-implemented system for automatically providing disease information about verifying a suspected diagnosis, the system comprising: at least one disease database, the disease database comprising a plurality of disease patterns about one or more diseases; at least one patient database, the patient database comprising a plurality of electronic patient files with patient information; a data processing unit, the data processing unit being configured to extract disease patterns for diagnosing a patient's disease from a disease database (KDB) based on the suspected diagnosis, matching the extracted disease patterns with patient information of the patient's electronic patient file loaded from a patient database (PDB) by using a mathematical model for automatically determining at least one disease information for the suspected diagnosis; and a user interface (I / O) for outputting the determined disease information.

[0010] The concept on which the present invention is based is to confirm or verify and / or disprove a suspected diagnosis for a patient fully automatically based on the existing patient information and the existing disease information by comparing the consistency of the signs and symptoms of the disease pattern associated with the suspected diagnosis based on the patient data for the specific patient. According to the present invention, this is done by using the additional intelligence of the mathematical model. Therefore, the suspected diagnosis can be confirmed and disprove with very little effort and relatively quickly, thereby providing the attending physician with additional diagnostic information.

[0011] Additionally or alternatively, it is also conceivable that, instead of a clear confirmation or falsification of a suspected diagnosis, there may also be no clear diagnosis. Thus, for example, there are usually signs indicating a specific disease, and there may be signs that cannot clearly indicate or deny the diagnosis at the same time. The particular advantage of the present invention is also that the system makes this transparent to the user: which signs represent it, which signs deny it, and, if necessary, which signs should be evaluated as boundary values. The task undertaken by the system is to identify which information is important for the suspected diagnosis. Data is provided and extracted from a patient database. The typical signs of (multiple) suspected diagnoses are matched, i.e. compared, with the symptoms of the observed patient. The results are made public transparently, and all extracted parameters may also be displayed, so that the user of the system can form his own image and may reject (überstimmen) the mathematical model as a whole.

[0012] Here and below, disease information should be understood as all information that is helpful for assessing the presence of a disease pattern (Krankheitsmusters) and thus for verifying a suspected diagnosis. This can be a sign of proof for the presence or absence of a disease pattern, such as when a characteristic feature exists, is missing, or is weaker or stronger than usual. The disease information can, but does not necessarily have to exist in binary, that is, the disease information can go beyond the pure confirmation or falsification of the disease pattern and the suspected diagnosis accompanying it. For example, disease information can also include the probability that there is a disease pattern or that the suspected diagnosis is correct. In general, disease information should be provided for providing decision assistance to users, such as to attending physicians. For example, this decision assistance can also include: based on the existing information, it is impossible to make a clear decision, which is of course usually helpful to users.

[0013] Here and in the following, a disease pattern is to be understood as a set of signs and symptoms of a disease, which is not necessarily patient-specific.

[0014] Here and hereinafter, a suspected diagnosis is to be understood as a suspicion of a disease.

[0015] Here and hereinafter, a disease database is to be understood as an information source that contains possible disease patterns corresponding to one or more diseases as well as symptoms, signs, criteria, etc. that are typical for these diseases. The database can be, for example, a medical standard book in text form or electronic form, a database of a medical institution (e.g. a doctor's office or hospital), information provided via the Internet, etc. The information can be in electronically readable form or in text form.

[0016] Here and in the following, a patient database is to be understood as any information source containing patient-specific information. Such information sources may be, for example, (electronic) patient files at a doctor's office or hospital, possible information from a health insurance, information on the patient's electronically readable medical card, voluntarily provided patient information, etc., wherein data protection must be taken into account with regard to the availability of this information. The information may be in electronically readable form or in text form.

[0017] Here and in the following, a user interface is to be understood as any interface that allows human-machine interaction. Preferably, this is a so-called HMI interface (human machine interface). The user interface may be an electronic interface, via which corresponding information can be input or output in an electronically readable manner. Additionally or alternatively, it is also conceivable that the user interface is configured as a screen, which directly displays the provided information to the user. However, a keyboard, such as a keyboard on a touch screen, is also conceivable, via which the user can operate the corresponding user input or even conduct corresponding questions and answers with the data processing unit.

[0018] The method according to the invention or the system according to the invention offers in particular the advantage that no image data are required.

[0019] Advantageous embodiments and developments emerge from the description with reference to the drawings.

[0020] In a particularly preferred design, disease information is automatically generated by using a mathematical model that is configured as a trained large language model, in particular a trained large language model. A large language model, also commonly referred to as a largelanguage model1 (LLM), represents a specialized mathematical model that models the order of elements in a text sequence in a natural language text, for example in the form of a random process. An example of such a language model is the so-called ChatGPT. It is particularly advantageous to use a language model that works by using artificial intelligence and that preferably utilizes a trainable or trained neural network. The neural network can be trained, for example, by using corresponding patient information from a patient database and / or disease patterns from a disease database as input vectors.

[0021] Preferably, the large language model used to determine the disease information has a patient-specific large language model. For the patient-specific large language model, only specific patient information associated with a preset patient group is considered. For example, the preset patient group can only contain patients with the same or at least similar suspected diagnoses. For example, the corresponding specific patient information can be extracted from the corresponding patient database. In this way, better verification of the generated and output disease information is feasible because the mathematical model or the large language model has been specifically trained for a specific suspected diagnosis and a corresponding disease pattern. For example, it is also conceivable that the preset patient group is associated with a specific gender, age group, etc. It is also conceivable that the preset patient group has diseases that are generally associated with the disease pattern of the suspected diagnosis, such as age, the presence of known metabolic diseases, etc.

[0022] Preferably, the disease information determined by the large language model contains at least one correct proof sign for the created suspected diagnosis and / or at least one incorrect counter-proof sign for the created suspected diagnosis. Only the quantitative evaluation of proof signs and counter-proof signs can realize the temporary information of whether the suspected diagnosis is correct. Additionally, the proof signs and counter-proof signs can be appropriately weighted as necessary with respect to their importance. By computer-supported evaluation of proof signs and counter-proof signs, its weight is also considered as necessary, and automatic evaluation can be provided. In addition, in some diseases, there are diagnostic criteria (i.e. proof signs) clearly defined by guidelines. For example, if all clearly defined diagnostic criteria or proof signs are met, it can be considered that a diagnosis is made. It is also conceivable that the system as a whole can identify such a diagnostic criteria group and inform the user that this is most likely a certain suspected diagnosis.

[0023] The disease patterns extracted from the disease database can preferably be selected from a list of predefined disease patterns with the same or similar criteria. Additionally or alternatively, the disease patterns extracted from the disease database can be selected based on the identified criteria of the suspected disease pattern of the suspected diagnosis. Such criteria can be, for example, a specific symptom, the size, shape, manifestation, etc. of the symptom, or, for example, characteristics (itching, burning, pain). Preferably, the extraction of disease patterns can also be supported by using mathematical models, such as large language models.

[0024] In a preferred design, a suspected diagnosis is automatically created from patient information from a patient database, from a user input such as a doctor, and / or from information provided by the patient. Preferably, the creation of a suspected diagnosis can also be supported by using a mathematical model, such as a large language model.

[0025] In order to create a suspected diagnosis, according to one design, a medical admission report (Aufnahmebericht), i.e. a medical diagnosis report (medical report), is created in text form by a user, such as an attending physician or a corresponding assistant. Typically, the user directly enters / selects the suspected diagnosis, or registers the creation of the admission report together. The analysis of the diagnosis of the completed admission report may then be considered as quality control. In order to automatically or automatically create a suspected diagnosis, the medical admission report is input via a second user interface so that the data processing unit can then process the information contained therein to generate disease information.

[0026] In a particularly preferred embodiment, the medical admission report is automatically created by registering the user's input in an input field specially provided for this purpose in the electronic admission report form. The medical admission report is thus already in a quasi-standardized or structured form, which makes it easier for the data processing unit to use and further process the information contained therein. This enables more reliable disease information to be generated.

[0027] In a particularly preferred improvement of the present invention, the generated disease information is output together with the specific patient information for creating a suspected diagnosis and / or used by the mathematical model. For example, it can be proposed that there is a link to the used patient data of the patient database used by the mathematical model in the generated and output disease information. This means that not all patient information is output here, but in particular only the patient information that is the basis for the provided result for the mathematical model is output. If symptoms, signs, omens, etc. are identified in the medical diagnosis report, not only the information is output together with the generated disease information, but also the medical diagnosis report or an important extract therein. Therefore, through these additionally provided information, the user obtains additional background information, thereby can very simply check and verify the result, i.e. the disease information generated by the mathematical model.

[0028] In another preferred improvement of the present invention, one or more differential diagnoses are automatically determined based on the sick pattern extracted from the disease database by using a mathematical model. Differential diagnosis can also be understood as other conceivable diagnoses, such as diagnoses that are very frequently confused with the suspected diagnosis created. For example, a differential diagnosis can also be a disease with similar or almost identical symptoms, which must be considered by the user or doctor as a possible cause of the patient's discomfort in addition to the real suspected diagnosis. The number of differential diagnoses decreases with the specificity of symptoms or syndromes. The so-called general symptoms (such as fever) that occur in a variety of diseases have the least significance. On the contrary, if the symptoms are pathognomonisch, they simplify the diagnosis. However, differential diagnosis does not necessarily have to be adapted to disease symptoms, that is, other criteria can also be considered.

[0029] There are different possibilities for determining possible differential diagnosis: differential diagnosis can be selected based on the matching of the diseased mode of suspected diagnosis with patient data. For example, unsatisfied signs or standards in the diseased mode can indicate differential diagnosis. Additionally or alternatively, different possible differential diagnoses can be selected based on suspected diagnosis. For the selection of possible differential diagnosis, the corresponding diseased mode is then used and matched with patient information. The selection of possible differential diagnosis can be carried out based on rules, for example, if there is a specific suspected diagnosis, similar suspected diagnosis can be selected to detect. Similarly, the following suspected diagnosis can also be selected, in which experience has shown that the suspected diagnosis is often confused with a specific suspected diagnosis. In general, this mode causes that the generated disease information is verified with greater security. Additionally or alternatively, different possible differential diagnoses can also be determined directly from patient information, for example, together with suspected diagnosis. Preferably, the disease information generated for one of the differential diagnoses determined automatically is output via the first user interface.

[0030] According to one refinement, at least one refined / finely selected disease database is provided as the disease database. In the refined disease database, the information and criteria required for the predefined disease pattern are already stored in a structured form, so that they can be processed significantly more simply and especially more reliably by the data processing unit or the mathematical model. However, it is also conceivable as an alternative that the mathematical model derives the information and criteria required for the predefined disease pattern from a partially structured and / or textual description in the disease database.

[0031] Preferably, the disease database from which the disease pattern is extracted for matching purposes is selected via a third user interface (I / O).

[0032] As long as it makes sense, the above-mentioned designs and improvements can be combined with each other arbitrarily. Other possible designs, improvements and implementations of the present invention also include combinations of features of the present invention described above or below with respect to the embodiments that are not explicitly mentioned. In particular, those skilled in the art also add individual aspects as improvements or supplements to the corresponding basic forms of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Furthermore, possible embodiments of the method according to the invention and the system according to the invention are explained in more detail with reference to the drawings.

[0034] Figure 1 A flow chart illustrating one embodiment of a computer-implemented method for automatically providing disease information regarding a suspected diagnosis is shown;

[0035] Figure 2An application example for explaining the method according to the present invention is shown; and

[0036] Figure 3 A block diagram for illustrating one embodiment of a system for automatically providing information about a suspected diagnosed disease according to the present invention is shown. DETAILED DESCRIPTION

[0037] In the figures of the drawing, identical elements, features and components which have the same function and act in the same way are provided with the same reference symbols in each case unless otherwise specified.

[0038] The elements depicted in the drawings are not necessarily shown in proportion. This is used to clearly and unambiguously disclose the basis and principles of the present invention. Similarly, in some cases, known structures and devices are shown in the form of block diagrams to illustrate possible designs of the present invention.

[0039] The numbering of the steps in the method should simplify its description. The numbering does not necessarily imply a specific order of the steps. In particular, multiple steps can be performed simultaneously.

[0040] The detailed description of the drawings includes specific details so that a thorough understanding of the present invention can be achieved. However, it is apparent to one skilled in the art that the present invention can be practiced without these specific details.

[0041] In one possible embodiment, a computer-implemented method for automatically providing information about a suspected disease includes a plurality of steps, such as: Figure 1 As schematically shown in the flow chart.

[0042] In a first step S1 of the computer-implemented method, at least one disease database KDB and at least one patient database PDB are provided. The disease database KDB contains a plurality of disease patterns for one or more diseases. The patient database PDB contains a plurality of electronic patient files with patient information. The disease database KDB can be formed, for example, by a clinical information platform.

[0043] In a subsequent or preceding step (not shown in the figures), at least one suspected diagnosis is created for the patient to be examined, automatically or by a user, for example by a treating physician.

[0044] In another step S2 , based on the suspected diagnosis created by the user for the patient, at least a disease pattern for diagnosing the disease of the patient is extracted from the provided disease database KDB.

[0045] The user is a medically trained person, such as a specialist (e.g. a radiologist). The specialist creates a suspected diagnosis for the patient being examined. For this purpose, the specialist usually creates a written medical report in text form. Additionally or alternatively, it is also conceivable that the system also forwards the acquired information to a system with artificial intelligence (AI), so that in this way a fully automatic system can be realized without user involvement.

[0046] In one possible embodiment, the disease database KDB has a refined disease database having structured disease information about one or more diseases. In one possible embodiment, the disease database KDB for extracting disease patterns is selected by a user via a user interface I / O of the system.

[0047] The disease database KDB can be formed by a clinical information platform, which is customized for the corresponding use site and maps one or more related topics in the field of medical technology. The medical information platform is, for example, a refined database called "Thieme eRef" provided by Thieme Publishing House. However, other libraries, such as "Radiopaedia", can be considered. The medical information platform includes relevant information in a structured form for different medical professional fields. In addition, information sources are, for example, review papers, professional books, professional articles, etc. However, it is also conceivable that no additional information is provided by the system.

[0048] In a possible embodiment, suspected diagnosis can have the medical diagnosis report in text form created by the user for the patient, especially written medical report.The diagnostic report of suspected diagnosis can be input by the user via the user interface I / O of system.Also feasible is, automatically derive suspected diagnosis from the input that the user registers together in the input column of user interface, for the creation of diagnostic report.For example, the user inputs in the input column of input screen window.In the simplest case, the user can directly input or select suspected diagnosis via the user interface I / O of system, or the creation of report is registered together.

[0049] In another step S3, by using a mathematical model, preferably a pre-trained and finely trained large language model (Large Language Model LLM), the extracted disease pattern is matched or adapted with the patient information of the electronic patient file loaded from the patient database PDB, so as to automatically generate at least one disease information for a suspected diagnosis created by a user and input via a user interface. In the matching, it can be proposed, for example, that the criteria and features of the disease pattern are compared with the corresponding criteria and features of the suspected diagnosis. In the case of preset consistency, the suspected diagnosis can be confirmed. The information and other information can be components of the generated disease information. Patient-specific documents and patient data sets that can be provided by the information system are stored in the patient database PDB.

[0050] The large language model LLM used is a large generative language model with artificial intelligence. The language model is based on a neural network with a transformer architecture and can understand, process and generate natural language. The neural network consists of multiple neural layers and has different components for specific functions. In contrast to simple recurrent neural networks (feedback neural networks), neural networks with a transformer architecture are equipped with encoders and decoders that are passed through successively and a so-called self-attention mechanism (Self-Attention-Mechanismus). An example of a large language model LLM is GPT (Generative Pre-trained Transformer, generative pre-trained transformer).

[0051] The disease information determined or generated by the large language model (LLM) in step S3 has at least one confirming sign and / or at least one disconfirming sign for the correctness of the suspected diagnosis created by the user. Additionally or alternatively, it is also conceivable that the determined or generated disease information contains information about which no conclusion can be drawn based on the current patient information.

[0052] The large language model (LLM) for determining disease information preferably has a patient-specific large language model (PS-LLM) for the available patient data of a selected group of patients. In a possible implementation, the user or a user committee has the possibility to select the patients. The selected group includes, for example, patients who are or have been involved in a medical study.

[0053] Large language models (LLMs) usually use so-called contexts, which are usually preset by the database used. This can be understood to a certain extent as a cache of the large language model, that is, information that the large language model can access and which was previously provided to the system as input.

[0054] The large language model (LLM) can be available in pre-trained form or can also be trained, for example, using suitable training data. The training of the large language model (LLM) can be carried out, for example, based on existing, so-called open source information. Training based on known or existing medical use cases is also conceivable.

[0055] Based on the extracted disease patterns, one or more differential diagnoses can be automatically determined by using the same or another correspondingly trained large language model (LLM). In one possible embodiment, the disease patterns extracted from the disease database KDB can be matched with the patient data of the electronic patient file loaded from the patient database PDB by using the trained large language model (LLM) to automatically determine at least one disease information for one of the differential diagnoses automatically determined.

[0056] Other possible diagnoses, such as diseases with similar or nearly identical symptoms, are called differential diagnoses.

[0057] In step S4 of the method, at least one piece of disease information of the suspected diagnosis for the patient created by the user and determined in step S3 is output via the user interface I / O.

[0058] In one possible embodiment, the disease information of the suspected diagnosis created by the user for the patient that is determined and output via the user interface I / O has a link to the patient data and / or an excerpt of the patient data in the electronic patient file of the patient.

[0059] Furthermore, in one possible embodiment of the method, in step S4 the differential diagnosis automatically determined in step S3 can additionally be output to a user via the user interface I / O.

[0060] In a possible implementation of the system SYS, users can share relevant information with each other through the sharing function of the disease database KDB or clinical information platform used. Users can access the disease database KDB via a data network (e.g., the Internet) with the help of a fixed or portable terminal device UE. The terminal device UE has, for example, a PC, a tablet computer, or a smartphone. With the help of an API (Application Programming Interface), users can access the digital content of the platform and temporarily store the digital content in a data memory for further data processing by a data processing unit DVE.

[0061] The content of the clinical information platform that is constituted as the disease database KDB includes information about specific medical states. These medical states are preferably associated with terminology standards, such as SNOMED (systematized medical terminology set) via standard coding. Meaningful units (Bedeutungstragende Einheiten) are also referred to as designs in SNOMED. Each design is characterized by a specific name (such as pharyngitis), a coding (such as 405737000), a standard and a partial text definition. For this reason, synonymous terms are preferably used. All terms are referred to as descriptions. Each design can be embedded in the semantic web via a relationship. Standard coding is also used in electronic medical data sets and in health information systems.

[0062] A medical state here may be a disease or a symptom of a disease. A medical state or a plurality of medical states may represent a disease pattern.

[0063] For a specific patient ID of a patient and about a suspected diagnosis existing for said patient, the data processing unit DVE of the system SYS can access fully structured patient information, partially structured patient information (e.g. medical textbook information) and unstructured patient information (e.g. doctor's letters). In the method according to the invention, a trained large language model (LLM) is used in order to evaluate or analyze the patient information available for the patient and his suspected diagnosis created by the user. For example, a correspondingly trained large language model (LLM) implemented on the data processing unit DVE of the system SYS can be used in order to search for corresponding information about the gender of the relevant patient (e.g. the patient is "female") in medical records or documents stored, for example, in the patient's electronic patient file, for information about gender prediction contained in a medical textbook database (e.g. a disease "occurs more often in men"). The corresponding information is derived not only from the patient database ("the patient is male") but also from the disease database ("the disease pattern is more common in men"). Even if the textbook database and the patient database contain only unstructured text snippets (e.g., "the most frequently manifesting symptoms of the disease pattern"), the trained large language model (LLM) is able to analyze the unstructured text and associate it with the relevant text snippets of the medical records within the patient's file in order to generate a chronologically classified list of patient symptoms that match the symptoms described or illustrated in the textbook database.

[0064] For example, the patient data set of the electronic patient file of the examined patient obtained or loaded from the patient database PDB of the health information system can be compared with the structured information of the refined text-based disease database KDB, such as in Figure 2 As shown in .

[0065] exist Figure 2The patient data from the health information system shown on the left side of the figure includes information about the patient. The patient data, for example, describes age, gender, living conditions, laboratory values, existing antibodies, etc. In addition, the patient data may also contain information about disease history, current and previous symptoms, patient medical history, other assessments (such as pathology, radiology, etc.). The patient data is compared with structured information obtained from at least one text-based refined disease database KDB. KDB entries may have references to documents or literature and / or links to documents. The patient data includes references found in the documents, which correspond to or match the required information in the refined disease database KDB with a high probability. Figure 2 On the right are shown three entries in the refined disease database KDB. Figure 2 The information shown in can be displayed on a graphical I / O user interface to the treating user or specialist.

[0066] The trained large language model LLM provides not only the patient's gender / symptoms in the example shown, but also the original text that may answer the question. The large language model LLM also contains background information, which is derived in particular from existing patient information. The specific patient information is then matched or adapted to the given conditions for the disease pattern using the trained large language model LLM. The refined disease database KDB already provides information for similar diseases, which can be used for differential diagnosis.

[0067] The present invention also includes a computer program product (not shown) having a stored program, the program comprising program instructions for executing the method for automatically providing information about a suspected diagnosed disease according to the present invention.

[0068] The computer program product is a non-volatile computer-readable data storage medium, which includes an executable program code, which is designed to perform the method according to the invention when it is run. The non-volatile computer-readable data storage medium may include or consist of any type of computer memory, in particular semiconductor memory, such as solid-state memory. The data storage medium may also include or consist of a CD, DVD, Blu-ray disc, USB memory stick, etc.

[0069] The present invention also implements a system SYS for automatically providing information about a suspected disease diagnosis, as schematically shown in FIG. Figure 3 is shown in the block diagram.

[0070] The system SYS has a data processing unit DVE which accesses at least one disease database KDB having structured disease information about one or more diseases stored therein. Figure 3 The data processing unit DVE of the system SYS shown in FIG. 1 also accesses at least one patient database PDB having electronic patient files of the patients stored therein.

[0071] exist Figure 3 The data processing unit of the system SYS shown in FIG. 1 has one or more processors and is configured to automatically extract disease patterns for diagnosing a patient's disease from a disease database KDB based on a suspected diagnosis created by a user for the patient. Figure 3 The data processing unit DVE of the system SYS shown in the figure is also configured to automatically determine at least one disease information for a suspected diagnosis created by a user or a specialist by using a trained large language model (LLM) to match or adapt the extracted disease pattern with the patient data of the electronic patient file of the patient loaded from the patient database PDB. The system SYS includes a user interface I / O of a user terminal device UE, which is used to output the at least one determined disease information to the corresponding user. For the disease information, an evaluation of the correctness probability can also be calculated and displayed.

[0072] Although the present invention has been completely described above on the basis of preferred exemplary embodiments, it is not restricted thereto but can be modified in many ways and methods.

[0073] Thus, the computer-implemented method according to the invention or the computer-implemented system according to the invention can be used in a variety of ways in the medical field, for example. For example, it can sometimes be difficult for a primary physician to match a patient with advanced cancer with a patient participating in one of the other ongoing medical studies, all of which have strict conditions for participation. In this case, the trained large language model LLM can derive an assessment that associates inclusion / exclusion criteria with the patient's status, wherein this provides a proof of the assessment by associating relevant medical text excerpts with the study criteria.

[0074] Regardless of the grammatical gender of a particular term, persons with either male or female gender identities are included.

Claims

1. A computer-implemented method for automatically providing disease information about verifying a suspected diagnosis, the method comprising the following steps: Providing (S1) at least one disease database (KDB) and at least one patient database (PDB), wherein the disease database (KDB) contains a plurality of disease patterns related to one or more diseases, and the patient database (PDB) contains a plurality of electronic patient files having patient information; Extracting (S2) a disease pattern for diagnosing the patient's disease from a provided disease database (KDB) based on the suspected diagnosis; Matching the extracted disease pattern with the patient information of the electronic patient file of the patient loaded from the patient database (PDB) by using a mathematical model for automatically determining at least one disease information for the suspected diagnosis (S3); The determined disease information is output (S4) via a first user interface (I / O).

2. The method according to claim 1, In this case, the disease information is automatically generated by using a mathematical model designed as a large language model (LLM), in particular a trained large language model (LLM).

3. The method according to claim 2, The large language model used for determining the disease information comprises a patient-specific large language model (PS-LLM), which takes into account specific patient information associated with a predefined patient group.

4. The method according to claim 2 or 3, The disease information determined by the large language model (LLM) includes at least one confirmatory sign indicating that the created suspected diagnosis is correct, and / or at least one disconfirmatory sign indicating that the created suspected diagnosis is incorrect.

5. The method according to claim 2 or 3, The disease information determined by the large language model (LLM) includes the following information: no conclusion can be drawn as to whether the created suspected diagnosis is correct and / or incorrect.

6. The method according to any one of the preceding claims, The disease pattern extracted from the disease database (KDB) is selected from a list of predefined disease patterns and / or selected according to criteria for identification of a disease pattern guessed at the suspected diagnosis.

7. The method according to any one of the preceding claims, In this case, the suspected diagnosis is automatically created from patient information from the patient database, user input and / or information provided by the patient, in particular when using a large language model.

8. The method according to claim 7, A medical admission report input via a second user interface (I / O) for automatically creating the suspected diagnosis is created in text form.

9. The method according to claim 8, The medical admission report is automatically created by including the user's input in an input field provided specifically for this purpose in an electronic admission report form.

10. The method according to any one of the preceding claims, The determined disease information is output together with specific patient information for creating the suspected diagnosis and / or used by the mathematical model.

11. The method according to any one of the preceding claims, One or more differential diagnoses are automatically determined by using the mathematical model based on the disease patterns extracted from the disease database (KDB) and information about the patient from the patient database.

12. The method according to any one of the preceding claims, In this case, at least one refined disease database is provided as the disease database, in which the information and criteria required for a predefined disease pattern are already stored in a structured form.

13. The method according to claim 12, A disease database is selected via a third user interface (I / O) from which the disease pattern is extracted for matching purposes.

14. A computer-implemented system (SYS) for automatically providing information about a disease that verifies a suspected diagnosis, the system (SYS) having: at least one disease database (KDB), the disease database (KDB) comprising a plurality of disease patterns related to one or more diseases; at least one patient database (PDB), said patient database (PDB) comprising a plurality of electronic patient files having patient information; A data processing unit (DVE), the data processing unit (DVE) is configured to - extracting disease patterns for diagnosing the patient's disease from the disease database (KDB) based on the suspected diagnosis, - matching the extracted disease pattern with the patient information of the electronic patient file of the patient loaded from the patient database (PDB) by using a mathematical model for automatically determining at least one disease information for the suspected diagnosis; and A user interface (I / O) for outputting determined disease information.

Images