Methods and devices for predicting refractive surprises in iol transplantation

Abstract

The invention relates to a computer-implemented method comprising a model (5) based on artificial intelligence and comprising the following method steps: reading information (1) characterizing a patient (3) into the model; reading information (7) characterizing defective vision (9) of an eye (11) of the patient (3) into the model (5); reading information (15) characterizing an intraocular lens (17) to be inserted; determining a probability (37) of a refractive surprise occurring when the intraocular lens (19) is inserted to correct the defective vision (9), by means of the model (5) based on artificial intelligence; and providing the probability (37) of a refractive surprise occurring. The invention also relates to a corresponding computer program product, to a device for data processing (19), to a computer-implemented method for training a model (5) based on artificial intelligence, to a corresponding trained model (6) based on artificial intelligence or the weighting matrix (51) of said model, to a computer-readable medium, and to a laser-therapy or transplantation device.

Description

[0001] Methods and devices for predicting refractive surprises in IOL transplantation

[0002] The present disclosure relates to a computer-implemented method comprising an artificial intelligence-based model for predicting refractive surprises during an implantation of an intraocular lens, a corresponding computer program product, a device for data processing, a computer-implemented method for training an artificial intelligence-based model, a corresponding trained artificial intelligence-based model or its weighting matrix, a computer-readable medium, and a laser therapy or transplantation device.

[0003] To treat reduced vision due to deterioration of the lens of the eye caused by disease or age, intraocular lenses (IOLs) can be used instead of the eye's lens. Especially in cases of cataracts, an IOL transplant can significantly improve quality of life. Transplanting an intraocular lens for cataracts is the most common eye surgery. Various types of IOLs can be used in such a transplant, for example, toric (which correct astigmatism) or multifocal IOLs. An IOL can also be used (in addition to the eye's lens) to correct refractive errors.

[0004] It is known that the position of the IOL to be implanted in the patient's eye and the refractive power of the IOL to be implanted are important implantation parameters. Ideally, the refractive power after IOL implantation corresponds exactly to the calculated refractive power required to enable the patient to see clearly without additional aids (such as glasses or contact lenses). A difference between the achieved and calculated refractive power can be referred to as the postoperative refractive error. For the purposes of this disclosure, the term "patient" refers to a person being treated, regardless of gender. Likewise, the term "doctor" refers to both male and female doctors.

[0005] Despite recent advances in IOL implantation, postoperative refractive errors remain unsatisfactorily high. Furthermore, it is currently not possible to reliably predict whether IOL implantation will be successful for a given combination of a patient's eye (and thus all possible parameters and characteristics of the eye) and an IOL (and all parameters and characteristics of the IOL). Successful is defined as no significant refractive errors remaining after implantation. Ideally, the postoperative refractive error after such an implantation is 0 diopters. It is of particular interest to physicians that implantations in which significant postoperative refractive errors remain are not performed or are performed separately.

[0006] The aim of the present invention is to identify precisely those combinations of a patient's eye to be treated and an IOL to be implanted that are highly likely to exhibit a significant postoperative refractive error after implantation.

[0007] Selecting the optimal IOL for the patient is a necessary step before IOL implantation. First, preoperative measurements are taken of the patient's eye. Then, one of the formulas for calculating IOL power is selected to determine the IOL power based on the patient's measured ocular dimensions and characteristics.

[0008] When planning an IOL implantation, the physician generally determines appropriate treatment parameters (e.g., type of IOL; location and / or orientation of the IOL) based on patient data (e.g., gender, age, etc.), diagnostic data (e.g., manifest refraction, pachymetry, etc.), and the data of the IOL to be implanted. The patient data, diagnostic data, and IOL data can be presented as numerical values ​​(e.g., measurements), text data, videos, and / or image data (e.g., diagnostic images).

[0009] Different approaches and formulas are known for calculating IOL refractive power, which differ, for example, in the number of variables involved. Purely examples include SRK I and II (linear regression), vergence formulas such as Holladay 1, SRK / T, Hoffer Q, Haigis, Lada's super formula, Barrett Universal II, Hoffer-H5, Holladay 2, and ray tracing (e.g., Okulix, Phacooptics, Z CALO), which can be used to calculate IOL parameters.

[0010] The next step is to determine the position and / or orientation of the IOL in the patient's eye. For example, an effective lens position can be estimated. The effective lens position corresponds to the effective distance between an anterior surface of the cornea and the lens plane. Furthermore, some calculations require the anterior chamber depth (ACD) of the treated eye to estimate the IOL position.

[0011] To improve the calculations of the IOL, its parameters, as well as its position and / or orientation in the patient's eye, machine learning-based methods are described, for example, in DE 10 2020 101 763 A1 and WO 2023 / 084 382 A1. The present invention aims to improve and / or enhance existing solutions from the prior art.

[0012] This object is achieved by the subject matter of the independent claims. Optional developments of the subject matter of the invention are specified in the subordinate claims.

[0013] The method according to the invention describes a method for predicting the probability of significant residual refractive errors occurring, also called refractive surprise, or in other words, the probability of failure of an IOL implantation. This probability is determined, among other things, depending on anatomical patient variables and IOL parameters using machine learning (ML) with a model or an ensemble of nonlinear models. The method can, in particular, consider a variety of preoperative measurements, patient data, and data from the IOL to be implanted and is used to predict refractive surprises, i.e., major errors that occur less frequently.

[0014] This object is achieved by a computer-implemented method comprising a model based on artificial intelligence, the method comprising the following method steps: reading information characterizing a patient into the model; reading information characterizing a refractive error of an eye of the patient into the model; reading information characterizing an intraocular lens to be inserted; determining a probability of an occurrence of a refractive surprise that occurs when inserting the intraocular lens to correct the refractive error by the model based on artificial intelligence; and providing the probability of the occurrence of a refractive surprise.

[0015] The computer program product described in this disclosure comprises instructions that, when executed on a data processing device, cause the data processing device to perform the method steps of the computer-implemented method.

[0016] Accordingly, the data processing device presented here is designed to carry out the computer-implemented method.

[0017] Furthermore, a computer-implemented method for training a model based on artificial intelligence is presented herein. This method comprises training the model using training data, wherein the training data comprises at least one piece of information characterizing a patient; one piece of information characterizing a refractive error of an eye of the respective patient; one piece of information characterizing an intraocular lens to be inserted; and a probability of a refractive surprise occurring when the intraocular lens is inserted into the respective patient's eye.After training, the model is designed to determine, based on: information characterizing a patient; information characterizing a refractive error of one of the patient's eyes; and information characterizing an intraocular lens to be inserted, a probability of an occurrence of a refractive surprise that occurs when the intraocular lens is inserted into the patient's eye to correct the refractive error, by the artificial intelligence-based model.

[0018] This disclosure also relates to a computer-readable medium comprising a model based on artificial intelligence, wherein the model is configured to, when the model is executed by a data processing device, cause the device to use the model to determine, based on: information characterizing a patient; information characterizing a refractive error of an eye of the patient; and information characterizing an intraocular lens to be inserted, a probability of an occurrence of a refractive surprise that occurs when the intraocular lens is inserted into the patient's eye to correct the refractive error, by the artificial intelligence-based model.

[0019] Furthermore, a trained artificial intelligence-based model or a weighting matrix of such a trained model is presented, wherein the trained model or its weighting matrix was obtained by a described computer-implemented method for training the artificial intelligence-based model.

[0020] The computer program product and / or the trained artificial intelligence-based model and / or the weighting matrix of such a trained model can be provided as a transferable and computer-readable data signal. Such a data signal can be transferable or transmitted centrally or decentrally, for example, from a server or client to another client, a data processing device, or a laser therapy or transplantation device, or can be obtained from such a source, i.e., have been transferred. The data signal can contain additional functional data that enable the interpretation, readability, or decryption of the other components of the data signal.

[0021] Finally, a laser therapy device and a transplantation device are provided, each of which comprises a described device for data processing or is configured to carry out the presented computer-implemented method. The laser therapy device or the transplantation device can thus make it possible to determine whether a refractive surprise occurs during an IOL transplantation to be performed. Such a laser therapy device or transplantation device can further be configured to perform the transplantation and, for example, enable the creation of an incision for introducing means for removing the eye lens and / or for inserting the intraocular lens.

[0022] The disclosure relates to an artificial intelligence-based model that is provided in the form of or as part of a computer program product. The artificial intelligence-based model is particularly designed to be used by a data processing device as part of or in the form of a computer program product. The model can, in particular, be a trained model that automatically generates output data based on input data.

[0023] Data input or to be input into the model can include, in particular, information identifying a patient, information identifying a refractive error in one of the patient's eyes, and information identifying an intraocular lens to be inserted. Such input or to be input data is processed by the model during execution of the computer program product by the data processing device in such a way that the output data is generated. The output data includes a probability of the occurrence of a refractive surprise when inserting the intraocular lens into the patient's eye to correct the refractive error.

[0024] The model can thus make it possible to identify an implantation that is highly unlikely to achieve the desired treatment outcome. This allows the physician to either avoid such an implantation or, after re-evaluating all circumstances and taking into account the possible occurrence of refractive surprise, to plan the procedure anew and / or with varied parameters.

[0025] The present computer-implemented method is therefore not a method for detecting outliers in an input data set, but rather supports the user in identifying regularities in constellations of patient data, data on their refractive error, and data on an intraocular lens to be implanted that would lead to a refractive surprise. This can allow such refractive surprises (applied to the application, a refractive surprise corresponds to a failed implantation of an intraocular lens) to be prevented, which is undoubtedly in the interest of the physician, but even more so in the interest of the patient.Assuming an appropriate data basis for training the presented artificial intelligence-based model, a trained model of the presented invention allows to determine, based on an input combination of patient data, data on his or her refractive error and data on the intraocular lens to be inserted, how likely this combination would lead to a refractive surprise.

[0026] A computer program product can be understood as a sequence of instructions or commands provided in a specific programming language. The computer program product enables certain functions or calculations related to the implantation of an IOL in a patient's eye to be performed or carried out with the aid of a data processing device.

[0027] A model based on artificial intelligence can, in particular, be one created using machine learning. In machine learning, an artificial system or model can learn from a large number of examples and, using algorithms, build a statistical model that evaluates the input data, for example, in a weighting matrix, to determine output data. After training, such a model is tested or validated with test or validation data. An artificial intelligence model is not based on memorized examples, but rather recognizes patterns and regularities in the training data and can apply these recognized patterns or regularities to unknown data.

[0028] The previously described computer program product enables an individualized prediction of the outcome of an IOL implantation in a patient's eye using an artificial intelligence-based model. The prediction is individualized because the artificial intelligence-based model calculates the probability of implantation failure based, among other things, on patient-specific information. This patient-specific information is among the information that serves as input data for the artificial intelligence-based model.

[0029] When determining the probability of a refractive surprise occurring (in the absence of such a surprise, i.e. the success of an implantation), the artificial intelligence-based model is not bound to a fixed set of rules, but can rely on knowledge acquired during the learning phase and relationships found when determining the probability.

[0030] Based on this acquired knowledge, the artificial intelligence-based model can also react to input data that has not previously occurred.

[0031] The artificial intelligence-based model thus allows the complex assessment process of an experienced physician to be modeled automatically and taking all input data into account, something that is not possible using a deterministic probability calculation. Furthermore, it is conceivable that the artificial intelligence-based model could integrate far more parameters into the complex assessment process and examine the complex interactions between all these parameters. The number of possible parameter combinations increases rapidly with the increasing number of parameters, so that a physician can no longer easily oversee all possible combinations, let alone recognize regularities and relationships that require a specific combination of many parameters.

[0032] The artificial intelligence-based model can also be further trained during operation, for example using feedback on a calculated probability from the physician using the artificial intelligence-based model.

[0033] In the following, optional further developments of the aforementioned methods and devices will be explained in detail. Features described in the course of explaining the method(s) can be applied to the corresponding device. Likewise, explanations of device features can also be mapped to corresponding process steps.

[0034] For example, if it is described that the method carries out a method step A, this implies that the corresponding device is designed to be able to carry out this method step A. Conversely, a corresponding embodiment of the method has the method step B if a correspondingly designed device has a feature C for carrying out B.

[0035] The information identifying the patient may include at least one piece of information from the following list: information about the biomechanical stability of the eye; information about a previous condition of the eye; information about a previous treatment of the eye; information about the shape of the eye to be treated; information about the origin and / or ethnicity of the patient; information about the age and / or gender of the patient. According to the invention, it can be provided that the information identifying the patient can be supplemented with further information.

[0036] The information characterizing the refractive error of the patient's eye may include at least one piece of information from the following list: information about an axial length of the eye; information about a preoperative depth of the anterior (ocular) chamber; information about a surface curvature and / or a corneal contour of the cornea of ​​the eye; information about a manifest refraction and / or a pachymetry; information about a position of the intraocular lens to be inserted; and information about a target refraction to be achieved with the IOL to be implanted.

[0037] The information characterising the intraocular lens to be inserted can comprise at least one piece of information from the following list: information about a type of intraocular lens, for example the material from which it is made, whether it is a monofocal, bifocal or trifocal intraocular lens, whether it corrects astigmatism and, for example, the shape and / or size of the holding and fixing elements of the intraocular lens; information about geometric dimensions of the intraocular lens (diameter, centre thickness, etc.); information about a predetermined position of the intraocular lens in the eye; and information about a refractive power of the intraocular lens.

[0038] The computer-implemented method according to the disclosure, using the artificial intelligence-based model, can assist a physician during intraocular lens implantation to avoid an implantation that is highly unlikely to be successful for the patient, i.e., to not perform the procedure under the given parameters. Likewise, in such a case, the model allows for a recalculation of the probability of a refractive surprise using modified parameters.

[0039] The computer-implemented method according to the disclosure can thus warn a physician if, given the constellation of the patient's eye, the calculated intraocular lens and the information characterizing the patient, there is a high probability that refractive surprises will occur.

[0040] The information to be read or read into the artificial intelligence-based model can be in any form, for example as image data, text data, or even video data (for example and not limited to, a video sequence of a color-coded Doppler image of blood flow in the eye). The information to be read can be requested from a patient file and read into the model. Likewise, the information to be read can be requested and received from a database or server. If the information to be read is available in the form of a table, for example, the order of the individual entries as well as their names and format are preferably consistent for both the training data and the data used to determine the probability.For example, it is relevant that the age or another parameter is always available in the same data format and, if necessary, must be converted into a predetermined data format before use. For example, to be able to use information about the age or another parameter with a period (.) as the decimal separator, as well as information about the age or another parameter with a comma (,) as the decimal separator. This allows expert knowledge to be used to train the artificial intelligence-based model regardless of location. The data processing device can be a local computing unit. This can be controlled wirelessly and / or via a cable and can receive or output data. The data processing device can be provided locally or remotely, or can operate partially locally and partially remotely.

[0041] For example, the artificial intelligence-based model can be stored entirely or partially locally, or entirely or partially in a cloud. For example, the information to be entered can be transferred from a device acting as a client to one or more servers for data processing, on which the artificial intelligence-based model is hosted.

[0042] The artificial intelligence-based model can thus be provided via cloud computing, also known as software-as-a-service (SaaS). The determined probability of a refractive surprise is calculated by the data processing device on the server (or servers) and output to the client, preferably via the internet. Such a configuration of the data processing device thus includes both the client and the server, as well as a corresponding connection between the two.

[0043] Through such centralized or decentralized provision of the artificial intelligence-based model, it can be further trained based on feedback from a large number of users of the model.

[0044] The data processing device may have an input interface to receive information identifying a patient, information identifying a refractive error of the respective patient's eye, and information identifying an intraocular lens to be inserted.

[0045] The input interface can be configured, for example, via a display device, to receive user input concerning information identifying the patient and / or information identifying a visual impairment of one of the patient's eyes and / or information identifying an intraocular lens to be inserted. In particular, it is conceivable for the user to complete previously incomplete information via the input interface. According to the invention, it can thus be provided that a check is carried out to determine whether the information to be read in meets a minimum requirement for information necessary to determine the probability of a refractive surprise occurring.

[0046] In particular, the input interface may comprise an input interface configured to receive a data signal comprising the information identifying the patient and / or the information identifying a visual impairment of an eye of the patient and / or the information identifying an intraocular lens to be inserted.

[0047] The input of information characterizing the patient and / or information characterizing a visual impairment of one of the patient's eyes and / or information characterizing an intraocular lens to be inserted into the artificial intelligence-based model can be done via this input interface and / or additionally by reading this information from at least one database. The information can thus be input from multiple resources.

[0048] The data processing device can further comprise an output interface for outputting the determined probability, at which an output signal representing the probability can be provided. If the input interface comprises a display, the output interface can also use this display to display the probability, for example.

[0049] Other known types of display devices can be connected to the output interface to display the probability. Alternatively or additionally, the output interface can be connected to a database and / or transfer the output signal and the determined probability of the occurrence of a refractive surprise represented by it to the database.

[0050] It is conceivable that the data processing device is configured to provide a warning signal based on the determined probability, wherein the warning signal represents a state in which the determined probability of the occurrence of a refractive surprise exceeds a predetermined threshold. The warning signal can cause a warning sign, e.g., in the form of a warning symbol, to be displayed on a display or a warning light to be activated in a display device. The warning signal thus triggers the output of the warning symbol.

[0051] Furthermore, the disclosure relates to a computer-implemented method for training a model based on artificial intelligence, wherein the method comprises training the model using training data at least comprising information characterizing a patient, information characterizing a visual impairment of an eye of the respective patient, information characterizing an intraocular lens to be inserted, and a probability of a refractive surprise occurring upon insertion of the intraocular lens into the eye of the respective patient, and wherein the model is designed after training, based on information characterizing a patient, information characterizing a visual impairment of an eye of the respective patient, and information characterizing an intraocular lens to be inserted, a probability of a refractive surprise occurring,which occurs when the intraocular lens is inserted into the patient's eye to correct the refractive error, is to be determined by the artificial intelligence-based model.

[0052] At least one method step of this and also of the computer-implemented method described above can be carried out at least partially by a data processing device.

[0053] The model is trained using a training dataset and subsequently validated using a test dataset. The training dataset preferably contains annotated data, i.e., a probability of the occurrence of a refractive surprise. This probability can be determined, for example, based on a residual refractive error. The test dataset also contains this probability; for validation, only the information characterizing a patient, the information characterizing a refractive error in one of the patient's eyes, and the information characterizing an intraocular lens to be inserted are input into the (trained) model. The thus calculated probability is compared with the probability stored in the test dataset.If this value is within a predefined confidence interval, the artificial intelligence-based model can be considered trained. If the confidence interval is not met, further training of the model may be necessary.

[0054] The model can also be further trained during operation, allowing previously calculated probabilities following implantation to be input into the model with an actual probability of a refractive surprise (e.g., calculated based on a residual refractive error). Thus, each implantation performed can improve the training dataset and the model's accuracy.

[0055] It is conceivable that the training of the model will be continued in particular if the actual probability of the occurrence of a refractive surprise of the implantation deviates significantly from a probability of the occurrence of a refractive surprise calculated by the model.

[0056] It can be particularly advantageous if a large number of datasets are available for training the model. It is also particularly advantageous if these datasets were acquired in more than one country. If a training dataset consists of subsets from different countries, it can be helpful to include the country in which the data was collected as an additional parameter in the training dataset. This way, for example, a higher average exposure to UV radiation in certain countries can be taken into account in the model's evaluation. On the other hand, it is also possible to consider social differences (purely as an example and not restrictive, media consumption at a young age and media availability).

[0057] A large number of data sets for training the model can also have the advantage that, for example, rare combinations of patient data and intraocular lens data can be captured during training and thus be known to the model.

[0058] The inventive, trained artificial intelligence-based model or the inventive weighting matrix is ​​obtained by the previously described computer-implemented method for training the artificial intelligence-based model. The weighting matrix can be understood as an evaluation of individual influencing variables of the artificial intelligence-based model. Depending on the application of the model, the different parameters are assigned a correspondingly greater or lesser influence on the result through different weightings in the weighting matrix. The different parameters are input into the model using information characterizing a patient, information characterizing a visual impairment of one of the patient's eyes, and information characterizing an intraocular lens to be inserted.

[0059] When training the artificial intelligence-based model, the importance and associated weighting of individual parameters are determined using well-known methods (gradient descent of the loss function or direct loss minimization). Parameters that have a strong influence on the result (the calculated probability of a refractive surprise occurring) are given a correspondingly higher weighting. During training, each parameter or combination of parameters can be assigned a weighting based on the information input into the artificial intelligence-based model. The total of these weights after training represents the weighting matrix, which can be used to evaluate new information input into the model.

[0060] The trained artificial intelligence-based model or its weighting matrix can be stored locally in a data processing device or in a distributed system, for example, on at least one server. The trained artificial intelligence-based model or its weighting matrix can thus be accessed locally or remotely when the model is used to determine the probability of a refractive surprise occurring. Alternatively, the trained artificial intelligence-based model or its weighting matrix can be requested by the data processing device and transferred to the data processing device in the form of a data signal. The trained artificial intelligence-based model or its weighting matrix can thus be stored in a volatile or non-volatile memory, or provided in the form of a data signal.

[0061] Typically, the weights of the weight matrix are initialized with values ​​from a normal distribution. After training, the weights can take on any value.

[0062] The artificial intelligence-based model can have an input and an output layer. Furthermore, the model can have one or more hidden layers. The model's layers can be fully networked linear layers, nonlinear layers (e.g., ReLU), or even transformer layers. However, the model's layers are not restricted to the types listed in this list.

[0063] The size of the input layer depends on the number of input parameters and their encoding. The encoding of the input parameters can, for example, include one-hot encoding for categorical variables or normalization. Categorical variables or data refer to variables that consist of label values; for example, a variable "country of origin" could have the values ​​"Germany," "Finland," "Australia," etc.

[0064] Some machine learning algorithms can work directly with categorical data, but most algorithms require that all input or output variables be a number or numeric value. This means it's advantageous to map all categorical data to integers.

[0065] One-hot encoding is a method of converting data to prepare it for an algorithm and obtain a better prediction.

[0066] Normalization has the advantage that the weights can be within a limited range of values.

[0067] The output layer can contain a neuron that outputs the probability of refractive surprises. Furthermore, it is possible to distinguish between multiple types of refractive surprises and expand the output layer accordingly with additional neurons.

[0068] The computer-readable (storage) medium according to the invention stores, in computer-readable form, instructions for carrying out the computer-implemented method for calculating the probability of a refractive surprise occurring, i.e., the computer-readable medium comprises a pre-described computer program product. The computer-readable medium can be any digital data storage device, such as an EPROM, a USB stick, a hard disk, a CD-ROM, a DVD, an SD card, or an SSD storage medium. The computer program product according to the invention can be stored on such a computer-readable (storage) medium in order to be made available to a data processing device. Alternatively, the computer program product can also be provided as a data signal via the Internet or otherwise.

[0069] In summary, an AI-based evaluation of the information input into the artificial intelligence-based model is provided. This evaluation can learn the relationship between input diagnostic, patient, or IOL data and the resulting probability of a refractive surprise occurring through training with data from IOL implantation experts and / or a respective device user.

[0070] The model thus trained can be made available for verifying the implantation of an intraocular lens and can provide a user with a prognosis for the occurrence of a refractive surprise based on the information entered.

[0071] The artificial intelligence-based model can thus provide event prediction. Various artificial intelligence-based models and methods are known for this purpose.

[0072] The aspects of the present invention will be explained in more detail below with reference to the accompanying drawings. The drawings show possible exemplary embodiments of the present invention purely by way of example, whereby the described features can be combined with one another or omitted as desired. Identical features or features with the same function are further identified by the same reference numerals. Repetitive descriptions of features are omitted, so that explanations of features described in previous figures can also be applied to other figures, unless differences are explicitly pointed out.

[0073] They show:

[0074] Fig. 1 is a schematic flow diagram of a computer-implemented method according to the disclosure;

[0075] Fig. 2 schematically shows a device for data processing and a data signal;

[0076] Fig. 3 is a schematic flow diagram of a training method according to the disclosure; Fig. 4 is a schematic representation of a data set;

[0077] Fig. 5 is a schematic flow diagram of a validation method according to the disclosure; and

[0078] Fig. 6 a schematic representation of a trained model and its weight matrix.

[0079] Figure 1 shows a schematic flow of the computer-implemented method. The method comprises a first method step S1, in which information 1 identifying a patient 3 is read into a model 5 based on artificial intelligence K1.

[0080] In a second method step S2, information 7, which characterizes a visual impairment 9 of an eye 11 of a patient 3, is read into the model 5.

[0081] Finally, in a third method step S3, information 15, which identifies an intraocular lens 17 to be inserted, is read into the model 5.

[0082] These first 3 process steps S1 to S3 represent an input E into model 5.

[0083] At this point, reference should be made to Figure 2, in which a data processing device 19 is shown schematically.

[0084] The data processing device 19 has an input interface 21, via which the information 1, 7 and 15 can be entered or read in. This information 1, 7, 15 is shown schematically in the form of a data array 23. The input interface 21 can be connected to different sources 24, for example to a local input unit 25, which, purely by way of example, can be a display with an input function 27. Furthermore, the input interface 21 can be connected to a remote computer 29. The connection can be wired or wireless. Likewise, the input interface 21 can be connected to a cloud 31 and receive the information 1, 7 and 15 from there. It is conceivable that the input interface 21 can be connected to further sources 24 in other embodiments.

[0085] The reception E of the information 1, 7, 15, as shown in the method steps S1, S2 and S3, can thus take place from one of the aforementioned sources 25, 29 or 31 or from any combination of these sources 25, 29 and 31. Likewise, the read-in information 1, 7, 15 can be partially or completely composed of one, two or three or all available sources 24.

[0086] This information 1, 7, 15 is subsequently transferred to the model 5 based on artificial intelligence K1. The model 5 can be stored partially or completely in a storage unit 33 of the data processing device 19. It is also conceivable that the model 5 is stored partially or completely in the cloud 31 in the sense of cloud computing.

[0087] Furthermore, it is conceivable that the model 5 is requested and received in the form of a data signal 35 from the cloud 31 and / or from the remote computer 29.

[0088] With reference to Figure 1, after the reading E, in a fourth method step S4, a probability 37 of an occurrence of a refractive surprise when inserting the intraocular lens 17 to correct the refractive error 9 is determined by the model 5 based on artificial intelligence K1.

[0089] This probability 37 is provided in a fifth method step S5.

[0090] With reference to Figure 2, the data processing device 19 further comprises an output interface 39, via which the probability 37 of the occurrence of a refractive surprise upon insertion of the intraocular lens 17 is output. The probability 37 can be in the form of an output signal 38, which is illustrated schematically. The output interface 39 can be connected to a wide variety of means that are not shown in Figure 2. Purely by way of example, a database, a warning light, or a display may be mentioned here. It is also possible for the aforementioned display with input function 27 to be used for the display.

[0091] The probability 37 can be represented as a numerical value or as binary information. Binary information can be considered, for example, an activated warning light that illuminates when the probability 37 exceeds a predetermined threshold. The warning light can be physically present or can be displayed on a display device.

[0092] Figure 3 schematically shows the sequence of a computer-implemented training method. In a first training step T1, training data 41 is input into model 5, and in a second training step T2, model 5 is trained. It should be noted that the second training step T2 includes various substeps of preferentially supervised training (not shown).

[0093] This means, for example, that a probability 37 is calculated by the model 5 based on an entry 47 of the training data 41 and if the calculated probability 37 deviates from the probability 37 entered into the model 5 by means of the entry 47 of the training data 41, individual weights of the weighting matrix (see Figure 6) are varied and the deviation of the entered probability 37 from the calculated probability 37 is iteratively minimized by adjusting the weighting matrix (known as backpropagation; for the weighting matrix see Figure 4).

[0094] Furthermore, a training step T3 (dashed line) is shown, in which the trained model 5 is validated with validation data 43. The validation is shown separately in Fig. 5 and shown in dashed lines in Fig. 3, since objectively speaking it cannot be assigned to the training of a model based on artificial intelligence, but is highly relevant for testing the training. Query T4, which asks whether the respective result of the validation data 43 lies within a predetermined confidence interval, is also dashed. If this is not the case (answer: n), the process should jump back to step T2. If this is the case, however, a trained model 6 based on artificial intelligence K1 and a weighting matrix 51 underlying this model 6 are available, which are stored in a preferred step T5.The process of training model 5, shown schematically in Figure 3, preferably occurs before using model 5 to calculate probability 37, whereby further learning of the trained model 6 based on new data obtained during operation of model 6 is conceivable. In this way, model 6 can be further improved during operation, and a more reliable calculation of probability 37 can be achieved.

[0095] Figure 4 schematically shows a data set 45 comprising a plurality of entries 47. Each of the entries 47 can contain the information 1, 7, and 15, i.e., information about the patient 3, about the visual impairment 9 or the intraocular lens 17, and additionally the probability 37. Such a data set 45 can be used to train the model 5 and can, for this purpose, preferably be divided into approximately (not restrictively) 75% training data 41 and 25% validation data 43. This division is purely exemplary and can also be chosen differently for other models 5, for example, 70% - 30% up to 90% - 10%.

[0096] The entries 47 can be coded in the form of a table in which a plurality of properties and / or measured values ​​of the patient 3, properties and / or measured values ​​relating to the refractive error 9 of the eye 11 of the patient 3 and properties and / or measured values ​​of the intraocular lens 17 are listed.

[0097] This information 1 characterizing the patient 3 can include information 1a representing a biomechanical stability of the eye 11, information 1b representing previous diseases of the eye 11, information 1c representing previous treatments of the eye 11, information 1d representing an eye shape and information 1e representing the age and / or information 1f representing the gender and / or information 1g representing the ethnicity of the patient 3.This information 7 characterizing the refractive error 9 of the eye 11 of the patient 3 can include information 7a representing the axial length of the eye, information 7b representing a preoperative depth of the anterior chamber, information 7c representing a surface curvature, information 7d representing a corneal contour, information 7e representing a manifest refraction, information 7f representing a pachymetry, information 7g representing a position of the intraocular lens to be inserted, and information 7h representing a target refraction.

[0098] This information 15 identifying the intraocular lens 17 to be inserted can include information 15a representing the type of intraocular lens 17, information 15b representing the geometric dimensions of the intraocular lens 17, information 15c representing a predetermined position of the intraocular lens 17 in the eye 11, and information 15d identifying the refractive power of the intraocular lens 17.

[0099] The above-mentioned information can be stored in a predefined format in entries 47. These formats can be a pure numerical format, for example in the form of a numerical value or a series of measurements, an image format in the form of a 2-dimensional array of pixel values ​​of a grayscale or color representation, a video format (this enables the model 6, for example, to analyze a time-dependent process on or in the eye 11 of the patient 3), or even a 3-D representation.

[0100] Figure 5 is considered below, which schematically shows a flow diagram. After the data set 45 has been used to train the model 5, the validation data 43 are entered into the trained model 5 in a method step V1, and in a method step V2 the value of the probability 37 of an occurrence of a refractive surprise determined by the model 5 is compared with a value of the probability 37' stored in the validation data 43. If the value of the calculated probability 37 lies within a confidence interval 49 of the probability 37' specified in the validation data 43, the model 5 can be referred to as the trained model 6, and the training is (provisionally, i.e., subject to further training during operation or additional training) completed.If this is not the case, a decision must be made in process step V3 as to whether the training should be repeated with other parameters or data sets 45 (the parameters or data sets 45 are changed in the optional process step V4).

[0101] Figure 6 schematically shows the trained model 6, which has a weighting matrix 51. The weighting matrix 51 can be viewed as a matrix of calculated weights 53 of individual neurons of the artificial intelligence-based model 5. The exact number of weights 53 (w11...w1x ...wy1 ...wyx) in each layer 55, the number of layers 55, the number and / or type of connections of the neurons in each layer 55, as well as the type of layer 55, are irrelevant for calculating the probability 37. According to the invention, the information 1, 7, 15 is input into the model 5, and the probability 37 is output.

[0102] List of reference symbols

[0103] I Information identifying a patient

[0104] 1a Information representing biomechanical stability of the eye

[0105] 1b Information representing previous eye diseases

[0106] 1c Information representing previous treatments of the eye

[0107] 1d Information representing an eye shape

[0108] 1e information representing age

[0109] 1f gender-representing information

[0110] 1g Information representing patient 3’s ethnicity.

[0111] 3 patients

[0112] 5 model based on artificial intelligence

[0113] 6 trained artificial intelligence-based model

[0114] 7 information characterizing a refractive error in one eye of a patient

[0115] 7a Information representing an axial length of the eye

[0116] 7b Information representing preoperative depth of the anterior chamber

[0117] 7c Information representing surface curvature

[0118] 7d Information representing a corneal contour

[0119] 7e Information representing a manifest refraction

[0120] 7f Information representing pachymetry

[0121] 7g Information representing the position of the intraocular lens to be inserted

[0122] 7h information representing a target refraction

[0123] 9 Visual impairment

[0124] II Eye

[0125] 15 information characterizing an intraocular lens to be inserted

[0126] 15a Information representing the type of intraocular lens

[0127] 15b Information representing the geometric dimensions of the intraocular lens

[0128] 15c Information representing a predetermined position of the intraocular lens in the eye

[0129] 15d Information characterizing the refractive power of the intraocular lens

[0130] 17 Intraocular lens

[0131] 19 Data processing device

[0132] 21 Input interface

[0133] 23 Data array

[0134] 24 Source

[0135] 25 local input unit

[0136] 27 Display with input function

[0137] 29 remote computers

[0138] 31 Cloud 33 Storage unit

[0139] 35 data signal

[0140] 37 Probability

[0141] 38 Output signal

[0142] 39 Output interface

[0143] 41 training data

[0144] 43 Validation data

[0145] 45 data sets

[0146] 47 entry

[0147] 49 Confidence interval

[0148] 51 Weighting matrix

[0149] 53 Weight

[0150] 55 layers

[0151] E Input

[0152] 51 first procedural step

[0153] 52 second procedural step

[0154] 53 third procedural step

[0155] 54 fourth procedural step

[0156] 55 fifth procedural step

[0157] T 1 first step of the training procedure

[0158] T2 second step of the training procedure

[0159] T3 (optional) third step of the training procedure

[0160] T4 (optional) fourth step of the training procedure / query

[0161] T5 fifth step of the training procedure

[0162] V1 first step of the validation process

[0163] V2 second step of the validation process

[0164] V3 third step of the validation process

[0165] V4 (optional) fourth step of the validation procedure w11...w1x...wy1...wyx weights of individual neurons

Claims

Patent claims 1. Computer-implemented method comprising a model based on artificial intelligence (5) and comprising the following method steps Reading information (1) characterizing a patient (3) into the model, reading information (7) characterizing a visual impairment (9) of an eye (11) of the patient (3) into the model (5); Reading information (15) identifying an intraocular lens (17) to be inserted into the model (5); Determining a probability (37) of an occurrence of a refractive surprise that occurs when inserting the intraocular lens (19) to correct the refractive error (9) by the artificial intelligence-based model (5); and Providing the probability (37) of the occurrence of the refractive surprise.

2. The method according to claim 1, wherein the information (1) characterizing the patient (3) comprises at least one piece of information from the list, comprising information (1a) about a biomechanical stability of the eye (11); information (1b) about a previous disease of the eye (11); information (1c) about a previous treatment of the eye (11); information (1d) about an eye shape of the eye to be treated (11); and information about an age (1e) and / or a gender (1f) and / or an ethnicity (1g) of the patient (3).

3. The method according to claim 1 or 2, wherein the information (7) characterizing the visual impairment (9) of the eye (11) of the patient (3) comprises at least one piece of information from the list, comprising information (7a) about an axial length of the eye (11); information (7b) about a preoperative depth of the anterior chamber; information (7c) about a surface curvature and / or a corneal contour (7d) of the cornea of ​​the eye (11); information (7e) about a manifest refraction and / or a pachymetry (7f); information (7g) about a position of the intraocular lens (17) to be inserted; and information (7h) about a target refraction; includes.

4. The method according to one of claims 1 to 3, wherein information (15) identifying the intraocular lens (17) to be inserted comprises at least one piece of information from the list, comprising information (15a) about a type of intraocular lens (17); information (15b) about geometric dimensions of the intraocular lens (17); information (15c) about a predetermined position of the intraocular lens (17) in the eye (11); and information (15d) about a refractive power of the intraocular lens (17).

5. The method according to any one of claims 1 to 4, wherein the probability (37) determined by the artificial intelligence-based model (5) represents the probability of the occurrence of a refractive surprise.

6. A computer program product comprising instructions which, when executed on a data processing device (19), cause the data processing device (19) to carry out the method steps of the method according to any one of claims 1 to 5 7. A data processing device (19), wherein the data processing device (19) is designed to carry out the method according to one of claims 1 to 5.

8. A computer-implemented method for training a model (5) based on artificial intelligence, wherein the method comprises training the model using training data (41) at least comprising information (1) characterizing a patient (3), information (7) characterizing a visual impairment (9) of an eye (11) of the respective patient (3); information (15) characterizing an intraocular lens (17) to be inserted; and a probability (37) of an occurrence of a refractive surprise that occurs when the intraocular lens (17) is inserted into the eye (11) of the respective patient (3); and wherein the model (5) is configured after training based on: information (1) characterizing a patient (3), information (7) characterizing a visual impairment (9) of an eye (11) of the patient (3); and information (15) identifying an intraocular lens (17) to be inserted; to determine a probability (37) of an occurrence of a refractive surprise that occurs when the intraocular lens (17) is inserted into the eye (11) of the patient (3) to correct the refractive error (9) by the artificial intelligence-based model (5).

9. A trained artificial intelligence-based model (6) or weighting matrix (51) of such a trained model (6), wherein the trained model (6) or its weighting matrix (51) was obtained by a computer-implemented method according to claim 8.

10. A computer-readable medium comprising a model (5, 6) based on artificial intelligence, wherein the model is designed, when the model (5, 6) is executed by a data processing device (19), to cause the device (19) to use the model (5, 6) to determine, based on: information (1) characterizing a patient (3), information (7) characterizing a visual impairment (9) of an eye (11) of the patient (3); and information (15) characterizing an intraocular lens (17) to be inserted, a probability (37) of an occurrence of a refractive surprise that occurs when the intraocular lens (17) is inserted into the eye (11) of the patient (3) to correct the visual impairment (9), by means of the artificial intelligence-based model (5, 6).

11. Laser therapy device which comprises a data processing device (19) according to claim 7 or is designed to carry out a method according to one of claims 1 to 5.

12. Transplantation device which is designed to carry out a transplantation of an intraocular lens (17) and which further comprises a device for data processing (19) according to claim 7 or is designed to carry out a method according to one of claims 1 to 5.

Images