Search

CN-122020425-A - Intelligent recommendation method and system for cornea shaping mirror test and allocation parameters

CN122020425ACN 122020425 ACN122020425 ACN 122020425ACN-122020425-A

Abstract

The invention discloses an intelligent recommendation method and system for cornea shaping lens test parameters, which belong to the technical field of ophthalmic medical instrument assistance, the method comprises six steps of cornea parameter acquisition, feature coding, anomaly detection, recommendation prediction, uncertainty quantification and effect simulation, a gradient lifting tree integrated model is adopted to realize multi-parameter collaborative prediction of lens base arc, adaptation curvature and lens diameter, and introducing quantile regression to output a confidence interval, setting an anomaly detection layer based on weighted multi-index risk scores to carry out four-level risk layering screening on the cornea morphology, providing visual preview of expected shaping effect and fluorescent dyeing mode, solving the problems of incomplete parameter recommendation, lack of uncertainty quantification and effect preview in the prior art, and shortening the average test and allocation period.

Inventors

  • Tai Yuqing
  • GE TIANQI

Assignees

  • 山东中医药大学

Dates

Publication Date
20260512
Application Date
20260304

Claims (10)

  1. 1. The intelligent recommendation method for the cornea shaping mirror fitting parameters is characterized by comprising the following steps: A cornea parameter acquisition step, namely acquiring cornea topography original data of an eye to be tested from a cornea topography instrument, and extracting a cornea geometric parameter set from the cornea topography original data, wherein the cornea geometric parameter set comprises a flat K value, a steep K value, cornea eccentricity, pupil diameter, cornea transverse diameter and central cornea thickness, and meanwhile, patient physiological characteristic data comprising diopter, eye axis length, age and astigmatism degree are acquired; a feature coding step, wherein the cornea geometric parameter set and the physiological feature data of the patient are subjected to standardized treatment, and the standardized data are spliced and fused to form a multidimensional input vector; an anomaly detection step of inputting the multidimensional input vector into an anomaly detection model, wherein the anomaly detection model calculates four sub-risk scores of an astigmatism risk score, an eccentric risk score, an irregular astigmatism risk score and a keratoconus risk score respectively; A recommendation prediction step, namely inputting the multidimensional input vector detected by abnormality into a gradient lifting tree integrated model, wherein the gradient lifting tree integrated model is obtained based on historical verification and matching success case training, and outputting a recommendation lens parameter combination, wherein the recommendation lens parameter combination comprises a lens base arc, a fitting curvature and a lens diameter; Uncertainty quantization, namely performing confidence interval estimation on each parameter in the recommended lens parameter combination based on a quantile regression model, and generating a boundary case prompt when the confidence interval width of any parameter exceeds a preset confidence threshold; And an effect simulation step, namely generating an expected shaping effect simulation diagram through a cornea deformation prediction model based on the recommended lens parameter combination and the cornea geometric parameter set, and generating a cornea fluorescence staining mode prediction diagram after wearing a lens through a lens-angle gap distribution model.
  2. 2. The method of claim 1, wherein in the cornea parameter acquisition step, the flat K value and the steep K value are measured in diopters ranging from 39.0D to 48.0D, the cornea eccentricity is measured in units of 0.3 to 0.8, the pupil diameter and the cornea transverse diameter are measured in millimeters ranging from 3.0mm to 7.0mm, the cornea transverse diameter is measured in units of 11.0mm to 13.0mm, and the central cornea thickness is measured in microns ranging from 480 μm to 600 μm.
  3. 3. The method according to claim 1, wherein in the feature encoding step, each parameter in the cornea geometrical parameter set is mapped to a range from 0 to 1 by using a Min-Max normalization method, discretized binning is performed on age parameters in the patient physiological feature data, the diopter and the ocular axis length are processed by using a Z-score normalization method, and all the processed parameters are spliced according to a preset sequence to form the multidimensional input vector with a dimension of N, wherein N is 10 to 15.
  4. 4. The method according to claim 1, wherein the astigmatism risk score is obtained by performing a context correction based on the astigmatism degree and combining an astigmatism type and a pupil diameter, the irregular astigmatism risk score is obtained by performing a weighted summation after assigning priority weights corresponding to clinical urgency to the four sub-risk scores respectively to obtain a comprehensive risk score, dividing the eye to be tested into four classes of normal, low-risk, medium-risk and high-risk according to a score interval of the comprehensive risk score, outputting an expert consultation prompt and terminating a subsequent recommended procedure when the eye to be tested is divided into a high-risk class, outputting a risk prompt when the eye to be tested is divided into a stroke risk class, attaching a risk class label to the multi-dimensional input vector, transmitting the multi-dimensional input vector to a recommended prediction step when the eye to be tested is divided into a normal or low-risk class, transmitting the multi-dimensional input vector to the recommended prediction step, setting the priority of the conic cornea to be a priority score of 0.35 to be normal, setting the priority of the astigmatism type to be 0.35 to be a standard of astigmatism, setting the priority of the astigmatism type to be 0.20 to be a standard of 0, setting the priority to be an astigmatism type to be 20 when the eye to be tested to be divided into a high-risk class, and setting the priority to be a standard of 0.20 to be an astigmatism type to be 20, the method comprises the steps of adjusting a reference threshold to be up-regulated by a preset correction amount when the pupil diameter is smaller than a preset pupil threshold, wherein the threshold of a cornea surface regularity index and the threshold of a cornea surface asymmetry index are 1.0 and 0.5 respectively in calculation of irregular astigmatism risk scores, starting irregular astigmatism risk score calculation when any one of the two indexes exceeds the corresponding threshold, and dividing the irregular astigmatism risk score into a high risk grade when the comprehensive risk score is larger than or equal to 0.8, a medium risk grade when the comprehensive risk score is larger than or equal to 0.5 and smaller than 0.8, a low risk grade when the comprehensive risk score is larger than or equal to 0.3 and smaller than 0.5, and a normal grade when the comprehensive risk score is smaller than 0.3.
  5. 5. The method according to claim 1, wherein in the recommending and predicting step, the training process of the gradient lifting tree integrated model comprises the steps of obtaining a sample set of successful cases from a multi-center verification database, wherein the judgment standard of the successful cases is that the eccentricity of a treatment area is not more than 1mm after three months of mirror wearing and the subjective comfort degree score of a patient is not lower than a preset comfort degree threshold value, dividing the sample set into a training set and a verification set according to a preset proportion, and performing iterative training on the training set by adopting a gradient lifting decision tree algorithm until loss of the verification set is converged.
  6. 6. The method according to claim 1, wherein in the uncertainty quantization step, the quantile regression model outputs a lower quantile predicted value and an upper quantile predicted value of each parameter, respectively, the lower quantile and the upper quantile corresponding to a lower bound and an upper bound of a preset confidence level, respectively, and the confidence interval width is a difference value between the upper quantile predicted value and the lower quantile predicted value.
  7. 7. The method according to claim 1, wherein in the effect simulation step, the cornea deformation prediction model calculates a curvature change distribution of the anterior surface of the cornea after wearing the lens by using a finite element analysis method based on a combination of the cornea biomechanical parameters and the recommended lens parameters, and generates the expected shaping effect simulation map presented in a pseudo-color form.
  8. 8. The method according to claim 1, wherein in the effect simulation step, the lens-angle gap distribution model calculates tear layer thickness distribution between the lens rear surface and the cornea front surface based on the recommended lens parameter combination and the cornea geometry parameter set, and generates the post-lens cornea fluorescence staining pattern prediction map according to a mapping relationship between tear layer thickness and fluorescence staining intensity.
  9. 9. The method of claim 1, further comprising the step of feedback updating, after the fitting is completed, of collecting actual lens effect data, associating the actual lens effect data with the recommended lens parameter combination to form a new training sample, and periodically adding the new training sample to the historical fitting success case to incrementally update the gradient-lifting tree integration model.
  10. 10. An intelligent recommendation system for cornea shaping and mirror fitting parameters, which is used for realizing the method of any one of claims 1-9, and is characterized by comprising the following steps: The cornea parameter acquisition module is used for acquiring cornea topographic map raw data of the eye to be tested from the cornea topographic map instrument, extracting a cornea geometric parameter set and acquiring physiological characteristic data of a patient; the feature coding module is used for carrying out standardized processing on the cornea geometric parameter set and the physiological feature data of the patient and constructing a multidimensional input vector; the anomaly detection module is used for respectively calculating an astigmatism risk score, an eccentric risk score, an irregular astigmatism risk score and a keratoconus risk score, obtaining a comprehensive risk score through weighted summation, dividing the eye to be tested into four grades of normal, low risk, medium risk and high risk according to the comprehensive risk score, and outputting expert consultation prompts for the high risk grade; The recommendation prediction module is used for inputting the multidimensional input vector passing through anomaly detection into a gradient lifting tree integrated model and outputting a recommendation lens parameter combination comprising a lens base arc, an adaptation curvature and a lens diameter; the uncertainty quantization module is used for estimating confidence intervals of the recommended lens parameter combinations based on a quantile regression model and generating boundary case prompts; and the effect simulation module is used for generating an expected shaping effect simulation image and a cornea fluorescence staining mode prediction image after wearing the lens.

Description

Intelligent recommendation method and system for cornea shaping mirror test and allocation parameters Technical Field The invention relates to the technical field of ophthalmic medical instrument assistance, in particular to an intelligent recommendation method and system for cornea shaping mirror test and allocation parameters. Background The cornea shaping lens is a specially designed air-permeable hard cornea contact lens, and the shape of the front surface of the cornea is temporarily changed by wearing at night, so that clear vision is obtained without wearing the lens in daytime, and the cornea shaping lens has the function of delaying the progression of myopia of teenagers. Cornea shaping lens fitting is a highly specialized clinical procedure requiring selection of appropriate lens parameters according to the patient's cornea morphology and refractive condition. In the prior art, CN113806908A discloses a parameter processing method based on cornea shaping lens and related equipment. According to the scheme, the cornea parameter map of the cornea to be measured after correction is evaluated by acquiring instantaneous refractive power characteristic data of the cornea to be measured and eye measurement parameters of a patient to be measured, and a height difference parameter prediction result is output according to a linear regression algorithm. According to the scheme, the XGBoost mathematical model is utilized to construct an objective function for training, so that the height difference parameters of the OK lens can be predicted. However, the prior art has the following technical problems that firstly, only a linear regression algorithm is adopted to predict a single height difference parameter, the collaborative optimization requirements of multidimensional test parameters such as a base curve, a fitting curvature, a lens diameter and the like are difficult to fully cover, and other parameters still need to be adjusted by a tester according to experience, secondly, quantitative evaluation on uncertainty of a prediction result is lacking, when an input parameter is positioned in a training sample sparse area, a model possibly gives a prediction result which is excessively self-trusted but has larger actual deviation, the risk of test failure is increased, thirdly, front screening is not carried out on a special cornea morphology, recommended parameters are directly given to complex cases such as high astigmatism, eccentric cornea and the like, the test effect is possibly poor, fourthly, the output result is only a numerical parameter, visual effect preview is lacking, and a fitter is difficult to evaluate the suitability of the recommended parameters before actual test. The problems still have the defects of more test wearing times, long test matching period, strong dependence on experience of a fitter and the like in the conventional cornea shaping lens test matching. Disclosure of Invention Aiming at the technical problems of incomplete recommendation of cornea shaping mirror fitting parameters, lack of uncertainty quantification, lack of screening of special cornea forms and lack of effect preview in the prior art, the invention provides an intelligent recommendation method and system for cornea shaping mirror fitting parameters. The invention provides an intelligent recommendation method for cornea shaping lens fitting parameters, which comprises a cornea parameter acquisition step, a characteristic coding step, an abnormality detection step and an abnormality detection step, wherein cornea topographic map raw data of an eye to be fitted are acquired from cornea topographic map raw data, cornea geometric parameter sets are extracted from cornea topographic map raw data, the cornea geometric parameter sets comprise a flat K value, a steep K value, cornea eccentricity, pupil diameter, cornea transverse diameter and central cornea thickness, patient physiological characteristic data are acquired at the same time, the patient physiological characteristic data comprise diopter, eye axis length, age and astigmatism degree, the characteristic coding step is performed on the cornea geometric parameter sets and the patient physiological characteristic data, the normalized data are spliced and fused to form multidimensional input vectors, the abnormality detection step is performed on the multidimensional input vectors, the abnormality detection models are respectively used for calculating four sub-risk scores of astigmatism risk scores, decentration risk scores and conical cornea risk scores, the irregular astigmatism risk scores are acquired by context correction based on astigmatism degree and combined astigmatism type and pupil diameter, the irregular astigmatism risk scores are correspondingly classified into four sub-risk scores when the eye surface indexes are combined to be matched with high-score to be matched with the clinical risk score to be normalized according to a high-score rule, and a hi