Systems and methods for selecting intraocular lenses

Abstract

A system and method for selecting an intraocular lens includes obtaining one or more preoperative measurements of an eye; selecting a subset of the past IOL implantation records from a plurality of past IOL implantation records to evaluate a first plurality of candidate prediction models; evaluating the first plurality of candidate prediction models; selecting a first prediction model from the first plurality of candidate prediction models based on the evaluation; calculating a plurality of estimated postoperative MRSE values ​​based on a set of IOL powers and one or more preoperative measurements of the eye using the selected first prediction model; identifying a first IOL power corresponding to the first estimated postoperative MRSE value that matches a predetermined postoperative MRSE value; and providing the identified first IOL power to a user to assist in selecting an IOL to implant in the eye.

Description

[Technical Field]

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS This application claims priority to and the benefit of U.S. Provisional Patent Application No. 62 / 613,927, filed January 5, 2018, entitled "SYSTEMS AND METHODS FOR VECTOR IOL POWER CALCULATION."

[0002] The present disclosure relates to systems and methods for assisting in the selection of an intraocular lens for implantation. [Background technology]

[0003] Cataract surgery involves removing the eye's natural lens and, in most cases, replacing it with an artificial intraocular lens (IOL). Good preoperative surgical planning is crucial to achieving optimal postoperative visual outcomes. Some of the key preoperative planning decisions are the selection of the appropriate IOL type and power to achieve the desired postoperative total refractive spherical equivalent (MRSE) after IOL implantation.

[0004] Typically, one or more predictive models may be used to identify an appropriate IOL type and power to achieve a desired MRSE based on, for example, a patient's preoperative eye measurements. However, predictive models may not produce accurate results under all circumstances. For example, a first predictive model may produce more accurate results than a second predictive model under one set of circumstances, while the second predictive model may produce more accurate results than the first predictive model under a different set of circumstances. Therefore, using a fixed predictive model to identify an appropriate IOL type and power under all circumstances may result in less than optimal visual outcomes for at least some patients. Summary of the Invention [Problem to be solved by the invention]

[0005] Therefore, there is a need in the art for techniques to better select intraocular lenses for implantation that will lead to optimized visual outcomes for patients. [Means for solving the problem]

[0006] According to some embodiments, a method includes, by one or more computing devices implementing a prediction engine, obtaining one or more pre-operative measurements of an eye to perform intraocular lens (IOL) implantation into the eye; selecting, by the prediction engine, a subset of past IOL implantation records from a plurality of past IOL implantation records to evaluate a first plurality of candidate prediction models based on at least the one or more pre-operative measurements of the eye, each of the first plurality of candidate prediction models estimating a post-operative total spherical refractive equivalent (MRSE) based on a set of pre-operative eye measurements and an IOL power; and selecting, by the prediction engine, an estimated post-operative MRSE generated by each of the first plurality of candidate prediction models using eye measurement data in the selected subset of past IOL implantation records and a predicted post-operative MRSE based on the selected past IOL power. The method includes evaluating a first plurality of candidate prediction models based on a deviation between the predicted postoperative MRSE and the actual postoperative MRSE indicated in the subset of L implant records; selecting, by a prediction engine, a first prediction model from the first plurality of candidate prediction models based on the evaluation; calculating, by the prediction engine, a plurality of estimated postoperative MRSE values ​​based on a set of available IOL powers and one or more preoperative measurements of the eye using the selected first prediction model; identifying, by the prediction engine, from the set of available IOL powers, a first IOL power from the plurality of estimated postoperative MRSE values ​​that corresponds to the first estimated postoperative MRSE value that matches the predetermined postoperative MRSE value; and providing, by the prediction engine, the identified first IOL power to a user to assist in selecting an IOL to implant in the eye.

[0007] According to some embodiments, the prediction engine includes one or more processors for obtaining one or more preoperative measurements of an eye to perform intraocular lens (IOL) implantation into the eye, selecting a subset of past IOL implantation records from a plurality of past IOL implantation records to evaluate a first plurality of candidate prediction models based on at least the one or more preoperative measurements of the eye, each of the first plurality of candidate prediction models estimating a postoperative total refractive spherical equivalent (MRSE) based on a set of preoperative eye measurements and an IOL power, and comparing the estimated postoperative MRSE generated by each of the first plurality of candidate prediction models using the eye measurement data in the selected subset of past IOL implantation records with the selected subset of past IOL implantation records. the first plurality of candidate prediction models based on the deviation between the predicted postoperative MRSE and the actual postoperative MRSE value indicated in the prediction engine; selecting a first prediction model from the first plurality of candidate prediction models based on the evaluation; calculating a plurality of estimated postoperative MRSE values ​​based on the set of available IOL powers and one or more preoperative measurements of the eye using the selected first prediction model; identifying a first IOL power from the set of available IOL powers and the plurality of estimated postoperative MRSE values ​​that corresponds to the first estimated postoperative MRSE value that matches the predetermined postoperative MRSE value; and providing the identified first IOL power to a user by the prediction engine to assist in selecting an IOL to implant in the eye.

[0008] According to some embodiments, a non-transitory machine-readable medium includes a plurality of machine-readable instructions, the plurality of machine-readable instructions being configured, when executed by one or more processors, to cause the one or more processors to perform a method, the method including: obtaining one or more pre-operative measurements of an eye to perform intraocular lens (IOL) implantation into an eye; selecting a subset of past IOL implantation records from a plurality of past IOL implantation records to evaluate a first plurality of candidate prediction models based on at least the one or more pre-operative measurements of the eye, each of the first plurality of candidate prediction models estimating a post-operative total refractive spherical equivalent (MRSE) based on a set of pre-operative eye measurements and an IOL power; and ... evaluating the estimated post-operative MRSE generated by each of the first plurality of candidate prediction models using the eye measurement data in the selected subset of past IOL implantation records and the selected candidate prediction models. the first plurality of candidate predictive models based on a deviation between the actual postoperative MRSE value indicated in the subset of the first plurality of candidate predictive models; selecting a first predictive model from the first plurality of candidate predictive models based on the evaluation; calculating a plurality of estimated postoperative MRSE values ​​based on the set of available IOL powers and one or more preoperative measurements of the eye using the selected first predictive model; identifying a first IOL power from the set of available IOL powers and the plurality of estimated postoperative MRSE values ​​that corresponds to the first estimated postoperative MRSE value that matches the predetermined postoperative MRSE value; and providing the identified first IOL power to a user to assist in selecting an IOL to implant in the eye.

[0009] For a more complete understanding of the present technology, its features, and advantages, reference is made to the following description taken in conjunction with the accompanying drawings, in which: [Brief explanation of the drawings]

[0010] [Figure 1] FIG. 1 is a diagram of an IOL selection system according to some embodiments. [Figure 2A] FIG. 1 is a diagram of a method for creating a database for use in implementing an IOL according to some embodiments. [Figure 2B] 1A-1C are diagrams of a method for performing a two-stage process for implanting an IOL according to some embodiments. [Figure 3] 1 is a diagram of an eye and eye features according to some embodiments. [Figure 4A-4B] FIG. 1 is a diagram of a processing system according to some embodiments. [Figure 5] FIG. 1 is a diagram of a multi-layer neural network according to some embodiments. DETAILED DESCRIPTION OF THE INVENTION

[0011] In the figures, elements with the same name have the same or similar functions.

[0012] This description and the accompanying drawings illustrating aspects, embodiments, implementations, or modules of the present invention should not be construed as limiting - the claims define the proposed invention. Various mechanical, compositional, structural, electrical, and operational changes may be made without departing from the spirit and scope of this description and claims. In some instances, well-known circuits, structures, or techniques have not been shown or described in detail so as not to obscure the invention. Like reference numerals in two or more figures represent the same or similar elements.

[0013] In this description, specific details are set forth to describe some embodiments according to the present disclosure. Many specific details are set forth to provide a thorough understanding of the embodiments. However, it will be apparent to one skilled in the art that some embodiments may be practiced without some or all of these specific details. The specific embodiments disclosed herein are intended to be illustrative, not limiting. One skilled in the art may recognize other elements not specifically described herein that are within the scope and spirit of the present disclosure. Additionally, to avoid unnecessary repetition, one or more embodiments shown and described in connection with one embodiment can be incorporated into other embodiments, unless specifically stated otherwise, or if one or more features render the embodiment inoperable.

[0014] The technology described below includes systems and methods for improving a patient's post-implant vision by identifying an appropriate intraocular lens (IOL) power for implantation into the patient's eye based on a target (e.g., desired) post-operative total refractive spherical equivalent (MRSE). The systems and methods use a two-stage process for identifying an appropriate IOL power for a patient. During the first stage, data related to a current IOL implant, such as one or more preoperative measurements of the patient's eye, may be obtained. A subset of past IOL implant records related to previously performed IOL implants (also referred to as past IOL implants) that are most similar to the current IOL implant may be selected from a database that stores multiple past IOL implant records related to previously performed IOL implants (also referred to as past IOL implants). During the second stage, multiple predictive models for estimating post-operative MRSE may be evaluated to identify the predictive model with the smallest deviation based on the selected subset of past IOL implant records. The identified predictive model may be used to generate an estimated post-operative MRSE value based on a set of available IOL powers. An available IOL power corresponding to the estimated post-operative MRSE value that matches the target post-operative MRSE may be selected. An intraocular lens corresponding to the selected IOL power may then be used for implantation into the patient's eye.

[0015] FIG. 1 illustrates a system 100 for IOL selection according to some embodiments. The system 100 includes an IOL selection platform 105 coupled to one or more diagnostic training data sources 110 via a network 115. In some examples, the network 115 may include one or more switching devices, routers, local area networks (e.g., Ethernet), and / or wide area networks (e.g., the Internet), etc. Each diagnostic training data source 110 may be a database and / or data repository provided by an ophthalmic surgery clinic, an ophthalmology clinic, a medical school, an electronic medical record (EMR) repository, etc. Each diagnostic training data source 110 may provide the IOL selection platform 105 with one or more forms of training data, such as multidimensional images and / or measurements of a patient's pre-operative and post-operative eye, surgical planning data, a surgical console parameter log, surgeon identification data, diagnostic device data, patient demographic data, a surgical complication log, a patient medical history, patient demographic data, and / or information about an implanted IOL. The IOL selection platform 105 may store the training data in one or more databases 155, which may be configured to anonymize, encrypt, and / or otherwise protect the training data.

[0016] IOL selection platform 105 includes a prediction engine 120 that processes received training data, extracts eye measurements, performs raw data analysis on the training data, trains machine learning algorithms and / or models to estimate postoperative MRSE based on preoperative measurements, and iteratively refines the machine learning to optimize the various models used to predict postoperative MRSE for improved use with future patients to improve postoperative visual outcomes (e.g., better optical properties of the eye with the IOL implanted) (as described in more detail below). In some examples, prediction engine 120 may evaluate multiple trained models (e.g., one or more neural networks) to select one model that identifies the appropriate IOL power for the patient.

[0017] The IOL selection platform 105 is further coupled to one or more devices at the ophthalmology clinic 125 via a network 115. The one or more devices include a diagnostic device 130. The diagnostic device 130 is used to obtain one or more multidimensional images and / or other measurements of the eye of a patient 135. The diagnostic device 130 can be any of several devices that obtain multidimensional images and / or measurements of the ophthalmic anatomy, such as an optical coherence tomography (OCT) device, a rotating camera (e.g., a Scheimpflug camera), a magnetic resonance imaging (MRI) device, a keratometer, a curvature meter, and / or an optical biometer.

[0018] The ophthalmology clinic 125 may also include one or more computing devices 140 that acquire data related to the current IOL implantation. For example, the one or more computing devices 140 may acquire multidimensional images and / or measurements of the patient 135 from the diagnostic device 130. In some embodiments, the one or more computing devices 140 may also acquire device configuration data related to the diagnostic device 130. For example, the one or more computing devices 140 may acquire the model number of the diagnostic device 130 and the software release number of the software installed on the diagnostic device 130. In some embodiments, the one or more computing devices 140 may also acquire other information, such as the identity of the surgeon assigned to perform the IOL implantation of the patient 135, the ethnicity of the patient 135, the gender of the patient 135, the height of the patient 135, and the age of the patient 135, for example, through a user interface of the one or more computing devices 140. The one or more computing devices 140 may transmit one or more acquired multidimensional images and / or measurements of the patient 135, device configuration data associated with the diagnostic device 130, surgeon identification data, and demographic data of the patient 135 to the IOL selection platform 105. The one or more computing devices 140 may be one or more of a standalone computer, a tablet and / or other smart device, a surgical console, and / or a computing device integrated into the diagnostic device 130, etc.

[0019] The IOL selection platform 105 may receive biometric data of the patient 135 (e.g., measurements of the patient 135 and / or calculations from the measurements) and may use the prediction engine 120 to generate estimates of post-operative MRSE for various IOL types and IOL powers. The prediction engine may then be used to provide the ophthalmology clinic 125 and / or surgeon or other user with the estimated post-operative MRSE for various IOL types and IOL powers to aid in the selection of an IOL type and IOL power for the patient 135.

[0020] Diagnostic device 130 may further be used to obtain post-operative measurements of patient 135 after patient 135 undergoes cataract removal and IOL implantation using the selected IOL. One or more computing devices 140 may then transmit post-operative multidimensional images and / or measurements of patient 135 and the selected IOL to IOL selection platform 105, for use in iterative training and / or updating of models used by prediction engine 120 to incorporate information from patient 135 for use with future patients.

[0021] The estimated postoperative MRSE, the selected IOL, and / or the selected IOL power may be displayed on the computing device 140 and / or another computing device, display, and / or surgical console, etc. Additionally, the IOL selection platform 105 and / or the one or more computing devices 140 may identify various features of the patient's 135 anatomy in the measurements, as described in more detail below. Additionally, the IOL selection platform 105 and / or the one or more computing devices 140 may create graphical elements that identify, highlight, and / or otherwise indicate the patient's anatomy and / or the measured features. The IOL selection platform 105 and / or the one or more computing devices 140 may supplement the measurements with graphical elements.

[0022] In some embodiments, the IOL selection platform 105 may further include a surgical planner 150 that may be used to provide one or more surgical plans to the ophthalmology clinic 125 using the estimated postoperative MRSE, the selected IOL, and / or the selected IOL power.

[0023] In some embodiments, the system 100 may further include a standalone surgical planner 160 and / or the ophthalmology clinic 125 may further include a surgical planner module 170 on one or more computing devices 140 .

[0024] As discussed above and further emphasized herein, FIG. 1 is merely an example that should not unduly limit the scope of the claims. Those skilled in the art will recognize many variations, alternatives, and modifications. According to some embodiments, IOL selection platform 105 and / or one or more components of IOL selection platform 150, such as database 155, prediction engine 120, and / or surgical planner 150, may be integrated into one or more devices at ophthalmology clinic 125. In some examples, computing device 140 may host IOL selection platform 105, database 155, prediction engine 120, and / or surgical planner 150. In some examples, surgical planner 150 may be combined with surgical planner 170.

[0025] 2A is a diagram of a method 200 for selecting an IOL power and creating a database for use in implementing an IOL according to some embodiments. One or more of processes 202-208 of method 200 may be embodied, at least in part, in the form of executable code stored on a non-transitory, tangible, machine-readable medium that, when executed by one or more processors (e.g., processor(s) of prediction engine 120, IOL prediction platform, diagnostic device 130, one or more computing devices 140, and / or one or more of surgical planners 150, 160, and / or 170), can cause the one or more processors to perform one or more of processes 202-208. According to some embodiments, process 208 is optional and may be omitted.

[0026] In process 202, a database of past IOL implantation records is constructed. For example, prediction engine 120 may obtain training data from diagnostic training data source 110 and store the training data in database 155. The training data may correspond to previously performed IOL implantation cases (also referred to as "past IOL implants"). In some embodiments, prediction engine 120 may organize database 155 based on past IOL implantation cases, such that each record in database 155 contains training data associated with a particular past IOL implantation. The training data in each record may include both categorical and numerical data. Categorical data in each record may include information such as the IOL model number, the identity of the surgeon who performed the IOL implantation, the patient's gender, the patient's ethnicity, and device configuration data (e.g., model number, software release number, etc.) associated with the biometric equipment used to obtain the patient's eye measurement data. Numerical data in each record may include preoperative eye measurements (e.g., corneal power, axial length, corneal thickness, anterior chamber depth, corneal horizontal diameter, preoperative MRSE, etc.), the patient's age, and the patient's height. Additionally, each prior IOL implantation record may include the actual IOL power selected for the case and the resulting actual postoperative MRSE. Both the IOL power and the postoperative MRSE may include quantized numerical data and can be treated as numerical or categorical data by the prediction engine 120. In some examples, IOL powers are known to typically be in 0.25 diopters (d) or 0.5 d steps (nearest 0.25 or 0.5 diopters (d)), and postoperative MRSEs are known to typically be in 0.125 d steps (nearest 0.125 d).

[0027] In process 204, a principal component analysis (PCA) matrix and a mean vector are calculated and stored for numerical feature vector normalization. For example, the prediction engine 120 may normalize the numerical data in the past IOL implant records so that different numerical features may have proportional weights (e.g., the same weight as each other) when the past IOL implant records are compared to each other or to a current IOL implant. Different embodiments use different techniques for normalizing the numerical data. In some embodiments, the prediction engine 120 generates a numerical feature vector for each record and uses PCA to calculate a linear transform to decorrelate and normalize the numerical feature vector. The numerical feature vector of a past IOL implant record may include some or all of the numerical values ​​contained in the record. An example numerical feature vector of a record may be expressed using Equation 1:

number

[0028] Although the formula shown above includes only three numerical features, in some embodiments, more features or other combinations of features may be included in the numerical feature vector.

[0029] To perform PCA, we can calculate the mean of the numerical feature vector components using Equation 2.

number

[0030] The prediction engine 120 then assigns an element C j,k We can calculate the covariance matrix C having

number

[0031] Next, LL T = C. The matrix square root may be calculated using, for example, Cholesky decomposition. Finally, the prediction engine 120 may use numerical linear algebra to calculate the lower triangular inverse of the covariance square root matrix, M=L. -1 To apply the PCA transform to the numeric feature vector, the prediction engine 120 of some embodiments may subtract the calculated mean vector (e.g., m) from the input numeric feature vector and multiply the result by M, as shown in Equation 4. y=M(xm) (4) where y is the output normalized feature vector, M is the lower triangular inverse of the square root covariance matrix, x is the input unnormalized feature vector, and m is the mean vector.

[0032] In process 206, one or more hyperparameters of a two-stage process for identifying an appropriate IOL power for a patient may be identified. As described above, during the first stage of the two-stage process, a subset of past IOL implantation records closest to the current IOL implantation is selected from the database. Thus, prediction engine 120 may determine the number of records (e.g., 70, 100, 400, etc.) (also referred to as the "hyperparameter K") to be selected such that the subset of past IOL implantation records is limited to the determined number K. Furthermore, for predictive models implemented using deep neural networks, prediction engine 120 may also determine the number of hidden layers for those predictive models. Hyperparameters may be identified offline, prior to training (e.g., learning) the predictive model. In some embodiments, prediction engine 120 may identify (e.g., optimize) the hyperparameters by testing over a predetermined range of hyperparameters using a brute-force approach. For example, to determine the number of records to select (hyperparameter K) during the first stage of a two-stage process, prediction engine 120 may test each number within a range (e.g., 50-100, 20-200, etc.). To determine the number of hidden layers in the neural network, prediction engine 120 may test each number of hidden layers within a range (e.g., 1-5, 2-10, etc.).

[0033] In some embodiments, prediction engine 120 may perform such optimization using three data sets. For example, prediction engine 120 may divide the past IOL implantation records in database 155 into three sets—a training set, a validation set, and a test set. The distribution of the three sets may vary, but the training set may contain a larger proportion than the validation and test sets. An example distribution may include 60% of the past IOL implantation records as the training set, 20% of the remaining past IOL implantation records as the validation set, and the remaining 20% ​​of the past IOL implantation records as the test set. As an example, a past IOL implantation database of 20,000 records may be divided by prediction engine 120 into 12,000 training cases, 4,000 validation cases, and 4,000 test cases. With that collection of 20,000 records, prediction engine 120 may determine, using the techniques disclosed above, that the optimal number of records to select during the first stage of the two-stage process (hyperparameter K) is approximately 100. However, for a different database containing different prior IOL implantation records and / or a different number of prior IOL implantation records, the prediction engine 120 may determine a different optimal number of records to select (a different hyperparameter K).

[0034] In some embodiments, prediction engine 120 may perform an exhaustive search over the hyperparameters within the determined range. For example, for each hyperparameter value (e.g., for each number of records selected within the range, for each number of hidden layers within the range, etc.), prediction engine 120 may use the hyperparameter value to train a predictive model using a training set. The trained predictive model is then evaluated by prediction engine 120 using a validation set, during which validation results for the hyperparameter values ​​are recorded. After evaluating all hyperparameter values ​​within the range, an embodiment of prediction engine 120 may perform a final estimation of generalization ability using a test set. Under this brute-force approach, hyperparameter optimization can take a long time, especially if the number of hyperparameter values ​​to search is large. Therefore, prediction engine 120 may perform hyperparameter optimization offline before executing the two-stage process.

[0035] In process 208, a predictive model may optionally be trained based on the records in the database. In some embodiments, multiple predictive models may be available for use by prediction engine 120 during the second stage of the two-stage process. Different predictive models may use different prediction techniques (e.g., regression techniques). For example, predictive models available for use by prediction engine 120 may include models that use at least one of the following recursive techniques: multiple linear regression, multiple polynomial regression, K-nearest neighbors with radial basis functions, neural networks, and other suitable regression techniques.

[0036] In some embodiments, the prediction models available to the prediction engine 120 include two types of prediction models. The first type of prediction model (also referred to as an F(.) model) generates a theoretical IOL power value based on a set of numerical feature data (e.g., a numerical feature vector) and a target postoperative MRSE. The F(.) model may be expressed according to Equation 5: IOL=F(x,T) (5) where x is the input vector of numerical features, T is the target post-operative MRSE, and IOL is the theoretical IOL power value that results in the target post-operative target refractive index T based on the patient's biometric data contained in the input vector x.

[0037] A second type of model (also called a G(.) model) estimates (e.g., predicts) a patient's postoperative MRSE based on a set of numerical features (e.g., a numerical feature vector) associated with the patient and a selected IOL power. The G(.) model may be expressed according to Equation 6: R x =G(x,SIOL) (6) where x is the input vector of numerical features, SIOL is the selected IOL power, and R x is the estimated postoperative MRSE.

[0038] In some embodiments, prediction engine 120 may train a predictive model offline, prior to performing the two-stage process, that can be used during the second stage of the two-stage process using the previous IOL implantation records stored in database 155. For example, prediction engine 120 may train both types of predictive models (F(.) model and G(.) model) using the previous IOL implantation records stored in database 155.

[0039] 2B is a diagram of method 210 for performing a two-stage process for implanting an IOL in a patient according to some embodiments. One or more of processes 212-230 of method 210 may be embodied, at least in part, in the form of executable code stored on a non-transitory, tangible, machine-readable medium that, when executed by one or more processors (e.g., processor(s) of prediction engine 120, IOL prediction platform, diagnostic device 130, one or more computing devices 140, and / or one or more of surgical planners 150, 160, and / or 170), can cause the one or more processors to perform one or more of processes 212-230. According to some embodiments, processes 228 and / or 230 are optional and may be omitted.

[0040] In process 212, data related to the current IOL implantation case may be received. For example, prediction engine 120 may obtain one or more pre-operative measurements of the patient's eye from diagnostic device 130. In some embodiments, one or more of the pre-operative eye measurements may be extracted from one or more pre-operative images of the patient's eye 135 obtained using a diagnostic device, such as diagnostic device 130, an OCT device, a rotational (e.g., Scheimpflug) camera, and / or an MRI device. In some examples, one or more of the pre-operative measurements may be determined using one or more measurement devices, such as diagnostic device 130, a keratometer, a keratometer, and / or an optical biometer. Process 210 is described in the context of FIG. 3, which is a diagram of an eye 300 and eye features according to some embodiments. As shown in FIG. 3, eye 300 includes a cornea 310, an anterior chamber 320, and a lens 330.

[0041] In some embodiments, one measurement of interest for the eye 300 is the lateral diameter of the cornea 310. In some examples, the lateral diameter of the cornea 310 may be measured using an optical biometer. In some examples, the lateral diameter of the cornea 310 may be determined by analyzing one or more images of the eye 300, for example, by measuring the distance between the sclera of the eye.

[0042] In some embodiments, one measurement of interest for the eye 300 is the mean curvature or roundness of the anterior surface of the cornea 310. In some examples, the mean curvature of the cornea 310 may be measured using one or more images of the eye 300 and / or a keratometer, etc. In some examples, the mean curvature of the cornea 310 may be based on an average of steep and shallow keratometry measurements of the cornea 310. In some examples, the mean curvature of the cornea 310 may be expressed as the radius of curvature (rc) of the cornea 310, which is 337.5 divided by the mean corneal curvature.

[0043] In some embodiments, one measurement of interest of the eye 300 is the axial length 370 of the eye 300, measured from the anterior surface of the cornea 310 to the retina along the central axis 380 of the eye 300. In some examples, the axial length 370 may be determined using one or more images of the eye 300 and / or biometry of the eye.

[0044] In some embodiments, one measurement of interest of the eye 300 is the pre-operative anterior chamber depth (ACD) 390 of the eye, which corresponds to the distance between the posterior surface of the cornea 310 and the anterior surface of the pre-operative lens 330. In some examples, the axial length 370 may be determined using one or more images of the eye 300 and / or biometry of the eye.

[0045] In some embodiments, one measurement of interest of the eye 300 is the corneal thickness 315 of the eye 300, measured from the posterior surface of the cornea 310 along the central axis 380 of the eye 300 to the anterior surface of the cornea 310. In some examples, the corneal thickness 315 may be determined using one or more images of the eye 300 and / or biometry of the eye.

[0046] In some embodiments, in addition to obtaining measurements of the patient's 135 eye, other status data related to the current IOL implant may also be obtained. For example, one or more computing devices 140 may obtain, from the diagnostic device 130, device configuration data related to the diagnostic device 130 that provided the measurements of the patient's 135 eye. The device configuration data may include the model number of the diagnostic device 130 and the software release number of the software installed on the diagnostic device 130. In some examples, the one or more computing devices 140 may obtain the configuration data by communicating with the diagnostic device 130 using an application programming interface (API) of the diagnostic device 130. Furthermore, the one or more computing devices 140 may also obtain other information related to the patient's 135 current IOL implant, such as the identity of the surgeon assigned to perform the IOL implantation of the patient 135, the ethnicity of the patient 135, the patient's gender, the patient's height, and the patient's age. In some embodiments, the one or more computing devices 140 may provide a user interface that allows a user to provide the surgeon identity and patient demographic data. The one or more computing devices 140 may then transmit the biometric and environmental data to the IOL selection platform 105.

[0047] Referring again to FIG. 2 , process 214 may select a subset of past IOL implant records (e.g., K-nearest neighbor records) based on the retrieved data. As described above, database 155 of IOL selection platform 105 may store training data retrieved from diagnostic training data source 110. The training data may include data related to IOL implants previously performed on other patients (e.g., past IOL implants). In particular, each record in database 155 may include numerical and categorical data related to a particular past IOL implant. The numerical data in each record may include preoperative eye measurements (e.g., corneal power, axial length, corneal thickness, anterior chamber depth, corneal diameter, preoperative MRSE, etc.), patient age, and / or patient height, etc. In some embodiments, the numerical data stored in each record may be normalized using the techniques described above with respect to process 204 and stored as a numerical feature vector y. The classification data within each record may include information such as the IOL model number, the identity of the surgeon who performed the IOL implantation, the patient's gender, the patient's ethnicity, and / or device configuration data (e.g., model number and software release number, etc.) associated with the biometric equipment used to obtain the patient's eye measurement data. Additionally, each prior IOL implantation record may include the actual IOL power selected for the case and the resulting actual postoperative MRSE.

[0048] In some embodiments, prediction engine 120 may select a subset of past IOL implant records from database 155 that are most similar to patient 135's current IOL implant based on the acquired biometric and / or environmental data. The selected subset of past IOL implant records may include (and be limited to) a number of records corresponding to a predetermined hyperparameter K (e.g., identified in process 206). In some embodiments, prediction engine 120 may use a K-nearest neighbor (KNN) algorithm to select the subset of records from database 155 that are closest to the current IOL implant using the predetermined hyperparameter K as the K parameter of the KNN algorithm. For example, prediction engine 120 may first generate a numerical feature vector for the current IOL implant based on some or all of the numerical data (e.g., patient 135's preoperative eye measurements, patient 135's age, patient 135's height, etc.). The numerical features included in the generated numerical feature vector for the current IOL implant may correspond to features included in a previously generated numerical feature vector associated with the past IOL implant records stored in database 155. In some examples, the numerical feature vector may include the preoperative corneal power of the eye of the patient 135, the axial length of the eye of the patient 135, and the corneal thickness of the eye of the patient 135, such as shown in Equation 1. The prediction engine 120 may then normalize the components in the numerical feature vector generated for the current IOL implant using the techniques described above, for example, using Equation 4.

[0049] The normalized vector may be used to calculate the Euclidean distance between the current IOL implant and each past IOL implant record in database 155. i and b i The Euclidean distance between vectors a and b, each having

number

[0050] Next, based on the calculated Euclidean distance, the prediction engine 120 may select a predetermined number of past IOL implantation records (e.g., K records) with the smallest distance values ​​to use in the second stage of the two-stage process.

[0051] After selecting the subset of past IOL implantation records, process 216 may evaluate multiple candidate prediction models based on the selected subset of past IOL implantation records. As described above, multiple prediction models (including the F(.) model and the G(.) model) may be available for use by prediction engine 120 during the second stage of the two-stage process. Prediction engine 120 may use these prediction models as candidate models to select one or more models to use during the second stage of the two-stage process based on performance. If prediction models were not trained offline (e.g., if process 208 was not performed), prediction engine 120 may train available prediction models using the past IOL implantation records in database 155 at this time. In some embodiments, prediction engine 120 may train a prediction model using only the selected subset of past IOL implantation records.

[0052] In some embodiments, prediction engine 120 may evaluate the F(.) models by using each F(.) model to generate a theoretical IOL power based on numerical feature data (e.g., a numerical feature vector) associated with a selected subset of past IOL implantation records. Prediction engine 120 may then generate a performance score for each F(.) model based on the deviation between the theoretical IOL power generated by the F(.) model and the actual IOL power used as indicated in the subset of past IOL implantation records. For example, the performance score for each F(.) model may be generated based on at least one of the mean error, the standard deviation error, the mean absolute error (MAE), the standard deviation of the absolute error, the percentage with an MAE less than a predetermined value (e.g., 0.5, 1.0, etc.), and the percentage of estimates within a range with a predetermined confidence (e.g., 95%).

[0053] In some embodiments, prediction engine 120 may also evaluate each G(.) model by using it to estimate (e.g., predict) a postoperative MRSE value based on the numerical feature data (e.g., numerical feature vector) and the actual IOL power used in a selected subset of prior IOL implantation records. Prediction engine 120 may then generate a performance score for each G(.) model based on the deviation between the estimated postoperative MRSE value generated by the G(.) model and the actual postoperative MRSE value exhibited in the subset of prior IOL implantation records. For example, the estimated error (e.g., deviation) between the estimated postoperative MRSE generated by the G(.) model and the actual postoperative MRSE exhibited in a given prior IOL implantation record n ​​may be expressed according to Equation 8: e[n]=ARx[n]-Rx[n] (8) where n is the record index of the past IOL implantation record from the selected subset of past IOL implantation records, ARx[n] is the actual postoperative MRSE shown in the past IOL implantation record with record index n, Rx[n] is the postoperative MRSE estimated by the G(.) model, and e[n] is the estimated error for record n.

[0054] In some embodiments, a performance score for each G(.) model may be generated based on at least one of the mean error, the standard deviation error, the mean absolute error (MAE), the standard deviation of the absolute error, the percentage with an MAE less than a predetermined value (e.g., 0.5, 1.0, etc.), and the percentage of estimates in the range with a predetermined confidence (e.g., 95%). For example, based on Equation 8 above, the performance score for a given G(.) model based on the estimated error standard deviation may be expressed according to Equation 9:

number

[0055] In process 218, the F(.) model and the G(.) model may be selected from the candidate models. For example, prediction engine 120 may select the F(.) model and the G(.) model from the candidate models based on performance scores generated for the candidate models. In some embodiments, prediction engine 120 may select the F(.) model with the best performance score (e.g., exhibiting the lowest standard deviation) from the candidate F(.) models, and may select the G(.) model with the best performance score (e.g., exhibiting the lowest standard deviation) from the candidate G(.) models.

[0056] In process 220, optionally, a theoretical IOL power for a target (e.g., desired) post-operative MRSE for patient 135 with the current IOL implantation may be calculated. For example, the prediction engine may generate the theoretical IOL power using the F(.) model selected in process 218 based on the numerical feature data associated with patient 135 and a predetermined target post-operative MRSE for patient 135. The target post-operative MRSE may be determined by the surgeon assigned to perform the current IOL implantation for patient 135. The generated IOL power represents the IOL power required for an intraocular lens implanted in the eye of patient 135 to achieve the post-operative MRSE for patient 135 after implantation.

[0057] In process 222, a selection table is calculated using the selected G(.) model based on a set of available IOL powers that are close to the theoretical IOL power. The set of IOL powers may be provided to the surgeon or user currently performing IOL implantation for patient 135. For example, intraocular lenses may be manufactured with a variety of but limited IOL powers. In one example, an intraocular lens manufacturer may manufacture lenses only according to a limited number of IOL powers. If a theoretical IOL power for the current IOL implantation is not generated (e.g., process 220 was not performed), prediction engine 120 may calculate an estimated postoperative MRSE value using the selected G(.) model (selected in process 218) based on the numerical feature data associated with patient 135 for all IOL powers currently available for IOL implantation. The result may include a table containing available IOL powers and the corresponding estimated postoperative MRSE value calculated by the selected G(.) model for patient 135.

[0058] Because the number of IOL powers currently available for IOL implantation may be large, in some embodiments, prediction engine 120 may reduce calculations by estimating the postoperative MRSE for only a subset of the available IOL powers (thereby improving the execution speed of this two-step process). For example, by performing process 220, a theoretical IOL power for patient 135 is generated using a selected F(.) model. However, the theoretical powers generated by the F(.) model may not be available in the set of powers provided by the manufacturer of a given IOL model; therefore, the surgeon or user must select from the set of available IOL powers. Therefore, prediction engine 120 may select a subset of available IOL powers within a predetermined threshold from the theoretical IOL powers generated by the F(.) model in order to calculate a selection table in process 222.

[0059] In some examples, if the generated theoretical IOL power is 20d, prediction engine 120 may select available IOL powers that are within the range of 18d and 22d. Prediction engine 120 may then use the selected G(.) model (selected in process 218) to calculate estimated post-operative MRSE values ​​based on the numerical feature data associated with patient 135 with only the subset of available IOL powers selected by prediction engine 120 based on the theoretical IOL powers generated by the F(.) model. The results may include a table that includes the subset of available IOL powers and the corresponding estimated post-operative MRSE values ​​calculated with the selected G(.) model for patient 135.

[0060] In some embodiments, in addition to calculating estimated post-operative MRSE values ​​for a subset of available IOL powers, prediction engine 120 may also generate predicted MRSE ranges (e.g., limits) with a predetermined confidence (e.g., 90% confidence, 95% confidence, 98% confidence, etc.) for each estimated MRSE value. The predicted MRSE ranges may be generated based on the standard deviation calculated for the selected G(.) model using Equation 9 above. For example, a predicted MRSE range with 95% confidence for a particular estimated post-operative MRSE Rx may be expressed according to Equation 10: Predicted MRSE range = Rx ± 1.96 SD (10)

[0061] The predicted MRSE ranges may also be included in a table, which the prediction engine 120 and / or surgical planner 150 may transmit over the network 115 to one or more computing devices 140 at the ophthalmology clinic 125 for display on the one or more computing devices 140 to assist the surgeon or user in implanting the IOL in the patient 135.

[0062] In process 224, available IOL powers corresponding to the target post-operative MRSE may be identified. For example, in addition to transmitting a table including available IOL powers and estimated post-operative MRSE values ​​to one or more computing devices 140, prediction engine 120 may also identify a particular available IOL power for use with patient 135's current IOL implantation based on the post-operative MRSE value and / or the predicted MRSE range. In some embodiments, prediction engine 120 may identify (e.g., select) a particular available IOL power having a corresponding estimated post-operative MRSE value that is closest to patient 135's target post-operative MRSE. In some embodiments, prediction engine 120 may identify (e.g., select) a particular available IOL power having a corresponding maximum and minimum MRSE value of the predicted MRSE range that is closest to patient 135's target post-operative MRSE (e.g., smallest deviation from the target post-operative MRSE value, where the deviation is the sum of the difference between the maximum MRSE value and the target post-operative MRSE value and the difference between the minimum MRSE value and the target post-operative MRSE value). The prediction engine 120 and / or surgical planner 150 may transmit the identified specific available IOL powers via the network 115 to one or more computing devices 140 at the ophthalmology clinic 125 for display on the one or more computing devices 140 to assist the surgeon or user in implanting the IOL in the patient 135.

[0063] In process 226, an intraocular lens having the identified available IOL power is implanted into the patient 135. In some examples, an intraocular lens having the IOL power identified during process 224 is implanted into the eye of the patient 135 by a surgeon.

[0064] In process 228, one or more post-operative measurements are obtained of the eye of patient 135. In some examples, the one or more post-operative measurements may include an actual post-operative ACD of the IOL after IOL implantation, an actual post-operative MRSE after IOL implantation, and / or an actual post-operative zone of emmetropia measurement, etc. In some examples, the actual post-operative ACD and / or actual post-operative MRSE may be determined based on one or more images of the post-operative eye and / or one or more physiological and / or optical measurements of the post-operative eye, etc.

[0065] In process 230, the prediction models available to prediction engine 120 are updated. In some examples, one or more preoperative measurements identified during process 212, actual postoperative ACD, and / or actual postoperative MRSE identified during process 228, etc., may be added to database 155 as new past IOL implantation records and used as additional training data for subsequent training of any prediction models. In some examples, the updates may include one or more of least squares approximations and / or feedback to a neural network (e.g., using backpropagation), etc. In some examples, one or more of the G(.) models may be trained using one or more loss functions based on their ability to accurately predict the postoperative MRSE of various IOL candidates.

[0066] 4A and 4B are diagrams of a processing system according to some embodiments. While two embodiments are shown in FIGS. 4A and 4B, one skilled in the art will readily appreciate that other system embodiments are also possible. According to some embodiments, the processing systems of FIGS. 4A and / or 4B are representative of computing systems that may be included in one or more of the IOL selection platform 105, the ophthalmology clinic 125, the prediction engine 120, the diagnostic device 130, the one or more computing devices 140, and / or any of the surgical planners 150, 160, and / or 170.

[0067] 4A shows a computing system 400, whose components are in electrical communication with one another using a bus 405. The system 400 includes a processor 410 and a system bus 405 that couples various system components to the processor 410, including memory in the form of read-only memory (ROM) 420 and / or random access memory (RAM) 425 (e.g., PROM, EPROM, flash EPROM, and / or any other memory chip or cartridge). The system 400 may further include a cache 412 of high-speed memory that is directly associated with, adjacent to, or integrated as part of the processor 410. The system 400 may access data stored in the ROM 420, the RAM 425, and / or one or more storage devices 430 through the cache 412 for fast access by the processor 410. In some examples, cache 412 may provide a performance boost that avoids delays by processor 410 in accessing data from memory 415, ROM 420, RAM 425, and / or one or more storage devices 430 that was previously stored in cache 412. In some examples, one or more storage devices 430 store one or more software modules (e.g., software modules 432, 434, 436, etc.). Software modules 432, 434, and / or 436 may control and / or be configured to control processor 410 to perform various operations, such as the processes of methods 200 and / or 210. And, while system 400 is shown with only one processor 410, it is understood that processor 410 may be representative of one or more central processing units (CPUs), multi-core processors, microprocessors, microcontrollers, digital signal processors (DSPs), field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), graphics processing units (GPUs), and / or tensor processing units (TPUs), etc. In some examples, system 400 may be implemented as a standalone subsystem and / or as a board added to a computing device or as a virtual machine.

[0068] To enable a user to interact with system 400, system 400 includes one or more communication interfaces 440 and / or one or more input / output (I / O) devices 445. In some examples, one or more communication interfaces 440 may include one or more network interfaces and / or network interface cards, etc., that provide communications according to one or more network standards and / or communication bus standards. In some examples, one or more communication interfaces 440 may include an interface for communicating with system 400 over a network, such as network 115. In some examples, one or more I / O devices 445 may include one or more user interface devices (e.g., a keyboard, a pointing / selection device (e.g., a mouse, touchpad, scroll wheel, trackball, and / or touchscreen, etc.), an audio device (e.g., a microphone and / or speaker), a sensor, an actuator, and / or a display, etc.).

[0069] Each of the one or more storage devices 430 may include non-transitory, non-volatile storage such as provided by a hard disk, optical media, and / or solid-state drive, etc. In some examples, each of the one or more storage devices 430 may be co-located with the system 400 (e.g., local storage) and / or remote from the system 400 (e.g., cloud storage).

[0070] 4B illustrates a computing system 450 based on a chipset architecture that may be used to perform any of the methods described herein (e.g., methods 200 and / or 210). System 450 may include processor 455, which represents any number of physically and / or logically distinct resources capable of performing software, firmware, and / or other computations, such as one or more CPUs, multi-core processors, microprocessors, microcontrollers, DSPs, FPGAs, ASICs, GPUs, and / or TPUs. As shown, processor 455 is backed by one or more chipsets 460, which may also include one or more CPUs, multi-core processors, microprocessors, microcontrollers, DSPs, FPGAs, ASICs, GPUs, TPUs, coprocessors, and / or coder-decoders (CODECs), etc. As shown, one or more chipsets 460 interface processor 455 with one or more I / O devices 465, one or more storage devices 470, memory 475, bridge 480, and / or one or more communication interfaces 490. In some examples, one or more I / O devices 465, one or more storage devices 470, memory, and / or one or more communication interfaces 490 may correspond to similarly named counterpart elements in FIG. 4A and system 400.

[0071] In some examples, bridge 480 may provide additional interfaces that provide system 450 with access to one or more user interface (UI) components, such as one or more keyboards, pointing / selection devices (e.g., a mouse, touchpad, scroll wheel, trackball, and / or touchscreen, etc.), audio devices (e.g., a microphone and / or speaker), and / or display devices.

[0072] According to some embodiments, systems 400 and / or 460 may provide a graphical user interface (GUI) suitable for assisting a user (e.g., a surgeon and / or other medical personnel) in performing the processes of methods 200 and / or 210. The GUI may include instructions regarding the next action to be performed, annotated and / or unannotated diagrams of the anatomical structure, such as pre-operative and / or post-operative images of the eye (e.g., such as shown in FIG. 3), and / or input prompts, etc. In some examples, the GUI may display true color and / or false color images of the anatomical structure, etc.

[0073] FIG. 5 is a diagram of a multi-layer neural network 500 according to some embodiments. In some embodiments, neural network 500 may represent a neural network used to implement at least some of the predictive models (including the F(.) model and the G(.) model) used by prediction engine 120. Neural network 500 processes input data 510 using an input layer 520. In some examples, input data 510 may correspond to input data provided to one or more models and / or training data provided to one or more models during processes 208 and 230 used to train one or more models. Input layer 520 includes multiple neurons used to condition input data 510, such as by scaling and / or range limiting. Each neuron in input layer 520 generates an output that is fed to an input of hidden layer 531. Hidden layer 531 includes multiple neurons that process the output from input layer 520. In some examples, each neuron in hidden layer 531 generates an output that then propagates through one or more additional hidden layers, terminating in hidden layer 539. Hidden layer 539 includes multiple neurons that process the output from the previous hidden layer. The output of hidden layer 539 is provided to output layer 540. Output layer 540 includes one or more neurons that are used to condition the output from hidden layer 539, such as by scaling and / or range limiting. It should be understood that the architecture of neural network 500 is merely representative, and that other architectures are possible, including neural networks with only one hidden layer, neural networks without input and / or output layers, and / or neural networks with recurrent layers.

[0074] In some examples, the input layer 520, the hidden layers 531-539, and / or the output layer 540 each include one or more neurons. In some examples, the input layer 520, the hidden layers 531-539, and / or the output layer 540 each may include the same or different numbers of neurons. In some examples, each neuron takes a combination (e.g., a weighted sum using a trainable weighting matrix W) of inputs x, adds an optional trainable bias b, and applies an activation function f to generate an output, as shown in Equation 11. In some examples, the activation function f may be a linear activation function, an activation function with upper and / or lower bounds, a log-sigmoid function, a hyperbolic tangent function, a normalized linear function, etc. In some examples, each neuron may have the same or different activation functions. a=f(Wx+b) (11)

[0075] In some examples, neural network 500 may be trained using supervised learning (e.g., during processes 208 and 230) using training data combinations including combinations of input data and ground truth (e.g., expected) output data. The output of neural network 500 is generated using input data 510, and output data 550 generated by neural network 500 is compared to the ground truth output data. Differences between the generated output data 550 and the ground truth output data may then be fed back to neural network 500 to correct various trainable weights and biases. In some examples, the differences may be fed back using backpropagation, such as stochastic gradient descent. In some examples, a large set of training data combinations may be presented to neural network 500 multiple times until an overall loss function (e.g., mean squared error based on the difference between each training combination) converges to an acceptable level.

[0076] Methods according to the above-described embodiments may be implemented as executable instructions stored on a non-transitory, tangible, machine-readable medium. The executable instructions, when executed by one or more processors (e.g., processor 410 and / or processor 455), may cause the one or more processors to perform one or more of the processes of methods 200 and / or 210. Some common forms of machine-readable media that may contain the processes of methods 200 and / or 210 are, for example, floppy disks, flexible disks, hard disks, magnetic tape, any other magnetic media, CD-ROMs, any other optical media, punch cards, paper tape, any other physical media with a pattern of holes, RAM, PROM, EPROM, Flash EPROM, any other memory chip or cartridge, and / or any other medium configured to be read by a processor or computer.

[0077] Devices implementing methods according to these disclosures may include hardware, firmware, and / or software and may take any of a variety of form factors. Typical examples of such form factors include laptops, smartphones, small form factor personal computers, and / or personal digital assistants. Portions of the functionality described herein may also be implemented in peripheral devices and / or add-in cards. Such functionality may also be implemented on a circuit board between different chips or different processes running on a single device, as further examples.

[0078] While exemplary embodiments have been shown and described, a wide range of modifications, changes, and substitutions are contemplated in the above disclosure, and in some cases, some features of the embodiments may be utilized without the corresponding use of other features. Those skilled in the art will recognize many variations, substitutions, and modifications. Accordingly, the scope of the present invention is to be limited only by the following claims, and it is appropriate that such claims be construed broadly and consistently with the scope of the embodiments disclosed herein.

Claims

1. 1. A prediction engine, comprising: one or more processors The prediction engine comprises: obtaining one or more pre-operative measurements of the eye that assist a user in performing an intraocular lens (IOL) implantation into the eye; selecting a subset of past IOL implantation records from a plurality of past IOL implantation records to evaluate a first plurality of candidate prediction models based on at least the one or more preoperative measurements of the eye, wherein each of the first plurality of candidate prediction models estimates a postoperative total spherical refractive equivalent (MRSE) based on a set of preoperative eye measurements and IOL power; and evaluating the first plurality of prediction models based on a deviation between an estimated postoperative MRSE generated by each of the first plurality of candidate prediction models using the eye measurement data in the selected subset of past IOL implantation records and an actual postoperative MRSE indicated in the selected subset of past IOL implantation records; selecting a first predictive model from the first plurality of candidate predictive models based on the evaluation; and calculating a plurality of estimated post-operative MRSE values ​​based on a set of available IOL powers and the one or more pre-operative measurements of the eye using the selected first prediction model; identifying a first IOL power from the set of available IOL powers and the plurality of estimated postoperative MRSE values ​​that corresponds to a first estimated postoperative MRSE value that matches a predetermined postoperative MRSE value; providing the user with the first IOL power identified by the prediction engine to assist in selecting an IOL to implant in the eye; configured to: the subset of past IOL implantation records is limited to a specific number of records, and the prediction engine is further configured to determine the specific number based on an offline analysis of the plurality of past IOL implantation records; the subset of past IOL implantation records that are closest to the current IOL implantation into the eye is selected from the plurality of past IOL implantation records using a K-nearest neighbor (KNN) algorithm, a parameter in the KNN algorithm corresponding to the specified number; Selecting from the plurality of previous IOL implantation records using the K-nearest neighbor (KNN) algorithm includes: generating a numerical feature vector including a preoperative corneal power of the eye, an axial length of the eye, and a corneal thickness of the eye based on the numerical data of the past IOL implantation record; normalizing the numerical feature vector and using the normalized numerical feature vector to calculate a Euclidean distance between the current IOL implantation and each of the past IOL implantation records; selecting the K previous IOL implantation records based on the calculated Euclidean distances; Including, a prediction engine.

2. the subset of IOL implant records is selected for evaluating a second plurality of candidate prediction models, each of the second plurality of candidate prediction models estimating an IOL power based on a set of preoperative eye measurements and a desired postoperative MRSE, and the prediction engine further comprises: evaluating the second plurality of candidate prediction models based on a deviation between an estimated IOL power generated by each of the second plurality of candidate prediction models using the eye measurement data in the selected subset of past IOL implantation records and an actual IOL power indicated in the selected subset of past IOL implantation records; selecting a second predictive model from the second plurality of candidate predictive models based on the evaluation; and calculating a first IOL power using the second prediction model based on the predetermined post-operative MRSE value and the one or more measurements of the eye; The prediction engine of claim 1 configured to:

3. 3. The prediction engine of claim 2, wherein the prediction engine is further configured to identify the set of IOL powers from a plurality of IOL powers based on the first IOL power, the set of IOL powers being within a predetermined threshold from the first IOL power.

4. 10. A non-transitory machine-readable medium comprising a plurality of machine-readable instructions, the plurality of machine-readable instructions being configured, when executed by one or more processors of a prediction engine according to claim 1 to claim 3, to cause the one or more processors to perform a method, the method comprising: obtaining one or more pre-operative measurements of the eye that assist a user in performing an intraocular lens (IOL) implantation into the eye; selecting a subset of past IOL implantation records from a plurality of past IOL implantation records to evaluate a first plurality of candidate prediction models based on at least the one or more preoperative measurements of the eye, wherein each of the first plurality of candidate prediction models estimates a postoperative total spherical refractive equivalent (MRSE) based on a set of preoperative eye measurements and IOL power; and evaluating the first plurality of candidate prediction models based on a deviation between an estimated postoperative MRSE generated by each of the first plurality of candidate prediction models using the eye measurement data in the selected subset of past IOL implantation records and an actual postoperative MRSE indicated in the selected subset of past IOL implantation records; selecting a first predictive model from the first plurality of candidate predictive models based on the evaluation; and calculating a plurality of estimated post-operative MRSE values ​​based on a set of available IOL powers and the one or more pre-operative measurements of the eye using the selected first prediction model; identifying a first IOL power from the set of available IOL powers and the plurality of estimated postoperative MRSE values ​​that corresponds to a first estimated postoperative MRSE value that matches a predetermined postoperative MRSE value; providing the identified first IOL power to the user to assist in selecting an IOL to implant in the eye; a non-transitory machine-readable medium, including

5. a subset of the IOL implant records is selected for evaluating a second plurality of candidate predictive models, each of the second plurality of candidate predictive models estimating an IOL power based on a set of preoperative eye measurements and a desired postoperative MRSE; evaluating the second plurality of candidate prediction models based on a deviation between an estimated IOL power generated by each of the second plurality of candidate prediction models using the eye measurement data in the selected subset of past IOL implantation records and an actual IOL power indicated in the selected subset of past IOL implantation records; selecting a second predictive model from the second plurality of candidate predictive models based on the evaluation; and calculating a first IOL power using the second prediction model based on the predetermined post-operative MRSE value and the one or more measurements of the eye; The non-transitory machine-readable medium of claim 4 further comprising:

6. 6. The non-transitory machine-readable medium of claim 5, further comprising identifying the set of IOL powers from a plurality of IOL powers based on the first IOL power, the set of IOL powers being within a predetermined threshold from the first IOL power.

7. 5. The non-transitory machine-readable medium of claim 4, wherein the one or more pre-operative measurements of the eye include at least one of corneal power, axial length, corneal thickness, anterior chamber depth, corneal diameter, or total refractive spherical equivalent.

8. The method comprises: Identifying a deviation value for the first prediction model based on a deviation between an estimated postoperative MRSE generated by the first prediction model using eye measurement data in the selected subset of past IOL implantation records and an actual postoperative MRSE indicated in the selected subset of past IOL implantation records; determining, for each estimated postoperative MRSE value, a predicted MRSE range based on the deviation value, wherein the first IOL power is determined further based on the predicted MRSE range; The non-transitory machine-readable medium of claim 4 further comprising:

9. obtaining, by one or more computing devices implementing a prediction engine, one or more pre-operative measurements of the eye that assist a user in performing an intraocular lens (IOL) implantation into the eye; selecting, by the prediction engine, a subset of past IOL implantation records from a plurality of past IOL implantation records to evaluate a first plurality of candidate prediction models based on at least the one or more preoperative measurements of the eye, wherein each of the first plurality of candidate prediction models estimates a postoperative total spherical refractive equivalent (MRSE) based on a set of preoperative eye measurements and IOL power; and evaluating, by the prediction engine, the first plurality of prediction models based on a deviation between an estimated postoperative MRSE generated by each of the first plurality of candidate prediction models using the eye measurement data in the selected subset of past IOL implantation records and an actual postoperative MRSE indicated in the selected subset of past IOL implantation records; selecting, by the prediction engine, a first predictive model from the first plurality of candidate predictive models based on the evaluation; calculating, by the prediction engine, a plurality of estimated post-operative MRSE values ​​based on a set of available IOL powers and the one or more pre-operative measurements of the eye using the selected first prediction model; identifying, by the prediction engine, from the set of available IOL powers and the plurality of estimated postoperative MRSE values, a first IOL power corresponding to a first estimated postoperative MRSE value that matches a predetermined postoperative MRSE value; providing the first IOL power identified by the prediction engine to the user to assist in selecting an IOL to be implanted in the eye; and Including, The subset of past IOL implantation records is limited to a specific number of records, and the method further includes determining the specific number based on an offline analysis of the plurality of past IOL implantation records, and the subset of past IOL implantation records that are closest to a current IOL implantation into the eye are selected from the plurality of past IOL implantation records using a K-nearest neighbor (KNN) algorithm, and parameters in the KNN algorithm correspond to the specific number; Selecting from the plurality of previous IOL implantation records using the K-nearest neighbor (KNN) algorithm includes: generating a numerical feature vector including a preoperative corneal power of the eye, an axial length of the eye, and a corneal thickness of the eye based on the numerical data of the past IOL implantation record; normalizing the numerical feature vector and using the normalized numerical feature vector to calculate a Euclidean distance between the current IOL implantation and each of the past IOL implantation records; selecting the K previous IOL implantation records based on the calculated Euclidean distances; A method comprising:

10. a subset of the IOL implant records is selected for evaluating a second plurality of candidate predictive models, each of the second plurality of candidate predictive models estimating an IOL power based on a set of preoperative eye measurements and a desired postoperative MRSE; evaluating the second plurality of candidate prediction models based on a deviation between an estimated IOL power generated by each of the second plurality of candidate prediction models using the eye measurement data in the selected subset of past IOL implantation records and an actual IOL power indicated in the selected subset of past IOL implantation records; selecting a second predictive model from the second plurality of candidate predictive models based on the evaluation; and calculating a first IOL power using the second prediction model based on the predetermined post-operative MRSE value and the one or more measurements of the eye; The method of claim 9 further comprising:

11. 10. The method of claim 9, wherein the one or more pre-operative measurements of the eye include at least one of corneal power, axial length, corneal thickness, anterior chamber depth, corneal diameter, or total refractive spherical equivalent.

12. Identifying a deviation value for the first prediction model based on a deviation between an estimated postoperative MRSE generated by the first prediction model using eye measurement data in the selected subset of past IOL implantation records and an actual postoperative MRSE indicated in the selected subset of past IOL implantation records; determining, for each estimated postoperative MRSE value, a predicted MRSE range based on the deviation value, wherein the first IOL power is determined further based on the predicted MRSE range; The method of claim 9 further comprising:

Images