Eye preoperative drug response prediction system based on iris image
By using an iris image-based preoperative ocular drug response prediction system, and leveraging data acquisition and deep learning technologies, the system addresses the inaccuracy of drug response prediction in traditional methods, enabling personalized drug recommendations and optimized preoperative preparation.
Patent Information
- Application Number
- CN202511030012.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-25
- Publication Date
- 2025-11-11
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Traditional drug selection relies on the doctor's clinical experience and the patient's past medication records, making it difficult to accurately predict an individual's preoperative drug response, which increases surgical risks.
A preoperative ocular drug response prediction system based on iris images is adopted. The system acquires patients' iris images and medical records through a data acquisition module, establishes a drug response prediction model, and combines a similarity completion mechanism and deep learning to make accurate predictions and recommend personalized drug regimens.
It enables accurate prediction of preoperative drug response, improves the safety and effectiveness of clinical drug use, reduces trial and error costs, and optimizes the preoperative preparation process.
Smart Images
Figure CN120932803A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical image analysis and drug response prediction technology, and in particular to a preoperative ocular drug response prediction system based on iris images. Background Technology
[0002] In the preoperative preparation for ophthalmic surgery, patients often require pre-treatment with medications, including the use of mydriatics, miotics, or local anesthetics to improve the intraoperative field of vision or reduce postoperative discomfort. However, due to differences in physiological structure, ocular history, and drug sensitivity, different patients often react differently to the same medication, such as poor pupil dilation, increased intraocular pressure, and blurred vision. This introduces uncertainty into surgical preparation and increases intraoperative risks.
[0003] Traditional drug selection relies primarily on physicians' clinical experience and patients' past medication records, making it difficult to accurately predict individual drug responses. Furthermore, the uncertainty of preoperative drug response can lead to surgical interruption or delay, increasing the risk of surgical failure and complications. Therefore, accurately predicting preoperative drug response based on patients' physiological characteristics, especially iris image features highly correlated with drug sensitivity, has become an important research direction for improving the quality of preoperative preparation and surgical safety. Summary of the Invention
[0004] The purpose of this invention is to address the shortcomings of current methods by proposing a preoperative ocular drug response prediction system based on iris images.
[0005] The present invention adopts the following technical solution:
[0006] A preoperative ocular drug response prediction system based on iris images is disclosed. The system includes a data acquisition module, a response prediction module, and a prediction output module. The data acquisition module is used to acquire preoperative iris images of the patient. The response prediction module is used to predict the patient's drug response based on the preoperative iris images. The prediction output module is used to provide a recommended preoperative drug regimen based on the prediction results of the response prediction module.
[0007] The data acquisition module includes an image acquisition unit, an image preprocessing unit, and a patient medical record acquisition unit; the image acquisition unit is used to acquire the patient's preoperative iris image, the image preprocessing unit is used to perform noise reduction, registration, and precise segmentation of the iris region on the patient's preoperative iris image; the patient medical record acquisition unit is used to acquire the patient's ocular medical history.
[0008] The response prediction module includes a model building unit, a prediction unit, and a feedback optimization unit. The model building unit is used to build a drug response prediction model, which is used to predict the patient's preoperative drug response. The prediction unit is used to input the patient's multidimensional physiological structural features into the drug response prediction model to complete the prediction of the patient's preoperative drug response. The feedback optimization unit is used to continuously optimize the drug response prediction model by combining real postoperative data.
[0009] Furthermore, the specific method for establishing the drug response prediction model is as follows:
[0010] S11: Obtain historical ocular medical information from a medical database. This historical ocular medical information includes the patient's ocular medical history record, preoperative iris image information, medication records, and post-medication reaction information. The post-medication reaction information includes postoperative pupillary dilation and constriction and adverse reaction records.
[0011] S12: Construct a training dataset from the historical medical information obtained in the previous step; each piece of training data in the training dataset includes an ocular medical history record, preoperative iris image information, medication record, and post-medication reaction information;
[0012] S13: Construct a drug response prediction model framework and train the model using the training dataset obtained in the previous step. Vectorize the ocular medical history record, preoperative iris image information, and medication record in each training data point as the model input, and vectorize the post-medication response information as the model output.
[0013] Furthermore, in the process of constructing the inspection dataset in step S12, the historical diagnosis and treatment information of the eyes is first classified and organized to obtain complete samples and incomplete samples, and data completion is performed for incomplete samples. The complete sample is a data record sample in which the data in the four dimensions of patient eye medical history record, preoperative iris image information, medication record and post-medication reaction information are complete and intact. The incomplete sample is a sample in which the data record is missing in at least one of the four dimensions of patient eye medical history record, preoperative iris image information, medication record and post-medication reaction information.
[0014] Furthermore, the data completion strategy for incomplete samples is as follows:
[0015] S121: Vectorization encoding is performed on both complete and incomplete samples in historical medical information;
[0016] S122: For each incomplete sample, calculate its similarity to each complete sample, and extract the top m complete samples based on the similarity scores as the basis for completing the incomplete sample; the similarity is calculated as follows:
[0017]
[0018] Among them, C(F) i ,F j F represents the similarity between an incomplete sample i and a complete sample j. i Let F be the feature vector of an incomplete sample i. j Let j be the feature vector of a complete sample j;
[0019] S123: For each incomplete sample, perform data completion using the following formula:
[0020]
[0021] in, F is the completion vector for the missing dimension of an incomplete sample i; mean Let be the average vector of the first m complete samples along the missing dimension of the incomplete sample i; ΔF is the vector of the incomplete sample i. Random numbers within the interval; where α is the perturbation strength coefficient, used to control the perturbation amplitude of the data; β is the similarity adjustment factor, used to control the nonlinear strength of the perturbation in combination with similarity; α and β are set through pre-experimentation; C i Let be the average similarity between the first m complete samples and the incomplete sample i.
[0022] Furthermore, the prediction output module selects multiple candidate medication records by combining conventional medication regimens, and inputs these multiple candidate medication records into the drug response prediction model by combining them with the patient's preoperative iris image and ocular medical history record to obtain post-medication response information for each medication record. By comparing the post-medication response information of multiple medication records with preset ideal post-medication response information, the optimal medication record is selected as the recommended preoperative medication regimen based on the degree of matching.
[0023] The beneficial effects achieved by this invention are:
[0024] This invention achieves accurate prediction of preoperative drug response by efficiently collecting and processing patient preoperative iris images, ocular medical history records, and routine medication regimens. It combines historical data expansion based on a similarity completion mechanism with a deep learning prediction model. Through a time decay weighting mechanism and dynamic updates of continuous feedback optimization units, the model can quickly adapt to the latest clinical data changes. By simulating and comparing multiple candidate medication records, it intelligently recommends personalized preoperative medication regimens, significantly improving the safety and effectiveness of clinical medication, reducing trial-and-error costs, and optimizing the preoperative preparation process. Attached Figure Description
[0025] The invention will be further understood from the following description taken in conjunction with the accompanying drawings. The components in the drawings are not necessarily drawn to scale, but rather the emphasis is on illustrating the principles of the embodiments. In different views, the same reference numerals designate corresponding parts.
[0026] Figure 1 This is a schematic diagram of the overall modules of the present invention.
[0027] Figure 2 This is a schematic diagram illustrating the process of establishing the drug response prediction model of the present invention.
[0028] Figure 3 This is a schematic diagram showing the function variation of the interval boundary of ΔF in this invention under different similarity adjustment factors β.
[0029] Figure 4 This is a schematic diagram of the workflow of the feedback optimization unit of the present invention. Detailed Implementation
[0030] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to its embodiments. It should be understood that the specific embodiments described herein are only for explaining the present invention and are not intended to limit the present invention. Other systems, methods, and / or features of this embodiment will become apparent to those skilled in the art after reviewing the following detailed description. It is intended that all such additional systems, methods, features, and advantages are included within this specification, are included within the scope of the present invention, and are protected by the appended claims. Further features of the disclosed embodiments are described in the following detailed description, and these features will be apparent from the following detailed description.
[0031] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or component referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present patent. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.
[0032] Example 1:
[0033] like Figure 1 As shown, this embodiment provides a preoperative ocular drug response prediction system based on iris images. The system includes a data acquisition module, a response prediction module, and a prediction output module. The data acquisition module is used to acquire the patient's preoperative iris image. The response prediction module is used to predict the patient's drug response based on the preoperative iris image. The prediction output module is used to provide a recommended preoperative drug regimen based on the prediction results of the response prediction module.
[0034] The data acquisition module includes an image acquisition unit, an image preprocessing unit, and a patient medical record acquisition unit; the image acquisition unit is used to acquire the patient's preoperative iris image, the image preprocessing unit is used to perform noise reduction, registration, and precise iris region segmentation on the patient's preoperative iris image; the patient medical record acquisition unit is used to collect the patient's ocular medical history record;
[0035] The response prediction module includes a model building unit, a prediction unit, and a feedback optimization unit. The model building unit is used to build a drug response prediction model, which is used to predict the patient's preoperative drug response. The prediction unit is used to input the patient's multidimensional physiological structural features into the drug response prediction model to complete the prediction of the patient's preoperative drug response. The feedback optimization unit is used to continuously optimize the drug response prediction model by combining real postoperative data.
[0036] Furthermore, such as Figure 2 As shown, the specific method for establishing the drug response prediction model is as follows:
[0037] S11: Obtain historical ocular medical information from a medical database. This historical ocular medical information includes the patient's ocular medical history record, preoperative iris image information, medication records, and post-medication reaction information. The post-medication reaction information includes postoperative pupillary dilation and constriction and adverse reaction records.
[0038] S12: Construct a training dataset from the historical medical information obtained in the previous step; each piece of training data in the training dataset includes an ocular medical history record, preoperative iris image information, medication record, and post-medication reaction information;
[0039] S13: Construct a drug response prediction model framework and use the training dataset obtained in the previous step to train and build the model. Vectorize the ocular medical history record, preoperative iris image information and medication record in each training data as the model input, and vectorize the post-medication response information as the model output.
[0040] Furthermore, in the process of constructing the inspection dataset in step S12, the historical medical information of the eyes is first classified and organized to obtain complete and incomplete samples, and data completion is performed for the incomplete samples. The complete sample is a data record sample in which all four dimensions of the patient's eye medical history record, preoperative iris image information, medication record, and post-medication response information are complete and intact. The incomplete sample is a sample in which at least one of the four dimensions of the patient's eye medical history record, preoperative iris image information, medication record, and post-medication response information is missing data. Figure 3 As shown, the data completion strategy for incomplete samples is as follows:
[0041] S121: Vectorization encoding is performed on both complete and incomplete samples in historical medical information;
[0042] S122: For each incomplete sample, calculate its similarity to each complete sample, and extract the top m complete samples based on the similarity scores as the basis for completing the incomplete sample; the similarity is calculated as follows:
[0043]
[0044] Among them, C(F) i ,F j F represents the similarity between an incomplete sample i and a complete sample j. i Let F be the feature vector of an incomplete sample i. j Let j be the feature vector of a complete sample j;
[0045] S123: For each incomplete sample, perform data completion using the following formula:
[0046]
[0047] in, F is the completion vector for the missing dimension of an incomplete sample i; meanLet be the average vector of the first m complete samples along the missing dimension of the incomplete sample i; ΔF is the vector of the incomplete sample i. Random numbers within the interval; where α is the perturbation strength coefficient, used to control the perturbation amplitude of the data; β is the similarity adjustment factor, used to control the nonlinear strength of the perturbation in combination with similarity; α and β are set through pre-experimentation; C i Let be the average similarity between the first m complete samples and the incomplete sample i.
[0048] By combining the similarity of training data with the above methods for data completion, the negative impact of incomplete samples on model training is reduced, and the completeness of the data is greatly improved. Furthermore, the introduction of non-linear perturbation based on similarity during the completion process avoids all completed data from being too concentrated on the mean, thereby improving the diversity of data and the generalization ability of the model. Filling in the missing data makes the model more robust during the model building phase and better able to cope with the lack of actual clinical data.
[0049] Furthermore, during the training process, the loss function of the drug response prediction model constructed in step S13 satisfies the following formula:
[0050]
[0051] Among them, L all Let n be the overall loss function of the model, and ω be the total number of training data. l The weight coefficients for the l-th training data are... The loss value for the l-th training data; y is the model's predicted output for the l-th training data. l This represents the actual output of the l-th training data, i.e., the post-medication response information in that training data; for ω l satisfy:
[0052]
[0053] Where, d l Let d be the number of feature vector dimensions of the l-th training data before data completion. total The number of dimensions of the feature vector of the complete sample;
[0054] The above methods effectively balance the contributions of the original data and the completed data during the training process, ensuring that the model maintains stable performance even in complex clinical data environments.
[0055] Furthermore, the prediction output module selects multiple candidate medication records by combining conventional medication regimens, and inputs these multiple candidate medication records into the drug response prediction model by combining them with the patient's preoperative iris image and ocular medical history record to obtain post-medication response information for each medication record. By comparing the post-medication response information of multiple medication records with preset ideal post-medication response information, the optimal medication record is selected as the recommended preoperative medication regimen based on the degree of matching.
[0056] Example 2:
[0057] This embodiment should be understood to include at least all the features of any of the foregoing embodiments, and to further improve upon them;
[0058] This embodiment provides a preoperative ocular drug response prediction system based on iris images. The system includes a data acquisition module, a feature extraction module, a response prediction module, and a prediction output module. The data acquisition module is used to acquire the patient's preoperative iris image. The feature extraction module is used to extract multidimensional physiological structural features from the patient's iris image. The response prediction module is used to predict the patient's drug response based on the multidimensional physiological structural features. The prediction output module is used to provide a recommended preoperative drug regimen based on the prediction results from the response prediction module.
[0059] The response prediction module includes a model building unit, a prediction unit, and a feedback optimization unit. The model building unit is used to build a drug response prediction model, which is used to predict the patient's preoperative drug response. The prediction unit is used to input the patient's multidimensional physiological structural features into the drug response prediction model to complete the prediction of the patient's preoperative drug response. The feedback optimization unit is used to continuously optimize the drug response prediction model by combining real postoperative data.
[0060] Furthermore, such as Figure 4 As shown, the feedback optimization unit updates and optimizes the drug response prediction model in the following ways:
[0061] S21: Store the training data with complete samples from the original medical database of historical eye diagnosis and treatment information into the benchmark sample library;
[0062] S22: Continuously acquire training data with complete samples from the medical system and store it in a temporary cache. When the accumulated training data in the temporary cache reaches a specified scale, trigger the update and optimization of the model.
[0063] S23: When model updates and optimizations are triggered, training data is extracted from the benchmark sample library and the temporary cache according to a preset ratio to jointly train the drug response prediction model. The loss function during training satisfies:
[0064]
[0065] Among them, L all,update Let τ be the loss function of the model during joint training, p be the total amount of training data during joint training, and τ be the loss function of the model during joint training. u This refers to the time decay weight of the u-th training data point during joint training. Let be the loss value of the u-th training data during joint training. y is the model's predicted output for the u-th training data. u For the u-th training data, this is the actual output, i.e., the post-medication response information in that training data; for τ u satisfy:
[0066]
[0067] Among them, T current T is the timestamp for the joint training of the model. u Let δ be the timestamp corresponding to the acquisition of the u-th training data, and δ be the time adjustment factor used to control the impact of the training data acquisition time on the time decay weight, which is set through pre-experimentation;
[0068] S24: After joint training is completed, transfer the training data in the temporary cache to the benchmark sample library and clear the temporary cache. Repeat steps S22 to S23 to continuously update and optimize the model.
[0069] By continuously acquiring training data with complete samples and continuously updating and optimizing the model, the accuracy and personalized adaptability of the model in drug response prediction are improved. By introducing a time decay weight mechanism to dynamically adjust the contribution ratio of new and old data, the influence of the latest data is highlighted, ensuring that the model can quickly adapt to clinical changes. Through a cyclic optimization mechanism, the model is continuously iterated and updated, thereby ensuring that the drug response prediction model always maintains high accuracy and robustness.
[0070] The content disclosed above is only a preferred and feasible embodiment of the present invention, and is not intended to limit the scope of protection of the present invention. Therefore, all equivalent technical changes made based on the content of the present invention specification and drawings are included within the scope of protection of the present invention. Furthermore, the elements therein can be updated as technology develops.
Claims
1. A preoperative ocular drug response prediction system based on iris images, characterized in that, The system includes a data acquisition module, a response prediction module, and a prediction output module; the data acquisition module is used to acquire the patient's preoperative iris image; the response prediction module is used to predict the patient's drug response based on the preoperative iris image; and the prediction output module is used to provide a recommended preoperative medication plan based on the prediction results of the response prediction module. The data acquisition module includes an image acquisition unit, an image preprocessing unit, and a patient medical record acquisition unit; the image acquisition unit is used to acquire the patient's preoperative iris image, the image preprocessing unit is used to perform noise reduction, registration, and precise iris region segmentation on the patient's preoperative iris image; the patient medical record acquisition unit is used to collect the patient's ocular medical history record; The response prediction module includes a model building unit, a prediction unit, and a feedback optimization unit. The model building unit is used to build a drug response prediction model, which is used to predict the patient's preoperative drug response. The prediction unit is used to input the patient's multidimensional physiological structural features into the drug response prediction model to complete the prediction of the patient's preoperative drug response. The feedback optimization unit is used to continuously optimize the drug response prediction model by combining real postoperative data.
2. The ocular preoperative drug response prediction system based on iris images according to claim 1, characterized in that, The specific method for establishing the drug response prediction model is as follows: S11: Obtain historical ocular medical information from a medical database. This historical ocular medical information includes the patient's ocular medical history record, preoperative iris image information, medication records, and post-medication reaction information. The post-medication reaction information includes postoperative pupillary dilation and constriction and adverse reaction records. S12: Construct a training dataset from the historical medical information obtained in the previous step; each piece of training data in the training dataset includes an ocular medical history record, preoperative iris image information, medication record, and post-medication reaction information; S13: Construct a drug response prediction model framework and train the model using the training dataset obtained in the previous step. Vectorize the ocular medical history record, preoperative iris image information, and medication record in each training data point as the model input, and vectorize the post-medication response information as the model output.
3. The ocular preoperative drug response prediction system based on iris images according to claim 2, characterized in that, In step S12, during the construction of the inspection dataset, the historical medical information of the eyes is first classified and organized to obtain complete and incomplete samples, and data completion is performed for the incomplete samples. The complete sample is a data record sample in which the data in the four dimensions of patient eye medical history record, preoperative iris image information, medication record and post-medication reaction information are complete and intact. The incomplete sample is a sample in which the data record is missing in at least one of the four dimensions of patient eye medical history record, preoperative iris image information, medication record and post-medication reaction information.
4. The ocular preoperative drug response prediction system based on iris images according to claim 3, characterized in that, The data completion strategy for incomplete samples is as follows: S121: Vectorization encoding is performed on both complete and incomplete samples in historical medical information; S122: For each incomplete sample, calculate its similarity to each complete sample, and extract the top m complete samples based on the similarity scores as the basis for completing the incomplete sample; the similarity is calculated as follows: Among them, C(F) i ,F j F represents the similarity between an incomplete sample i and a complete sample j. i Let F be the feature vector of an incomplete sample i. j Let j be the feature vector of a complete sample j; S123: For each incomplete sample, perform data completion using the following formula: in, F is the completion vector for the missing dimension of an incomplete sample i; mean Let be the average vector of the first m complete samples along the missing dimension of the incomplete sample i; ΔF is the vector of the incomplete sample i. Random numbers within the interval; where α is the perturbation strength coefficient, used to control the perturbation amplitude of the data; β is the similarity adjustment factor, used to control the nonlinear strength of the perturbation in combination with similarity; α and β are set through pre-experimentation; C i Let be the average similarity between the first m complete samples and the incomplete sample i.
5. The ocular preoperative drug response prediction system based on iris images according to claim 1, characterized in that, The prediction output module selects multiple candidate medication records by combining conventional medication regimens, and inputs these multiple candidate medication records into the drug response prediction model by combining them with the patient's preoperative iris image and ocular medical history record. The module obtains the post-medication response information of each medication record by comparing the post-medication response information of multiple medication records with the preset ideal post-medication response information, and selects the optimal medication record as the recommended preoperative medication regimen based on the degree of matching.