Post ptosis surgery corneal intervention regimen recommendation system based on multidimensional assessment

Abstract

The application discloses a post-ptosis corneal intervention scheme recommendation system based on multi-dimensional evaluation, and relates to the technical field of personalized service recommendation. The system comprises a state acquisition module, a monitoring module, an environment acquisition module, a recording module, an analysis module and an output module. Objective data of four dimensions, including a corneal injury grade, an eyelid gap quantitative grade, an environmental humidity and a compliance score, are combined and input. A gradient boosting decision tree model is used to calculate a continuous corneal comprehensive state index, so that the one-sidedness of single-index evaluation is overcome. A feedback learning module is set, which can use reevaluation data of a patient, add new samples to a training set to update a regression model, and dynamically adjust a risk decision threshold. The mechanism enables the system to continuously evolve based on individual reaction data, and continuously improve the matching degree of a recommended scheme and an actual condition of the patient.

Description

Technical Field

[0001] This invention relates to the field of personalized service recommendation technology, specifically a corneal intervention recommendation system based on multidimensional assessment after ptosis surgery. Background Technology

[0002] Corneal exposure risk is common in patients with incomplete eyelid closure, such as those with facial nerve palsy, deep coma, eyelid trauma, or eyelid closure disorders during the postoperative recovery period. Prolonged corneal exposure can lead to excessive tear evaporation, corneal epithelial shedding, and consequently, serious complications such as exposure keratitis, corneal ulcers, and even perforation. For these patients, the key is to promptly select appropriate interventions based on the level of risk, such as physical eye shielding devices, antibacterial ointments, or moisturizing gels.

[0003] Currently, the assessment of corneal exposure risk relies primarily on the subjective judgment of healthcare professionals. Routine methods include observing corneal fluorescein staining morphology under a slit lamp and estimating the degree of corneal damage based on the physician's personal experience. The evaluation of eyelid closure is often done through visual observation or simple ruler-and-compass measurements, lacking continuous and quantitative monitoring methods. Clinical intervention plans typically rely on general guidelines or physicians' empirical choices, lacking auxiliary tools to integrate and analyze multidimensional data such as the objective degree of corneal damage, eyelid incompleteness parameters, environmental humidity, and actual patient medication adherence, and to stratify risk accordingly. Subjective assessment methods struggle to guarantee consistency and timeliness, potentially leading to high-risk patients not receiving timely and effective interventions, or overuse of medications in low-risk patients.

[0004] Therefore, there is a need for a computer-aided system that can integrate multidimensional objective parameters, automatically quantify corneal exposure risk, and assist in recommending graded intervention plans, so as to provide standardized decision-making references for medical staff while reducing subjective dependence. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a corneal intervention recommendation system based on multidimensional assessment after ptosis surgery, thus solving the problems mentioned in the background section.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows: Firstly, a recommendation system for corneal intervention programs after ptosis surgery based on multidimensional assessment includes: The state acquisition module is used to acquire corneal fluorescein staining images. In the corneal fluorescein staining images, the green channel image is extracted, and the green channel image is adaptively thresholded by the maximum inter-class variance method to obtain the fluorescent stained region. By r=A s / A c Calculate the area A of the fluorescently colored region. sThe ratio r of A to the total corneal area, where A c This represents the total area of ​​the cornea. Based on the preset first-level mapping table, the ratio r is converted into corneal fluorescence staining feature level G. c The first hierarchical mapping table is defined as follows: when r < 0.01, G c =1; when 0.01≤r<0.05, G c =2; when 0.05≤r<0.15, G c =3; when 0.15≤r<0.30, G c =4; when r≥0.30, G c =5; The monitoring module acquires the upper eyelid displacement signal sequence of the target object within a preset time period, and determines the quantification level of the eyelid gap based on the upper eyelid displacement signal sequence, as follows: The difference signal between the strain-sensitive capacitor and the reference capacitor is acquired in real time. Based on the differential reference obtained from zero-point calibration and the pre-calibrated capacitance-displacement mapping model, the capacitance difference is converted into the vertical distance d of the upper eyelid relative to the closed position. t1 ; Within a preset time period, a sequence of effective displacement signals of length T1 is acquired at a fixed sampling frequency, and P1=N1 is used to... over / T1, calculate vertical distance d t1 Exceeding the preset closure threshold d th The percentage of valid sampling points P1 out of the total number of valid sampling points T1, and the percentage of valid sampling points exceeding the preset closure threshold d. th The vertical distance d corresponding to the sampling point t1 average value d P1 Among them, N1 over For distances exceeding d th The number of sampling points; where t1=1,2,......,T1; The eyelid gap quantification level G is generated based on the preset second-level mapping table. e When P1 > 0.30 and d P1 When >4 mm, G e =3; when P1 > 0.30 and d P1 When ≤4 mm, or 0.05≤P1≤0.30 and d P1 When >3 mm, G e =2; in other cases, G e =1; if there is insufficient valid data, then G e Set to invalid.

[0007] Alternatively, by acquiring an eye image sequence, extracting consecutive T2 frames, and detecting the vertical distance d between the midpoints of the upper and lower eyelid margins in each frame. t2Where t2 = 1, 2, ..., T2; By P2=N2 over / T2, calculate vertical distance d t2 Exceeding the preset closure threshold d th The percentage of frames P2 that exceed the total number of valid frames T2, and the percentage of frames exceeding the preset closure threshold d. th The vertical distance d corresponding to the frame t2 average value d P2 N2 over For distances exceeding d th The number of frames; According to the preset second-level mapping table, based on the frame percentage P2 and the mean d P2 Generate eyelid gap quantification level G e The second hierarchical mapping table is defined as follows: when P2 > 0.30 and d P2 When >4 mm, G e =3; when P2 > 0.30 and d P2 When ≤4 mm, or when 0.05≤P2≤0.30 and d P2 When >3 mm, G e =2; in other cases, G e =1; The environment acquisition module is used to acquire ambient relative humidity data H. The recording module is used to acquire record data containing medication timestamps and to count the actual number of medication uses (n) within the evaluation period. a According to the preset number of medication doses n y ,pass Calculate compliance score S; Analysis module, used to classify corneal fluorescence staining feature level G c Quantitative grade of eyelid space G e The environmental humidity data H and compliance score S are concatenated into an input feature vector X; the input feature vector X is then input into a pre-trained regression model based on gradient boosting decision tree to calculate the corneal comprehensive state index R. Based on the threshold range of the corneal comprehensive condition index R, the target intervention program is determined from multiple intervention programs that are pre-associated with different combinations of items; The threshold interval is divided by the first threshold TY1=70 and the second threshold TY2=40. The multiple intervention plans are corresponding to the first intervention plan, the second intervention plan and the third intervention plan. When R≥TY1, the first intervention plan is determined as the target intervention plan; when R<TY2, the second intervention plan is determined as the target intervention plan; when TY2≤R<TY1, the third intervention plan is determined as the target intervention plan. The output module is used to output the target intervention plan and to generate reassessment time interval information based on the threshold range of the corneal comprehensive condition index R.

[0008] As a further aspect of the present invention, the eyelid gap quantification level is calculated based on the upper eyelid displacement signal sequence. The monitoring module includes a flexible capacitive strain sensing patch attached to the upper eyelid skin. The flexible capacitive strain sensing patch is composed of a first flexible electrode layer, an elastic dielectric layer, and a second flexible electrode layer stacked from top to bottom, and its overall thickness is less than 200 micrometers. Flexible capacitive strain sensor patch is used to generate capacitance changes based on the stretching and deformation of the upper eyelid skin; The contact surface of the flexible capacitive strain sensor patch is also provided with multiple contact impedance detection electrodes arranged in an array. The sensing patch includes a strain-sensitive capacitor region and a non-strain-sensitive reference capacitor region integrated with the substrate. The elastic dielectric layer of the reference capacitor region is constrained to be incompressible and is only sensitive to temperature and humidity.

[0009] As a further aspect of this invention, the training method for the regression model based on gradient boosting decision trees is as follows: Obtain a historical sample set, each sample containing corneal fluorescence staining feature level G. c,i Quantitative grade of eyelid space G e,i Environmental humidity data H i Compliance score S i The sample input vector X formed by splicing i And the corneal risk value Y, as indicated by the physician based on the follow-up results. i i = 1, 2, ..., N, where N is the total number of samples; The residuals of the current model are fitted by iteratively constructing a regression decision tree. The output of the decision tree generated in each iteration is multiplied by the preset learning rate and then added to the model to obtain the final regression model.

[0010] As a further aspect of the present invention, the re-evaluation time interval information generated by the output module includes: when R≥TY1, the generated monitoring suggestion interval is no more than the preset interval time TJ; when TY2≤R<TY1, the generated monitoring suggestion interval is γ1×TJ; when R<TY2, the generated monitoring suggestion interval is no less than γ2×TJ; wherein, γ1 is greater than or equal to 1, and γ2 is greater than γ1, and γ1 and γ2 are preset re-evaluation interval amplification coefficients.

[0011] As a further aspect of the invention, a feedback learning module is also included, which is used to acquire secondary corneal fluorescein staining images of the patient after intervention according to the target intervention plan, and to determine the secondary corneal fluorescein staining feature level G. c '; Utilizing secondary corneal fluorescence staining characteristic grade Gc 'and the corresponding eyelid space quantification level G for this reassessment' e ', Environmental humidity data H' and compliance score S', update the regression model by adding new samples to the historical sample set and re-executing the training process, and / or adjust the first threshold TY1 and / or the second threshold TY2; The method for adjusting the second threshold TY2 is as follows: calculate the adjustment amount. , To adjust the step size coefficient as preset, update TY2 to Furthermore, if the difference between the updated TY2 and the first threshold TY1 is less than a preset minimum interval, then the first threshold TY1 is simultaneously adjusted to... Add the preset minimum interval.

[0012] Secondly, a method for recommending corneal intervention plans after ptosis surgery based on multidimensional assessment is provided. This method is implemented through a system for recommending corneal intervention plans after ptosis surgery based on multidimensional assessment, and includes the following steps: Step S1: Obtain the corneal fluorescein staining image. Extract the green channel image from the corneal fluorescein staining image. Use the maximum inter-class variance method to perform adaptive threshold segmentation on the green channel image to obtain the fluorescent stained region; then, using r=A... s / A c Calculate the area A of the fluorescently colored region. s The ratio r of A to the total corneal area, where A c The total corneal area; according to the preset first-level mapping table, the ratio r is converted into the corneal fluorescence staining feature level G. c ; Step S2: Obtain the upper eyelid displacement signal sequence of the target object within a preset time period, and calculate the quantization level of the eyelid gap based on the upper eyelid displacement signal sequence; Alternatively, acquire an eye image sequence, extract consecutive T2 frames, and detect the vertical distance d between the midpoints of the upper and lower eyelid margins in each frame. t2 Where t2 = 1, 2, ..., T2; through P2 = N2 over / T2, calculate vertical distance d t2 Exceeding the preset closure threshold d th The percentage of frames P2 that exceed the total number of valid frames T2, and the percentage of frames exceeding the preset closure threshold d. th The vertical distance d corresponding to the frame t2 average value d P2 N2 over For distances exceeding d th The number of frames; based on the preset second-level mapping table, and based on the frame percentage P and the mean d P2 Generate eyelid gap quantification level G e; Step S3: Obtain the ambient relative humidity data H.

[0013] Step S4: Obtain record data containing medication timestamps and count the actual number of medication uses n within the evaluation period. a According to the preset number of medication doses n y ,pass Calculate compliance score S; Step S5: Assign corneal fluorescence staining feature level G c Quantitative grade of eyelid space G e The environmental humidity data H and compliance score S are concatenated to form the input feature vector X; Step S6: Input the input feature vector X into the pre-trained regression model based on gradient boosting decision tree, and calculate the corneal comprehensive state index R; Step S7: Based on the threshold range of the corneal comprehensive state index R, determine the target intervention plan from multiple intervention plans that are pre-associated with different combinations of items; Step S8: Output the target intervention plan and generate information on the reassessment time interval.

[0014] Thirdly, a computer-readable storage medium containing a program that, when executed by a processor, implements a corneal intervention recommendation system based on multidimensional assessment after ptosis surgery.

[0015] This invention provides a system for recommending corneal intervention plans after ptosis surgery based on multidimensional assessment. Compared with existing technologies, it has the following advantages: By combining objective data from four dimensions—corneal damage level, eyelid gap quantification level, environmental humidity, and compliance score—a gradient boosting decision tree model is used to calculate a continuous comprehensive corneal status index. This overcomes the limitations of single-indicator assessment and improves the accuracy and comprehensiveness of risk assessment.

[0016] The designed feedback learning module utilizes patient reassessment data to add new samples to the training set to update the regression model and dynamically adjust the risk decision threshold. This mechanism enables the system to continuously evolve based on individual response data, constantly improving the match between recommended plans and the patient's actual condition.

[0017] In addition to providing intervention plans, the output module also generates differentiated reassessment interval suggestions based on the risk index, providing a reference for medical staff to reasonably arrange subsequent observations. Attached Figure Description

[0018] Figure 1 This is a system block diagram of the corneal intervention program recommendation system based on multidimensional assessment after ptosis surgery, which is based on the present invention.

[0019] Figure 2 This is a flowchart illustrating the corneal intervention recommendation system based on multidimensional assessment after ptosis surgery, as described in this invention.

[0020] Figure 3 This is a schematic diagram of the structure of the flexible capacitive strain sensing patch in this invention.

[0021] In the figure, A: non-strain-sensitive reference capacitor region; B: strain-sensitive capacitor region; C: contact impedance detection electrode. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please see Figures 1 to 3 As shown, the embodiments of the present invention provide the following technical solutions: Example 1 This invention is a corneal intervention recommendation system based on multidimensional assessment after ptosis surgery, comprising: a status acquisition module, a monitoring module, an environment acquisition module, a recording module, an analysis module, and an output module.

[0024] The state acquisition module is used to acquire corneal fluorescein staining images. In the corneal fluorescein staining images, the green channel image is extracted, and the green channel image is adaptively thresholded by the maximum inter-class variance method to obtain the fluorescent stained region. By r=A s / A c Calculate the area A of the fluorescently colored region. s The ratio r of A to the total corneal area, where A c This represents the total area of ​​the cornea. Based on the preset first-level mapping table, the ratio r is converted into corneal fluorescence staining feature level G. c In this embodiment, the first hierarchical mapping table is defined as follows: when r < 0.01, G c =1; when 0.01≤r<0.05, G c =2; when 0.05≤r<0.15, G c =3; when 0.15≤r<0.30, G c =4; when r≥0.30, G c =5.

[0025] In practice, the state acquisition module is connected to the slit-lamp microscope digital imaging system to receive RGB format corneal fluorescein-stained images. The G channel component is extracted from the RGB image to obtain a green channel grayscale image. The optimal threshold for this grayscale image is calculated using the maximum inter-class variance method, and binarization is performed based on the optimal threshold. The bright pixels are the fluorescently stained regions. The area A is obtained by calculating the number of pixels in this fluorescently stained region. s .

[0026] The corneal limbus is located using Hough circular transform, the circular region of the cornea is determined, and its pixel area is calculated as the total corneal area A. c By r=A s / A c The ratio is calculated, and the corneal fluorescence tinting feature level Gc is output based on the preset first-level mapping table.

[0027] The monitoring module is used to acquire the upper eyelid displacement signal sequence of the target object within a preset time period, and calculates the quantification level of the eyelid gap based on the upper eyelid displacement signal sequence as follows: The difference signal between the strain-sensitive capacitor and the reference capacitor is acquired in real time. Based on the differential reference obtained from zero-point calibration and the pre-calibrated capacitance-displacement mapping model, the capacitance difference is converted into the vertical distance d of the upper eyelid relative to the closed position. t1 ; Within a preset time period, a sequence of effective displacement signals of length T1 is acquired at a fixed sampling frequency, and P1=N1 is used to... over / T1, calculate vertical distance d t1 Exceeding the preset closure threshold d th The percentage of valid sampling points P1 out of the total number of valid sampling points T1, and the percentage of valid sampling points exceeding the preset closure threshold d. th The vertical distance d corresponding to the sampling point t1 average value d P1 Among them, N1 over For distances exceeding d th The number of sampling points; where t1=1,2,......,T1; The eyelid gap quantification level Ge is generated according to the preset second-level mapping table: when P1 > 0.30 and dP1 > 4 mm, Ge = 3; when P1 > 0.30 and dP1 ≤ 4 mm, or 0.05 ≤ P1 ≤ 0.30 and dP1 > 3 mm, Ge = 2; in other cases, Ge = 1; if there is insufficient valid data, Ge is set to invalid.

[0028] In this embodiment, the pre-calibration method of the capacitance-displacement mapping model is as follows: A sensor patch integrating a strain-sensitive capacitor region and a non-strain-sensitive reference capacitor region was attached to the surface of a standard eyelid model, and the temperature was set to 35°C and the relative humidity to 50% to simulate the microenvironment of the eyelid skin.

[0029] To control the upper eyelid of the standard eyelid model to be in a midline closed position, by... Determine the differential reference ΔC under the initial state. 0，c Among them, C a，0 The capacitance measurement value of the strain-sensitive capacitance region is given in the initial state, i.e., when the upper eyelid of the model is in the midline closed position; C b，0 The capacitance measurement values ​​of the non-strain-sensitive reference capacitance region under the same initial state; Then, the upper eyelid is lifted in 0.5 mm increments up to 12 mm. After each step is stabilized, the lifting displacement x and C are simultaneously collected. a and C b The steady-state value, through The differential capacitance change ΔC1 is calculated in real time; where x is the lifting displacement of the upper eyelid model relative to the closed position, increasing from 0 in steps of 0.5 mm to 12 mm; C a This refers to the real-time capacitance values ​​of the strain-sensitive capacitance region acquired after each displacement x has stabilized; C b This represents the real-time capacitance value of the non-strain-sensitive reference capacitor region acquired under the same conditions.

[0030] A set of (x, ΔC1) data pairs is obtained, and cubic spline interpolation is performed to fit the data, yielding the mapping function d = f(ΔC1), which is the capacitance-displacement mapping model. This function f is stored in the microcontroller. Specifically, the piecewise polynomial coefficients obtained from cubic spline interpolation or the lookup table obtained through dense sampling can be stored in the microcontroller's non-volatile memory. During actual operation, after inputting the real-time differential signal, the corresponding upper eyelid vertical distance is directly output through table lookup or polynomial calculation.

[0031] In practical use, the real-time differential signal ΔC(t) is subtracted from the field zero-point differential reference ΔC0 and then directly substituted into the function, i.e. Where ΔC(t) is the differential capacitance signal acquired in real time, and ΔC(t) = C sen (t)-C ref (t), which is the difference between the strain-sensitive capacitor and the reference capacitor at time t; ΔC0 is the on-site zero-point differential reference, which represents the ΔC(t) value measured when the patient keeps their eyes completely closed after wearing the patch, and represents the initial differential offset in the usage scenario.

[0032] It should be noted that the mapping function f obtained during the calibration process reflects the change in differential capacitance relative to the initial zero point ΔC. 0,cThe relationship with upper eyelid displacement. In practical use, the zero-point differential reference measured when the patient wears the flexible capacitive strain sensor patch and keeps their eyes completely closed is ΔC0. Since the capacitance-displacement characteristics of the sensor patch have good linear consistency within the working range, the vertical distance of the upper eyelid relative to the closed position can be obtained by subtracting ΔC0 from the real-time acquired differential signal and directly substituting it into the function f.

[0033] During use, after the patient wears the flexible capacitive strain sensor patch in the standard manner, they are instructed to gently close their eyes and hold this position for 3-5 seconds. In this fully closed-eye state, differential capacitance signals are collected for a specified duration, and their average value is calculated and stored as the on-site zero-point differential reference ΔC0. Subsequent real-time displacement calculations are all based on and corrected for this ΔC0.

[0034] Please see Figure 3 The monitoring module includes a flexible capacitive strain sensing patch attached to the skin of the upper eyelid. The flexible capacitive strain sensing patch is composed of a first flexible electrode layer, an elastic dielectric layer and a second flexible electrode layer stacked from top to bottom, and its overall thickness is less than 200 micrometers. A flexible capacitive strain sensor patch is used to detect capacitance changes based on the stretching deformation of the upper eyelid skin. The contact surface of the patch also features an array of multiple contact impedance detection electrodes. The sensor patch includes a strain-sensitive capacitor region and a non-strain-sensitive reference capacitor region integrated into the substrate, such as... Figure 3 As shown, point A is the non-strain-sensitive reference capacitor region, point B is the strain-sensitive capacitor region, and point C is the contact impedance detection electrode. Among them, the contact impedance detection electrode is used to monitor the contact status between the patch and the skin in real time; when the contact impedance detected by any contact impedance detection electrode exceeds the preset contact impedance threshold, a poor contact prompt signal is generated to remind the patch position to be adjusted.

[0035] In this embodiment, the strain-sensitive capacitor region occupies most of the area of ​​the flexible capacitor strain sensing patch. The elastic dielectric layer in this region can be freely stretched and deformed, and the capacitance value changes with eyelid movement and temperature and humidity changes.

[0036] The non-strain-sensitive reference capacitance region is located at one end of the flexible capacitive strain sensing patch and occupies a small area. The elastic dielectric layer of the reference capacitance region is constrained to be incompressible and is only sensitive to temperature and humidity. A rigid mesh framework or a thickened array of photoresist pillars is embedded within the elastic dielectric layer of this region, preventing it from compressing or deforming when the flexible capacitive strain sensing patch is stretched. Therefore, its capacitance value is almost unaffected by eyelid movement and is only sensitive to temperature and humidity changes.

[0037] The environment acquisition module is used to acquire relative humidity data H. In this embodiment, the relative humidity data is the measurement value read by an electronic hygrometer in the ward, expressed as a percentage.

[0038] The recording module is used to acquire record data containing medication timestamps and to count the actual number of medication uses (n) within the evaluation period. a According to the preset number of medication doses n y ,pass Calculate the compliance score S; In specific implementation, the recording module acquires recording data including medication timestamps. In this embodiment, the recording data is acquired in two ways. First, it is automatically generated and uploaded by a smart medicine bottle device that wirelessly communicates with the system. This smart medicine bottle has a built-in cap opening detection sensor and communication module. When the patient opens the bottle to take medication, it automatically records the timestamp of this opening and sends it to the recording module as a medication record. The second method is to acquire recording data from a medication log manually entered by the patient. (Substituting the actual number of medication uses into...) This allows you to calculate the compliance score S, which ranges from 0 to 1.

[0039] In this embodiment, the compliance score is normalized and non-negative to ensure that the score is always between 0 and 1, thus avoiding outliers caused by overdosing.

[0040] Analysis module, used to classify corneal fluorescence staining feature level G c Quantitative grade of eyelid space G e The environmental humidity data H and compliance score S are concatenated into an input feature vector X; the input feature vector X is then input into a pre-trained regression model based on gradient boosting decision tree to calculate the corneal comprehensive state index R. Based on the threshold range of the corneal comprehensive condition index R, the target intervention program is determined from multiple intervention programs that are pre-associated with different combinations of items; In this embodiment, the threshold interval is divided by a first threshold TY1=70 and a second threshold TY2=40; the multiple intervention programs include a first intervention program, a second intervention program, and a third intervention program; When R≥TY1, the first intervention plan is determined as the target intervention plan; in this embodiment, it is assumed that the item combination corresponding to the first intervention plan includes the use of an eye physical occlusion device and instructions for artificial tear rinsing; When R < TY2, the second intervention plan is determined as the target intervention plan; in this embodiment, it is assumed that the item combination corresponding to the second intervention plan includes antibacterial cream; When TY2≤R<TY1, the third intervention plan is determined as the target intervention plan; in this embodiment, it is assumed that the item combination corresponding to the third intervention plan includes moisturizing gel.

[0041] In actual use, the corneal fluorescence staining characteristic level G c Quantitative grade of eyelid space G e The environmental humidity data H and compliance score S are normalized to form the input feature vector X, which is then input to a gradient boosting decision tree-based regression model deployed on an FPGA or GPU. The regression model consists of 100 regression trees with a learning rate of 0.1 and outputs a continuous numerical value R.

[0042] The decision unit reads the first threshold TY1=70 and the second threshold TY2=40, and determines the threshold range into which R falls: if R≥70, output the first intervention plan instruction; if R<40, output the second intervention plan instruction; if 40≤R<70, output the third intervention plan instruction.

[0043] The identification codes for the item combinations are pre-stored in a lookup table.

[0044] The output module is used to output the target intervention plan and to generate reassessment time interval information based on the preset threshold range of the corneal comprehensive condition index R.

[0045] The output module displays the determined target intervention plan in text and icon form on the nurse station terminal screen and pushes it to the attending physician's handheld device simultaneously.

[0046] The re-evaluation time interval information generated by the output module includes: when R≥TY1, the generated monitoring recommendation interval is no more than the preset interval time TJ; when TY2≤R<TY1, the generated monitoring recommendation interval is γ1×TJ; when R<TY2, the generated monitoring recommendation interval is no less than γ2×TJ, where γ1 and γ2 are preset re-evaluation interval amplification coefficients.

[0047] In this embodiment, TJ is 24 hours, γ1 is 2, and γ2 is 3. The generated re-evaluation time interval information is as follows: when R≥TY1, the monitoring recommendation interval is no more than 24 hours; when TY2≤R<TY1, the monitoring recommendation interval is 48 hours; and when R<TY2, the monitoring recommendation interval is no less than 72 hours.

[0048] In the embodiments of the present invention, the collection of patient data is based on informed consent, and the data processing strictly follows relevant privacy protection regulations.

[0049] This embodiment also provides a method for recommending corneal intervention plans after ptosis surgery based on multidimensional assessment. This method is implemented through a corneal intervention plan recommendation system after ptosis surgery based on multidimensional assessment, and includes the following steps: Step S1: Obtain the corneal fluorescein staining image. Extract the green channel image from the corneal fluorescein staining image. Use the maximum inter-class variance method to perform adaptive threshold segmentation on the green channel image to obtain the fluorescent stained region; then, using r=A... s / A c Calculate the area A of the fluorescently colored region. s The ratio r of A to the total corneal area, where A c The total corneal area; according to the preset first-level mapping table, the ratio r is converted into the corneal fluorescence staining feature level G. c ; Step S2: Obtain the upper eyelid displacement signal sequence of the target object within a preset time period, and calculate the eyelid gap quantization level G based on the upper eyelid displacement signal sequence. e ; Step S3: Obtain the ambient relative humidity data H.

[0050] Step S4: Obtain record data containing medication timestamps and count the actual number of medication uses n within the evaluation period. a According to the preset number of medication doses n y ,pass Calculate compliance score S; Step S5: Assign corneal fluorescence staining feature level G c Quantitative grade of eyelid space G e The environmental humidity data H and compliance score S are concatenated to form the input feature vector X; Step S6: Input the input feature vector X into the pre-trained regression model based on gradient boosting decision tree, and calculate the corneal comprehensive state index R; Step S7: Based on the preset threshold range of the corneal comprehensive condition index R, determine the target intervention plan from multiple intervention plans that are pre-associated with different combinations of items; Step S8: Output the target intervention plan and generate information on the reassessment time interval.

[0051] In this embodiment, the intervention plan ultimately generated by the method is only used as auxiliary decision-making information for doctors' reference and does not constitute a direct medical instruction.

[0052] Example 2 This embodiment proposes to replace the method for determining the quantification level of eyelid space in Embodiment 1 with the following method: The monitoring module is used to acquire eye image sequences, extract consecutive T2 frames, and detect the vertical distance d between the midpoints of the upper and lower eyelid margins in each frame. t2 Where t2 = 1, 2, ..., T2; By P2=N2 over / T2, calculate vertical distance d t2Exceeding the preset closure threshold d th The percentage of frames P2 that exceed the total number of valid frames T2, and the percentage of frames exceeding the preset closure threshold d. th The vertical distance d corresponding to the frame t2 average value d P2 N2 over For distances exceeding d th The number of frames; According to the preset second-level mapping table, based on the frame percentage P2 and the mean d P Generate eyelid gap quantification level G e The second hierarchical mapping table is defined as follows: when P2 > 0.30 and d P2 When >4 mm, G e =3; when P2 > 0.30 and d P2 When ≤4 mm, or when 0.05≤P2≤0.30 and d P2 When >3 mm, G e =2; in other cases, G e =1.

[0053] In practice, the monitoring module receives video streams of the patient's eyes with their eyes closed from a nighttime surveillance camera and extracts 300 consecutive images with a T2 value.

[0054] A face landmark localization algorithm based on ensemble regression trees is used to locate eyelid margin landmarks. The method is as follows: the input eye region image is grayscaled and its size normalized, and the histogram of oriented gradients is extracted; through a pre-trained cascaded regressor, the precise coordinates of the midpoints of the upper and lower eyelid margins are estimated iteratively from the initial shape; then, the vertical distance d between the two midpoints is calculated using the Euclidean distance formula. t2 The cascaded regressor is trained using a dataset of eye images labeled with the coordinates of the eyelid margin midpoint, comprising 500 sets of cascaded regression trees. In this embodiment, the vertical distance is expressed in millimeters.

[0055] In this embodiment, a preset closure threshold d is used. th Set to 2 millimeters. Statistical d t2 >d th Frame count N2 over By P2=N2 over / T2, calculate the frame percentage P2, and obtain the average distance d between these frames. P2 Based on the logical judgment rules of the second-level mapping table, the eyelid space quantification level G is generated. e .

[0056] It should be noted that in actual products, the monitoring module may only include the flexible capacitive strain sensor patch, or it may also include a night monitoring camera. The flexible capacitive strain sensor patch is suitable for long-term monitoring in hospitals or at home, while the night monitoring camera is used for rapid assessment in the clinic.

[0057] This embodiment also provides a method for recommending corneal intervention plans after ptosis surgery based on multidimensional assessment. The difference between this method and the method for recommending corneal intervention plans after ptosis surgery based on multidimensional assessment provided in Embodiment 1 is that step S2 in this embodiment is replaced by: acquiring an ocular image sequence, extracting consecutive T2 frames, and detecting the vertical distance d between the midpoints of the upper and lower eyelid margins in each frame. t2 Where t2 = 1, 2, ..., T2; through P2 = N2 over / T2, calculate vertical distance d t2 Exceeding the preset closure threshold d th The percentage of frames P2 that exceed the total number of valid frames T2, and the percentage of frames exceeding the preset closure threshold d. th The vertical distance d corresponding to the frame t2 average value d P2 N2 over For distances exceeding d th The number of frames; based on the preset second-level mapping table, and based on the frame percentage 2P and the mean d P2 Generate eyelid gap quantification level G e .

[0058] Example 3 This embodiment proposes a method for constructing a regression model based on gradient boosting decision trees. This regression model is trained in the following way: Obtain a historical sample set, each sample containing corneal fluorescence staining feature level G. c,i Quantitative grade of eyelid space G e,i Environmental humidity data H i Compliance score S i The sample input vector X formed by splicing i And the corneal risk value Y, as indicated by the physician based on the follow-up results. i i = 1, 2, ..., N, where N is the total number of samples; The residuals of the current model are fitted by iteratively constructing a regression decision tree. The output of the decision tree generated in each iteration is multiplied by the preset learning rate and then added to the model to obtain the final regression model.

[0059] In this embodiment, complete data from 1500 patients undergoing mechanical ventilation or eyelid surgery were selected from the ophthalmology information system. For each case, the corresponding G... c,i G e,i Hi S i Constructing the sample input vector X i Three attending physicians, based on a pre-set corneal epithelial damage severity scoring scale, jointly assigned a quantitative corneal risk value Y ranging from 0 to 100. i .

[0060] The training uses the gradient boosting regression tree algorithm, and the loss function is the mean squared error.

[0061] In this embodiment, the above-mentioned settings of 100 regression decision trees, a learning rate of 0.1, and 8 leaf nodes per regression decision tree are used to initialize the constant prediction value. In the m-th iteration, the residual is calculated. Construct a regression decision tree h with 8 leaf nodes. m (X), whose leaf region R j,m The output c j,m This represents the average residual of samples falling within this region.

[0062] The model was then updated. Where v is the preset learning rate, which is 0.1, and the total number of iterations M is determined to be 100 through five-fold cross-validation.

[0063] Finally, the model structure and parameters of each node are saved for use by the analysis module.

[0064] Example 4 This embodiment, based on Embodiments 1 and 2, further includes a feedback learning module. This module acquires secondary corneal fluorescein staining images of the patient after intervention according to the target intervention plan and determines the secondary corneal fluorescein staining feature level G. c '; Utilizing secondary corneal fluorescence staining characteristic grade G c 'and the corresponding eyelid space quantification level G for this reassessment' e ', Environmental humidity data H' and compliance score S', update the regression model by adding new samples to the historical sample set and re-executing the training process, and / or adjust the first threshold TY1 and / or the second threshold TY2; The method for adjusting the second threshold TY2 is as follows: calculate the adjustment amount. , To adjust the step size coefficient as preset, update TY2 to Furthermore, if the difference between the updated TY2 and the first threshold TY1 is less than a preset minimum interval, then the first threshold TY1 is simultaneously adjusted to... Add the preset minimum interval.

[0065] In this embodiment, the initial feedback occurs the day after the recommended solution is implemented, followed by a reassessment.

[0066] That is, to re-acquire corneal stained images and obtain the secondary corneal fluorescence staining feature level G according to the method of Example 1. c ', and at the same time, the average level G within this period is calculated. e The average humidity (H) and compliance score (S) are used. For regression model updates, the regression model is updated by adding new samples to the historical sample set and re-executing the complete training process of Example 2. For threshold adjustment, the following logic is executed: the adjustment amount is calculated based on the deviation of Gc' from the ideal level 1. Preset adjustment step size coefficient The value was determined to be 5 based on preliminary clinical trials. Subtracting this increment from the original TY2 yields To avoid the high and low threshold ranges being too close, a preset minimum interval is set to 15. If If the difference between TY1 and the current TY1 is less than 15, then TY1 will be forcibly updated to [new value]. +15.

[0067] Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.

[0068] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0069] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0070] The various embodiments of this application have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A corneal intervention recommendation system based on multidimensional assessment after ptosis surgery, characterized in that, include: The status acquisition module is used to acquire corneal fluorescein staining images and extract green channel images from the corneal fluorescein staining images to obtain the fluorescent stained areas. Calculate the ratio of the area of ​​the fluorescently stained region to the total area of ​​the cornea, and convert the ratio into a corneal fluorescent staining feature level according to a preset first-level mapping table; The monitoring module is used to acquire the upper eyelid displacement signal sequence of the target object within a preset time period, and calculate the quantification level of the eyelid gap based on the upper eyelid displacement signal sequence; The environment acquisition module is used to acquire relative humidity data. The recording module is used to acquire record data containing medication timestamps, count the actual number of medications used within the evaluation period, and determine the compliance score based on the preset number of medications used. The analysis module is used to concatenate corneal fluorescence staining feature level, eyelid gap quantification level, environmental humidity data and compliance score into an input feature vector; input the input feature vector into a pre-trained regression model based on gradient boosting decision tree to calculate the corneal comprehensive state index; and determine the target intervention program from multiple intervention programs with different pre-associated item combinations based on the preset threshold range of the corneal comprehensive state index. The output module is used to output the target intervention plan and to generate reassessment time interval information based on the preset threshold range of the corneal comprehensive condition index.

2. The corneal intervention recommendation system based on multidimensional assessment after ptosis surgery as described in claim 1, characterized in that, The method for calculating the quantization level of the eyelid space based on the upper eyelid displacement signal sequence is as follows: The difference signal between the strain-sensitive capacitor and the reference capacitor is acquired in real time. Based on the differential reference obtained by zero-point calibration and the pre-calibrated capacitance-displacement mapping model, the capacitance difference is converted into the vertical distance of the upper eyelid relative to the closed position. Within a preset time period, a valid displacement signal sequence of length T1 is acquired at a fixed sampling frequency. The percentage of valid sampling points whose vertical distance exceeds a preset closure threshold is calculated, along with the average vertical distance corresponding to the sampling points that exceed the preset closure threshold. The eyelid gap quantization level is generated based on a preset second-level mapping table.

3. The corneal intervention recommendation system based on multidimensional assessment after ptosis surgery as described in claim 1, characterized in that, in, The preset threshold range consists of a first threshold and a second threshold, with the first threshold being greater than the second threshold. The reassessment time interval information generated by the output module includes: when the corneal comprehensive condition index is greater than or equal to the first threshold, the generated monitoring suggestion interval is no more than the preset interval time TJ; when the second threshold is less than or equal to the corneal comprehensive condition index and less than the first threshold, the generated monitoring suggestion interval is γ1×TJ; when the corneal comprehensive condition index is less than the second threshold, the generated monitoring suggestion interval is no less than γ2×TJ; where γ1 is greater than or equal to 1 and γ2 is greater than γ1, and γ1 and γ2 are preset reassessment interval amplification coefficients.

4. The corneal intervention recommendation system based on multidimensional assessment after ptosis surgery as described in claim 1, characterized in that, The first grade mapping table is as follows: when the ratio of the area of ​​the fluorescent stained region to the total area of ​​the cornea is <0.01, the corneal fluorescence staining feature level = 1; When 0.01 ≤ ratio < 0.05, the corneal fluorescence staining characteristic grade = 2; when 0.05 ≤ ratio < 0.15, the corneal fluorescence staining characteristic grade = 3. When 0.15 ≤ ratio < 0.30, the corneal fluorescence staining characteristic grade = 4; when the ratio ≥ 0.30, the corneal fluorescence staining characteristic grade = 5.

5. The corneal intervention recommendation system based on multidimensional assessment after ptosis surgery as described in claim 2, characterized in that, The second classification mapping table is as follows: when the percentage of effective sampling points is greater than 0.30 and the average value is greater than 4 mm, the quantization level of the eyelid gap is 3; when the percentage of effective sampling points is greater than 0.30 and the average value is less than or equal to 4 mm, or when 0.05 ≤ the percentage of effective sampling points ≤ 0.30 and the average value is greater than 3 mm, the quantization level of the eyelid gap is 2; in other cases, the quantization level of the eyelid gap is 1.

6. The corneal intervention recommendation system based on multidimensional assessment after ptosis surgery as described in claim 1, characterized in that, The monitoring module includes a flexible capacitive strain sensor patch attached to the skin of the upper eyelid. The flexible capacitive strain sensor patch is composed of a first flexible electrode layer, an elastic dielectric layer and a second flexible electrode layer stacked from top to bottom. Flexible capacitive strain sensor patch is used to generate capacitance changes based on the stretching and deformation of the upper eyelid skin; The contact surface of the flexible capacitive strain sensor patch is also provided with multiple contact impedance detection electrodes arranged in an array. The sensing patch includes a strain-sensitive capacitor region and a non-strain-sensitive reference capacitor region integrated with the substrate. The elastic dielectric layer of the reference capacitor region is constrained to be incompressible and is only sensitive to temperature and humidity.

7. The corneal intervention recommendation system based on multidimensional assessment after ptosis surgery as described in claim 1, characterized in that, The monitoring module is also used to calculate the eyelid gap quantization level by acquiring eye image sequences. Specifically, it extracts T2 consecutive frames from the acquired eye image sequence, detects the vertical distance between the midpoints of the upper and lower eyelid margins in each frame, calculates the percentage of frames with vertical distances exceeding a preset closure threshold out of the total number of valid frames, and the average distance of the frames exceeding the threshold, and generates the eyelid gap quantization level based on the frame percentage and the average distance according to the preset second classification mapping table. When the percentage of frames with a vertical distance exceeding a preset closure threshold is greater than 0.30% of the total number of valid frames and the average distance of the frames exceeding the threshold is greater than 4 mm, the eyelid gap quantization level is 3; when the percentage of frames exceeding the threshold is greater than 0.30% and the average distance is less than or equal to 4 mm, or when 0.05 ≤ percentage of frames ≤ 0.30% and the average distance is greater than 3 mm, the eyelid gap quantization level is 2; otherwise, the eyelid gap quantization level is 1.

8. The corneal intervention recommendation system based on multidimensional assessment after ptosis surgery according to claim 1, characterized in that, It also includes a feedback learning module, which is used to acquire secondary corneal fluorescein staining images of patients after intervention according to the target intervention plan, and determine the secondary corneal fluorescein staining feature level; using the secondary corneal fluorescein staining feature level and the eyelid gap quantification level, environmental humidity data and compliance score corresponding to the reassessment, the regression model is updated by adding new samples to the historical sample set and re-executing the training process, and / or the first threshold and / or the second threshold are adjusted. The fluorescent stained region was obtained by adaptive threshold segmentation of the green channel image using the maximum inter-class variance method.

9. The corneal intervention recommendation system based on multidimensional assessment after ptosis surgery as described in claim 8, characterized in that, The feedback learning module adjusts the second threshold by multiplying the preset adjustment step size coefficient by the difference between the secondary corneal fluorescence staining feature level and one, and adding this adjustment amount to the original second threshold to obtain the new second threshold. If the difference between the updated second threshold and the first threshold is less than the preset minimum interval, the first threshold is simultaneously adjusted to the updated second threshold plus the preset minimum interval.

10. The corneal intervention recommendation system based on multidimensional assessment after ptosis surgery according to claim 1, characterized in that, The training method for a regression model based on gradient boosting decision trees is as follows: Obtain a historical sample set. Each sample contains a sample input vector formed by splicing corneal fluorescence staining feature level, eyelid gap quantification level, environmental humidity data, and compliance score, as well as a corneal risk value marked by the supervisor based on the follow-up results. The residuals of the current model are fitted by iteratively constructing a regression decision tree. The output of the decision tree generated in each iteration is multiplied by the preset learning rate and then accumulated into the model to obtain the final regression model. The compliance score is calculated by subtracting the absolute value of the difference between the actual number of medications and the preset number of medications from 1, and dividing the result by the preset number of medications. The compliance score is 0 when it is negative.

Images