Vision quality assessment based on machine learning model and wavefront analysis

Abstract

To provide vision quality assessment based on a machine learning model and wavefront analysis.SOLUTION: There is provided a system 10 and method for assessing vision quality of an eye E. A controller C includes a processor P and a tangible, non-transitory memory M in which instructions are recorded, and is configured to selectively execute at least one machine learning model 35, 36, 38. Execution of the instructions by the processor P causes the controller C to receive wavefront aberration data of an eye and express the wavefront aberration data as a collection of Zernike polynomials. The controller C is configured to obtain a plurality of input coefficients based on the collection of Zernike polynomials, which are fed to at least one machine learning model 35, 36, 38 trained to analyze the plurality of input coefficients. The machine learning model 35, 36, 38 generates at least one vision correction coefficient based in part on the plurality of input coefficients.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present disclosure generally relates to systems and methods for assessing the vision quality of an eye based on at least one machine learning model and wavefront analysis. [Background technology]

[0002] Humans have five basic senses: sight, hearing, smell, taste, and touch. Vision gives us the ability to visualize the world around us and connects us to our surroundings. Several scientific reports have shown that the brain devotes more space to processing and storing visual information than the other four senses combined, highlighting the importance of vision. Many people around the world suffer from vision quality problems, primarily due to refractive errors. Refractive errors of the eye can generally be categorized into low-order aberrations and high-order aberrations. Low-order aberrations include myopia, hyperopia, and astigmatism. Higher-order aberrations include many different aberrations, such as coma, trefoil, and spherical aberration. Traditional eye examination procedures result in assessments of vision quality that only evaluate the eye's low-order aberrations. Summary of the Invention [Means for solving the problem]

[0003] Disclosed herein are systems and methods for assessing vision quality of an eye, wherein the controller includes a processor and a tangible, non-transitory memory having instructions stored thereon. The controller is configured to selectively execute at least one machine learning model. Execution of the instructions by the processor causes the controller to receive wavefront aberration data of the eye and represent the wavefront aberration data as a collection of Zernike polynomials. The controller is configured to obtain a plurality of input coefficients based on the collection of Zernike polynomials. The plurality of input coefficients are provided to at least one machine learning model trained to analyze the plurality of input coefficients. The machine learning model generates at least one vision correction coefficient based in part on the plurality of input coefficients. The vision correction coefficients can be programmed into a laser device for reshaping the eye during vision correction procedures / refractive correction surgery. The vision correction coefficients can be used to aid in the selection of eyeglasses, contact lenses, and / or intraocular lenses for the eye.

[0004] The plurality of input coefficients may include respective wavefront coefficients for defocus, primary spherical aberration, oblique astigmatism, and vertical astigmatism. The vision correction coefficient may be a manifest refraction spherical equivalent. The vision correction coefficient may be an uncorrected vision coefficient in log MAR (logarithm of the minimum angle of resolution). The at least one machine learning model may incorporate a neural network and / or a support vector regression model.

[0005] The machine learning models may include a first machine learning model and a second machine learning model. Training the first machine learning model may include receiving a first training data set having respective wavefront aberration measurements and respective measured subjective refraction-based spherical equivalents for a first set of patients. First training input values ​​are obtained based on the respective wavefront aberration measurements and applied to respective input layers of the first machine learning model. The respective measured subjective refraction-based spherical equivalents may include pre-operative data and post-operative data. The respective measured subjective refraction-based spherical equivalents may be provided to respective output layers of the first machine learning model.

[0006] The first training input values ​​can be used to generate a first plurality of weights associated with each node of the first machine learning model. The first set of patients in the first training dataset can be characterized by respective health conditions and / or respective biometric parameters that fit within a first predefined maximum and a first predefined minimum. The respective biometric parameters can be anterior chamber depth, lens thickness, lens diameter, or other dimension.

[0007] Training the second machine learning model may include receiving a second training data set having respective wavefront aberration measurements and respective measured subjective refraction-based spherical equivalents for the second set of patients. Second training input values ​​are obtained based on the respective wavefront aberration measurements. The second training input values ​​are applied to respective input layers of the second machine learning model. The respective measured subjective refraction-based spherical equivalents are provided to respective output layers of the second machine learning model. The second training input values ​​may be used to generate a second plurality of weight values ​​associated with respective nodes of the second machine learning model. The second set of patients in the second training data set may be characterized by respective health conditions and / or respective biometric parameters that fit within a second predefined maximum value and a second predefined minimum value.

[0008] The above and other features and advantages of the present disclosure will become readily apparent from the following detailed description of the best mode for carrying out the disclosure, when read in conjunction with the accompanying drawings. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 1 is a schematic diagram of a system for assessing eye vision quality having a controller. [Figure 2] FIG. 2 is a schematic flow chart of a method that can be performed by the controller of FIG. [Figure 3] FIG. 3 is a schematic example of a neural network that can be implemented by the controller of FIG. [Figure 4] FIG. 4 is a schematic example of a support vector regression (SVR) network that can be implemented by the controller of FIG. [Figure 5] FIG. 5 is a schematic graph showing the measured (vertical axis) and predicted (horizontal axis) spherical equivalent powers based on subjective refraction tests relative to preoperative data. [Figure 6] FIG. 6 is a schematic graph showing the measured (vertical axis) and predicted (horizontal axis) spherical equivalent powers based on subjective refraction tests for postoperative data. [Figure 7] FIG. 7 is a schematic graph showing measured (vertical axis) and predicted (horizontal axis) uncorrected visual acuity in log MAR (logarithm of the minimum angle of resolution) for pre- and post-operative data. DETAILED DESCRIPTION OF THE INVENTION

[0010] Referring to the drawings, in which like reference numerals refer to like components, FIG. 1 schematically illustrates a system 10 for assessing vision quality of an eye E. As described below, the system 10 employs a robust approach that utilizes one or more machine learning models to optimize the assessment of vision quality, resulting in a higher success rate of predicting vision quality. Referring to FIG. 1, the system 10 includes a refractive device 12 having a light source 14 configured to project a beam of light 16 onto the eye E. The beam 16 is reflected by the retina R. Referring to FIG. 1, the reflected light 20 passes through a lens 22 before exiting the eye E as a wavefront 24. The wavefront 24 is characterized by distortions specific to the physical structure of the eye E.

[0011] Referring to FIG. 1 , a wavefront 24 is captured by a lenslet array 26 and detected by a sensor 30. An aberration map of eye E is created by comparing the shape of the captured wavefront 24 with the shape of a pre-programmed reference wavefront having the same pupil size (as the wavefront 24 passes through the pupil of eye E). For example, a point of difference between the two may be obtained at a particular point. The refractive device 12 may include associated beam guiding elements (not shown), electronic components, and other components available to those skilled in the art. It will be understood that the refractive device 12 may take many different forms and may include multiple and / or alternative components.

[0012] Referring to FIG. 1 , the system 10 includes a controller C configured to receive data from the sensor 30. The controller C may be incorporated into the refraction device 12. Referring to FIG. 1 , the controller C may be configured to communicate with the refraction device 12 and other entities via a short-range network 32. The short-range network 32 may be wireless or may include physical components. The short-range network 32 may be a bus implemented in various ways, such as a serial communication bus in the form of a local area network. The local area network may include, but is not limited to, a Controller Area Network (CAN), Controller Area Network with Flexible Data Rate (CAN-FD), Ethernet, Bluetooth®, Wi-Fi, and other forms of data connectivity. The short-range network 32 may be a Bluetooth® connection, which is defined as a short-range radio technology (or wireless technology) intended to simplify communication between Internet devices and between devices and the Internet. Bluetooth® is an open wireless technology standard for transmitting data over short distances between fixed and mobile electronic devices and creating personal networks that operates in the 2.4 GHz band. Other types of connections may also be employed.

[0013] Referring to FIG. 1 , the controller C can communicate with a user interface 34, which may include a display unit. Additionally, the controller C can be configured to communicate with a remote server 40 and / or a cloud unit 42 via a long-distance network 44. The remote server 40 may be a private or public information source maintained by an organization such as a research institute, a company, a university, and / or a hospital. The cloud unit 42 may comprise one or more servers hosted on the Internet for data storage, management, and processing. The long-distance network 44 may be a wireless local area network (LAN) that connects multiple devices in a wirelessly distributed manner, a wireless metropolitan area network (MAN) that connects several wireless LANs, or a wireless wide area network (WAN) that covers a large geographic area such as a nearby city or town. Other types of connections may also be employed.

[0014] The controller C may be configured to receive and transmit wireless communications with the remote server 40 via a mobile application 46 shown in FIG. 1. The mobile application 46 may communicate with the controller C over the short-range network 32 to access data from the controller C. In one example, the mobile application 46 is physically connected (e.g., wired) to the controller C. In another example, the mobile application 46 is embedded in the controller C. Circuits and components for the remote server 40 and the mobile application 46 ("app") available to those skilled in the art may be employed.

[0015] The controller C includes at least one processor P and at least one memory M (or a tangible, non-transitory, computer-readable storage medium) having instructions recorded thereon for executing the method 100. The method 100 is illustrated in FIG. 2 and will be described below with reference thereto. The controller C is specifically programmed to selectively execute one or more machine learning models 35 (hereinafter, "one or more" will be omitted), such as the first machine learning model 36 and the second machine learning model 38 illustrated in FIG. 1. The controller C can access the machine learning models 35 via the short-range network 32, the long-range network 44, and / or a mobile application 46. Alternatively, the machine learning models 35 may be embedded in the controller C. The machine learning models 35 may be configured to find parameters, weights, or structures that minimize respective cost functions and incorporate respective regression models.

[0016] Referring now to Figure 2, there is shown a flow diagram of a method 100 executable by the controller C of Figure 1. The method 100 does not have to be applied in the particular order described herein, and some blocks may be omitted. The memory M may store a controller-executable instruction set, and the processor P may execute the controller-executable instruction set stored in the memory M.

[0017] 2, the controller C is configured to receive wavefront aberration data of the eye E and transform or express it in terms of a collection of Zernike polynomials. The wavefront aberration data (A) is

number

number

[0018] The controller C is configured to obtain a plurality of input coefficients based on the collection of Zernike polynomials, the plurality of input coefficients being one or more of the respective wavefront coefficients measured on the eye E. In one example, the controller C obtains two input coefficients, namely, the respective wavefront coefficients for defocus and

number

number

number

number

number

number

[0019] Per block 120 of FIG. 2 , the method 100 includes providing a plurality of input coefficients to a machine learning model 35 trained to analyze the plurality of input coefficients. Per block 130 of FIG. 2 , the controller C is configured to generate at least one vision correction coefficient by executing the machine learning model 35 based in part on the plurality of input coefficients. The vision correction coefficient may include components of refraction, i.e., sphere, cylinder, and spherical equivalent, and may be expressed as a subjective refraction-based spherical equivalent (MRSE). The vision correction coefficient may be expressed as an uncorrected visual acuity coefficient in log MAR (logarithm of the minimum angle of resolution). The vision correction coefficient may be used to establish an ablation profile for refractive correction surgery and to assist in the selection of glasses, contact lenses, and / or an intraocular lens for the eye. Additionally, the controller C may be configured to create a patient profile for the patient (of eye E) in the cloud 42 and / or a remote server 40 via a long-distance network 44 and upload or “store” the vision correction coefficient in the patient profile.

[0020] The machine learning model 35 of FIG. 1 may include a neural network, an example of which is shown in FIG. 3. Referring to FIG. 3, the neural network 200 is a feedforward artificial neural network having at least three layers, including an input layer 201, at least one hidden layer 220, and an output layer 240. Each layer is composed of a respective node N configured to perform an affine transformation of a linear sum of inputs. Each node N is characterized by a respective bias and a respective weighted link. The parameters of each node N may be independent of others, i.e., characterized by a unique set of weights. The input layer 201 may include a first input node 202, a second input node 204, a third input node 206, a fourth input node 208, a fifth input node 210, and a sixth input node 212. Each node N in the input layer 201 receives inputs, normalizes them, and forwards them to a respective node N in the hidden layer 220.

[0021] Referring to FIG. 3 , hidden layer 220 may include first hidden node 222, second hidden node 224, third hidden node 226, fourth hidden node 228, and fifth hidden node 230. Each node N in a subsequent layer computes a linear combination of the outputs of the previous layer. A network with three layers forms an activation function f(x) = f(3)(f(2)(f(1)(x))). The activation function f may be linear for each node N in output layer 240. The activation function f may be a sigmoid in hidden layer 220. A linear combination of sigmoids may be used to approximate a continuous function that characterizes the output vector y. Patterns recognized by neural network 200 may be translated or converted into numerical form and organized into vectors or matrices.

[0022] The machine learning model 35 may include a support vector regression model 300, an example of which is shown in FIG. 4. The support vector regression model 300 is configured to find a function (hyperplane 304 in FIG. 4) such that a data point 302 lies within a margin 306 from this function, i.e., inside a first boundary line 308 and a second boundary line 310. Referring to FIG. 4, the hyperplane 304 may be defined as a line that matches an input vector x to an output vector y, i.e., predicts a target value. The hyperplane 304 is personalized to maximize the margin 306 and minimize a predefined error. A penalty may be incorporated into the support vector regression model 300 for points that fall outside the margin 306 (e.g., an outer point 312). Before determining the hyperplane 304, the support vector regression model 300 may employ a kernel function to match a low-dimensional dataset to a high-dimensional dataset. Other machine learning models available to those skilled in the art may also be employed.

[0023] The machine learning model 35 can match an input vector x to an output vector y by using a deep learning map to learn an activation function f such that f(x) maps to y. The training process allows the machine learning model 35 to correlate the appropriate activation function f(x) to transform the input vector x into the output vector y. For example, in a simple linear regression model, two parameters are learned: bias and slope. The bias is the level of the output vector y when the input vector x is set to 0, and the slope is the predicted rate of increase or decrease of the output vector y for each unit increase in the input vector x. Once the machine learning model 35 is trained, an estimate of the output vector y can be calculated with each new value of the input vector x.

[0024] Referring to FIG. 1 , the controller C may be configured to obtain one or more training data sets from a remote server 40 via a long-distance network 44. Training the first and second machine learning models 36 and 38 may include receiving first and second training data sets, respectively, having wavefront aberration measurements and spherical equivalent values ​​based on measured subjective refraction tests for a first set of patients and a second set of patients. The training data sets may be stratified based on ocular biometric parameters. In other words, the process may be optimized by grouping training data sets for similar eye size dimensions or other health factors (e.g., grouping patients affected by glaucoma into a first set of patients and patients affected by a history of retinal detachment into a second set of patients).

[0025] In one non-limiting example, a first set of patients in a first training data set may be characterized by each biometric parameter falling within a first predefined maximum and a first predefined minimum. Each biometric parameter may be an anterior chamber depth, a lens thickness, a lens diameter, or other physical dimension of the eye. A second set of patients in a second training data set may be characterized by each biometric parameter falling within a second predefined maximum and a second predefined minimum.

[0026] First and second training input values ​​may be obtained based on the respective wavefront aberration measurements and applied to respective input layers of the first machine learning model 36 and the second machine learning model 38. The respective measured subjective refraction-based spherical equivalents may include pre-operative data and post-operative data. The respective measured subjective refraction-based spherical equivalents may be provided to respective output layers of the first machine learning model 36 and the second machine learning model 38. The first and second training input values ​​may be used to generate first and second plurality of weights associated with respective nodes of the first and second machine learning model 36 and the second machine learning model 38. This may be done by a training program separate from the refractive device 12 and / or the controller C.

[0027] Referring now to Figures 5, 6, and 7, schematic graphs depicting various examples of fitting models using topographically guided laser refractive correction studies are shown. Figure 5 shows a model fit line 400 for preoperative data, with the vertical axis Y1 representing measured subjective refraction-based spherical equivalent (MRSE) values ​​and the horizontal axis X1 representing predicted MRSE values. Figure 6 shows a model fit line 500 for postoperative data, with the vertical axis Y2 representing measured MRSE values ​​and the horizontal axis X2 representing predicted MRSE values. Note that Figures 5 and 6 are on different scales, with Figure 6 having a smaller range.

[0028] 7 shows model fit lines 600 and respective contours 610 for both pre- and post-operative data, with the vertical axis Y3 representing measured log MAR (logarithm of the angle of resolution) uncorrected visual acuity values ​​and the horizontal axis X3 representing predicted log MAR (logarithm of the angle of resolution) uncorrected visual acuity values. Tables 1 and 2 below show a comparison of the fitted models of FIGS. 5 and 6, which are linear sums of second-order and fourth-order Zernike polynomials, respectively.

[0029] [Table 1]

[0030] [Table 2]

[0031] As shown above in Tables 1 and 2, the machine learning model 35 improves both the mean absolute prediction error and the prediction success rate for assessing vision quality. Additionally, the system 10 eliminates the need to rescale pupil diameter when viewing distant objects.

[0032] The controller C of FIG. 1 includes computer-readable media (also referred to as processor-readable media), including non-transitory (e.g., tangible) media involved in providing data (e.g., instructions) that may be read by a computer (e.g., by a computer processor). Such media may take many forms, including, but not limited to, non-volatile and volatile media. Non-volatile media may include, for example, optical or magnetic disks and other persistent memory. Volatile media may include, for example, dynamic random access memory (DRAM), which may constitute a primary storage device. Such instructions may be transmitted over one or more transmission media, including coaxial cables, copper wire, and optical fiber, including the wires that comprise a system bus coupled to the computer's processor. Some forms of computer-readable media include, for example, floppy disks, flexible disks, hard disks, magnetic tape, other magnetic media, CD-ROMs, DVDs, other optical media, punch cards, paper tape, other physical media with patterns of holes, RAM, PROMs, EPROMs, Flash EEPROMs, other memory chips or cartridges, or other computer-readable media.

[0033] The lookup tables, databases, data repositories, or other data stores described herein may include various types of mechanisms for storing, accessing, and retrieving various types of data, including a hierarchical database, a set of files in a file system, a proprietary application database, a relational database management system (RDBMS), etc. Each such data store may be contained within a computing device employing a computer operating system such as those described above, or may be accessed over a network in one or more of a variety of ways. A file system is accessible from a computer operating system and may include files stored in various formats. An RDBMS may employ Structured Query Language (SQL) in addition to a language for creating, saving, editing, and executing stored procedures, such as the PL / SQL language described above.

[0034] While the detailed description and drawings or figures support and explain the present disclosure, the scope of the present disclosure is defined solely by the claims. While the best mode and some other embodiments for carrying out the claimed disclosure have been described in detail, various alternative designs and embodiments exist for carrying out the disclosure defined in the appended claims. Furthermore, the features of the various embodiments shown in the drawings or described herein should not necessarily be understood as independent embodiments of one another. Rather, each of the characteristics described in one of the example embodiments can be combined with one or more other desirable characteristics from other embodiments, resulting in other embodiments not described in words or with reference to the drawings. Accordingly, such other embodiments are encompassed within the scope of the appended claims. According to aspect (1), there is provided a system for assessing the quality of vision of an eye, comprising: a controller having a processor and a tangible, non-transitory memory having instructions recorded thereon, the controller being configured to selectively execute at least one machine learning model; Execution of the instructions by the processor causes the controller to: receiving wavefront aberration data of the eye and expressing the wavefront aberration data as a collection of Zernike polynomials; obtaining a plurality of input coefficients based on the collection of Zernike polynomials; providing the plurality of input coefficients to at least one of the machine learning models trained to analyze the plurality of input coefficients; The system generates at least one vision correction coefficient based in part on the plurality of input coefficients via the at least one machine learning model. According to aspect (2), the plurality of input coefficients include wavefront coefficients for defocus, primary spherical aberration, oblique astigmatism, and vertical astigmatism. According to aspect (3), at least one of the vision correction coefficients is a spherical equivalent coefficient based on a subjective refraction test. According to aspect (4), at least one of the vision correction coefficients is an uncorrected vision coefficient of log MAR (logarithm of the minimum angle of resolution). According to aspect (5), at least one of the machine learning models incorporates a neural network. According to aspect (6), at least one of the machine learning models incorporates a support vector regression model. According to aspect (7), the at least one machine learning model includes a first machine learning model, and training the first machine learning model includes: receiving a first training data set having respective wavefront aberration measurements and respective measured subjective refraction-based spherical equivalent values ​​for a first set of patients; obtaining first training input values ​​based on the respective wavefront aberration measurements and applying the training input values ​​to respective input layers of the first machine learning model; providing each of the measured subjective refraction-based spherical equivalent values ​​to an output layer of each of the first machine learning models; generating a first plurality of weight values ​​associated with respective nodes of the first machine learning model based in part on the first training input values. According to aspect (8), the spherical equivalent power values ​​based on the respective measured subjective refraction tests include respective preoperative data and respective postoperative data. According to aspect (9), the first set of patients in the first training data set are characterized by respective biometric parameters and / or respective health conditions, and the respective biometric parameters fit within a first predefined maximum value and a first predefined minimum value. According to aspect (10), the at least one machine learning model includes a second machine learning model, and training the second machine learning model includes: receiving a second training data set having the respective wavefront aberration measurements and the respective measured subjective refraction-based spherical equivalent values ​​of a second set of patients; obtaining second training input values ​​based on the respective wavefront aberration measurements and applying the second training input values ​​to the respective input layers of the second machine learning model; providing the respective measured subjective refraction-based spherical equivalent values ​​to the respective output layers of the second machine learning model; generating a second plurality of weight values ​​associated with each node of the second machine learning model based in part on the second training input values. According to aspect (11), the second set of patients in the second training dataset are characterized by respective biometric parameters and / or respective health conditions, and the respective biometric parameters fit within a second predefined maximum value and a second predefined minimum value. According to aspect (12), there is provided a method for assessing the quality of vision of an eye, comprising: receiving wavefront aberration data of the eye via a controller having a processor and a tangible, non-transitory memory, the controller configured to selectively execute at least one machine learning model; expressing the wavefront aberration data as a collection of Zernike polynomials; obtaining a plurality of input coefficients based on the collection of Zernike polynomials; providing the plurality of input coefficients to at least one of the machine learning models trained to analyze the plurality of input coefficients; and generating, via at least one of the machine learning models, at least one vision correction coefficient based in part on the plurality of input coefficients. According to aspect (13), the at least one machine learning model includes a first machine learning model, and training the first machine learning model includes: receiving a first training data set having respective wavefront aberration measurements and respective measured subjective refraction-based spherical equivalent values ​​for a first set of patients; obtaining first training input values ​​based on the respective wavefront aberration measurements and applying the training input values ​​to respective input layers of the first machine learning model; providing each of the measured subjective refraction-based spherical equivalent values ​​to an output layer of each of the first machine learning models; generating a first plurality of weight values ​​associated with respective nodes of the first machine learning model based in part on the first training input values. According to aspect (14), the method further includes including each preoperative data and each postoperative data in the spherical equivalent power value based on each measured subjective refraction test. According to aspect (15), the method further includes characterizing the first set of patients in the first training dataset by their respective biometric parameters and / or their respective health conditions, wherein the respective biometric parameters fit within a first predefined maximum value and a first predefined minimum value. According to aspect (16), the at least one machine learning model includes a second machine learning model, and training the second machine learning model includes: receiving a second training data set having the respective wavefront aberration measurements and the respective measured subjective refraction-based spherical equivalent values ​​of a second set of patients; obtaining second training input values ​​based on the respective wavefront aberration measurements and applying the second training input values ​​to the respective input layers of the second machine learning model; providing the respective measured subjective refraction-based spherical equivalent values ​​to the respective output layers of the second machine learning model; generating a second plurality of weight values ​​associated with each node of the second machine learning model based in part on the second training input values. According to aspect (17), the method further includes characterizing the second set of patients in the second training dataset by their respective biometric parameters and / or their respective health states, wherein the respective biometric parameters fit within a second predefined maximum value and a second predefined minimum value.

Claims

1. 1. A system for assessing ocular vision quality, comprising: a controller having a processor and a tangible, non-transitory memory having instructions recorded thereon, the controller being configured to selectively execute two or more machine learning models; Execution of the instructions by the processor causes the controller to: receiving wavefront aberration data of the eye and expressing the wavefront aberration data as a collection of Zernike polynomials; obtaining a plurality of input coefficients based on the collection of Zernike polynomials; providing the plurality of input coefficients to two or more of the machine learning models trained to analyze the plurality of input coefficients; generating at least one vision correction coefficient based in part on the plurality of input coefficients via two or more of the machine learning models; the plurality of input coefficients include wavefront coefficients for defocus, primary spherical aberration, oblique astigmatism, and vertical astigmatism; At least one of the vision correction coefficients can be expressed as a spherical equivalent coefficient based on subjective refraction; The two or more machine learning models include a first machine learning model having an input layer and an output layer, and training the first machine learning model includes: receiving a first training data set having wavefront aberration measurements for each patient of a first set of patients and spherical equivalent values ​​based on measured subjective refraction for each patient; obtaining first training inputs based on the wavefront aberration measurements for each patient and applying the training inputs to the input layer of each of the first machine learning models; providing the measured spherical equivalent power value based on the subjective refraction test for each patient to the output layer of each of the first machine learning models; generating a first plurality of weight values ​​associated with respective nodes of the first machine learning model based in part on the first training input values; the first set of patients in the first training data set are characterized by patient-specific biometric parameters and patient-specific health conditions, the patient-specific biometric parameters fitting within a first predefined maximum value and a first predefined minimum value, the biometric parameters being physical dimensions of the eye including at least one of anterior chamber depth, lens thickness, and lens diameter; the first set of patients in the first training data set are characterized by the physical dimension of the eye falling within a range between the first predefined maximum and the first predefined minimum; The two or more machine learning models include a second machine learning model having an input layer and an output layer, and training the second machine learning model includes: receiving a second training data set having the wavefront aberration measurements for each patient of a second set of patients and the measured subjective refraction-based spherical equivalent values ​​for each patient; obtaining second training inputs based on the wavefront aberration measurements for each patient, and applying the second training inputs to the input layer of each of the second machine learning models; providing the measured spherical equivalent power value based on the subjective refraction test for each patient to the output layer of each of the second machine learning models; generating a second plurality of weight values ​​associated with each node of the second machine learning model based in part on the second training input values; the second set of patients in the second training data set are characterized by the biometric parameters for each patient and by a health condition for each patient, the biometric parameters for each patient fitting within a second predefined maximum value and a second predefined minimum value; the second set of patients in the second training data set are characterized by the physical dimensions of the eyes falling within a range of a second predefined maximum value and a second predefined minimum value that are different from the first predefined maximum value and the first predefined minimum value; The first set of patients and the second set of patients are grouped based on different patient health conditions. system.

2. The system of claim 1 , wherein at least one of the vision correction factors is an uncorrected vision factor of log MAR (logarithm of the minimum angle of resolution).

3. The system of claim 1 , wherein at least one of the two or more machine learning models incorporates a neural network.

4. The system of claim 1 , wherein at least one of the two or more machine learning models incorporates a support vector regression model.

5. The system of claim 1 , wherein the measured subjective refraction-based spherical equivalent for each patient includes pre-operative data for each patient and post-operative data for each patient.

6. 1. A method for assessing visual acuity quality of an eye, comprising: receiving wavefront aberration data of the eye via a controller having a processor and a tangible, non-transitory memory, the controller configured to selectively execute two or more machine learning models; expressing the wavefront aberration data as a collection of Zernike polynomials; obtaining a plurality of input coefficients based on the collection of Zernike polynomials; providing the plurality of input coefficients to two or more of the machine learning models trained to analyze the plurality of input coefficients; generating at least one vision correction coefficient based in part on the plurality of input coefficients via two or more of the machine learning models; The two or more machine learning models include a first machine learning model having an input layer and an output layer, and training the first machine learning model includes: receiving a first training data set having wavefront aberration measurements for each patient of a first set of patients and spherical equivalent values ​​based on measured subjective refraction for each patient; obtaining first training inputs based on the wavefront aberration measurements for each patient and applying the training inputs to the input layer of each of the first machine learning models; providing the measured subjective refraction-based spherical equivalent value for each patient to the output layer of each of the first machine learning models; generating a first plurality of weight values ​​associated with respective nodes of the first machine learning model based in part on the first training input values; characterizing the first set of patients in the first training data set by patient-specific biometric parameters and patient-specific health conditions, wherein the patient-specific biometric parameters fit within a first predefined maximum value and a first predefined minimum value, and the biometric parameters are physical dimensions of the eye including at least one of anterior chamber depth, lens thickness, and lens diameter; the first set of patients in the first training data set are characterized by the physical dimension of the eye falling within a range between the first predefined maximum and the first predefined minimum; The two or more machine learning models include a second machine learning model having an input layer and an output layer, and training the second machine learning model includes: receiving a second training data set having the wavefront aberration measurements for each patient of a second set of patients and the measured subjective refraction-based spherical equivalent values ​​for each patient; obtaining second training inputs based on the wavefront aberration measurements for each patient, and applying the second training inputs to the input layer of each of the second machine learning models; providing the measured spherical equivalent power value based on the subjective refraction test for each patient to the output layer of each of the second machine learning models; generating a second plurality of weight values ​​associated with each node of the second machine learning model based in part on the second training input values; characterizing the second set of patients in the second training data set by the biometric parameters for each patient and by a health condition for each patient, wherein the biometric parameters for each patient fit within a second predefined maximum value and a second predefined minimum value; the second set of patients in the second training data set are characterized by the physical dimensions of the eyes falling within a range of a second predefined maximum value and a second predefined minimum value that are different from the first predefined maximum value and the first predefined minimum value; The first set of patients and the second set of patients are grouped based on different patient health conditions. method.

7. 7. The method of claim 6, further comprising including pre-operative data for each patient and post-operative data for each patient in the measured subjective refraction-based spherical equivalent value for each patient.

Images