Ai-based conversion of patient free-text into quantitative evaluation

Abstract

The invention relates to a computer-implemented method for controlling a selection of an intraocular lens to be inserted. The method comprises receiving and storing subjective patient information in free-text form, determining numerical, patient-relevant parameter values for predefined parameter categories from the subjective patient information in free-text form by means of a trained machine-learning system, which comprises a trained machine-learning model, and generating a control parameter value for selecting the intraocular lens to be inserted on the basis of the numerical, patient-relevant parameter values.

Description

AI-BASED TRANSFER OF PATIENT FREE TEXT INTO QUANTITATIVE ANALYSIS Field of invention

[0001] The invention relates to a method for selecting an intraocular lens to be used, and more precisely to a computer-implemented method for controlling the selection of an intraocular lens to be used. The invention further relates to an analysis system for controlling the selection of an intraocular lens to be used, and to a computer program product. Technical background

[0002] A typical area of ​​ophthalmology involves replacing the natural lens of the eye with an intraocular lens (IOL) in patients with conditions such as cataracts. A wide range of tools exists for determining the appropriate IOL for a given patient, from classical mathematical formulas to artificial intelligence systems. The methods for determining the required IOL are relatively straightforward.

[0003] On the other hand, there is the patient history, during which information is collected in various forms. On the one hand, specific ophthalmological values ​​and / or medical conditions can be measured. On the other hand, this and / or further information can be elicited from the patient. This can occur in complex and diverse ways, for example, in written free text, as patient statements during the history interview, or in other formats. A typical characteristic of this type of patient information is that it is difficult to analyze and may not be readily integrated into a quantitatively usable context for treatment. Furthermore, subjective misjudgments can occur when completing a questionnaire because a patient, for example, does not understand a question, misinterprets it, or simply misjudges their own condition or behavior. Such errors, or...Misjudgments are difficult to identify in a standardized way. As a result, the doctor's interview time is not used optimally and effectively.

[0004] Therefore, there is a need for better automated classification of patient data collected during the medical history, for example for upcoming Cataract surgery or to correct refractive error when patients no longer wish to wear glasses. Overview of the invention

[0005] This problem is solved by the method, the corresponding system, and the associated computer program product proposed here, in accordance with the independent claims. Further embodiments are described by the respective dependent claims.

[0006] According to one aspect of the present invention, a computer-implemented method for controlling the selection of an intraocular lens to be used is presented. The method comprises receiving and storing subjective patient information in free text form, determining numerical, patient-relevant parameter values ​​for predefined parameter categories from the subjective patient information in free text form using a trained machine learning system, which includes a trained machine learning model, and generating a control parameter value for selecting the intraocular lens to be used based on the numerical, patient-relevant parameter values.

[0007] According to another aspect of the present invention, an analysis system for controlling the selection of an intraocular lens to be used is presented. The system comprises a processor and a memory operationally coupled to the processor, wherein the memory stores program code elements which, when executed by the processor, cause the processor to: receive and store subjective patient information in free text form; determine numerical, patient-relevant parameter values ​​for predefined parameter categories from the subjective patient information in free text form using a trained machine learning system comprising a trained machine learning model; and generate a control parameter value for selecting the intraocular lens to be used based on the numerical, patient-relevant parameter values.

[0008] The proposed computer-implemented method for controlling the selection of an intraocular lens to be used has several advantages and technical effects that can also apply to the associated system:

[0009] The presented concept solves the task outlined above; in particular, it enables the quantification or structured examination of unstructured patient information, which is recorded, for example, during a medical history, in order to make an assisted or automated selection of an intraocular lens to be implanted in a patient.

[0010] The presented concept utilizes machine learning functions to evaluate the collected patient data / patient information and to ensure better quality and higher information content of the collected patient data through interactive data collection. This is achieved at least partially through automated mapping of information or patient details that consist at least partially or predominantly of free text.

[0011] This also fulfills the requirement that this information extraction be performed in the form of predefined quantitative parameters relevant to medical treatment. Mapping and relating the data to the information ensures efficient use of free-text information, optimization of the obtained information with regard to the planned treatment, and an (optional) consistency check of the patient's assessment. In this way, contradictory patient statements can be identified and, if necessary, corrected, resulting in a consistent overall picture of the patient's requirements for their future intraocular lens. Consequently, the time invested by both the physician and the patient is optimally utilized to derive the greatest possible amount of reliable quantitative data for selecting the intraocular lens from the patient's information.

[0012] This takes place against the following background: Currently, a patient's medical history is typically obtained through an interview. This can optionally be supplemented by a questionnaire. The interview information cannot yet be evaluated in a standardized way—or only to a very limited extent. The proposed solution creates the possibility of a structured evaluation of the medical history information, which also provides the option of a (optional) systematic consistency check of the patient's statements. In this way, the proposed procedure allows the direct use of free-text information in a standardized and quantified context for planning surgery (in short: surgical planning) for the implantation of an intraocular lens. This goes far beyond known methods of postoperative satisfaction surveys through the structuring of patient feedback. This can also be achieved, among other things, by... Preoperative quantification of the available information about the patient can be achieved.

[0013] The advantage is achieved, among other things, through a typically AI-based text processing algorithm that is sufficiently trained and / or must have been initiated with a prompt to convert the free-text information into a quantitative scheme suitable for surgical planning. The free-text information can be evaluated either alone or in combination with other structured information—such as a patient questionnaire—to then perform the necessary quantification interactively or subsequently. This process can then include consistency checks, filling in information gaps, clarifying uncertainties, and / or correcting typos and / or incorrect entries.This allows patient statements to be checked against specific criteria – such as clarity, completeness, relevance, minimum quality, conscious weighting – and the identified deficiencies to be eliminated through defined questioning strategies.

[0014] It is therefore relatively easy to recognize the potential that arises from the difference compared to established approaches, which—if they rely at all—on purely questionnaire-based solutions. This typically restricts the patient or respondent unnecessarily in their responses, as unusual life situations are potentially not captured by the questionnaire. The advantage of free-text patient responses is eliminated by these limitations, thus allowing for better categorization with regard to patient-relevant numerical parameter values ​​and ultimately the control parameter value.

[0015] Furthermore, the procedure can also prepare or carry out the classification of the patient into a treatment group.

[0016] In this way, greater patient satisfaction can be achieved with regard to their subjective vision or the usability of their vision in their individual life circumstances after the operation.

[0017] Furthermore, the procedure allows information to be combined to reach a conclusion that cannot be derived from the individual pieces of information and that can easily be overlooked or not drawn in the same way in a classic medical history.

[0018] The following are further examples of implementation that may be valid both in connection with the method and with the corresponding system.

[0019] (2) According to a supplementary embodiment of the method, receiving the subjective patient information in free text form can involve capturing the subjective patient information. The capturing of the subjective patient information can occur in response to a set of questions. This can be a single, e.g., open-ended question, or it can be a pre-printed questionnaire containing open-ended and / or closed-ended questions. The set of questions can therefore comprise one or more questions. Electronic capture can be performed by input from the patient, an authorized assistant (e.g., a mother for a child), or—in the case of handwritten answers—also by handwriting recognition via OCR (optical character recognition).

[0020] Alternatively, transcribed audio (or video) recordings are also possible. Alternatively, the questions can also be asked by a treating physician or a medical assistant. This allows, for example, pre-scoring for verification and / or adjustment of each question. The physician or medical assistant can also ask follow-up clarifying questions. At least some of the answers should be in free text format. Furthermore, a single question can address several sets of questions or answers simultaneously. This allows the patient to easily describe their daily life without having to ask questions for every single instance. The subsequent AI-driven analysis can then separate the individual sets of questions and answers.

[0021] Furthermore, according to a supplementary embodiment of the method, the patient data can be collected interactively. For example, a chatbot can be used that also generates follow-up questions – perhaps in a generative manner. Alternatively, a physician or other interviewee can ask follow-up questions and / or clarifying questions in a multi-stage process. In such a case, a free-text form would not necessarily be required, although this would normally be easier for a patient to use. Preferably, at least some of the subjective patient information is provided or recorded in free-text form in response to the question(s) and any follow-up questions.

[0022] Consequently, an advantageous embodiment of the method may include the recording of subjective patient information—particularly in free text form—using at least one of the following aids: a free text input field, a questionnaire, a chatbot, and a recording of the conversation, e.g., as an audio stream, via text recognition on a handwritten note on paper (e.g., a photograph), or as handwritten input on a touch-sensitive electronic device (e.g., a tablet computer). Anything that makes it easy for the patient to provide their input in free text form can be used.

[0023] Another possible embodiment of the procedure involves capturing subjective patient information – particularly in free text form – by additionally recording patient information based on a multiple-choice procedure. This is generally done as a supplement to the free text input, since the free text can also contain unforeseen content.

[0024] According to a further advantageous embodiment of the method, the collection of subjective patient information—particularly in free text form—can be carried out in multiple stages. A first stage can consist of a set of questions—especially at least one question—for the subjective patient information in free text form. In the second stage, an interactive query can be generated in response to the subjective patient information collected in free text form in the first stage. The answer to this query can be used to clarify and / or correct—and possibly also specify—the subjective patient information from the first stage. A specially trained chatbot, for example, can be used to generate the interactive query. Repeated iterations of this process would be possible.The process could therefore consist of a set of questions, whereby at least one question (and / or patient input in response to a question) would generate an interactive follow-up question, which could then lead to another set of questions, which would again be considered interactive follow-up questions. This cycle can be repeated multiple times. Furthermore, it is possible that for a (predefined) set of questions, consisting of at least one question, (at least) one interactive follow-up question is generated, followed by another (predefined) set of questions, for which (at least) one interactive follow-up question is again generated, and so on. In this way, several predefined sets of questions can be processed in a structured manner, with (at least) one interactive follow-up question being generated after / for each set of questions, preferably in response to the respective subjective patient information recorded in free text form.

[0025] According to an advantageous further development of the method, an embodiment is also possible that further enables the detection of at least one content-related inconsistency in the subjective patient information—particularly in free text form—and / or between the subjective patient information and reference patient information. This embodiment allows for a variety of technical implementations. The inconsistencies can arise, for example, from differing information regarding different question categories. The term "inconsistency" can be interpreted broadly in this context, so that unrealistic information or information gaps can also be detected. If, for example, a patient states that they work 40 hours per week, but on the other hand pursues leisure activities that take up at least three days per week, this results in an obvious inconsistency.The inconsistency can arise directly from the free-text information, from a potential contradiction between free-text statements and answered questions, or from other information sources concerning the patient. Furthermore, reference patient data, i.e., information from a patient reference group, can be used. For this purpose, a database search can be conducted with an analysis of patient data of any kind. This analysis can focus on specific numerical, patient-relevant parameter values ​​or a direct analysis of the received patient data. For reference patient data, a cross-section of subjective patient information from a large number of patients or a selected subgroup assigned to the patient can be used. Alternative assignment mechanisms are possible.

[0026] Accordingly, an additional embodiment of the method can include identifying a content gap within the subjective patient information. This can be achieved by comparing the patient information with a defined minimum requirement. Furthermore, an additional embodiment of the method can include identifying unrealistic information in the subjective patient information using a reference. The use of a verification engine based on machine learning mechanisms can be employed to identify content inconsistencies, information gaps, and / or unrealistic information, just like other available data, such as from a database or directly from the internet (e.g., social media data).

[0027] According to a further developed embodiment, the method can – for example, in response to a detected inconsistency – involve adjustment (possibly correcting, replacing, inserting [e.g., in the case of an information gap]), e.g., by means of (interactive) questioning. (so) or via auto-correction) of the subjective patient information. The identified information gap and / or inconsistency and / or unrealistic information can be eliminated before the extraction or interpretation (predicting) of numerical, patient-relevant parameter values ​​using machine learning takes place. The resulting advantage is that the identified inconsistency and / or the content gap and / or the unrealistic information are not reflected in the numerical, patient-relevant parameter values.

[0028] According to another elegant embodiment, the method can further involve weighting the determined, numerical, patient-relevant parameter values—using, for example, predefined weighting parameter values—to determine outcome parameter values. Typically, the outcome parameter values ​​are reflected in the control parameter value described above. This allows for fine-tuning when determining outcome parameter values ​​or the control parameter value.

[0029] According to another interesting embodiment, the method can also include processing the result parameter values ​​into a recommendation vector using predefined rules. This processing can also be understood as a condensation—in the sense of a mathematical combination of values ​​into new values—of the result parameter values. The recommendation vector can then be used to determine the control parameter value. A specially trained machine learning system with a corresponding machine learning model can also be used for this purpose.

[0030] Regardless of the above, in a particular embodiment, the method can also include the direct prediction of the control parameter value by the trained machine learning system using the trained machine learning model. This would eliminate the need for "detours" via result parameter values ​​and recommendation vectors. This could accelerate the intended process or require fewer technical resources.

[0031] On this basis, a further advantageous embodiment may consist in the machine learning model of the trained machine learning system of the presented concept (e.g. according to claim 1) being trained with training data comprising the following: a plurality of patient details in free text form, each with associated Numerical, patient-relevant parameter values ​​for predefined parameter categories, as well as any corresponding control parameter values, would be used. Training with such a triplet would eliminate the need for further calculations, thus saving resources.

[0032] Optionally, according to a further possible embodiment, the machine learning model of the proposed method can be based on a trained large-language model. Generally, a language model would be used instead of the large-language model.

[0033] According to an extended embodiment, the method can further include the deselection of intraocular lenses to be used—or even implantable—which may be stored, for example, in an automated warehouse or available in an electronic ordering system, based on the control parameter value. The deselection of intraocular lenses to be used can thus be taken into account, for example, in an automated procurement process and / or an automated ordering process and / or in an electronic ordering system. In general, the method can include the deselection of intraocular lenses to be used from a predefined set of intraocular lenses based on the control parameter value. In this case, a deselected intraocular lens would not be proposed when applying the extended method. Thus, deselection can be understood as a de-selection or...Removal from a selectable set, particularly from a set of intraocular lenses (IOLs), is understood as the process of deselecting an intraocular lens. There can be several reasons for deselecting an IOL. For example, it might be known (at least partially from the patient's information) that the selected insurance company does not cover the corresponding IOL model (e.g., from a specific manufacturer and / or lens type), and the patient is unwilling to pay for it, thus ruling out that IOL model.

[0034] The procedure can also take other parameters into account. For example, deselection can occur independently of the control parameter value. For instance, certain intraocular lenses may be unavailable or out of stock. For example, a drawer where the specific intraocular lens should be located may be empty. Such lenses can also be deselection.

[0035] According to a supplementary embodiment of the method, the Control parameter value as input value for an intraocular lens prediction system for a The intraocular lens to be implanted can be used. The intraocular lens prediction system – for example, in the form of a trained machine learning model, possibly based on a trained neural network – can additionally use measured ophthalmological data from a patient assigned to the patient's information as input values. In this way, a completely automated pipeline from subjective patient information to the automated selection of the intraocular lens to be implanted could be realized. In particular, the lens type and power of the intraocular lens to be implanted can be determined in this way.

[0036] According to a further developed version of the supplementary embodiment of the method just described, the measured ophthalmological data can be given greater weight as input values ​​for the intraocular lens prediction system than the control parameter value. In short, objectivity trumps the subjectivity of subjective patient information; or, put another way, measured ophthalmological data takes precedence over numerical values ​​or control parameter values ​​based on subjective patient information. For example, this greater weighting could be implemented by assigning a higher value to the measured ophthalmological data. Furthermore, it could be implemented by considering the measured ophthalmological data earlier in the process, for example, in a sequential execution.For example, a two-stage approach could be used, whereby the intraocular lens prediction system takes into account the (specific) measured ophthalmological data in a first stage and the subjective patient information (using the control parameter value) in a second stage.

[0037] According to another further developed form of the procedure, the intraocular lens prediction system can, in a first step, consider the control parameter value to select a lens type of the intraocular lens to be inserted, and in a second step, consider the measured ophthalmological data to select a power of the intraocular lens to be inserted.

[0038] According to an extended embodiment of the method, the intraocular lens prediction system can predict several IOL parameter values ​​for an implanted IOL and, using the predicted IOL parameter values, generate an output selected from the group consisting of: a (computer-aided / -generated) simulation – e.g., using a digital eye twin – an image representing a visual impression with an IOL implanted in the patient, which which represents the predicted IOL parameter values; and a display of a focus curve and / or defocus curve for one eye of the patient with an implanted IOL that exhibits the predicted IOL parameter values.

[0039] It is worth emphasizing that the digital Eye-Twins, with its (de)focus curve, takes into account the entire optical system of the eye, right up to the image being projected onto the retina. This includes consideration of the IOL, the cornea, and distances within the eye, as well as the natural lens, if present.

[0040] Should the intraocular lens prediction system suggest more than one intraocular lens based on the measured ophthalmological data, the overall system according to the invention, which is also used to weight the possibilities and recommend the intraocular lens to be selected, can be implemented in the form of a decision tree. In this scenario, the measured ophthalmological data would provide an initial set of possible intraocular lenses, and subsequently, subjective patient information would be used to propose a final recommendation for the intraocular lens to be implanted, based on this initial set of results.

[0041] Consequently, and according to a further developed embodiment of the method, it can include automated selection and / or procurement of the intraocular lens to be used from an automated warehouse. If the selection and / or procurement is semi-automated, an optical signal (e.g., LED) can be triggered on a container holding the corresponding intraocular lens, based on the control parameter value. Additionally or alternatively, the method can include automated selection in an ordering system, whereby the selected lens can be displayed to the user for placing an order for the intraocular lens to be used.

[0042] According to a supplementary embodiment, the control parameter value for selecting the intraocular lens to be used can be directly determined by the numerical, patient-relevant parameter values. In this case, generating the control parameter value would therefore consist of directly using the numerical, patient-relevant parameter values.

[0043] Based at least partially on the preceding aspects and the described embodiments, a further developed embodiment can also be based on the following concept. Implementation: If an ideal IOL is unavailable for an upcoming IOL implantation, the extended procedure can automatically evaluate the available IOLs. These IOLs are assigned a corresponding rating factor. This information can be presented to a user, such as clinical staff, for example, via a user-friendly interface (LLM) using alphanumeric display and / or speech output. The user can then decide which of the suitable IOLs should be used. A system output could read: "Only IOL X is available, which does not perfectly match your preference."<Liste der vorhandenen IOL(en)> This would mean the following differences from the desired visual acuity. Is this tolerated? (readily available second best choice).

[0044] The procedure would also include receiving a control signal - especially one triggered by the user - which would then directly influence the value of the control parameter.

[0045] Furthermore, embodiments may relate to a computer program product that can be accessed from a computer-usable or computer-readable medium containing program code for use by, or in conjunction with, a computer or other instruction processing systems. In the context of this description, a computer-usable or computer-readable medium may be any device suitable for storing, communicating, transmitting, or transporting the program code. Overview of the characters

[0046] It should be noted that embodiments of the invention may be described with reference to different implementation categories. In particular, some embodiments are described in relation to a method, while other embodiments may be described in the context of corresponding devices. Regardless, unless otherwise indicated, a person skilled in the art can recognize and combine possible combinations of the features of the method and possible combinations of features with the corresponding system from the preceding and following description, even if they belong to different claim categories.

[0047] Aspects already described above, as well as additional aspects of the present invention, arise, among other things, from the described embodiments and from the additional specific embodiments described with reference to the figures.

[0048] Preferred embodiments of the present invention are described by way of example and with reference to the following figures: Fig. 1 shows a flowchart-like representation of an embodiment of the computer-implemented method according to the invention for controlling the selection of an intraocular lens to be used. Fig. 2 shows elements of an embodiment of the proposed method using individual functional blocks. Fig. 3 shows elements of an embodiment of the proposed method in the context of several interactions. Fig. 4 shows various elements of the proposed method from a different perspective. Fig. 5 illustrates the proposed process again based on the different data types used. Fig. 6 shows an embodiment of an analysis system for controlling the selection of an intraocular lens to be used. Fig. 7 shows an embodiment of a computer system which has the system according to Fig. 6. Detailed character description

[0049] In the context of this description, conventions, terms and / or expressions should be understood as follows:

[0050] In the context of this document, the term 'selection control' describes, on the one hand, the use of at least one automatically generated suggestion or recommendation for a user, such as clinical staff like a surgeon or surgical assistant, regarding the selection of an intraocular lens (IOL) to be implanted in a patient, based on the patient's subjective information. A selection can also include a pre-selection. The selection can suggest exactly one specific IOL or several IOLs. Furthermore, the selection can also be automated, with a system using control parameter values ​​to select a specific IOL—again based on the patient's subjective information—directly from a stock and / or an ordering system. The IOL selected directly from the stock or ordering system can then be suggested to and / or made available to a surgeon.

[0051] The term 'intraocular lens' (abbreviated IOL) describes an artificial lens that is surgically implanted into the eye to replace the natural lens. A typical application is cataract surgery. In addition to conventional intraocular lenses, plOLs (phakic intraocular lenses) will also be considered here.

[0052] Selecting an IOL to be implanted can be understood as choosing one or more of the following options / parameters: an IOL type (monofocal, monofocal plus / enhanced monofocal, multifocal [e.g., bifocal, trifocal], extended depth of focus (EDOF), modular, accommodating, toric, light-adjustable lens), a power (spherical refractive power, cylinder power) or refractive index of the IOL, a working distance of the IOL (position of the focal point or focal points in the case of a multifocal IOL), etc.

[0053] For selecting the IOL to be implanted, a set of IOLs, particularly those with discrete sets of parameter values, may be available. Each IOL can have a set of parameter values, which can be discrete, i.e., not continuously selectable. To select an IOL, IOL parameter values ​​can generally be predicted, allowing for the selection of a suitable IOL. Alternatively, a subset of the available set of IOLs can be chosen. This subset may include one or more IOLs that are suitable for implantation.

[0054] The term 'subjective patient information' – sometimes also referred to simply as 'patient information' – can encompass everything a patient expresses, regardless of the medium, to describe their daily life, habits, visual requirements, work demands, etc. This can occur in response to one or more questions posed to the patient, with at least one question requiring the patient to provide free-text input. This can be done via keyboard or voice input. The audio track of a video could also be used for this purpose. Alternatively, the patient could provide a handwritten answer to the at least one question, with the text then automatically recognized and digitized using OCR. Voice input would be transcribed, so the information would also be available in digital text form.“Subjective” can mean that the patient information is given from the patient's perspective; for example, the patient (or their guardian or caregiver) provides information about themselves based on their own perceptions. These subjective patient statements are distinct from a mere reproduction of medical reports and / or measurement data.

[0055] This also provides a relatively good description of the term 'free text form'. For example, in addition to one or more questions to which the patient responds freely (i.e., in free text form), one or more follow-up questions are possible to further specify or correct a described circumstance. Furthermore, a questionnaire is also possible, either additionally or alternatively. A multiple-choice procedure can also be used to quickly gather certain facts, for example. Nevertheless, at least some of the patient's information should always be in free text form.

[0056] The term 'parameter values' is used here to describe numerical values ​​that are automatically determined from at least part of the patient information using a trained machine learning system, starting from the free text input.

[0057] The term 'predefined parameter categories' describes categories for which numerical, patient-relevant parameter values ​​are to be determined. These categories relate to the patient's practical activities and may include, for example: "need for distance vision in everyday life," "need for near vision in everyday life," "dominance of one eye," and "pre-operative limitations in visual acuity," to name just a few. These parameter categories may, for example, allow values ​​in the range of 0-10 (other scale values ​​are possible) and may also be subject to a weighting factor.

[0058] The term 'machine learning system' describes a system that generates predictions based on input data. A machine learning system is not procedurally programmed, but rather "learns" its behavior through training with training data. The machine learning system is provided with input data as well as corresponding desired output data, known as "ground truth data." Typically, machine learning systems in this context are based on neural networks. These can be implemented as software and data constructs, in hardware, or a combination of both. Furthermore, machine learning systems can provide a quality assessment alongside the prediction data, for example, "the predicted value is correct with a probability of x%."

[0059] In this context, and for the purposes of this document, the difference between a machine learning system and a "machine learning model" should be emphasized. The machine learning system represents the fundamental architecture of the learning system, while the machine learning model is developed during training with training data. Consequently, it is a trained ML system with a trained ML model. Additionally, it is also possible to implement the fully trained ML system and its trained ML model in hardware (e.g., as an FPGA - Field Programmable Gate Array). This allows for comparatively fast predictions.

[0060] In the context of this document, the term 'control parameter value' describes a numerical or digital value that can be automatically processed to select the intraocular lens to be used. Controlling an inventory management system or an ordering system for intraocular lenses would be one possible application of the control parameter value.

[0061] The term 'Large Language Model' (LLM) can generally describe a computer-based model adapted for general language output / generation. Alternatively, these systems / models can also be used for other natural language processing or for classification. A typical—and well-known—application is the generation of new texts. However, the property needed in the context of this text is more specifically classification, in order to generate the parameter values ​​for the various parameter categories from the patient's free-text input. Such systems are trained with a large number of texts, some of which are domain-specific, whereby—to put it simply—a multidimensional space is defined through word embedding. Upon request (corresponding to `mind`, an input term), the system determines the relevant parameters. The system then calculates the shortest multidimensional distance between the embedding vector of the input term and the embedding vectors of the multidimensional space in order to generate an output (i.e., response) that is interpreted as an answer to the input term. This response need not be—but can be—new text; it can also be a classification based on preferred, predefined parameter categories.

[0062] The term 'intraocular lens prediction system' describes a system that, based on pre-operative ophthalmological measurements, can predict which intraocular lens should be implanted – for example, after a cataract diagnosis. Typically, available measured eye parameters and a target corrected visual acuity (target refraction) are used as input values ​​for the intraocular lens prediction system to obtain guidelines for selecting a suitable intraocular lens.

[0063] A detailed description of the figures follows. It should be understood that all details and instructions in the figures are shown schematically. First, a flowchart-like representation of an embodiment of the computer-implemented method according to the invention for controlling the selection of an intraocular lens to be used is presented. Subsequently, further embodiments, or embodiments for the corresponding system, are described:

[0064] Fig. 1 shows a flowchart-like representation of a preferred embodiment of the computer-implemented method 100 for controlling the selection of an intraocular lens to be inserted. The method 100 includes the receipt, 102, of subjective patient information in free text form. This can involve interactive data capture (question-answer follow-up, etc.) or linear capture (batch capture). The patient can provide the information independently or receive assistance from an interviewer (for example, a mother for her child), a physician, or a chat system. The method can also include subsequent storage of the subjective patient information. Any audio recordings made—e.g., using an app—can be transcribed and converted into digital form. The same applies—via OCR—to handwritten text.The information collected in this way can then be compared in a further step with a scheme filled out manually by / for the patient in order to check the consistency of the patient's information and / or be used directly for further treatment planning.

[0065] Furthermore, procedure 100 involves the determination, 104 – for example, also in the sense of extraction or interpretation, prediction, all using a machine learning system – of numerical, patient-relevant parameter values. The parameter values ​​are determined for predefined parameter categories, such as specific behaviors (e.g., in everyday life or other predefined environments) and / or specific preferences. The parameter values ​​are structured data that can be represented predominantly as numerical values. These can be stored as a database record, as a vector, or as a matrix (or tensor). Since the parameter values ​​are determined from the subjective patient information in free text form using a trained machine learning system (e.g., an LLM), they are not explicit measured values, e.g., not measured ophthalmological data.Furthermore, the machine learning system is based on a trained machine learning model.

[0066] It is also worth noting that this involves not only the conversion of unstructured information (such as free text) into structured information (according to predefined categories), but also the processing / absorption of textual information into quantified conclusions not directly contained in the text. These conclusions can then be used within a standardized framework for further treatment planning. Furthermore, in another embodiment, the free text information can be obtained not beforehand, but during the interactive process of using a chat system based on a machine learning system. In this process, the system can either communicate directly with the patient or be operated by a physician. This interactive process enables improved information gathering, allowing gaps, errors, and inaccuracies to be identified and precisely addressed through follow-up questions.

[0067] Furthermore, the parameter values ​​can be specified in tabular form or an equivalent thereof, have a predefined range of values, and be provided with weighting factors.

[0068] Finally, procedure 100 involves generating, 106 – in the sense of determining, ascertaining, and / or calculating – a control parameter value for selecting the intraocular lens to be used, based on the numerical, patient-relevant parameter values. The control parameter value(s) typically consist of one or more individual numerical values, which may also be organized as a vector. This step of the procedure can also be performed using the aforementioned machine learning system or a second, specifically trained machine learning system.

[0069] These numerical values ​​can in turn be used as data input values ​​for an automated intraocular lens selection system.

[0070] As a concrete example, consider the implantation of a specific lens type, such as a trifocal IOL, for cataract surgery. A patient's suitability for this lens type is assessed, among other things, through a patient interview. For the present invention, information about the patient is collected in the form of free text regarding their work, hobbies, habits, and other preferences, as well as medically relevant information. In addition, the patient can also complete a questionnaire asking, for example, about their assessment of the importance of near and far vision, occupation, hobbies, etc.

[0071] Additionally, a schema can be prepared by listing relevant quantitative characteristics for planning the IOL to be implanted. The free-text information can then be interpreted using an intelligent text processing system (such as ChatGPT) and transferred into this schema.

[0072] An example can further illustrate this: The patient is asked how important distance vision is to them in everyday life. They might answer with a low score, for example, because they work in an office job and don't fully consider the question or understand the entire context. During the free-text analysis, it then turns out that they play amateur soccer, from which the machine learning system infers a greater need for good distance vision, records a higher score for this question within the scheme, and detects the discrepancy. A more intelligent question design can reduce such misunderstandings, but subjective misjudgments are always possible, and a consistency check can mitigate this risk or, ideally, eliminate the need for the patient's subjective assessment altogether. The following table provides an example of the aforementioned scheme: Table 1:

[0073] This scheme is intended as a draft, and the weighting of the categories can be adjusted and / or further optimized by the user according to their own preferences. The created scheme can be designed to allow the grouping of patients into predefined subgroups with the same treatment pattern.

[0074] The resulting score values ​​are processed individually for each of the attributes listed below in order to determine a trend: Table 2:

[0075] Alternatively, the schema can also consist directly of the described attributes (see Table 2), allowing the system to directly generate the respective categorization through intelligent text processing. This can, for example, be encoded in numerical form (e.g., one-shot encoding) and thus correspond to a quantification.

[0076] Two examples below can further illustrate the principle. Example 1:

[0077] The following is an example of a scoring process in which a score value is created in two steps and converted into a concrete evaluation of certain attributes.

[0078] The following scoring values ​​are examples of values ​​determined from free text using the proposed method / system. The weights were predefined: Table 3:

[0079] The following transfer functions are used for mapping the scoring. This mapping serves as an example and does not necessarily represent a mapping function that actually works in a clinical context. Furthermore, it is possible to design the mapping in various other ways, for example, also using a machine learning system.

[0080] Table 4:

[0081] This then results in the following for the specific example: Table 5:

[0082] This information, extracted from the free text in this way, can be passed on in categorized form, for example as a numerical code, to the next system for IOL selection. This example also demonstrates that the patient-relevant numerical parameter value can consist of a plurality of parameter values, which can be elegantly represented as a vector, matrix, or tensor. From this, the control parameter value described above can then be determined using a predefined algorithm or a dedicated machine learning system.

[0083] Example 2:

[0084] The following is another example of the scoring process where the quantification (or categorization) is performed directly by the machine learning system. This classification into a scheme thus bypasses the intermediate step of a simpler pre-categorization of the statements, but directly provides the final categorization: Table 6:

[0085] The text interpretation system, which is based on the previously described machine learning system, can also be specifically trained for this task. This training can be carried out, for example, by collecting historical information on previous treatments, the results of the operation, patient satisfaction, and other relevant postoperative information, which can then be used as labels or ground truth for the training. If the patient has completed a questionnaire concurrently, the quantitative analysis can be performed simultaneously with the information from the questionnaire; either the questionnaire is already present in the quantitative structure or is converted into it. The questionnaire can optionally be passed on as further input to the machine learning-based text processing system to support the quantification of the information (see Fig. 2).Alternatively, a consistency analysis can be performed using the questionnaire. The extracted information is then considered for planning the cataract surgery and selecting the IOL to be implanted. This can, for example, help to verify and, if necessary, correct inaccurate subjective assessments made by the patient.

[0086] Fig. 2 shows elements of an embodiment 200 of the proposed method, based on individual functional blocks already mentioned in the preceding paragraph. Here, the patient data 202 in free text form are fed to the text processing / text recognition system 206, which is based on a machine learning system with a correspondingly trained machine learning model. In a parallel branch, a questionnaire 204 and / or input options for a multiple-choice procedure can be used, in which the supplementary data determined in this way are used. Patient data are processed manually or algorithmically, 208. Both the output data from text processing 206 and that from manual or algorithmic processing are subjected to a consistency check 210. The subsequent quantification 212 then ultimately leads to the output data 214, which can be used for surgical planning. This can also be the control signal for selecting an IOL to be implanted.

[0087] It should also be noted that the patient information collected through the questionnaire can be used both as further input for the proposed procedure or system and as a basis for a consistency check. Of course, the consistency check is also possible using only the free text responses.

[0088] Fig. 3 shows elements of another embodiment 300 of the proposed method in the context of several interactions. Here, too, the patient information in free-text form—shown here as a free-text interview 302—and further patient information collected by means of a questionnaire 204 are subjected to a consistency check 210. This can, however, be a multi-stage or cyclical process controlled by a text processing module 206, which in turn is based on a machine learning system. In this way, contradictions between the information from the free-text interview and the supplementary patient information can be reliably identified, so that the entirety of the patient information can be passed on to the quantification module 212 without contradictions and largely corrected. From this, the data 214 are then used for surgical planning or... Control parameter values ​​are generated for selecting an IOL from an IQL stock. For example, a selection can be an automated selection of an IOL from an automated IOL stock.

[0089] The proposed system can also actively implement the process or procedure outlined in Figures 2 and 3 during an interview with a physician or other person (for example, a refractive manager or optometrist) to support the physician or manager and conduct the interview in a way that allows for the best possible and most accurate quantification (see Figure 3). The proposed procedure or system can check and correct patient statements based on, among other things, the following criteria:

[0090] Accuracy: The patient describes a situation—e.g., everyday activities—but is not specific enough to allow for quantification. The system then suggests follow-up questions to the interviewer—or asks them itself—to obtain the necessary specification. Example: "You say you enjoy playing tennis occasionally. How many hours a week do you play, and how important is this activity to you?"

[0091] Discrepancy: During the interview, the patient makes contradictory statements. The proposed procedure or system recognizes this contradiction and then asks specific questions to resolve the discrepancy. Example: "At the beginning of the interview, you said you have no difficulties with 3D vision. However, you have now indicated that you find driving difficult. What is the nature of this difficulty? Do you perhaps have problems judging distances?"

[0092] Future changes: A patient's circumstances may change over the course of their life. These potential changes can be explicitly inquired about, especially if changes become apparent during the interview. Example: "You indicated that you don't drive much, but you will soon be changing jobs to a neighboring city. Could this lead to significantly increased car use?"

[0093] Relative importance: If the patient indicates several potentially conflicting preferences, a weighting can be asked to determine where the actual priority lies. Example: "You stated that optimal near vision for reading is very important to you. At the same time, you indicated that you enjoy playing football, which requires sufficient distance vision. Where would you place the priority in terms of percentage weighting?"

[0094] Context: Some patient statements may show significant statistical deviations from the general patient population. In such cases, the system can ask further questions. Example: "You indicated that you only want to be optimized for near vision. However, 90% of patients prefer good distance vision. Are you therefore certain that you gave the correct information?" This can be done following the model of a compliance engine that catches outliers.

[0095] In another variant, instead of a reference to the general population, a reference can also be made to a comparison group that is tailored to the individual patient based on patient-specific criteria, provided that corresponding data are available, such as gender, activity profile (employed or not?; predominantly desk work or predominantly outdoor?), same disease or similar visual impairment, etc.

[0096] Fig. 4 shows various elements of the proposed procedure from a different perspective, 400. The process begins with asking a question to a patient or a person representing them, 402. The patient information is then received, 404. At least one piece of patient information is in free text form. This patient information can be supplemented by further patient information – for example, also with the aid of a multiple-choice procedure or a questionnaire.

[0097] Following this, gaps, inconsistencies, or similar potentially unclear or implausible patient information are identified using a machine learning system, 406. This can then generate follow-up questions to correct or clarify the patient information provided so far, 408. Patient information is then received again via one of the available channels. Based on this, the necessary correction and / or adjustment is then made, 410.

[0098] Next, it is determined whether there is (sufficient) consistency in the patient data, 412. This can also be achieved using one or the machine learning system already described. Alternatively, known algorithmic methods can be used. Subsequently, the numerical, patient-relevant parameter value(s) for predefined parameter categories are determined, 414. These form the basis for generating the control parameter value, 416 (see 106 in Fig. 1). The control parameter value(s) can then be passed on to a selection system for the intraocular lens to be implanted, 418. The control parameter values ​​can, for example, define the lens type of the IOL to be implanted. In this way, a subset can be selected from a large set of IOLs, whereby the elements of the subset are eligible as IOLs to be implanted. Optionally, from here, using the control parameter values ​​(if necessary,Additionally, using measured ophthalmological data, an IOL selection 418 can also be made directly in an IOL stock, 420. Alternatively, optionally, using the control parameter values ​​(possibly additionally using measured ophthalmological data), an IOL selection 418 can also be made for an ordering process.

[0099] In this way, based on patient information in free text form, a completely automated IOL selection from an IOL stock is ultimately possible with the help of a machine learning system, without the need for manual process interventions.

[0100] Figure 5 further illustrates the proposed process 500 based on the different data types used. First, the free-text inputs are received as patient information by a receiving unit 502 and stored in a memory / storage system module 504. The data format here is therefore a free-text format 506. This is used as input data for determining a numerical patient-related parameter value 510 with the aid of the machine learning system or the determination unit 508. These parameter values ​​510 are then used to generate the actual control parameter value 514 (multiple values ​​in vector form are also possible) with the help of a generator module 512, in order to optionally supply it to the IOL selection system 516 as input data.

[0101] Fig. 6 shows an embodiment of an analysis system 600 for controlling the selection of an intraocular lens to be used.The system comprises a processor 602 and a memory 604 operationally coupled to the processor 602, wherein the memory 604 stores program code elements which, when executed by the processor 602, cause the processor to receive and store – for example, with a receiving unit 502 and a memory unit 504 – subjective patient information in free text form, to determine – by means of a determination unit 508 – numerical, patient-relevant parameter values ​​for predefined parameter categories from the subjective patient information in free text form by means of a trained machine learning system which has a trained machine learning model, and to generate – by means of a generation / generator module 512 – a control parameter value for the selection of the intraocular lens to be used based on the numerical, patient-relevant parameter values.Optionally, an IOL selection system 516 can also be connected to the system's internal bus system 606. Furthermore, the memory 604 and the storage unit 504 can be configured as an integrated system. Alternatively, the receiver unit 502 and the storage unit 504 can also be designed as a single unit.

[0102] It should be expressly noted that the modules and units – in particular the processor 602, the memory 604, the receiver unit 502, the storage unit 504, the detection module 508 and the generator module 512 – are equipped with electrical signal lines or can be connected via an internal bus system 606 for the purpose of signal or data exchange.

[0103] The analysis system 600 can, in turn, be part of a computer system 700, which has a plurality of general-purpose functions, as shown in Fig. 7. The computer system can be a tablet computer, a laptop / notebook computer, another portable or mobile electronic device, a microprocessor system, a microprocessor-based system, a smartphone, a computer system with specially configured functions, or even a component of a microscope system. The computer system 700 can be configured to execute instructions—such as program modules—that can be executed to implement functions of the concepts proposed here. The program modules can include routines, programs, objects, components, logic, data structures, etc., to implement specific tasks or specific abstract data types.

[0104] The components of the computer system can include: one or more processors or processing units 702, a memory system 704, and a bus system 706, which connects various system components, including the memory system 704, to the processor 702. Typically, the computer system 700 has a plurality of volatile or non-volatile storage media accessible to the computer system 700. In the memory system 704, the data and / or instructions from the storage media can be stored in volatile form—such as in a RAM (random access memory) 708—for execution by the processor 702. This data and these instructions implement one or more functions or steps of the concept presented here.Other components of the 704 memory system can be a permanent memory (ROM) 710 and a long-term memory 712, in which the program modules and data (reference numeral 716), as well as workflows, can be stored.

[0105] The computer system features a number of dedicated devices for communication and interaction (keyboard 718, mouse / pointing device (not shown), screen 720, etc.). These dedicated devices can also be combined in a touch-sensitive display. A separate I / O controller 714 ensures smooth data exchange with external devices. A network adapter 722 is available for communication via a local or global network (LAN, WAN, WLAN, for example, via the internet). Other components of the computer system 700 can access the network adapter via the bus system 706. It is understood that - although not shown - other devices may also be connected to the Computer System 700.

[0106] Additionally, at least parts of the analysis system 600 can be connected to the bus system 706 for controlling the selection of an intraocular lens to be used (see Fig. 6). The analysis system 600 and the computer system 700 can optionally share the memory and / or the processor(s) or the bus system.

[0107] The description of the various embodiments of the present invention is provided for better understanding, but does not serve to directly limit the inventive idea to these embodiments. Further modifications and variations will be apparent to those skilled in the art. The terminology used here has been chosen to best describe the fundamental principles of the embodiments and to make them easily accessible to those skilled in the art.

[0108] The principle presented here can be embodied as a system, a method, a combination thereof, and / or as a computer program product. The computer program product can comprise one or more computer-readable storage media containing computer-readable program instructions to instruct a processor or control system to execute various aspects of the present invention.

[0109] Electronic, magnetic, optical, electromagnetic, infrared media, or semiconductor systems can be used as transmission media; for example, SSDs (solid-state devices / drives), RAM (random access memory), and / or ROM (read-only memory), EEPROM (electrically erasable ROM), or any combination thereof. Propagating electromagnetic waves, electromagnetic waves in waveguides or other transmission media (e.g., light pulses in optical cables), or electrical signals transmitted in wires can also be used as transmission media.

[0110] The computer-readable storage medium can be an embodied device that contains or stores instructions for use by an instruction execution device. The computer-readable program instructions described here can also be downloaded to a corresponding computer system, for example, as a (smartphone) app from a service provider via a wired connection or a mobile network.

[0111] The computer-readable program instructions for executing operations of the invention described herein can be machine-dependent or machine-independent instructions, microcode, firmware, status-defining data, or any source code or object code written, for example, in C++, Java, or similar languages, or in conventional procedural programming languages ​​such as C or similar. The computer-readable program instructions can be executed entirely by a computer system.In some embodiments, electronic circuits, such as programmable logic circuits, field-programmable gate arrays (FPGAs) or programmable logic arrays (PLAs), can also be used to execute the computer-readable program instructions by using status information from the computer-readable program instructions in order to configure or customize the electronic circuits according to aspects of the present invention.

[0112] Furthermore, the invention presented here is illustrated with reference to flowcharts and / or block diagrams of processes, devices (systems), and computer program products according to exemplary embodiments of the invention. It should be noted that virtually every block of the flowcharts and / or block diagrams can be designed as computer-readable program instructions.

[0113] The computer-readable program instructions can be provided to a general-purpose computer, a special-purpose computer, or any other programmable data processing system to manufacture a machine, such that the instructions, executed by the processor, computer, or other programmable data processing device, generate the means to implement the functions or operations depicted in the flowchart and / or block diagrams. These computer-readable program instructions can also be stored on a computer-readable storage medium.

[0114] In this sense, each block in the depicted flowchart or block diagrams can represent a module, a segment, or portions of instructions, which in turn represent several executable instructions for implementing the specific logic function. In some embodiments, the functions depicted in the individual blocks can be executed in a different order—possibly even in parallel.

[0115] The structures, materials, processes and equivalents of all means and / or steps with associated functions described in the claims below are intended to apply all structures, materials or processes as expressed in the claims.

Claims

PATENT CLAIMS 1. A computer-implemented method for controlling the selection of an intraocular lens to be used, the method comprising - Receiving and storing subjective patient information in free text format, - Determining numerical, patient-relevant parameter values ​​for predefined parameter categories from subjective patient information in free text form using a trained machine learning system that has a trained machine learning model, and - Generating a control parameter value for selecting the intraocular lens to be used based on the numerical, patient-relevant parameter values.

2. The method according to claim 1, further comprising - Selecting a lens type for the intraocular lens to be used based on the control parameter value.

3. The method according to claim 1 or 2, wherein receiving the subjective patient information in free text form comprises capturing the subjective patient information in response to a set of questions.

4. The method according to one of the preceding claims, wherein receiving the subjective patient information comprises capturing the subjective patient information with the aid of at least one of the following tools: a free text input field, a questionnaire, a chatbot and a conversation recording.

5. The method according to one of claims 3 or 4, wherein the recording of subjective patient information is carried out in multiple stages, wherein a first stage consists of a set of questions for the subjective patient information in free text form, and wherein during a second stage, in response to the subjective patient information recorded in free text form in the first stage, an interactive query is generated, the answer to which can be used to clarify and / or correct the subjective patient information of the first stage.

6. The method according to one of the preceding claims, further comprising - Identifying a content inconsistency in the subjective patient information and / or between the subjective patient information and reference patient information.

7. The method according to claim 6, further comprising - Adjusting the subjective patient information, eliminating the identified inconsistency, before determining numerical, patient-relevant parameter values.

8. The method according to any of the preceding claims, further comprising - Weighting of the determined, numerical, patient-relevant parameter values ​​using predefined weight parameters to determine result parameter values.

9. The method according to any of the preceding claims, further comprising - Deselecting intraocular lenses to be used, which are stored in an automated warehouse or available in an electronic ordering system, based on the control parameter value.

10. The method according to one of the preceding claims, wherein the control parameter value is used as an input value for an intraocular lens prediction system for an intraocular lens to be inserted, and wherein the intraocular lens prediction system additionally uses measured ophthalmological data of a patient assigned to the patient information as input values.

11. The method according to claim 10, wherein the measured ophthalmological data as input values ​​for the intraocular lens prediction system have a higher weighting than the control parameter value as input value for the intraocular lens prediction system.

12. The method according to one of claims 10 or 11, wherein the intraocular lens prediction system, in a first step, takes into account the control parameter value to select the lens type of the intraocular lens to be used, and, in a second step, takes into account the measured ophthalmological data to select a power of the intraocular lens to be used.

13. The method according to any one of claims 10 to 12, wherein the intraocular lens prediction system predicts several intraocular lens parameter values ​​for the intraocular lens to be implanted and / or, using the predicted intraocular lens parameter values, generates an output selected from: a simulation of an image representing a visual impression with an intraocular lens implanted in the patient, which has the predicted intraocular lens parameter values, and a display of a focus curve and / or defocus curve for one eye of the patient with an implanted intraocular lens, which has the predicted intraocular lens parameter values.

14. The method according to any of the preceding claims, additionally comprising - Selecting and / or procuring the intraocular lens to be used from an electronic ordering system and / or an automated warehouse.

15. The method according to one of the preceding claims, wherein the control parameter value for a selection of the intraocular lens to be used is given directly by the numerical, patient-relevant parameter values.

16. An analysis system for controlling the selection of an intraocular lens to be used, comprising the analysis system - a processor and a memory operationally coupled to the processor, wherein the memory stores program code elements which, when executed by the processor, cause the processor to: - Receiving and storing subjective patient information in free text format, - Determining numerical, patient-relevant parameter values ​​for predefined parameter categories from subjective patient information in free text form using a trained machine learning system that has a trained machine learning model, and - Generating a control parameter value for selecting the intraocular lens to be used based on the numerical, patient-relevant parameter values.

17. A computer program product for controlling the selection of an intraocular lens to be used, wherein the computer program product is a comprising a computer-readable storage medium containing program instructions, wherein the program instructions are executable by one or more computer systems or controllers to cause the one or more computer systems to - Receiving and storing subjective patient information in free text format, - Determining numerical, patient-relevant parameter values ​​for predefined parameter categories from subjective patient information in free text form using a trained machine learning system that has a trained machine learning model, and - Generating a control parameter value for selecting the intraocular lens to be used based on the numerical, patient-relevant parameter values.

Images