CN110675929B

CN110675929B - Data processing system based on corneal topography

Info

Publication number: CN110675929B
Application number: CN201910910067.1A
Authority: CN
Inventors: 张哲�; 张素华
Original assignee: Individual
Current assignee: Individual
Priority date: 2019-09-25
Filing date: 2019-09-25
Publication date: 2020-09-01
Anticipated expiration: 2039-09-25
Also published as: CN110675929A

Abstract

The present disclosure discloses a data processing system based on corneal topography, comprising: the data entry unit is used for entering corneal topography maps before and after operations of a plurality of patients who have performed AK operations and consider the operations successful at the N-th follow-up visit after the operations, and AK arc length and axial position of the patients before the operations; a training model unit, configured to train the regression model established by the model establishing unit according to the training set established by the training set establishing unit and the AK arc length and axial position of the patient before the surgery, where: taking a training set as the input of a regression model, and taking the relation between the training set and the AK arc length and axial position during AK operation as a training standard; and the checking unit is used for checking the regression model trained by the training model unit according to the test set created by the test set unit. Therefore, a system for performing regression, inspection and even prediction on the AK arc length and the axial position through historical disease data is realized, and the AK surgical planning is facilitated by utilizing a data processing technology.

Description

Data processing system based on corneal topography

Technical Field

The disclosure belongs to the field of medical data processing, and particularly relates to a data processing system based on a corneal topography.

Background

Astigmatism is a major factor affecting the naked eye vision and visual quality of patients after refractive cataract surgery. Recently, a newly developed femtosecond laser cataract surgery (note: 1 femtosecond equals 10)^-15Second, light speed at 3 × 10⁸1 femtosecond time, the distance of light transmission in vacuum is 3 × 10^-7Meter, namely 0.3 micron, namely the precision can reach micron level), can make accurate cornea bow-shaped incision and correct astigmatism while cataract surgery, arouse the attention.

Arcus Keratotomy, its full name of academic Keratotomy, also referred to as AK surgery for short. In the practice of AK surgery, a device (for short, corneal topographer, not limited to devices for obtaining corneal topographes based on various principles) for obtaining corneal topography is often not available, and is a novel device for presenting corneal morphology through computer assistance, which can analyze the morphology and curvature of the whole corneal surface more accurately, so that it is possible to systematically, objectively, and accurately analyze corneal morphology. Fig. 1 is a picture of a device of this type in a combined operation of AK surgery and cataract surgery, in which an arcuate incision, an axial position of the arcuate incision and a main cataract incision are marked, fig. 2-1 is an axial curvature diagram of a corneal topography, and fig. 2-2 is a corneal thickness diagram of the corneal topography. Wherein, the color information of the axial curvature diagram represents the refractive power of the part of the cornea, the warm color refractive power is large, the cold color refractive power is small, and in the thickness diagram: a warm color represents a relatively thin cornea, and a cool color represents a relatively thick cornea; through the color distribution of the axial curvature diagram, the cornea shape can be seen whether is regular, the cornea refractive power is the largest at what position, the cornea refractive power is the smallest at what position, and the distribution and the shape of astigmatism are shown; the skilled practitioner can also consider the location of the post-operative incision edema and how the edema changes, and presumably how it affects the change in astigmatism, by corneal pachymetry analysis.

With the development of technology, although the corneal topography apparatus can accurately obtain the corneal topography of a patient before an operation, the image processing system can perform digital analysis on the corneal morphology, and express the obtained information by using a color image with different characteristics, and the image is called a corneal topography map because of the apparent topographic surface height state in geography. Such devices are expected to measure and analyze the curvature of any point on the total corneal surface with very high accuracy, and detect the refractive power of the cornea.

While corneal topography devices are constantly being developed, pre-AK planning is still obtained based on manual arcuate incision reference tables and the experience of the physician: during AK surgery, the AK arc length and axis of the incision made on the cornea are shown in table 1 below:

TABLE 1 reference Table based on Manual AK (sourced from wood and cock)

K Value(D)	Incision,Number(°)
		≤0.75	1(45)
≤1.00	2(44each)
		≤1.25	2(44and 45)
≤1.50	2(45each)
		≤2.0	2(47each)
≤2.5	2(49each)

K＝keratometry

That is, there is a need in the art for a solution for performing surgery-assisted planning by using data processing technology in the computer field.

Disclosure of Invention

To solve the above problem, the present disclosure discloses a data processing system based on corneal topography, comprising:

the device comprises a data entry unit, a training set creation unit, a test set creation unit, a model establishment unit, a training model unit and a test unit, wherein:

a data entry unit for entering: a plurality of patients who have undergone AK surgery and are considered successful in surgery at visit of day N after surgery, pre-operative corneal topography maps and post-operative corneal topography maps, and AK arc lengths and axial positions of the plurality of patients at the time of surgery before; wherein,

the corneal topography before the operation is obtained by a device which can obtain the corneal topography;

the post-operative corneal topography is obtained by multiple visits during 7 to N days post-operatively using the corneal topography's device on multiple time nodes, and 7< N < 730;

the AK arc lengths and axial positions of the patients during the operation are obtained through the record document of the operation;

the training set creating unit is used for creating a training set according to the corneal topography before the operation and the corneal topography on the Nth day after the operation of the patients obtained by the data entry unit;

the test set creating unit is used for creating a test set according to the corneal topography before the operation and the corneal topography on the Nth day after the operation of the patients obtained by the data entry unit;

a model establishing unit for establishing a regression model;

a training model unit to: training the regression model established by the model establishing unit according to the training set established by the training set establishing unit and the AK arc length and axial position of the patients in the operation, wherein: taking a training set as the input of a regression model, and taking the relation between the training set and the AK arc length and axial position during AK operation as a training standard;

and the checking unit is used for checking the regression model trained by the training model unit according to the test set created by the test set unit.

Preferably, the system further comprises:

a first prediction unit that employs a regression model as a first prediction model for: and for a patient who does not make an operation scheme, taking a corneal topography before an operation as an input, and outputting the predicted AK arc length and axial position according to the first prediction model for reference of a doctor.

Preferably, the system further comprises:

a second prediction unit employing a second prediction model different from the regression model for: for patients with an operation scheme not yet formulated, taking the corneal topography before the operation as input, and outputting corneal topography of different time nodes after the expected operation according to the second prediction model for reference of doctors;

wherein the second predictive model is obtained by:

s1, at least taking the axial curvature maps of the corneal topography maps of a plurality of patients before operation as the input of a regression model, thereby outputting the AK arc length and axial position corresponding to the patients according to the regression model;

s2, establishing a second prediction model;

s3, using the AK arc length and the axis position output by the regression model in step S1 as the first input and the second input of the second prediction model, and: using at least the axial curvature maps of the plurality of patient preoperative corneal topography maps as a third input of the second predictive model, and: and training to obtain a second prediction model by taking the axial curvature diagrams of at least 1 post-operative corneal topography diagram obtained at a plurality of time nodes and subjected to multiple follow-up visits in 7 days to N days after the operation of the patients as the standard of training of the second prediction model.

More preferably, it is a mixture of more preferably,

the regression model includes any one of: a polynomial regression model, a neural network based regression model, a logarithmic regression type, or other regression model.

More preferably, the system further comprises:

an adjustment unit for: and judging whether the regression model needs to be further adjusted for improving the precision or not according to the result of the checking unit.

More preferably, it is a mixture of more preferably,

the inputs to the regression model further include: the age or age group of the patient;

the training model unit is further configured to: training the regression model established by the model establishing unit according to the training set established by the training set establishing unit, the ages or the age groups of the patients and the AK arc lengths and axial positions of the patients in the operation, wherein: and fitting the training set, the ages or the age groups of the patients and the regression model by taking the training set, the ages or the age groups of the patients as input of the regression model: the relationship between AK arc length and axial position during AK operation.

More preferably, it is a mixture of more preferably,

the first prediction unit is further configured to: for a patient for which an operation scheme is not established, taking a corneal topography before an operation as a first input, taking the age or age group of the patient as a second input, and outputting the predicted AK arc length and axial position for reference of a doctor according to a first prediction model.

More preferably, it is a mixture of more preferably,

a second prediction unit further to: for patients for whom an operation scheme is not established, at least taking the axial curvature graph of the corneal topography graph before the operation as input, and outputting the axial curvature graphs of the corneal topography graphs of different time nodes after the expected operation according to the second prediction model for reference of doctors;

wherein the second predictive model is obtained by:

s11, at least taking the axial curvature maps of the corneal topography maps of a plurality of patients before operation and the ages or age groups of the patients as the first input and the second input of a regression model, thereby outputting AK arc length and axial position corresponding to the patients according to the regression model;

s21, establishing a second prediction model;

s31, using the AK arc length and the axis position output by the regression model in step S11 as the first input and the second input of the second prediction model, and: using at least the axial curvature maps of the plurality of patient preoperative corneal topography maps as a third input to the second predictive model, and: taking the ages or age groups of the plurality of patients as a fourth input to a second predictive model, and: and training to obtain a second prediction model by taking the axial curvature diagrams of at least 1 post-operative corneal topography diagram obtained at a plurality of time nodes and subjected to multiple follow-up visits in 7 days to N days after the operation of the patients as the standard of training of the second prediction model.

More preferably, the system further comprises:

an optimization unit, configured to determine, according to the output of the second prediction model, an AK arc length and an axis position output by the regression model in step S1, and/or: and optimizing the regression model.

Preferably, the first and second liquid crystal materials are,

when the regression model is based on a neural network, the regression model takes a ResNet convolution neural network as a basic network for extracting corneal topography characteristic information, and two full connection layers are added behind the ResNet network as a regressor for fitting the characteristic information and the internal relation between AK arc length and axial position.

Through the technical scheme, the system for performing regression, inspection and even prediction on the AK arc length and the axial position through historical disease data is realized by utilizing a computer data processing technology, so that the AK surgical planning is assisted by utilizing the data processing technology.

Drawings

FIG. 1 is a screen shot of a femtosecond laser cataract surgery combined with AK surgery in the prior art;

FIG. 2-1 is an axial curvature view of a corneal topography of a patient;

FIG. 2-2 is a thickness map of a corneal topography of a patient;

FIG. 3 is a schematic diagram of a regression model in one embodiment of the present disclosure;

FIG. 4 is a schematic diagram of a second predictive model in one embodiment of the disclosure;

figure 5 is an axial curvature and thickness map of a corneal topography of a patient at various time nodes after surgery, according to one embodiment of the present disclosure.

Detailed Description

In order to make those skilled in the art understand the technical solutions disclosed in the present disclosure, the technical solutions of the various embodiments will be described below with reference to the embodiments and the related drawings, and the described embodiments are a part of the embodiments of the present disclosure, but not all of the embodiments. The terms "first," "second," and the like as used in this disclosure are used for distinguishing between different objects and not for describing a particular order. Furthermore, "include" and "have," as well as any variations thereof, are intended to cover and not to exclude inclusions. For example, a process, method, system, or article or apparatus that comprises a list of steps or elements is not limited to only those steps or elements but may alternatively include other steps or elements not expressly listed or inherent to such process, method, system, article, or apparatus.

Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the disclosure. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. It will be appreciated by those skilled in the art that the embodiments described herein may be combined with other embodiments.

a model establishing unit for establishing a regression model;

It can be understood that the data entry unit of the above embodiment may be a PC or a notebook with a processor and an IO interface, or may be various lower computers with IO interfaces; the format of the corneal topography may be JPEG or JPG format, or other formats derived by devices of the corneal topography (abbreviated as corneal topographer, and not limited to such devices of various working principles), including but not limited to RAW format, vector diagram format, PDF format, or other original data formats proprietary to manufacturers of the corneal topographer, which can be processed to form an axial curvature map and a thickness map of the corneal topography; when the proprietary format is related, the proprietary format can communicate with a manufacturer to communicate with the manufacturer according to the protocol or specification, and the corresponding API or software is developed to be recorded into the data recording unit; after the input, the corneal topography can be stored in a file form or a database form;

and once the data is input, the training set creating unit, the test set creating unit, the model establishing unit, the training model unit and the inspection unit are arranged, wherein each unit can be a data processing unit with a corresponding function in a PC (personal computer), a notebook or a lower computer, or a data processing unit with a corresponding function in an upper computer, a server or a cluster, and can be located at a local end or a cloud end. It will be appreciated that the above units, including the data entry unit, often do not work apart from the processor and memory.

For the above embodiment, the size of the training set may be a part of the corneal topography before and after the operation, for example, 50% of the corneal topography data, or 70% of the corneal topography data, of a plurality of patients who have succeeded in the operation. The training set creating unit creates the training set according to the corneal topography before the operation and the corneal topography of the N day after the operation of the plurality of patients obtained by the data entry unit. Accordingly, the test set is sized to the data of the remaining 50% or 30% of patients, and the test set is created by the test set creation unit according to the data of the data entry unit.

As far as the above embodiments are concerned, they do not exclude any possible regression models. It is understood that for purposes of this disclosure, both the corneal topography before and after an AK procedure, as well as the AK arc length and axis at the time of the procedure, are historical data that have already occurred, and regression through the historical data, whether linear or non-linear, is contemplated as long as the regression model can help fit the corresponding input and output relationships. Whether the fit satisfies the usability can be checked by corresponding thresholds, e.g. bias, variance, least squares estimation, maximum likelihood estimation, etc. This part of the functionality may be implemented by the checking unit described above. The inventor of the present disclosure can know, according to mathematical principles: a regression model is always implemented on the premise of meeting a certain threshold (e.g., deviation, variance), i.e., a certain precision, so as to perform relatively accurate regression on historical data. Generally, the higher the accuracy, the more complex the regression model itself and its acquisition may be, and therefore, accuracy and cost are often tradeoffs in terms of regression.

That is, for the purposes of this disclosure, the regression models and the concepts tested thereon are themselves subject to mathematical knowledge, and these concepts or concepts are not themselves the focus of this disclosure, nor are they innovations of this disclosure. However, it can be understood that the adoption of existing concepts or concepts is a general phenomenon of the invention and is characterized in that: the present disclosure provides a technical solution capable of utilizing computer data processing technology and regression in the aspect of AK surgery assisted planning through the above embodiments, which combines with the practice of AK surgery, and implements a data processing system by a specific technical means by operating and processing various data (including but not limited to various corneal topography, AK arc length, axial position, etc.) according to various units including a data entry unit, thereby making use of computer processing capability to improve the technical progress in the aspect of AK surgery planning, implementing a data processing system in the aspect of AK surgery assisted planning by using only historical data, and thus solving the problem in the prior art that the AK surgery assisted planning must be performed by a manual arcuate incision reference table and relying on rich experience of doctors. This is just a technical contribution of the present disclosure to the art of ophthalmic AK surgery. For example, the above-described embodiments of the present disclosure typically innovatively introduce corneal topography as an input to a regression model and the basis for the overall solution. This enables the above-described embodiments to take full advantage of the information that is abundant in the corneal topography, whether it be an axial curvature map of the corneal topography or a thickness map of the corneal topography, or both maps of the corneal topography. This is because the pachymetry map of the corneal topography relates to the actual pachymetry throughout the cornea, and thus the pachymetry map of the corneal topography resembles a contour map, which correlates curvature and pachymetry, either of which may be theoretically possible, or better utilized. However, from the perspective of engineering practice, the axial curvature map of the corneal topography is sufficient, and only one corneal topography can reduce the time cost and the amount of calculation for data processing. Thus, preferably, the present disclosure takes corneal topography as input, by default referring to: axial curvature map of corneal topography.

It should be noted that, so far, about 13 international units have reported this new technology in the case of AK surgery for making corneal arcual incisions by femtosecond laser. It is one of the teams to which the inventor of the present disclosure belongs, and the papers published by this team and the Femtosecond laser arcuate incision reference table provided therein based on the correction of the manual arcuate incision reference table relating to three factors of the type of astigmatism, age and target correction, published in J CATARACT REFRACT SURG, one of the top journals in the field of international Cataract refractive surgery, are widely adopted by practitioners in the art, and the literature information of the manual table is as follows, namely, femtocell laser surgery and communicating with the cooperation with the patient surgery J.

The table after correction is shown in table 2 below:

TABLE 2 modified femtosecond laser penetration type AK reference table

However, the manual arcuate incision reference table only highlights the technical lack of AK surgery assisted planning using historical surgical data and computer data processing techniques in the art. Therefore, in view of the state of the art, it is reasonable to consider: the technical scheme disclosed by the disclosure makes corresponding technical contribution to the field. In addition, applicants have conducted extensive research and have not discovered prior art techniques in which this aspect is highly similar to the embodiments described above.

In another embodiment, the system further comprises:

For this embodiment, it is intended to use a regression model for the actual surgery-assisted planning, and therefore it uses the regression model as the first prediction model for: and for a patient who does not make an operation scheme, taking a corneal topography before an operation as an input, and outputting the predicted AK arc length and axial position according to the first prediction model for reference of a doctor.

As previously mentioned, the present disclosure takes corneal topography as input, by default refers to: axial curvature map of corneal topography. For example for the regression model and the first prediction model described above:

since the axial curvature map of the corneal topography contains information of corneal astigmatism, therefore:

(1) when the regression model is used for AK planning, its purpose is to correct the related astigmatism problem, so the input to the regression model is by default an axial curvature map. However, this does not exclude the further use of the pachymetry map of the corneal topography, and it can be understood that when the axial curvature map and the pachymetry map are used in combination, the data processing system of the present disclosure can not only help to correct the astigmatism problem of the patient by the curvature information of each part of the cornea, but also comprehensively consider the thickness information of each part of the cornea when the data processing system of the present disclosure plays a role in AK planning.

(2) When the first prediction model also employs the same model as the regression model, then it can also be understood that:

if only the axial curvature map is taken as input, it also helps to predict the possible curvature everywhere the cornea is implemented by the AK planning scheme given by the data processing system of the present disclosure, in addition to implementing the function and effect of the first prediction model;

how to jointly use the axial curvature map and the thickness map as inputs can not only help to predict the possible curvatures everywhere after the AK planning scheme given by the data processing system of the present disclosure is implemented, but also comprehensively consider the thickness information everywhere in the cornea.

Since some cataract patients need to be combined with astigmatism correction, the astigmatism correction can be AK operation, however, for such combined operation, according to the research of the inventor, the corneal thickness map can also assist in predicting the change trend of the postoperative corneal axial curvature map of the combined operation and analyzing the reason. The inventor believes that this is because, for such combined surgery, the post-operative corneal thickness map can see the part and thickness of the edema of the corneal main incision of the cataract surgery, the edema can cause the corneal curvature at the part to become smaller (the axis becomes flat), and as the edema of the corneal main incision recovers, the change of the corneal morphology (astigmatism) after the judgment can be assisted, and therefore, the effect in a period after such combined surgery can be estimated through the corneal thickness map. In addition to the above-described case of the combined surgery, for the surgery involving the cornea, the corneal topography-based data processing system disclosed in the present disclosure may also consider the pachymetric map of the corneal topography when estimating the effect thereof in the later stage, in addition to the axial curvature map of the corneal topography. It should be noted that for the above-mentioned combined operation, the factors affecting the edema of the main corneal incision after the operation include: hardness of cataract nucleus, energy used in phacoemulsification surgery. These factors can theoretically be further used for other inputs of the second predictive model.

That is, the present disclosure applies an axial curvature map for AK planning by default, such as application in a regression model and a first prediction model: on the one hand, this is because the arc length and axial position of the AK cut can be planned using the axial curvature map alone as input; on the other hand, if the axial curvature map is further combined with the thickness map, the possible thicknesses of the cornea can be considered/predicted as much as possible during planning so that the data processing system is more comprehensive during AK planning.

In another embodiment, the system further comprises:

wherein the second predictive model is obtained by:

s2, establishing a second prediction model;

This embodiment is clearly different from the previous embodiment in that the second prediction model does not fully adopt or clone the aforementioned regression model, but utilizes the output of the regression model as an input to the second prediction model, and in addition utilizes a third other input. It can be appreciated that the second prediction model herein focuses on predicting corneal topography at various post-operative time points. This is because, the surgery makes a certain or several axial incisions cut on the cornea with corresponding arc length (in practice, generally, there are no more than two incisions, because the cornea needs to be protected to prevent the biomechanical properties of the cornea from being greatly changed due to too many and too long incisions, which results in a great reduction in the ability to resist external force), and as the surgery is recovered, the degree of recovery is different at different time points, the cornea will change with the lapse of time, and it is predicted that the axial curvature map and even the thickness map of the corneal topography map at different time nodes are exactly the surgery effect at the corresponding time points after the surgery as accurately as possible. Therefore, using the preoperative 1-axis curvature map and the postoperative at least 1-axis curvature map of each patient in a plurality of patients, which have been verified to be successful in surgery, can be trained to develop the postoperative trend compared with the preoperative trend. It can be understood that more axial curvature graphs of the post-operation and other time nodes are more beneficial to the accuracy of the second prediction model.

Because the thickness maps at different time points after the operation can well reflect the time-sequence change of the corneal thickness, the axial curvature map time-sequence prediction of the corneal topography map after the operation can be more conveniently carried out by further combining the thickness maps. Also for this reason, it is more preferable that:

the above step S3 may be further optimized as follows: "taking at least the axial curvature map and the corneal thickness map of the plurality of patients' preoperative corneal topography maps as a third input, a fourth input of a second prediction model, and: and training to obtain a second prediction model by taking the combination of the axial curvature graph of at least 1 postsurgical corneal topographic map acquired at a plurality of time nodes and the thickness graph of the 1 postsurgical corneal topographic map corresponding to the time nodes, which are visited for a plurality of times from 7 days to N days after the operation of the patients, as a standard for training the second prediction model. More preferably, if more stringent and meaningful criteria are adopted, the field can generally use at least 2 post-operative corneal topography maps obtained at multiple time nodes with multiple follow-up visits at 1 week post-operative to 2 years post-operative (i.e., day 730 post-operative) as the criteria for training the second predictive model.

So far, according to the study of the axial curvature diagram and the thickness diagram by the inventor, it can be further understood that:

(1) "S1, using at least the axial curvature maps of the corneal topography maps of a plurality of patients before surgery as the input of the regression model" means: the input in step S1 can also include a thickness map, and if so, step S1 can include two types of inputs, one being an axial curvature map and the other being a thickness map; the use of an axial curvature map in combination with a thickness map has been described previously;

(2) in step S3, "using at least the axial curvature maps of the pre-operative corneal topography maps of the plurality of patients as the third input of the second prediction model" also means: in step S3, the thickness map may also be included as an input, and if so, in addition to the axial curvature map as a third input, the thickness map is also included as another kind of combined third input, and in this case, the thickness map is not numbered/defined as a fourth input; the advantages and the characteristics of the thickness map in the aspect of time sequence prediction are utilized;

(3) it should be noted that, where corneal topography is involved, "at least … … and … …" throughout the present disclosure may be understood with reference to (1) and (2) above.

In another embodiment of the present invention, the substrate is,

As previously mentioned, the present disclosure is not particularly exclusive of or limited to the types of regression models. However, in practice, polynomial regression models or neural network-based regression models are often preferred. This is mainly because: these two models can solve most of the problems in different domains. Taking polynomial regression as an example, theoretically, any curve can be combined by curves corresponding to a plurality of sections of different polynomial functions, wherein the curves comprise primary, secondary or higher-order polynomials; the regression model based on neural network is due to the development of deep learning in recent years, especially its wide application in image processing. Similarly, it is also understood that the second predictive model may include any of the following: a polynomial regression model, a neural network based regression model, a logarithmic regression type, or other regression model.

In another embodiment, the system further comprises:

As described above, a regression model can be implemented on the premise that a certain threshold (e.g., a threshold that can be set in terms of deviation, variance, least square estimation, maximum likelihood estimation, etc.) is satisfied, that is, a certain accuracy is met, so that the regression model can be used for performing relatively accurate regression on historical data. Generally, the higher the accuracy, the more complex the regression model itself and its acquisition may be, and therefore, accuracy and cost are often tradeoffs in terms of regression. That is, the checking unit may determine whether the regression model needs to be further adjusted to improve accuracy according to different threshold conditions. Adjusting the regression model, including adjusting its polynomial parameters or parameters related to the neural network, may also be understood as further training the regression model. Similarly, it can also be understood that the second prediction model can also be adjusted by a similar second adjustment unit. It should be noted that, both the adjusting unit and the second adjusting unit can be considered to further include the postoperative CI or the postoperative naked eye vision and optometry results to optimize the regression model.

In another embodiment of the present invention, the substrate is,

More preferably, it is a mixture of more preferably,

wherein the second predictive model is obtained by:

s21, establishing a second prediction model;

s31, using the AK arc length and the axis position output by the regression model in step S11 as the first input and the second input of the second prediction model, and: using at least the axial curvature maps of the plurality of patient preoperative corneal topography maps as a third input of the second prediction model, and: taking the ages or age groups of the plurality of patients as a fourth input to a second predictive model, and: and training to obtain a second prediction model by taking the axial curvature diagrams of at least 1 post-operative corneal topography diagram obtained at a plurality of time nodes and subjected to multiple follow-up visits in 7 days to N days after the operation of the patients as the standard of training of the second prediction model.

It can be understood that the above embodiments related to the age or age group of the patient are to consider the age or age group of the patient as a factor in the data processing system of the present disclosure, which facilitates more accurate aided planning and corresponding data processing of the AK surgery according to the age or age group of the patient. This is true in the field, as in table 2 above. In short, age or age group often reflects the healing condition of the cornea and other physiological characteristics such as cornea biomechanics.

It is further noted that in addition to the age or age range of the patient, the regression model and the second prediction model may also take into account other factors, such as: the preoperative average astigmatism numerical information (or the average astigmatism size in different areas of the preoperative cornea) and the preoperative astigmatism axis (or the astigmatism axis in different areas of the preoperative cornea) are used as related input. Furthermore, the second prediction model may also take into account: and after training is finished, besides outputting corneal topography maps of different time nodes, the correction rate is used as another output and is also used as a predicted reference result, and even used as a basis for optimizing AK arc length and/or axial position. Wherein, the astigmatism and axial position before and after operation can be obtained according to the axial curvature diagram of the cornea topographic map before and after operation, and the correction rate (CI) of a period of time after operation, such as 1 month and 3 months after operation, can be calculated by further applying a vector analysis method. It should be noted that the second prediction model may also be considered to further include the postoperative naked eye vision and the optometry results for tuning.

In another embodiment, the system further comprises:

It will be appreciated that this embodiment is intended to optimise the output of the regression model using the results output by the second predictive model, and/or: the regression model itself is optimized. Optimization of the regression model itself, including but not limited to tuning or retraining the regression model.

Referring to fig. 3, in another embodiment,

when the regression model is based on a neural network, the regression model takes a ResNet convolution neural network as a basic network for extracting corneal topography characteristic information, and at least two full connection layers are added behind the ResNet network as a regressor for fitting the characteristic information and the internal relation between AK arc length and axial position.

For this embodiment, specific options are given for the relevant models that may employ neural networks.

Referring to fig. 4, in another embodiment,

when in addition to age, AK arc length and axis position as inputs to the second prediction model, additional considerations may be taken: the pre-operative mean astigmatism value information (or mean astigmatism magnitude in different regions of the pre-operative cornea), the pre-operative astigmatism axis (or astigmatism axis in different regions of the pre-operative cornea) as correlation inputs, and the second prediction model may also consider as output: and (3) taking the postoperative correction rates (CI) of different time nodes (such as 1 month after operation, 3 months after operation and the like) as training standards, outputting the correction rates of the different time nodes after the training is finished, and optimizing the input of AK arc length.

It can be appreciated that this embodiment is directed to: and on the premise that the AK arc length and the axial position can be given by the regression model, determining the axial position within a certain time, further predicting the postoperative correction rate by only the second prediction model, and optimizing the AK arc length by using the correction rate. For example, the astigmatism correction rate closest to 1 is predicted by traversing the whole interval in a mode of taking 1 ° as a scale and 20-120 ° as an AK arc length range, and the corresponding arc length is obtained, namely the ideal AK arc length, so that the postoperative astigmatism correction effect is predicted by adjusting the AK arc length and the axial position of the operation.

For this embodiment, as shown in connection with fig. 4, the second predictive model includes a fully connected form of Deep Belief Network (DBN, Deep Belief Network, also known as Deep Belief Network) taking into account the correlation between the various variables. The constituent elements of the DBN are constrained boltzmann machines. This example is well consistent with the characteristics of preoperative and postoperative data in the field. This is because: for the present disclosure, patient data is acquired in a discrete form; the limited Boltzmann machine is used as a probability generation model, and when the limited Boltzmann machine is used for the disclosure, the difference of probability distribution between the limited Boltzmann machine and discrete data can be reduced as much as possible, so that the model can express probability information contained in the data; and replacing original random initial points with data of a large amount of corneal topography maps rich in information, training a second prediction model layer by layer, and finally predicting and obtaining the postoperative (such as 1 month and 3 months after operation) correction rate through the age of a patient, the preoperative astigmatism and axial position, the AK arc length and the axial position.

Furthermore, since the second prediction model is also aimed at: according to the pre-operation corneal topography and AK arc length and axial position of the patient, the axial curvature diagram of the post-operation corneal topography of different time nodes within a period of time (such as 7 days, 1 month, 3 months, half a year, 1 year, 2 years and the like) after the operation of the patient is predicted, so that the characteristic of a long-time and short-time memory network is also met. Therefore, preferably, the second prediction model may further employ: and memorizing the network at long time and short time.

Further, in another embodiment:

the second prediction model is obtained by:

s12, at least taking the axial curvature maps of the corneal topography maps of a plurality of patients before operation as the input of a regression model, thereby outputting the AK arc length and axial position corresponding to the patients according to the regression model;

s22, establishing a second prediction model;

s32, according to the AK arc length and the axis position output by the regression model in the step S12, the AK arc length and the axis position map of the corresponding patient are obtained;

taking the AK arc length and axis maps of the patients and the preoperative corneal topography maps of the patients as a first input and a second input of a second prediction model, and: and training to obtain a second prediction model by taking the axial curvature diagrams of at least 1 post-operative corneal topography diagram obtained at a plurality of time nodes and subjected to multiple follow-up visits in 7 days to N days after the operation of the patients as the standard of training of the second prediction model.

For this embodiment, since all the inputs of the second prediction model are graphs, the second prediction model may also employ: a net neural network. Further, preferably, in order to fully utilize the characteristics of the long-term memory network and the Unet neural network, the second prediction model may adopt: long and short time memory networks and Unet neural networks.

Preferably, the step S32 of obtaining the AK arc length and axial position map of the corresponding patient includes: and taking the outline of the preoperative corneal topography of the corresponding patient as the outline of the AK arc length and axial diagram, constructing a white or transparent canvas by the outline, and drawing the AK arc length and the axial diagram on the canvas. Thus, AK arc length and axis map of the corresponding patient are prepared.

Further, in another embodiment:

the second prediction model is obtained by:

s13, at least taking the axial curvature maps of the corneal topography maps of a plurality of patients before operation and the ages or age groups of the patients as the first input and the second input of a regression model, thereby outputting AK arc length and axial position corresponding to the patients according to the regression model;

s23, establishing a second prediction model;

s33, according to the AK arc length and the axis position output by the regression model in the step S13, the AK arc length and the axis position map of the corresponding patient are obtained;

taking the AK arc length and axis maps of the patients, the pre-operative corneal topography maps of the patients as the first input and the second input of a second prediction model, the ages or age groups of the patients as the third input of the second prediction model, and: and training to obtain a second prediction model by taking the axial curvature diagrams of at least 1 post-operative corneal topography diagram obtained at a plurality of time nodes and subjected to multiple follow-up visits in 7 days to N days after the operation of the patients as the standard of training of the second prediction model.

It is understood that this embodiment further incorporates the age or age group of the patient. This can further improve the technical effect of the data processing system.

It should be noted that not only the second prediction model but also the regression model may be based on the deep confidence network, and: when based on a deep belief network, the present disclosure can digitize, vectorize corneal topography, train in conjunction with vector analysis methods, rather than read patterns. As for the second prediction model, a long-term memory network may be adopted, and even when a pnet neural network is further utilized, the second prediction model may be trained by reading the graph in view of that the second prediction model can directly process image information.

The embodiments described above with respect to fig. 3 and 4 are described with respect to neural networks, which can be read to extract characteristic information therein when utilized. Just as in the prior art, the face recognition is performed by images shot by a camera, the relevant models can be trained by reading and through a neural network. As previously mentioned, the training model unit of the present disclosure is configured to: training the regression model established by the model establishing unit according to the training set established by the training set establishing unit and the AK arc length and axial position of the patients in the operation, wherein: and fitting the relation between the training set and the AK arc length and axial position during AK surgery by taking the training set as the input of the regression model, wherein the training set can be derived from a corneal topography, such as an axial curvature diagram of the corneal topography referred to by default in the disclosure. In this case, the regression model may be established by a neural network and used as a training target.

Fig. 5 is an axial curvature and thickness map of a corneal topography of a patient at a pre-operative time point and at different post-operative time nodes, as shown in fig. 5. It should be noted that, since fig. 5 relates to a graph of 1 week, 1 month, 3 months after operation, and a time point before operation, and a total of 4 time points, considering that fig. 5 is inserted into the drawings of the specification in a direction of turning left by 90 degrees, the following is described according to the direction of turning left by 90 degrees:

fig. 5 is, from bottom to top: preoperative, postoperative 1 week, postoperative 1 month, postoperative 3 month's picture, wherein one is the axial curvature map of each time point in the left side list, and the thickness map of each time point in the right side list:

(1) AK surgery assisted planning

An arch-shaped incision: axial position 55 degree arc length @180 degree

Axial position 55 degree arc length @0 degree

As for cataract surgery combined with this AK surgery, it was additionally judged by the doctor that: the main cataract incision is at 120 °

(2) Table for showing vision and astigmatism at each time point before and after operation

TABLE 3 patient naked eye and corneal astigmatism changes

The following is described for polynomial regression:

as mentioned above, the regression model may also be implemented by polynomial regression.

The goal of the regression analysis is to model the expected value of the dependent variable y from the value of the independent variable (or independent variable vector) x. In a simple linear regression, the model is used:

y＝β₀+β₁x+ (1)，

where no observed random error is observed, conditioned on a scalar x with a mean of zero in this model, for each unit increase in the value of x, the condition for y is expected to increase β₁And (4) units. In many cases, this linear relationship may not hold. In this case, there is a quadratic model as shown below:

y＝β₀+β₁x+β₂x2+ (2)，

in this model, y varies when increasing from x to x +1 units:

β₁+β₂(2x+1)·β₁+β₂(2x+1) (3)，

for an infinitesimal change in x, the effect on y is given by the derivative with respect to x:

β₁+β₂(2x+1) (4)，

the fact that y varies depending on the above-described regression model of x makes the relationship between x and y a non-linear relationship even though the model is linear in the parameters to be estimated.

More generally, the expectation of y can be modeled as an nth order polynomial, resulting in a general polynomial regression model, as shown in equation (5) below:

y＝β₀+β₁x+β₂x²+β₃x³+…+β_nxⁿ+ (5)

for the convenience of data processing by a computer (note: the computer referred to in this disclosure refers to a device or system with data processing capability, and is not limited to a portable device, a PC, a notebook computer, a server, a cluster, etc.), further, the polynomial regression model may be as follows:

can be determined by designing matrix X, response vector

Vector parameters

And a random error vector

To indicate. X and in the ith row

The x and y values for the ith data sample. The model can then be written as a system of linear equations:

when a pure matrix representation is used, it is written as:

the vector of polynomial regression coefficients is estimated (e.g., using least squares estimation) as:

for least squares, assume m<n is a requirement that the matrix be reversible, then since X is a Van der Waals matrix, if all X's are_iThe values are all different, and the condition of reversibility is ensured to be satisfied, which is the only solution of the least square method.

It should be noted that the objective of polynomial regression is to model the nonlinear relationship between independent variables and dependent variables (between the conditional mean values of independent variables and dependent variables). This is similar to the goal of nonparametric regression, which aims to capture the nonlinear regression relationship. Thus, nonparametric regression methods such as smoothing may be an effective alternative to polynomial regression. Some of these methods utilize a local form of classical polynomial regression. One advantage of conventional polynomial regression is that the inference framework of multivariate regression can be used (also when other families of basis functions are used, such as splines). In modern statistics, polynomial basis functions can be used with new basis functions, such as splines, radial basis functions, and wavelets. These families of basis functions provide a more concise fit for many types of data.

For purposes of this disclosure, in another embodiment, the AK arc length and axis of the procedure may be taken as x in the training of the regression model₁，x₂I.e., m is 2, and can be obtained from the axial curvature map and thickness map of the preoperative corneal topography: (1) the pre-operative total astigmatism, (2) the axis of the pre-operative total astigmatism, (3) the curvature of each point in the axial curvature map of the pre-operative corneal topography and/or (4) the thickness of each point in the thickness map of the pre-operative corneal topography, (5) the curvature of each point in the axial curvature map of the post-operative corneal topography and/or (6) the thickness of each point in the thickness map of the post-operative corneal topography, may then be taken as the corresponding y₁To y_iThus, it can be seen that: n is at least 4. This obviously satisfies the aforementioned m<n is the same as the formula (I). That is, for the regression model, the corresponding variables can be obtained by training the preoperative corneal topography and the postoperative corneal topography in the set, and the following units are implemented:

a model establishing unit for establishing a regression model; and the number of the first and second groups,

a training model unit to: training the regression model established by the model establishing unit according to the training set established by the training set establishing unit and the AK arc length and axial position of the patients in the operation, wherein: and fitting the relation between the training set and the AK arc length and axial position during AK surgery by taking the training set as the input of a regression model.

As mentioned above, the regression model may be based on a polynomial or on a neural network. Similarly, the second predictive model may be based on a polynomial or on a neural network. It will be appreciated that the regression model is not limited to a polynomial or neural network, but may be derived by other suitable fitting means.

Preferably, if to further improve the technical effect, in another embodiment:

and comparing according to a traditional manual reference table, and optimizing the regression model or optimizing the AK arc length and the axial position output by the regression model according to the second prediction model based on the depth confidence network and the second prediction model based on the long-time and short-time memory network.

Furthermore, for further tuning, whether based on a polynomial or neural network or other suitable fitting:

after the second prediction model is trained, respectively calculating the flatness of a 3mm area and a 5mm area in the center of each corneal topography map according to the corneal topography maps of different time nodes after the operation, which are output by the second prediction model, so as to obtain corresponding 3mm area scores and 5mm area scores; weighting and calculating the 3mm region score and the 5mm region score according to the weights of flatness of the central 3mm region and 5mm region of the corneal topography in the postoperative effect evaluation counted in advance to obtain a final postoperative score, wherein if the score is qualified, AK arc length and axial position adjustment are not needed; if not, adjusting the AK arc length and the axial position, and then making a new AK arc length and axial position diagram, and repeating the above embodiment until obtaining the optimal AK arc length and axial position.

It can be appreciated that this embodiment further adjusts the AK arc length and axis by scoring the central region. Furthermore, if the AK is not qualified, the regression model can be adjusted to adjust the AK arc length and the axial position, and a new AK arc length and axial position graph is prepared according to the AK arc length and the axial position graph, and the above embodiment is repeated until the optimal AK arc length and the optimal axial position are obtained. Wherein, the flatness reflects the residual astigmatism of the area, and can be calculated by curvature or thickness; this is because, according to the present disclosure, the corneal topography refers by default to an axial curvature map of the corneal topography, but may further include a thickness map of the corneal topography. It should be noted that the center of the corneal topography may be centered on the corneal vertex or the pupil center, and since the central 3mm region of the cornea is one of the most central considerations, the above embodiment may be simplified to include only the flatness of the central 3mm region of the corneal topography, but not the flatness of the central 5mm region.

In addition, in addition to the above-described central 3mm region and central 5mm region, it is also possible to consider: naked eye vision and optimal corrected vision at different time nodes after operation, AK arc length and axial position are further optimized, and even the related model, such as a regression model, is optimized. As the central 3mm area, the central 5mm area, the naked eye vision and the optimal corrected vision at different time nodes after operation all belong to postoperative data, the method can also be used for optimizing a second prediction model.

The units in the system of the embodiment of the present disclosure can be merged, divided, and deleted according to actual needs. It should be noted that, for simplicity of description, the foregoing steps are described as a series of acts or combinations of acts, but it should be understood by those skilled in the art that the present disclosure is not limited by the order of acts or acts described, as some steps may occur in other orders or concurrently with other steps in accordance with the disclosure. Further, those skilled in the art should also appreciate that the embodiments described in the specification are preferred embodiments and that the acts, modules, and elements referred to are not necessarily required by the disclosure.

In the foregoing embodiments, the descriptions of the respective embodiments have respective emphasis, and for parts that are not described in detail in a certain embodiment, reference may be made to related descriptions of other embodiments.

In the several embodiments provided in the present disclosure, it should be understood that the disclosed system, module, and unit may be implemented in other ways. For example, the above-described embodiments are merely illustrative, and for example, the division of the units is only one type of division of logical functions, and there may be other divisions when actually implemented, for example, a plurality of units or components may be combined or may be integrated into another system, or some features may be omitted, or not executed. In addition, the coupling or direct coupling or communication connection between the units or components may be through some interfaces, and the indirect coupling or communication connection between the devices or units may be in an electrical or other form.

The units described as separate parts may or may not be physically separate, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.

In addition, functional units in the embodiments of the present disclosure may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.

The integrated unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present disclosure may be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for causing a computer device (which may be a smartphone, a personal digital assistant, a wearable device, a laptop, a tablet, a server, a cluster) to perform all or part of the steps of the method according to the embodiments of the present disclosure. The storage medium includes various media capable of storing program codes, such as a U-disk, a Read-only Memory (R0M), a Random Access Memory (RAM), a removable hard disk, a magnetic disk, or an optical disk.

As described above, the above embodiments are only used to illustrate the technical solutions of the present disclosure, and not to limit the same; although the present disclosure has been described in detail with reference to the foregoing embodiments, it should be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present disclosure.

Claims

1. A corneal topography based data processing system comprising:

a model establishing unit for establishing a regression model;

the test unit is used for testing the regression model trained by the training model unit according to the test set created by the test set unit;

an adjustment unit for: judging whether the regression model needs to be further adjusted for improving the precision or not according to the result of the inspection unit;

wherein,

the regression model includes any one of: polynomial regression models, neural network based regression models, logarithmic regression types.

2. The data processing system of claim 1, further comprising:

3. The data processing system of claim 1, further comprising:

a second prediction unit employing a second prediction model different from the regression model for: for patients for whom an operation scheme is not established, at least taking the axial curvature graph of the corneal topography graph before the operation as input, and outputting corneal topography graphs of different time nodes after the expected operation according to the second prediction model for reference of doctors;

wherein the second predictive model is obtained by:

s2, establishing a second prediction model;

4. The data processing system of claim 1, wherein:

5. The data processing system of claim 2, wherein:

6. The data processing system of claim 3, wherein:

wherein the second predictive model is obtained by:

s21, establishing a second prediction model;

7. The data processing system of claim 3, further comprising:

8. The data processing system of claim 1, wherein: