Methods for training predictive models, methods for determining laser irradiation parameters, and devices for treating the retina.

Abstract

The present invention relates to a method for training predictive models 1 and 2 that assign one or more laser irradiation parameters 10a, 10b, and 10c to current state parameters 3a, 3b, 3c, 4a, 4b, and 4c of a patient 5 for retinal treatment, wherein, as training data for a plurality of patients 9a, 9b, and 9c, in each case, at least one parameter 6a, 6b, and 6c independently determined from the patient's retina, particularly skin parameters, hair parameters, iris parameters, the patient's age, physiologically relevant lifestyle or medical history, and at least one retinal image 7a, 7b, and 7c of the patient acquired by an imaging method are used as training input parameters, and on the other hand, thresholds 8a, 8b, and 8c of the laser irradiation parameters, above which permanent tissue damage occurs, are used as training result parameters. The present invention further relates to such predictive models and their use in retinal irradiation of patients.

Description

Technical Field

[0001] The present invention relates to the fields of medical technology, measurement technology, laser technology, and data processing. The present invention can be used particularly advantageously in the treatment of the retina of patients having various diseases or age-related developments.

Background Art

[0002] In laser treatment of the retina, it is desirable to appropriately select and control the irradiation intensity in order to achieve a physiologically beneficial effect while remaining below the threshold of the effect that causes serious and undesired changes in the retina. This threshold, in addition to other boundary conditions such as laser parameters, also depends greatly on the individual retinal physiological characteristics of the patient, such as the level of pigmentation. Efforts are being made to individually record the physiological parameters of the patient to be treated so that optimized retinal treatment can be performed. For example, it is known from the paper of Non-Patent Document 1 to determine the age and skin type of a patient according to the so-called Fitzpatrick scale, but this method is based on the assumption that these parameters are related to the properties of the retina.

Prior Art Documents

Non-Patent Documents

[0003]

Non-Patent Document 1

Summary of the Invention

Problem to be Solved by the Invention

[0004] Building upon prior art, the present invention aims to create a method and device for retinal treatment that enables the recording of necessary patient parameters with as little effort as possible, and thus ensures safe and effective treatment of the retina. [Means for solving the problem]

[0005] The above objectives are achieved by using the present invention through a method for training a predictive model, a method for determining laser irradiation parameters, and a device for treating a patient's retina with laser irradiation. Advantageous embodiments of these aspects are also provided.

[0006] Accordingly, the present invention relates in part to a method for training a predictive model, particularly a neural network or algorithm, or a method for acquiring data from regression analysis or multivariate analysis to generate a mapping function, wherein the predictive model, particularly a neural network, algorithm or mapping function, assigns one or more laser irradiation parameters for retinal treatment to the patient's current state parameters, and in each case, as training data for multiple patients, on the one hand, at least one retinal image of the patient acquired by an imaging method, and in particular, at least one parameter determined independently of the patient's retina, particularly one or more retina-independent parameters independent of the patient's eye, and / or one or more retina-independent parameters of the patient's eye, further, parameters relating to one or more of the elements of skin parameters, hair parameters, iris parameters, cornea, lens and vitreous humor, the patient's age, physiologically related lifestyle, occupation, regular medications or medical history are used as training input parameters, and on the other hand, thresholds for the laser irradiation parameters (which may be multiple) above which permanent tissue damage occurs are used as training result parameters.

[0007] Parameters relating to one or more of the elements of the cornea, lens, and vitreous humor can also be interpreted to mean, for example, whether surgery was performed on the eye, the type of surgery, when the surgery was performed, and which element of the eye was modified or replaced. Replacement of the lens of the eye is a specific example of this parameter.

[0008] Furthermore, the parameters may relate to the occupation the patient is engaged in or the occupation that may have a specific effect on the eyes. For example, for this purpose, classes of occupations with similar stress levels on the eyes can be formed, such as occupations performed outdoors with high UV exposure and occupations that tend to be performed indoors with lower UV exposure. Individual occupations can also be specified / selected from a list, and these occupations can be automatically assigned certain specific exposures, such as farmers, roofers or bricklayers (with high UV exposure), or office workers, accountants, or cooks (with low UV exposure), or steelworkers (with high IR radiation exposure).

[0009] In the case of physiologically related lifestyle habits, for example, you can ask / inform the patient whether they are a smoker or non-smoker, or how long they smoked regularly. Also, some specific medications can cause changes in the eyes if taken regularly for a certain period of time, and these can be considered as parameters.

[0010] Certain diseases, such as diabetes, can also be inquired about or identified and can be considered as parameters, as can the habit of participating in sports and the regular amount of sports activity. Furthermore, certain regularly performed sports, such as surfing, sailing, and mountain climbing, which involve high UV exposure, and swimming, which involves moderate UV exposure but may also involve some chemical exposure from water additives, can also be inquired about or identified.

[0011] A predictive model can be designed, for example, as a programmable and / or trainable data processing device.

[0012] A mapping function can be interpreted as any mechanism that assigns one or more laser irradiation parameters for retinal treatment to individual variables, groups of variables, or ranges of variables. A mapping function in the form of an assignment table with discrete values ​​can also be created for this purpose. For example, age groups can be created (e.g., groups of the same size: 10-20 years, 20-30 years, 30-40 years, etc., or groups of heterogeneous sizes with smaller groups around the ages of 34, 60, and 78, where more significant changes tend to occur, and larger groups in the remaining areas), and each of these age groups can be assigned a set of irradiation parameters.

[0013] Similarly, the absorption intensity can be divided into ranges (for example, absorption = 0-0.19, 0.2-0.39, 0.4-0.59, etc.).

[0014] In this regard, the laser irradiation parameters should be interpreted as parameters that describe the intensity of irradiation to the treatment area, i.e., the area of ​​the retina being irradiated.

[0015] Excluding areas of the retina from treatment, such as blood vessels, should not be interpreted here as meaning that these areas are not treated and therefore laser parameters do not need to be determined in those areas. Laser parameters, or laser irradiation parameters, therefore describe the active operation of the therapeutic laser in which the therapeutic laser actually emits radiation such as laser power, power density, spot diameter, pulse duration, pulse frequency, duty cycle, number of pulses, and total energy density introduced into the tissue.

[0016] It is already possible to derive a large amount of information linked to each threshold determined by a predictive model, using individual retinal images or groups of different retinal images with partially overlapping information content.

[0017] Pigmentation intensity or density can be used as skin, hair, and iris parameters because they correlate to some extent with retinal pigmentation intensity. Pigmentation intensity can be determined, for example, by reflectance measurement. On the other hand, skin parameters, and even hair and iris parameters, can also be determined by color measurement, for example, by analyzing photographic images, especially color images or IR images or live views, and using the intensity in each color channel to analyze the color in a color space, such as the RGB or HSV color space.

[0018] Physiologically related lifestyle habits may include, for example, smoking, regular alcohol or other drug use, or obesity, while pre-existing conditions may include diabetes or hypertension.

[0019] By optionally combining patient parameters, independently determined from the patient's retina, with at least one retinal image as training input parameters, it becomes possible to collect data objectively, quickly, and easily. Using the retinal image itself, it becomes possible, for example, to detect and treat retinal damage or changes compared to previous conditions, or to exclude certain patient datasets from training due to certain atypical symptoms.

[0020] Patient retinal independent parameters can therefore be interpreted as parameters relating to one or more of the following elements, particularly skin parameters, hair parameters, iris parameters, cornea, lens, and vitreous humor, as well as the patient's age, physiologically relevant lifestyle, or medical history. Skin parameters / skin color, hair parameters / hair color, and iris parameters / iris color may mean the color and brightness or reflectivity of the skin, hair, or iris. Retinal independent parameters include parameters that are completely independent of the patient's eye, such as skin parameters and hair parameters, but also include parameters that can be manually entered into the system / predictive model or parameters that can be directly measured with respect to the eye, such as iris color or brightness, and further include parameters that affect one or more of the cornea, lens, and vitreous humor. This could mean, for example, that retinal independent parameters include whether or not the patient has an artificial lens, and optionally, how long ago the corresponding surgery to replace the patient's own lens was performed, which may be determined by measurement, manually entered, or obtained from patient data. This parameter has been shown to have a specific effect on the absorption and / or scattering of therapeutic radiation before it reaches the retina. In addition, this effect also acts when recording retinal images, and as a result, the scattering and absorption of radiation reflected in the retinal region and subsequently recorded, for example, by the imaging system, are further doubly affected by this scattering and absorption.

[0021] Therefore, in general, retinal independent parameters can be considered, at their discretion, for training predictive models, and further optionally, among these retinal independent parameters, parameters independent of the eye, and / or retinal independent parameters of the patient's eye, such as the properties of the iris and / or cornea and / or lens and / or vitreous humor, can be considered. Considering retinal independent parameters of the eye can be particularly useful because these parameters can be measured directly in the patient's eye.

[0022] These parameters can then be combined with one of other parameters such as hair color or skin color as the input of the system to be trained.

[0023] The scattering and / or absorption of radiation in the cornea and / or lens and / or vitreous body can be determined, for example, by selecting retinal regions known to have different reflectance, such as the optic disc of the retina, blood vessels carrying arterial blood, blood vessels carrying venous blood, and other normally reflective high-sensitivity regions, and measuring the reflectance of different retinal regions. If the intensity of the reflected radiation passing through the vitreous body, lens and cornea is measured in two or more of these retinal regions, and the ratio of the reflectance in these regions of the retina is known, the radiation loss caused by scattering and absorption when the radiation passes through the eye can be calculated as an offset.

[0024] The above description regarding parameters that can be considered in addition to retinal images when training a prediction model also refers to parameters that are determined and used in the procedure for determining laser irradiation parameters for treating a patient's retina during the operation / use of the prediction model.

[0025] A threshold value of laser irradiation parameters (which may be plural), above which permanent tissue damage changes occur, is used as a training result parameter for a patient's specific training input parameters. To determine the threshold value for each individual retina that provides training data, the patient's retina can be treated at different positions with different titration shots of varying intensities, that is, with laser treatments of different intensities. Thereafter, image analysis can be used to determine the laser treatment intensity at which adverse effects in the form of visible tissue changes can be detected. For this purpose, each retina providing training data is analyzed after one or more laser treatments of different intensities using one or more imaging methods, to determine which laser intensity is the lowest laser intensity at which tissue damage is already visually detectable. This value of the laser intensity can be set as the threshold value.

[0026] Even after different waiting times after treatment, several different imaging methods can be cumulatively used for the visual detection of tissue damage in each retina, and the lowest laser intensity threshold that can be used to determine damage with one of these methods and that is detectable in one of these methods can be adopted as the determined treatment threshold.

[0027] The mapping of the retina and the determination of location-dependent threshold values above which permanent tissue damage changes occur can be performed independently of the training of the prediction model, for example, immediately while acquiring other training data, but can also be performed in advance weeks or months earlier. For example, this can be done by different people at different locations and at different times than the education / training of the prediction model. This mapping can also be performed, for example, on the basis of available results of previous laser treatments, or by assignment of retinal images and literature values or empirical values.

[0028] Both thresholds obtained as part of the training data and thresholds or irradiation parameters obtained by the predictive model can be made available and processed according to the model as location-independent variables applied to the patient's retina as a whole. This results in easier handling while adhering to certain safety limits. On the other hand, due to location-dependent correction variables based on fundus images recorded near the treatment site, the design may consequently become location-dependent during treatment. However, in principle, the predictive model should not be limited to location-independent variables so that the laser treatment threshold can also be determined for specific regions of the retina.

[0029] When using a predictive model in a situation where patient-specific laser irradiation parameters are actually determined from currently recorded state parameters, it can be assumed, in principle, that the same or similar selections of input parameters are recorded for the patient being treated as training input parameters recorded in the training data. The assignment of radiation parameters then benefits from the advantages of numerous empirical values ​​using well-studied patients.

[0030] On the other hand, a particular advantage of the predictive model according to the present invention may be that the input data collected before the actual treatment of the patient is significantly less than the data used to train the model, saving time, effort, and cost. A further advantage of this model may be that it can find correlations that exist between partially redundant training data or data collected with respect to their information content, thereby partially replacing or eliminating data that is not collected from the patient being treated in a particular application. This is particularly true for retinal images, where several images are complex. It is advantageous if a single image or a few images are sufficient to record the patient's current state data before actual treatment, and it may be particularly advantageous if a single technical variation of the retinal image (e.g., color retinal photography or IR photography) is sufficient, and if necessary, only images of a single image type taken under different conditions are required. This saves time and effort for both the patient and the treating physician. Different conditions for recording types can be interpreted as, for example, several recordings being made at different illumination intensities or different illumination colors / wavelengths.

[0031] Specific examples of input parameters that have proven to be well-suited and important for training predictive models include one or more retinal images, as well as the patient's age (and possibly life stage) and retina-independent eye parameters, particularly those describing the status of the lens, i.e., whether the lens is natural or artificial, and / or whether and when certain surgeries have been performed on the eye.

[0032] To enable scaling of illuminance during retinal image generation and the sensitivity of the imaging system, such as a camera, and to make recorded retinal images comparable both during training the predictive model and during the actual application of the predictive model in the context of preparing for patient treatment, it is reasonable to determine these variables, namely the illuminance during retinal image generation and the sensitivity of the imaging system, and to fit them together or consider them using a scaling function.

[0033] The statement that a predictive model assigns one or more laser irradiation parameters for retinal treatment to the patient's current state parameters may mean that the predictive model specifies / assigns several laser irradiation parameters for different treatment methods or several sets of laser irradiation parameters for different treatment goals, or different sets of laser irradiation parameters corresponding to different probabilities or risk probabilities of tissue damage.

[0034] For example, a predictive model can be used to assign one set of therapeutic laser irradiation parameters that remain safely below the damage threshold, and a second set of therapeutic laser irradiation parameters where tissue coagulation / partial tissue damage is acceptable. In the same sense, retinal treatment devices can also be set up for such different treatment approaches.

[0035] In this specification, laser treatment or laser irradiation can be interpreted to mean not only treatment / irradiation using a laser, but also treatment / irradiation using a light source that has the same effect as a laser, such as a light-emitting diode with sufficient power, and it should be noted that therapeutic laser can be interpreted to mean a light source that has the same effect as a laser, such as a light-emitting diode.

[0036] A particular embodiment of this method can be defined as using at least two retinal images acquired at different points in time, particularly before and after titration, and / or using different imaging methods and / or under different technical imaging conditions, as training input parameters for each patient.

[0037] The following methods may be used as imaging methods to determine training input parameters: photographic color images, photographic infrared images (IR photography), OCT-B scan (depth cross-sectional images of images generated using optical coherence tomography), BAF (blue autofluorescence, fluorescence images without dye, excitation wavelength 488 nm), scanning laser ophthalmoscope, OCT in combination with scanning laser ophthalmoscope, and FAG (fluorescence angiography, 488 nm).

[0038] Therefore, for example, retinal images obtained using color retinal photography can be combined with IR images, OCT B images, blue autofluorescence images, or scanning laser ophthalmoscopy (SLO) images; or several color retinal photographs taken at different illumination intensities or different illumination spectra can be combined with each other; or IR images can be combined with OCT images or SLO. OCT images can be combined with SLO, or one of these can be combined with BAF. Each of the above methods can also be combined with, for example, FAG images. Several color retinal photographs or IR images taken under different imaging conditions can be combined; or several color retinal photographs or IR images taken under different imaging conditions can be combined with another of the above methods.

[0039] Of course, it is also possible to generate training data by combining three or more of these methods.

[0040] This method not only provides more data for training or regression analysis, but also identifies correlations and redundancies / overlaps in informational content between different types of retinal images and feeds them into predictive models. For example, if the same description can be derived from two different types of imaging, when applying the predictive model to a specific treatment case, the current set of patient data can be limited to one of the two types of imaging, or at least to fewer records than those used to generate the training data, without substantially compromising quality.

[0041] Furthermore, one or a group of less complex methods for generating retinal images to obtain data on the patient's current condition, namely, methods that are less costly or time-consuming or less prone to error, can be used in the application and can be performed without expert workers or by at least semi-skilled workers.

[0042] In the method of the type described above, the respective thresholds for each patient's laser irradiation parameters (which may be multiple) are determined by a series of titration shots onto the patient's retina with different laser irradiation parameters and subsequent detection of tissue changes based on multiple retinal images acquired by different imaging methods. In particular, the lowest threshold determined by one of the imaging methods may be further specified to be used as a training result parameter for training a predictive model, especially for training a neural network, or for acquiring data for regression or multivariate analysis to generate a mapping function.

[0043] When generating training data, thresholds for laser irradiation parameters (which may be multiple) for each patient can be determined with increased safety, sensitivity, and reliability, such as thresholds that, when reached or exceeded, result in permanent tissue damage. It cannot be ignored that some specific detection methods may be more sensitive in individual patient groups than in others. In addition, some specific methods may be more sensitive than others when detecting tissue changes after a certain waiting time following a titration shot.

[0044] In each of the methods used, retinal images can be automatically analyzed, for example, by digital image analysis. This results in an objectified standard.

[0045] To specifically determine the thresholds for acquiring training data, it may be stipulated that the thresholds for each laser irradiation parameter (which may be multiple), particularly laser power, power density, spot diameter, pulse duration, pulse count or duty cycle, or any combination of two or more of the aforementioned physical quantities, are determined for each patient by a series of titration shots onto the patient's retina using different laser irradiation parameters, and subsequent detection of tissue changes using multiple retinal images acquired by different imaging methods in each case. The thresholds determined by the imaging method that show the lowest threshold for most of the patients examined are used as training result parameters for training predictive models, particularly neural networks, for all patients examined.

[0046] Specifically, in one implementation of the method, the threshold values ​​for each patient's laser irradiation parameter(s) may be defined as training result parameters obtained by a series of titration shots onto the patient's retina using different laser irradiation parameters, and subsequent detection of tissue changes using optical coherence tomography in combination with IR scanning laser ophthalmography.

[0047] In the above type of method, the power of the irradiating laser, or the power density of the laser spot using otherwise fixed laser irradiation parameters, or the total energy deposited on the laser spot over a fixed irradiation time, may be further specified to vary during the titration shot, and the threshold thereof may be selected as the laser irradiation parameter to be obtained as the training result parameter.

[0048] Each of these variables may change during laser irradiation under otherwise constant boundary conditions, potentially leading to permanent tissue changes if the threshold is exceeded. The threshold for tissue changes can be determined by the described series of tests.

[0049] When determining training input parameters, a particular challenge may be objectively identifying the retinal independent parameters for each patient. On the one hand, these parameters can be obtained as part of a medical history and can be entered via an input device. This procedure is suitable, for example, for specifying age, illness, or habits. However, data regarding skin, hair, and iris parameters are particularly meaningful and important. These can be evaluated on a scale, for example, by placing them on the well-known Fitzpatrick scale or separate color or lightness scales for individual parameters. Alternatively, technical measurement methods can be used to determine the lightness or color of the skin, hair, and iris, for example, through color photography or IR photography under defined lighting conditions. The lightness or color distribution in defined color channels can be objectively measured and specified as parameter values. When evaluating photographs, a trained neural network, trained using multiple photographs and assigned parameterized evaluations, can also be used for this evaluation. Such a trained neural network may be identical to the trained neural network described above; that is, the predictive model described above can be additionally trained to classify patient photographs into different skin / hair / iris types.

[0050] Thus, the method described above can also be prepared so that skin parameters and / or hair parameters and / or iris parameters are determined by optical measurement methods, in particular by measuring the color and / or reflectance of the patient's skin, hair, or iris.

[0051] In addition to methods for forming the type of predictive model described above, the present invention also relates in particular to a programmed and / or trained data processing device that assigns one or more laser irradiation parameters for retinal treatment to the patient's current state parameters, and to a predictive model formed in accordance with the methods described above, in particular by training a neural network or regression analysis or multivariate analysis and generation of a mapping function.

[0052] Such predictive models can take the form of trained neural networks or data processing programs or data processing devices, and can be stored, for example, on a data carrier. If such a data processing program is executed by a processor or data processing system, the resulting method can assign laser irradiation parameters for retinal treatment to the patient's current state parameters.

[0053] The present invention also relates to a method for determining laser irradiation parameters for treating a patient's retina, wherein in a first step, the patient's current state parameters are determined, and in a further step, one or more laser irradiation parameters for retinal treatment are assigned to the state parameters by a predictive model of the type described above. These state parameters include one, two, or three or more retinal images of the patient acquired by imaging methods, where, in the case of two or more retinal images, the retinal images are acquired by different imaging methods and / or under different imaging conditions. The assignment of laser irradiation parameters for retinal treatment to the current state parameters can be performed by a predictive model described above, in particular a neural network trained according to the method described above.

[0054] In addition, in such a method, the condition parameter may be defined as including at least one parameter independently determined from the patient's retina, in particular one or more retinal independent parameters independent from the patient's eye and / or one or more retinal independent parameters of the patient's eye, and more particularly parameters relating to one or more of the elements of skin or hair parameters, iris parameters, cornea, lens and vitreous humor, the patient's age, physiologically related lifestyle, occupation, regular medications or medical history.

[0055] The retinal independent parameters of the patient's eye may include the following parameters, namely, iris color or brightness, and parameters relating to one or more of the cornea, lens, and vitreous humor, such as one or more of the absorption and / or scattering of radiation by these elements.

[0056] Parameters relating to one or more of the elements of the cornea, lens, and vitreous humor may describe, for example, whether and when surgical intervention was performed on one of these elements, and / or whether and when one of these elements was replaced or altered.

[0057] Many of the parameters mentioned above, such as skin, hair, or iris color, or whether the patient has an artificial lens or has undergone surgery on the eye, and what kind of surgery was performed on the eye, can be manually entered into the system / predictive model. Some of the parameters, such as iris color or brightness, and parameters relating to one or more of the cornea, lens, and vitreous humor, can also be measured directly on the eye. This could mean, for example, that as a retinal independent parameter, whether the patient has an artificial lens, and optionally, how long ago the corresponding surgery to replace the patient's own lens was performed, can be determined by measurement, manually entered, or obtained from patient data. This parameter has been shown to have a particular effect on the absorption and / or scattering of therapeutic radiation before it reaches the retina. In addition, this effect also acts when recording retinal images, and as a result, the scattering and absorption of radiation reflected in the retinal region and subsequently recorded, for example, by the imaging system, is further doubly affected by this scattering and absorption.

[0058] Generally, when using a predictive model, retinal independent parameters can be optionally determined, measured, or input and considered, and among these retinal independent parameters, ocular independent parameters and / or retinal independent parameters of the patient's eye, such as the characteristics of the iris and / or cornea and / or lens and / or vitreous humor.

[0059] Considering retinal independent parameters of the eye can be particularly useful because these parameters can be measured directly in the patient's eye.

[0060] These parameters can then be combined with one of several other parameters, such as hair color or skin color, as input to the system being trained.

[0061] The scattering and / or absorption of radiation in the cornea and / or lens and / or vitreous humor can also be determined when determining the irradiation parameters by measuring the reflectivity of different retinal regions, for example, using a predictive model. Retinal regions known to have different reflectivity, such as the optic nerve head, arterial blood vessels, venous blood vessels, and other regions, are selected for this purpose. If the intensity of the reflected radiation is measured in two or more of these retinal regions and the ratio of the reflectivity in these regions is determined, the radiation loss resulting from scattering and absorption as the radiation passes through the eye can be calculated as an offset.

[0062] Generally, the required laser irradiation parameters relate to the irradiation parameters of the radiation source / treatment laser that is actively operating when irradiating the area of ​​the retina to be treated, i.e., in the phase in which radiation is actually emitted. The required laser irradiation parameters can be, for example, laser power, maximum duration of irradiation at the laser spot, and laser power density, which are, in some cases, related to the duty cycle or the total amount of energy introduced into the retinal tissue at the spot by the laser intensity. It should be possible to select these parameters in such a way that laser irradiation with these parameters either does not cause permanent tissue damage to the retina or causes precisely calculated tissue changes.

[0063] One or more of the following methods can be used as imaging methods to generate the current retinal image to be used as the current state parameter, and in principle, this can also be used to generate the training dataset for training the prediction model described above. The following methods are: These include photographic color imaging, or several sets of illumination conditions, such as several color images with illumination intensity or illumination spectrum, photographic infrared imaging (IR imaging), and, in some cases, several images with different illumination conditions, such as illumination intensity or illumination spectrum, OCT-B-scan (depth cross-sectional images of images generated using optical coherence tomography), BAF (blue autofluorescence, acquisition of fluorescence images without dye addition, excitation wavelength 488 nm), scanning laser ophthalmoscope, OCT+SLO IR (optical coherence tomography used in combination with IR scanning laser ophthalmoscope, confocal imaging method in the infrared region), and FAG (fluorescence angiography, fluorescence imaging method using fluorescein dye, excitation wavelength 488 nm).

[0064] Particularly advantageous are imaging methods such as color photography and IR photography, which can be performed easily and quickly using user-friendly and inexpensive equipment without the need for specialized personnel. The BAF method is also fast and easy to operate because it does not use contrast agents. For example, it is possible to combine two of the three methods mentioned above. In principle, a single retinal image from the above methods may be sufficient, and in this case as well, the three methods mentioned above are particularly low-cost and easy to evaluate.

[0065] Pigmentation intensity or density can also be used as skin, hair, and iris parameters when determining a patient's current state variables. This is because it correlates to some extent with retinal pigmentation intensity. Pigmentation intensity can be determined, for example, by reflectance measurement. However, hair and iris parameters, as well as skin parameters, can also be determined by color measurement, for example, by recording the intensity in several color channels of an RGB system.

[0066] Specifically, it may be specified that the patient's skin parameters and / or hair parameters and / or iris parameters are determined by optical measurement methods, in particular by measuring the color and / or reflectance of the patient's skin, hair or iris, and / or by measuring photographic images of the patient's skin, hair or iris, and / or by analyzing photographic images of the patient's skin, hair or iris, especially color images, IR images or live views, to determine state parameters, and the color is analyzed using the intensity in each color channel, evaluated, for example, by a predictive model, in a color space, such as an RGB color space or an HSV color space.

[0067] Physiologically related lifestyle habits may include, for example, smoking, regular alcohol or other drug use, or obesity, while pre-existing conditions may include diabetes, obesity, and hypertension.

[0068] By combining parameters independently determined from the patient's retina with at least one retinal image as parameters representing the patient's current status, objective, rapid, and easy data acquisition becomes possible.

[0069] Specific examples of patient parameters that have proven to be well-suited and important when determining irradiation parameters using predictive models include one or more retinal images, as well as the patient's age (and possibly life stage) and independent retinal eye parameters, particularly parameters describing the status of the lens, i.e., whether the lens is the patient's own lens or an artificial lens, and / or whether and when certain surgeries have been performed on the eye.

[0070] These parameters can be made available to predictive models before treating a patient's retina by measuring or imaging the retina and inputting or transferring known patient data from a database. In many cases, this data alone is already sufficient to successfully determine promising radiation parameters by the predictive model. These radiation parameters can then be used for subsequent treatment.

[0071] In addition, a scaling function can be used to account for the illuminance and sensitivity ratio of the imaging system / camera when training the predictive model and when actually using the predictive model to obtain patient measurements. The predictive model can provide illuminance and sensitivity data of the imaging system used to train the predictive model.

[0072] A further embodiment of the method described above may specify that from at least one retinal image of the patient, areas whose reflectivity deviates by at least 50% from the average reflectivity of the retina, particularly areas whose reflectivity is more than 50% higher than the average reflectivity of the retina or 50% lower than the average reflectivity of the retina, are determined and excluded from subsequent laser treatment.

[0073] Areas of the retina with significantly higher or lower reflectivity than other areas are assumed to be areas that are practically free of pigmentation and / or do not require treatment. When using color fundus photography or infrared photography, these are particularly the optic nerve head, which has high reflectivity, and the areas above the major blood vessels supplying the retina, which have significantly lower reflectivity than other areas of the retina (the latter due to hemoglobin absorbing radiation). When other imaging methods are used, the contrast may differ. For example, in images acquired by fluorescence angiography, areas of blood vessels appear clearly, i.e., at least 50% brighter than other areas of the retina, whereas when using autofluorescence, these areas appear significantly darker than other areas of the retina.

[0074] The above method for determining laser irradiation parameters can also be designed so that the variance of the reflectance value, which is less than 50% of the mean reflectance, can be obtained from the patient's retinal image, and the correction value of the laser irradiation parameter determined by the predictive model can be obtained from this variance, and / or a loss parameter can be obtained from the reflectance obtained at the optic nerve head, and this parameter can be designed to take into account the loss due to light absorption and scattering in the vitreous humor of the eye, and this loss parameter can be taken into account when determining the laser irradiation parameters.

[0075] By removing areas where the average reflectance is more than 50% higher than the average reflectance of other areas, the non-pigmented areas of the retina described above are excluded from the calculation. By calculating the dispersion of reflectance in the remaining areas, corrected values ​​for laser irradiation parameters that enhance safety against damaging laser effects can be obtained.

[0076] From the reflectance values ​​measured in the optic nerve head region, the intensity loss occurring along the path of the laser beam through the vitreous humor to the retina and back through the vitreous humor can be calculated, taking into account residual retinal absorption independent of pigment deposition. This consists of absorption loss and scattering in the vitreous humor, both of which depend, among other things, on the health of the eye, particularly the presence of cataract disease. If there is an intensity loss due to these effects, the laser intensity can be increased during treatment to compensate for it.

[0077] The loss of laser radiation during its passage through the eye to the retina can also be determined by reflectance measurement.

[0078] The following formula applies to absorption A. A = 1 - (Refl MAC / REfl ONH ) In this formula, Refl MAC This is the reflectance intensity of the macula, REfl ONHThis is the reflectivity of the optic nerve head. Alternatively, one can use a formula to calculate the absorption by comparing the reflectivity of the retina in the macula region with and without blood vessels behind it.

[0079] When measuring retinal reflectance, drusen, or deposits beneath the retinal pigment epithelium (RPE), must also be considered. In the retinas of patients with macular degeneration, the number of drusen, which affect the backscattering of light, is increased. Therefore, when analyzing retinal reflectance to determine the radiation intensity required for treatment, and when measuring reflectance to determine the absorption of radiation as it passes through the eye to the retina, measurements in the drusen region should be excluded, or the presence of drusen should be taken into account in the evaluation model.

[0080] Finally, the present invention also relates to a device for laser irradiation treatment of a patient's retina, comprising at least one therapeutic laser, at least one device for generating a retinal image, and a processing device having a predictive model of the type described above. The predictive model assigns one or more laser irradiation parameters for controlling the therapeutic laser to at least one retinal image currently being generated by the device, in particular to additional, currently desired state parameters.

[0081] Devices for laser irradiation of a patient's retina can be designed in a known manner and structurally connected to one or more devices for generating retinal images relative to the optical system used. This facilitates the creation of a common reference system for the retina being treated, for measurement and treatment, and, if necessary, for the treatment itself. This also makes it possible to ensure that certain individually detected areas of the retina, such as major blood vessels and the optic nerve head, are excluded from irradiation. In addition, such a system can also be used to observe the retina during treatment and, if necessary, for eye tracking, i.e., observation and compensation of eye movements during treatment.

[0082] For example, a system for color retinal photography and / or IR photography and / or FAG (fluorescence angiography) can be connected to a treatment laser.

[0083] In addition, a measuring device for determining skin parameters, hair parameters, or iris parameters can be integrated into the device. This device may include, for example, a color camera or IR camera or color image scanner for capturing images of the patient's skin, hair / hairline / eyebrows, or iris, and an analysis device for assigning status parameters to the image by evaluating reflectivity or color. For example, the camera or scanner may be integrated into a chamber having an illumination device and a window, in which the patient's skin, hair, or iris can be positioned. The illumination may be variable with respect to intensity and / or color spectrum in order to obtain additional information by forming the difference or quotient of different images under different illumination conditions.

[0084] In the following, the present invention will be described by means of embodiments shown in the drawings. [Brief explanation of the drawing]

[0085] [Figure 1] This is a diagram illustrating the training process for a predictive model. [Figure 2] This is an explanatory diagram showing how to determine laser irradiation parameters using a trained predictive model. [Figure 3] This is a further diagram illustrating the training of the predictive model. [Modes for carrying out the invention]

[0086] Figure 1 schematically illustrates the process for training a predictive model, specifically the process for training neural network 1 using training input data 6a, 6b, 6c, 7a, 7b, 7c and training result data 8a, 8b, 8c. Each of the above data is determined based on a larger group of people 9a, 9b, 9c and made available to neural network 1 or a regression analysis device for training or evaluation.

[0087] The training input data can initially consist of image data from one, two, or more images 7a, 7b, 7c, particularly retinal images of each person 9a, 9b, 9c. For example, fundus color photographs or IR photographs, each with one or more illumination conditions, can be used as the recording mode. Here, the illumination conditions can be defined by illuminance, illumination angle, and a specific illumination spectrum. In addition, OCT-B-scan (depth cross-sectional images of images generated using optical coherence tomography), BAF (blue autofluorescence, fluorescence images without added dye, excitation wavelength 488 nm), OCT+SLO IR (optical coherence tomography in combination with IR scanning laser ophthalmography, confocal imaging method in the infrared region), and FAG (fluorescence angiography, fluorescence images using fluorescein dye, excitation wavelength 488 nm) can be considered. In Figure 1, the imaging devices for different images are represented by the symbols 13 and 14. These cameras are aligned with each other so that the reference points for different images can relate to each other.

[0088] In addition, skin parameters and / or hair parameters and / or iris parameters, as well as other health-related parameters such as age, certain specific lifestyle habits, and medical history, can be obtained as further training input parameters 6a, 6b, and 6c, and can be used as training input parameters for neural networks or regression analysis. Skin parameters and / or hair parameters and / or iris parameters can be obtained by a measurement device 11, for example, a color camera, an IR camera, or a hyperspectral camera, in any case accompanied by an evaluation device. These parameters and additional parameters can also be entered using a data input device 12, for example, a keyboard, or a voice recognition device using a microphone.

[0089] The aforementioned training input parameters initially provide the predictive model with a database for a comprehensive description of a person's state.

[0090] The threshold data obtained for the laser irradiation parameters of each individual 9a, 9b, and 9c, which form the training outcome parameters 8a, 8b, and 8c, are linked to the aforementioned data / training input parameters. This threshold data is obtained for each person by using a laser to emit titration shots with different laser irradiation parameters, i.e., different intensities, into each person's retina. Which of the titration shots caused permanent tissue change is individually evaluated for each person, thereby determining the threshold or laser threshold for each person 9a, 9b, and 9c, and making it available to the predictive model as a training outcome parameter.

[0091] To detect tissue changes, the recording device 15 can record each person's retina using different recording modes 15a, 15b, and 15c. In principle, all of the above recording modes can be used, and some can be used in combination. The blue autofluorescence method has been found to be particularly sensitive. This method can be combined with one or more of the other methods described above to detect tissue changes with the highest possible sensitivity in each individual case. The results of the titration shots can be evaluated by experts in different images generated by different acquisition modes, for example, using the threshold that resulted in recognizable tissue changes in any of the images in the titration shot with the lowest intensity as a standard. It is also possible to wait a certain amount of time, for example, more than a week, before examining the retina after the titration shots.

[0092] On the other hand, in principle, it is possible to determine which of the recording modes is most sensitive for most test subjects, and then use this mode for evaluating all subjects. Tissue changes can also be detected using digital image processing.

[0093] Instead of a neural network, regression analysis can also be performed using the determined and collected data to determine and weight the influence of individual features of the training input parameters on the laser threshold. Regression analysis can then be used to train / learn an algorithm that functions as a predictive model.

[0094] Figure 2 schematically illustrates the system and method for using predictive model 1 in the actual treatment of patient 5. Patient 5 uses recording devices 18, 19 for this purpose, which take photographs 4a, 4b, 4c of the patient's retina or multiple parts of the retina.

[0095] In some cases, further recording devices 16 and / or 17, for example in the form of a color camera, IR camera, or hyperspectral camera, record the reflectance, brightness, or color of the patient's skin, hair, or iris. These reflectances, etc., are examples of condition data 3a, 3b, 3c that can be determined independently from the patient's retina. In the simplest case, this can also be done using a photosensitive sensor and illumination device placed in a common measurement chamber, thereby having a window, and the patient's skin, hair, or eye is positioned in front of the window.

[0096] For example, the imaging device may consist of only one technical device for capturing a single imaging mode from the above list, such as a fundus color camera, a fundus infrared camera, or a device for blue autofluorescence imaging. In principle, the recording device may consist of only one camera for recording different recording modes under different lighting conditions, because the technical effort required in this case remains very low. Alternatively, two or more cameras can be combined to capture different recording modes. In principle, when recording the current state data of patient 5 before treatment, the recording can be expected to be performed using fewer different recording modes than are needed to train the predictive models 1 and 2 used, and to be used for analysis.

[0097] The current status data acquired using local devices 16, 17, 18, and 19 can be further processed at the treatment site, but this data can also be further processed by a distributed computer network with, for example, a remote server, thereby allowing predictive models 1 and 2 to be deployed at the treatment site or remotely from the treatment site. Predictive models 1 and 2 assign one or more laser irradiation parameters, such as the laser intensity for tissue alteration treatment and the upper limit / threshold of the laser intensity at which tissue alteration is safely avoided, to the patient 5 data / current status data recorded by local devices 16, 17, 18, and 19. In this regard, laser intensity refers to all laser and beam guidance parameters that have an effect on possible tissue damage, such as laser power, power density, duty cycle, laser dwell time on the spot, and total energy introduced into the laser spot.

[0098] In principle, the required laser threshold can be uniform across the entire retina being treated, but it can also be location-dependent, for example, excluding the optic nerve head and major blood vessels from treatment. In addition, certain specific regions of the retina can be excluded depending on the retinal image used.

[0099] The current retinal image may be used to determine the dispersion / variability of retinal pigment deposition / absorption intensity, i.e., the degree of deviation of the average pigment deposition / absorption intensity from the determined average, and it may be specified that, when the retina as a whole is treated with position-independent laser irradiation parameters, a correction value for the laser irradiation parameter that reduces the laser intensity at high dispersion so as not to damage high-absorption areas on the retina may be determined from the dispersion. When determining the average retinal pigment deposition / absorption, the locations of the optic nerve head and major blood vessels can be excluded.

[0100] It is also possible to determine the loss rate representing the absorption and scattering of laser radiation in the vitreous humor of the eye from the reflectance measured and determined from the current image of retinal singularities such as ganglia compared to other areas of the retina. Using a detectable loss rate, the laser intensity can be given a correction factor for laser irradiation intensity that is greater than 1, for example, between 1 and 1.5, in the case of position-independent treatment.

[0101] As can be seen in Figure 2, patient data / current status data recorded by local devices 16, 17, 18, and 19 are fed to predictive models 1 and 2, which use this data to determine laser irradiation parameters and transmit them to the treatment device 20. This treatment device can be positioned in the same location as local devices 16, 17, 18, and 19 so that, in principle, patient 5 can be treated at the same location using the same device that captured the current retinal image after the patient's current status data has been recorded. This has the advantage that the imaging method used for the retinal image can, in some cases, also be used for controlling, aligning, and guiding the treatment laser. For example, the treatment laser can pass through the same optical elements such as the aperture and lens used to record the current status data. An eye-tracking device 21 can also be provided and, as indicated by arrow 23, can be connected to at least one of devices 18 and 19. This device 21 detects eye movements and guides the treatment laser 22 in a known manner.

[0102] Figure 3 shows a more detailed diagram illustrating the generation of predictive models 1 and 2 using training data 6a, 6b, 6c, 7a, 7b, and 15a, which are combined in integration step 24 and supplied to the predictive models for training. The results of the predictive models when analyzing individual patients then become a “risk curve” 25, from which the laser irradiation parameters 10a, 10b, and 10c can be obtained as a function of the desired or acceptable probability of tissue damage / modification. Before extracting specific laser irradiation parameters from the risk curve 25 for a given treatment case, an optional approval step 26 can be provided. This step is shown in Figure 2 and can be performed, for example, by an ophthalmologist.

[0103] In many cases, the selection of parameters described above may be sufficient as state parameters for training a predictive model, and may also be sufficient when using the model in relation to treatments, such as detecting / inputting hair color, eye color, medical history, and previous eye surgeries. Further information about the profession in which the individual is engaged or primarily engaged, and / or the sport in which they participate intensively, may also be added.

[0104] Optionally, information regarding whether the patient is or was a smoker can be added to these parameters.

[0105] The system described above enables simple, less complex, safer, and more harmless treatment of the patient's retina.

Claims

1. A method for training a predictive model (1, 2), particularly a neural network (1) or an algorithm (2), or a method for acquiring data from regression analysis or multivariate analysis to generate a mapping function, wherein the predictive model, particularly the neural network, the algorithm, or the mapping function assigns one or more laser irradiation parameters (10a, 10b, 10c) for retinal treatment to the current state parameters (3a, 3b, 3c, 4a, 4b, 4c) of the patient (5), As training data for multiple patients (9a, 9b, 9c), in each case, On the one hand, at least one retinal image of the patient (7a, 7b, 7c) acquired by the imaging method, and in particular at least one parameter (6a, 6b, 6c) independently determined from the patient's retina, in particular one or more retinal independent parameters independent from the patient's eye, and / or one or more retinal independent parameters of the patient's eye, and in particular parameters relating to one or more of the elements of skin parameters, hair parameters, iris parameters, cornea, lens and vitreous humor, the patient's age, physiologically related lifestyle, occupation, regular medications or medical history are used as training input parameters. On the other hand, the thresholds (8a, 8b, 8c) of the laser irradiation parameters, above which permanent tissue damage occurs, are used as training result parameters. A method characterized by the following features.

2. The method according to claim 1, characterized in that at least two retinal images (7a, 7b, 7c) are used as training input parameters for training for each patient, and the at least two retinal images are acquired using different imaging methods or under different technical imaging conditions.

3. The following method, namely, Photographic color images, photographic infrared images, OCT-B-scan (depth cross-sectional images of images generated using optical coherence tomography), BAF (blue autofluorescence, fluorescence images without dye, excitation wavelength 488 nm), OCT + SLO IR (optical coherence tomography in combination with IR scanning laser ophthalmography, confocal imaging method in the infrared region), FAG (fluorescence angiography, fluorescence images using fluorescein dye, excitation wavelength 488 nm), or scanning laser ophthalmography using one or more wavelengths, especially wavelengths intended for post-treatment. The method according to claim 1 or 2, characterized in that one or two of the methods are used as an imaging method for determining the training input parameters (7a, 7b, 7c).

4. The respective thresholds (8a, 8b, 8c) for the laser irradiation parameters for each patient (9a, 9b, 9c) are determined by a series of titration shots onto the retina of the patient with different laser irradiation parameters, and subsequent detection of tissue changes based on multiple retinal images acquired by different imaging methods. In particular, the lowest threshold obtained by one of the imaging methods is used as a training result parameter for training the prediction model, especially for training a neural network, or for acquiring data for regression analysis or multivariate analysis to generate a mapping function. The method according to any one of claims 1 to 3, characterized in that

5. The respective thresholds (8a, 8b, 8c) for the laser irradiation parameters for each patient (9a, 9b, 9c) are determined by a series of titration shots onto the retina of the patient with different laser irradiation parameters, and subsequent detection of tissue changes based on several retinal images acquired by different imaging methods in each case. The threshold obtained by the imaging method, which shows the lowest threshold for most of the patients examined, is used as the training result parameter for training the predictive model for all patients examined, particularly for training the neural network. The method according to any one of claims 1 to 3, characterized in that

6. The respective thresholds (8a, 8b, 8c) for the laser irradiation parameters, particularly the laser power, power density, spot diameter, pulse duration, number of pulses, or duty cycle, or any combination of two or more of the variables, are determined as training result parameters for each patient (9a, 9b, 9c) by a series of titration shots onto the retina of the patient with different laser irradiation parameters, and subsequent detection of tissue changes using optical coherence tomography in conjunction with IR scanning laser ophthalmography. The method according to any one of claims 1 to 3, characterized in that

7. The method according to any one of claims 1 to 6, characterized in that the power of the irradiation laser, or the power density of the laser spot using otherwise fixed laser irradiation parameters, or the total energy deposited on the laser spot over a fixed irradiation time, changes during the titration shot, and the threshold (8a, 8b, 8c) is selected as the laser irradiation parameter to be obtained as a training result parameter.

8. The method according to any one of claims 1 to 7, characterized in that skin parameters and / or hair parameters and / or iris parameters (6a, 6b, 6c) are determined by an optical measurement method, and the color and / or reflectance of the patient's skin, hair, or iris is measured in particular.

9. Predictive model, in particular a programmed and / or trained data processing device, designed in accordance with any one of claims 1 to 8, by assigning one or more laser irradiation parameters (10a, 10b, 10c) to current state parameters (3a, 3b, 3c, 4a, 4b, 4c) of a patient (5) for retinal treatment, in particular by training a neural network (1) or generating a mapping function by regression analysis or multivariate analysis.

10. A method for determining laser irradiation parameters (10a, 10b, 10c) for treating the retina of a patient (5), In the first step, the patient's condition parameters (3a, 3b, 3c, 4a, 4b, 4c) are determined. In a further step, one or more laser irradiation parameters (10a, 10b, 10c) for retinal treatment are assigned to the state parameters by the predictive model (1, 2) described in claim 9. The state parameter includes one, two, or three or more retinal images (4a, 4b, 4c) of the patient (5) acquired by the imaging method, In the case of two or more retinal images, the retinal images are acquired by different imaging methods and / or under different imaging conditions. A method characterized by the following features.

11. A method for determining the laser irradiation parameters (10a, 10b, 10c) according to claim 10, The state parameters (3a, 3b, 3c, 4a, 4b, 4c) include at least one parameter (6a, 6b, 6c) independently determined from the patient's retina, in particular one or more retinal independent parameters independent from the patient's eye and / or one or more retinal independent parameters of the patient's eye, and more particularly parameters relating to one or more of the elements of skin or hair, iris, cornea, lens and vitreous humor, the patient's age, physiologically related lifestyle, occupation, regular medications or medical history. A method characterized by the following features.

12. The skin parameters and / or hair parameters and / or iris parameters (3a, 3b, 3c) of the patient (5) are determined by an optical measurement method. In particular, the color and / or reflectance of the patient's skin, hair, or iris are measured, and / or photographic images of the patient's skin, hair, or iris are evaluated by the predictive model in each case to determine state parameters (3a, 3b, 3c). The method according to claim 11, characterized in that

13. The method according to any one of claims 10 to 12, characterized in that the region having high reflectivity is determined from at least one retinal image (4a, 4b, 4c) of the patient and is excluded from irradiation for subsequent laser treatment, and the region having high reflectivity is defined in particular as a region whose reflectivity is at least 50% higher than the average reflectivity value of the retina.

14. The method according to any one of claims 10 to 13, characterized in that the variance of the reflectance value, which is less than 50% of the difference from the average reflectance, is determined from the patient's retinal images (4a, 4b, 4c), and the corrected values ​​of the laser irradiation parameters (10a, 10b, 10c) determined by the prediction models (1, 2) are determined from the variance, and / or the loss parameter is determined from the reflectance determined at the optic nerve head, and the loss parameter describes the loss due to the absorption and scattering of light in the vitreous humor of the eye, wherein the loss parameter is taken into consideration when determining the laser irradiation parameters (10a, 10b, 10c).

15. A device for treating a patient's retina by laser irradiation, comprising at least one treatment laser (22) and at least one device (18, 19) for generating a retinal image, and a processing device having a predictive model (1, 2) according to claim 9, which assigns one or more laser irradiation parameters (10a, 10b, 10c) for controlling the treatment laser to at least one retinal image (4a, 4b, 4c) currently being generated by the device and in particular additional currently desired state parameters (3a, 3b, 3c).

Images