Intraoperative intraocular pressure measurement method based on array pressure sensor and machine learning

Abstract

This invention relates to the field of medical testing and perioperative monitoring technology, and discloses a method for measuring intraocular pressure (IOP) during surgery in the prone position based on array pressure sensing and machine learning. The invention collects periocular array pressure signals through an eye mask integrating a flexible array pressure sensor. After baseline correction, the spatial distribution characteristics of the pressure are extracted. Physical inverse intermediate features are constructed by combining an equivalent elastic model of the eyelid and periocular soft tissue. Individual patient parameters are fused to construct a feature vector, which is then input into a pre-trained random forest regression model to obtain an estimated IOP value. After time-series smoothing, an abnormal IOP warning is achieved. This invention solves the problems of existing technologies that cannot adapt to prone position general anesthesia surgery scenarios, and that prone position intraoperative IOP monitoring is not continuous and lacks sufficient quantitative monitoring.

Description

Technical Field

[0001] This invention relates to the field of medical testing and perioperative monitoring technology, and in particular to a method for measuring intraocular pressure in the prone position during surgery based on array pressure sensing and machine learning. Background Technology

[0002] Intraocular pressure is a core indicator reflecting the tension of the eyeball and the perfusion status of the optic nerve. In prone neurosurgery and spinal surgery, pressure on the patient's eye and obstruction of venous return can easily lead to increased intraocular pressure, which is a core risk factor for postoperative visual function impairment. However, the lack of effective continuous monitoring methods during surgery poses a significant clinical safety hazard.

[0003] Currently, the mainstream clinical intraocular pressure measurement devices are represented by the Goldmann applanation tonometer and the non-contact air jet tonometer. Both require the patient to be awake and open their eyes to cooperate, and are operated by professionals in a sitting or supine position. They can only obtain intermittent single-point measurements before and after surgery, and cannot meet the long-term continuous monitoring needs of general anesthesia with eyes closed and prone position surgery scenarios, and cannot provide real-time data support for intraoperative intervention.

[0004] Existing headrests and pressure-reducing pads for prone positioning can only distribute pressure on the face and prevent direct contact with the eyes, but cannot achieve quantitative collection of periocular pressure and intraocular pressure estimation. Existing flexible pressure-sensing intraocular pressure estimation solutions are mostly designed for daily health monitoring scenarios and are not adapted to the special working conditions of complex contact boundaries and easily fluctuating force during prone surgery. Moreover, they mostly use the raw pressure signal directly as the input of machine learning models without combining the mechanical properties of ocular tissues to construct physical prior features. This results in poor individual adaptability, insufficient stability and accuracy of continuous measurement, and cannot meet the reliability requirements of clinical intraoperative monitoring.

[0005] In summary, there is currently no mature solution suitable for prone general anesthesia surgery that can achieve continuous, non-invasive, and accurate intraocular pressure monitoring under passive contact with eyes closed, leaving a significant gap in clinical application. Summary of the Invention

[0006] The present invention aims to provide a method for measuring intraocular pressure during prone surgery based on array pressure sensing and machine learning, in order to solve the problems in the prior art that cannot be adapted to the prone general anesthesia surgery scenario, and that the intraocular pressure during prone surgery cannot be continuously and quantitatively monitored.

[0007] To achieve the above objectives, the present invention provides the following method:

[0008] The intraoperative intraocular pressure measurement method based on array pressure sensing and machine learning provided by this invention is as follows:

[0009] S1. Array pressure signal acquisition and baseline correction: An eye mask integrating a flexible array pressure sensor is worn around the patient's eyes, so that the flexible array pressure sensor covers the patient's eyelids and surrounding soft tissue areas; a zero-load pressure matrix is ​​acquired in a no-load state before the patient is placed in a prone head support device; under the conditions that the patient has completed general anesthesia and established a prone position, the head is fixed to the support device, the eyes are in a closed, purely passive contact state and no additional active pressure is applied, the array pressure matrix is ​​continuously acquired at a preset sampling frequency throughout the entire operation, and the array pressure matrix is ​​baseline corrected to obtain a corrected pressure matrix;

[0010] S2. Pressure space feature extraction: Based on the corrected pressure matrix, the array pressure space distribution features are calculated;

[0011] S3. Construction of intermediate physical inverse solution features: Based on the preset equivalent elastic model of eyelid and periorbital soft tissue, the correction pressure matrix is ​​regarded as an external load generated by passive contact in the prone position. The contact problem between the eyelid and the support device is solved physically to obtain intermediate physical inverse solution features for auxiliary regression.

[0012] S4. Feature Fusion and Intraocular Pressure Regression Estimation: The array pressure spatial distribution features and the physical inverse solution intermediate features are fused with the patient individual parameters collected before surgery in a preset order to construct a feature vector for intraocular pressure estimation; the feature vector is input into a pre-trained random forest regression model to output the intraocular pressure estimation value at the current time.

[0013] S5. Intraocular Pressure Time Series Processing and Abnormal Warning: The intraocular pressure time series composed of the estimated intraocular pressure values ​​in the continuous time dimension is smoothed to extract the trend of intraocular pressure change; an individualized intraocular pressure safety range is set according to the patient's preoperative baseline intraocular pressure value; when the smoothed intraocular pressure value continuously exceeds the preset upper limit of the individualized intraocular pressure safety range and the duration of exceeding the limit exceeds the preset threshold, a warning signal is triggered.

[0014] Furthermore, in step S1, the sensing units of the flexible array pressure sensor are arranged in a matrix, the array resolution is not less than 8×8, and the preset sampling frequency is 10Hz.

[0015] Furthermore, in step S1, the no-load state refers to the patient not having a prone head support device placed on them, not having external force applied to the area around their eyes, and being in a preparatory posture close to a prone position or a supine position, with the flexible array pressure sensor in a state of no load or only slight contact; the zero-load pressure matrix is ​​obtained by sampling the flexible array pressure sensor in the no-load state multiple times and averaging it over time; the baseline correction is calculated by subtracting the zero-load pressure matrix from the array pressure matrix collected at each moment; for sensing units with obvious output abnormalities after correction, data correction is performed using neighborhood interpolation or channel shielding.

[0016] Furthermore, in step S2, the spatial distribution characteristics of the array pressure include total pressure, maximum pressure, average pressure, pressure standard deviation, pressure centroid coordinates, second moment along the horizontal direction, and second moment along the vertical direction; the formula for calculating the pressure centroid coordinates is:

[0017] ;

[0018] in, For the coordinates of the pressure centroid, The first in the flexible array pressure sensor The spatial coordinates of each sensing unit For the first The calibration pressure value corresponding to each sensing unit This represents the total number of sensing units in the flexible array pressure sensor.

[0019] Furthermore, in step S3, the equivalent elastic model of the eyelid and periocular soft tissue abstracts the eyelid and periocular soft tissue into an equivalent elastic body region with a preset thickness. The equivalent elastic body region includes an upper surface in contact with the cornea, a lower surface in contact with the flexible array pressure sensor, sides, and an internal enclosed space. During the physical inverse solution process, the correction pressure matrix is ​​used as a discrete surface load acting on the lower surface of the equivalent elastic body. The intermediate features of the physical inverse solution are obtained by numerical solution, table lookup, or flattening approximation model calculation. The intermediate features of the physical inverse solution include the equivalent contact area, the equivalent contact radius, the equivalent total force, and the preliminary intraocular pressure estimate based on the elasticity model.

[0020] Furthermore, in step S4, the individual patient parameters include the patient's age, central corneal thickness, and preoperative baseline intraocular pressure; when constructing the feature vector for intraocular pressure estimation, each feature component is normalized or standardized.

[0021] Furthermore, in step S4, the random forest regression model consists of multiple regression decision trees, and the model is pre-trained offline. During the training phase, a mapping relationship between the input feature vector and the actual intraocular pressure value is constructed using multi-source datasets, including finite element simulation datasets, simulated eye experimental datasets, and clinical calibration datasets. During the running phase, each regression decision tree outputs a local intraocular pressure prediction value for the input feature vector, integrates and calculates all the local intraocular pressure prediction values, and outputs the intraocular pressure estimate value at the corresponding time through nonlinear inversion.

[0022] Furthermore, in step S5, the smoothing process adopts first-order exponential smoothing to obtain a smooth intraocular pressure curve. While suppressing sensor noise and instantaneous fluctuations, the process retains the trend characteristics of slow or sustained high intraocular pressure, ensuring a sensitive response to abnormal changes in intraocular pressure.

[0023] Furthermore, in step S5, when the smoothed intraocular pressure value is consistently lower than the preset lower limit of the individualized intraocular pressure safety range and the duration exceeds the preset threshold, it can be determined as abnormally low intraocular pressure according to clinical needs, triggering a corresponding reminder signal; the preset lower limit is set according to the patient's preoperative baseline intraocular pressure value.

[0024] Furthermore, in step S5, the warning signal is an audible and visual alarm signal, and the reminder signal is an audible and visual reminder signal.

[0025] The beneficial effects of this invention are reflected in:

[0026] Non-invasive passive monitoring with strong clinical adaptability: This invention achieves indirect intraocular pressure measurement by integrating a flexible array pressure sensor into the eye mask, allowing the patient to passively contact the device with their eyes closed without any additional active pressure. It does not require the patient to be awake and cooperate, does not expose the cornea, and does not change the existing surgical position or procedure. It is fully adaptable to the clinical scenario of prone general anesthesia surgery, filling the clinical gap in continuous intraocular pressure monitoring during surgery.

[0027] The fusion of physical priors and machine learning inversion results in high measurement robustness and accuracy: This invention integrates the spatial distribution characteristics of array pressure with intermediate features from the physical inversion based on an equivalent elasticity model, constructs a feature vector by combining individual parameters, and achieves nonlinear inversion of intraocular pressure through a random forest regression model. It not only improves the interpretability of the model through physical priors, but also solves the measurement error problems caused by differences in individual tissue mechanical properties and intraoperative contact boundary fluctuations through machine learning, maintaining stable and accurate measurement results even under complex surgical conditions.

[0028] Individualized correction and continuous trend monitoring offer strong clinical applicability: This invention incorporates individual parameters such as patient age, central corneal thickness, and preoperative baseline intraocular pressure into the feature vector to achieve individualized correction, adapting to patient groups of different ages and with different ocular physiological characteristics. By extracting the trend of intraocular pressure changes through time series smoothing and setting individualized safety intervals, it enables automatic identification and early warning of abnormal intraocular pressure, providing real-time and reliable data support for intraoperative positioning adjustments and intervention measures, effectively reducing the risk of postoperative visual function impairment in prone positions. Attached Figure Description

[0029] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0030] Figure 1 This is a schematic diagram illustrating the passive pressure relationship between the eyelid and the array pressure sensor provided in an embodiment of the present invention.

[0031] Figure 2 A schematic diagram of the passive pressure contact mechanics model of the eye provided in an embodiment of the present invention;

[0032] Figure 3 A schematic diagram of the intraoperative intraocular pressure measurement method based on array pressure sensing and machine learning provided in an embodiment of the present invention;

[0033] Figure 4 This is a schematic diagram of the random forest regression model algorithm provided in an embodiment of the present invention. Detailed Implementation

[0034] To enable those skilled in the art to better understand the present invention, 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.

[0035] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or end that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or ends.

[0036] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0037] Currently, the mainstream clinical intraocular pressure measurement devices are represented by the Goldmann applanation tonometer and the non-contact air jet tonometer. Both require the patient to be awake and open their eyes to cooperate, and are operated by professionals in a sitting or supine position. They can only obtain intermittent single-point measurements before and after surgery, and cannot meet the long-term continuous monitoring needs of general anesthesia with eyes closed and prone position surgery scenarios, and cannot provide real-time data support for intraoperative intervention.

[0038] Existing headrests and pressure-reducing pads for prone positioning can only distribute pressure on the face and prevent direct contact with the eyes, but cannot achieve quantitative collection of periocular pressure and intraocular pressure estimation. Existing flexible pressure-sensing intraocular pressure estimation solutions are mostly designed for daily health monitoring scenarios and are not adapted to the special working conditions of complex contact boundaries and easily fluctuating force during prone surgery. Moreover, they mostly use the raw pressure signal directly as the input of machine learning models without combining the mechanical properties of ocular tissues to construct physical prior features. This results in poor individual adaptability, insufficient stability and accuracy of continuous measurement, and cannot meet the reliability requirements of clinical intraoperative monitoring.

[0039] In summary, there is currently no mature solution suitable for prone general anesthesia surgery that can achieve continuous, non-invasive, and accurate intraocular pressure monitoring under passive contact with eyes closed, leaving a significant gap in clinical application.

[0040] The present invention aims to provide a method for measuring intraocular pressure during prone surgery based on array pressure sensing and machine learning, in order to solve the problems in the prior art that cannot be adapted to the prone general anesthesia surgery scenario, and that the intraocular pressure during prone surgery cannot be continuously and quantitatively monitored.

[0041] like Figure 1-4 As shown, a specific embodiment of the present invention provides a method for intraoperative intraocular pressure measurement in the prone position based on array pressure sensing and machine learning, including the following steps:

[0042] Preoperative preparation stage

[0043] Before surgery, the patient's intraocular pressure was measured using the Goldmann applanation tonometer, which is the clinical gold standard, to obtain the patient's preoperative baseline intraocular pressure value. At the same time, individual parameters such as the patient's age and central corneal thickness were collected and entered into the monitoring system.

[0044] A sterile medical goggle with an integrated flexible array pressure sensor is selected. The sensing unit of the flexible array pressure sensor is arranged in an 8×8 matrix and the sampling frequency is set to 10Hz, which can completely cover the patient's eyelids and periocular soft tissue area. The goggle is connected to the monitoring system to complete the device self-test and calibration.

[0045] Baseline acquisition: Before anesthesia induction, the patient wears an eye shield to ensure that the flexible array pressure sensor is in contact with the eyelids and surrounding soft tissues. In a supine, unloaded state, without the placement of a prone head support device, without external force applied to the periorbital area, the flexible array pressure sensor is continuously sampled 20 times. The sampling results are then averaged over time to obtain the zero-load pressure matrix, which serves as the baseline matrix for subsequent measurements and is stored in the monitoring system.

[0046] Intraoperative monitoring implementation phase

[0047] After the patient completes general anesthesia induction and endotracheal intubation, a prone position is established, and the head is fixed on a clinically standard prone gel headrest to ensure that the patient's eyes are naturally closed. The flexible array pressure sensor inside the eye mask maintains a purely passive contact with the eyelids, without applying any additional active pressure, and does not change the routine clinical surgical procedure.

[0048] Continuous acquisition and correction of array pressure signals: Throughout the entire surgical procedure, the array pressure matrix output by the flexible array pressure sensor at each moment is continuously acquired at a preset sampling frequency of 10Hz; for each frame of the acquired array pressure matrix, baseline correction is performed by subtracting the pre-stored zero-load pressure matrix to obtain the corrected pressure matrix; for sensing units with obvious output abnormalities in the corrected pressure matrix, neighborhood interpolation is used to correct the data to ensure the accuracy of the pressure data.

[0049] Pressure spatial feature extraction: For each frame of the corrected pressure matrix, the corresponding array pressure spatial distribution features are calculated, specifically including total pressure, maximum pressure, average pressure, pressure standard deviation, pressure centroid coordinates, second moment along the horizontal direction, and second moment along the vertical direction; among which, the pressure centroid coordinates are used to characterize the location and force eccentricity of the pressure area, and their calculation formula is as follows:

[0050] ;

[0051] in, For the coordinates of the pressure centroid, The first in the flexible array pressure sensor The spatial coordinates of each sensing unit For the first The calibration pressure value corresponding to each sensing unit, and the total number of sensing units in the flexible array pressure sensor. =64.

[0052] Physical inverse solution intermediate feature construction: Based on a pre-defined equivalent elastic model of the eyelid and periocular soft tissue, the correction pressure matrix is ​​regarded as an external load generated by passive contact in the prone position, and the physical inverse solution is performed on the contact problem between the eyelid and the headrest support device. The equivalent elastic model abstracts the eyelid and periocular soft tissue into an equivalent elastic body region with a pre-defined thickness h. This region includes the upper surface in contact with the cornea, the lower surface in contact with the flexible array pressure sensor, the sides, and the internal enclosed space. The correction pressure matrix is ​​regarded as a discrete surface load acting on the lower surface of the equivalent elastic body. Through numerical solution, the equivalent contact area, equivalent contact radius, equivalent total force, and preliminary intraocular pressure estimate based on the elasticity model are obtained, which constitute the physical inverse solution intermediate features to assist subsequent machine learning regression.

[0053] Feature fusion and intraocular pressure regression estimation: The aforementioned spatial distribution features of pressure and intermediate features from physical inversion are concatenated with the patient's age, central corneal thickness, and preoperative baseline intraocular pressure values ​​collected before surgery in a preset order to construct a multidimensional feature vector. After standardizing each component of the feature vector, it is input into a pre-trained offline random forest regression model. The random forest regression model consists of 100 regression decision trees. During the training phase, training is completed using multi-source datasets, including eyelid-eye contact finite element simulation datasets, simulated eye experiment calibration datasets, and clinical volunteer prone intraocular pressure calibration datasets. Through training, a nonlinear mapping relationship between the input feature vector and the actual intraocular pressure value is constructed. During the model operation phase, each regression decision tree traverses the input feature vector node by node, outputting the local intraocular pressure prediction value at the leaf node. The mean of the local prediction values ​​of all decision trees is integrated and calculated, and the estimated intraocular pressure value at the current moment is output through nonlinear inversion.

[0054] Intraocular pressure (IOP) time series processing and abnormal warning: The IOP time series, composed of IOP estimates over a continuous time dimension, is smoothed using first-order exponential smoothing to obtain a smooth IOP curve. This process suppresses sensor noise and instantaneous fluctuations while preserving the trend characteristics of slowly increasing or persistently high IOP. Based on the patient's preoperative baseline IOP value, an individualized IOP safety range is set. The upper limit of the safety range is 1.5 times the preoperative baseline IOP value, and the lower limit is 0.5 times the preoperative baseline IOP value. The preset threshold for exceeding the limit is 30 seconds. When the smoothed IOP value consistently exceeds the upper limit of the safety range for more than 30 seconds, it is considered a risk of abnormally high IOP, and the system triggers an audible and visual alarm signal, prompting medical staff to adjust the patient's prone position or head support method. When the smoothed IOP value consistently falls below the lower limit of the safety range for more than 30 seconds, it can be considered abnormally low IOP based on clinical needs, and the system triggers a corresponding audible and visual alert signal.

[0055] Postoperative recovery phase

[0056] After the surgery, pressure signal acquisition is stopped, and the system automatically saves the intraocular pressure monitoring data and pressure distribution data throughout the entire surgery, generating an intraocular pressure change trend report for clinical postoperative evaluation; the patient's eye mask is removed, and the equipment is cleaned and disinfected.

[0057] Example 1

[0058] like Figure 1-4 As shown, this embodiment of the invention uses prone general anesthesia surgery as an application scenario to illustrate the intraocular pressure measurement method of the present invention. This method, with the patient's eyes closed and in a prone position, acquires the pressure distribution around the eye using a flexible array pressure sensor, and combines a contact mechanics model with a random forest regression algorithm to achieve indirect and continuous estimation of intraocular pressure during surgery.

[0059] like Figure 1 As shown, after establishing the prone position, a flexible array pressure sensor is positioned between the patient's face and the contact interface, with the array sensing surface located below the eyelids or adjacent to the periorbital region. In the initial state ( Figure 1 As shown in (a), with the eyelids naturally covering the eyeballs, a light touch on the array pressure sensor causes the sensor output to approach the zero baseline; once the prone position is stable ( Figure 1 (b) As shown, the eyeball protrudes slightly forward under the influence of body position and gravity, applying passive force to the array through the eyelid, causing the eyelid to deform and forming a certain range of pressure zone on the array; the array pressure sensor discretizes this pressure state into a two-dimensional pressure matrix, and its spatial intensity distribution can be represented as a pressure cloud map ( Figure 1 (c) is shown.

[0060] Before surgery, intraocular pressure is measured using a standard clinical tonometer (such as a Goldmann applanation tonometer) to obtain the preoperative baseline intraocular pressure value. Simultaneously, patient age, central corneal thickness, and other optional individual parameters are acquired and recorded in the case information. Without applying external force to the periocular region, the patient is positioned in a near-prone or supine position, with the array pressure sensors in a no-load or only slightly contact state. Zero-load outputs from each sensing unit are obtained through multiple samplings, and the zero-load pressure matrix is ​​obtained after time averaging. This serves as the baseline matrix for subsequent measurements.

[0061] After establishing and stabilizing the prone position, ensure the patient's eyes are naturally closed, maintaining a purely passive contact between the array pressure sensor and the eyelids. Continuously collect data from the array pressure sensor at various times according to the preset sampling frequency of 10Hz. The output matrix For each frame of the pressure matrix By subtracting the zero-load pressure matrix Baseline correction is performed to obtain the corrected pressure matrix:

[0062]

[0063] Corrected matrix It reflects the actual pressure distribution in the periorbital region under passive pressure conditions with eyes closed in a prone position; for sensing units that show obvious abnormalities, corrections can be made by methods such as neighborhood interpolation and channel shielding.

[0064] like Figure 2 As shown, the eyelid and periocular soft tissue are abstracted as an equivalent elastomer region with a thickness h, consisting of the upper surface in contact with the cornea. The lower surface that contacts the array sensor is side and interior space Composition. Corrected pressure matrix Considered as discrete surface loads acting on the lower surface of the elastic body, the following can be obtained through numerical solutions or table lookups: equivalent contact area and equivalent contact radius; equivalent total force (obtained by integrating over the load distribution); and preliminary intraocular pressure estimates calculated based on the approximation model or empirical formulas. The aforementioned intermediate features, derived from physical inverse kinematics, are used to assist machine learning models in improving their robustness and interpretability.

[0065] like Figure 4 As shown, at each moment This involves combining pressure space characteristics, physical inverse intermediate characteristics, and preoperative individual parameters (including age, central corneal thickness, and...) (etc.) are fused and spliced ​​in a predetermined order to form a multidimensional feature vector. If necessary, normalization or standardization can be performed on each component. A pre-trained offline random forest regression model is called, and the feature vectors are... Input model. The random forest regression model consists of multiple regression decision trees, each of which partitions according to its internal rules. The node traversal is performed, and the local intraocular pressure prediction value is output at the leaf node. The processing unit integrates and calculates the prediction results of all decision trees to obtain the comprehensive intraocular pressure estimate at the current time.

[0066] To suppress instantaneous measurement noise and possess trend recognition capabilities, intraocular pressure estimation sequences over continuous time are analyzed. Time series smoothing is performed. Preferably, first-order exponential smoothing is used to obtain a smoothed intraocular pressure curve. This allows the system to respond sensitively to slowly rising or persistently high intraocular pressure. Based on the preoperative baseline intraocular pressure... In conjunction with clinically established safety thresholds, determine an individualized safe intraocular pressure range. , When the smoothed intraocular pressure value consecutively exceeding Furthermore, if the duration of the excess exceeds a preset threshold, it is determined that there is a risk of abnormally elevated intraocular pressure, and a notification may be displayed on the monitoring interface or an independent alarm may be triggered; when Long-term lower than In some cases, it can also be determined that the intraocular pressure is abnormally low, and appropriate reminders can be given as needed.

[0067] The beneficial effects of this invention are reflected in:

[0068] Non-invasive passive monitoring with strong clinical adaptability: This invention achieves indirect intraocular pressure measurement by integrating a flexible array pressure sensor into the eye mask, allowing the patient to passively contact the device with their eyes closed without any additional active pressure. It does not require the patient to be awake and cooperate, does not expose the cornea, and does not change the existing surgical position or procedure. It is fully adaptable to the clinical scenario of prone general anesthesia surgery, filling the clinical gap in continuous intraocular pressure monitoring during surgery.

[0069] The fusion of physical priors and machine learning inversion results in high measurement robustness and accuracy: This invention integrates the spatial distribution characteristics of array pressure with intermediate features from the physical inversion based on an equivalent elasticity model, constructs a feature vector by combining individual parameters, and achieves nonlinear inversion of intraocular pressure through a random forest regression model. It not only improves the interpretability of the model through physical priors, but also solves the measurement error problems caused by differences in individual tissue mechanical properties and intraoperative contact boundary fluctuations through machine learning, maintaining stable and accurate measurement results even under complex surgical conditions.

[0070] Individualized correction and continuous trend monitoring offer strong clinical applicability: This invention incorporates individual parameters such as patient age, central corneal thickness, and preoperative baseline intraocular pressure into the feature vector to achieve individualized correction, adapting to patient groups of different ages and with different ocular physiological characteristics. By extracting the trend of intraocular pressure changes through time series smoothing and setting individualized safety intervals, it enables automatic identification and early warning of abnormal intraocular pressure, providing real-time and reliable data support for intraoperative positioning adjustments and intervention measures, effectively reducing the risk of postoperative visual function impairment in prone positions.

[0071] The above descriptions are merely embodiments of the present invention. Commonly known technical solutions or characteristics are not described in detail here. It should be noted that those skilled in the art can make various modifications and improvements without departing from the present invention, and these should also be considered within the scope of protection of the present invention. These modifications and improvements will not affect the effectiveness of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A method for intraoperative intraocular pressure measurement in the prone position based on array pressure sensing and machine learning, characterized in that, The method includes: S1. Array pressure signal acquisition and baseline correction: An eye mask integrating a flexible array pressure sensor is worn around the patient's eyes, so that the flexible array pressure sensor covers the patient's eyelids and surrounding soft tissue areas; a zero-load pressure matrix is ​​acquired in a no-load state before the patient is placed in a prone head support device; under the conditions that the patient has completed general anesthesia and established a prone position, the head is fixed to the support device, the eyes are in a closed, purely passive contact state and no additional active pressure is applied, the array pressure matrix is ​​continuously acquired at a preset sampling frequency throughout the entire operation, and the array pressure matrix is ​​baseline corrected to obtain a corrected pressure matrix; S2. Pressure space feature extraction: Based on the corrected pressure matrix, the array pressure space distribution features are calculated; S3. Construction of intermediate physical inverse solution features: Based on the preset equivalent elastic model of eyelid and periorbital soft tissue, the correction pressure matrix is ​​regarded as an external load generated by passive contact in the prone position. The contact problem between the eyelid and the support device is solved physically to obtain intermediate physical inverse solution features for auxiliary regression. S4. Feature Fusion and Intraocular Pressure Regression Estimation: The array pressure spatial distribution features and the physical inverse solution intermediate features are fused with the patient individual parameters collected before surgery in a preset order to construct a feature vector for intraocular pressure estimation; the feature vector is input into a pre-trained random forest regression model to output the intraocular pressure estimation value at the current time. S5. Intraocular Pressure Time Series Processing and Abnormal Warning: The intraocular pressure time series composed of the estimated intraocular pressure values ​​in the continuous time dimension is smoothed to extract the trend of intraocular pressure change; an individualized intraocular pressure safety range is set according to the patient's preoperative baseline intraocular pressure value; when the smoothed intraocular pressure value continuously exceeds the preset upper limit of the individualized intraocular pressure safety range and the duration of exceeding the limit exceeds the preset threshold, a warning signal is triggered.

2. The method for intraoperative intraocular pressure measurement in the prone position based on array pressure sensing and machine learning according to claim 1, characterized in that: In step S1, the sensing units of the flexible array pressure sensor are arranged in a matrix, the array resolution is not less than 8×8, and the preset sampling frequency is 10Hz.

3. The method for intraoperative intraocular pressure measurement in the prone position based on array pressure sensing and machine learning according to claim 1, characterized in that: In step S1, the no-load state means that the patient is not in a prone head support device, no external force is applied around the eyes, and is in a preparatory posture close to the prone position or in a supine position, and the flexible array pressure sensor is in a no-load or only slightly contacted state. The zero-load pressure matrix is ​​obtained by sampling the flexible array pressure sensor under no-load conditions multiple times and averaging the samples over time. The baseline correction is calculated by subtracting the zero-load pressure matrix from the array pressure matrix collected at each moment; for sensing units that show obvious output abnormalities after correction, data correction is performed by neighborhood interpolation or channel shielding.

4. The method for intraoperative intraocular pressure measurement in the prone position based on array pressure sensing and machine learning according to claim 1, characterized in that: In step S2, the array pressure spatial distribution characteristics include total pressure, maximum pressure, average pressure, pressure standard deviation, pressure centroid coordinates, second moment along the horizontal direction, and second moment along the vertical direction. The formula for calculating the coordinates of the pressure centroid is: ； in, For the coordinates of the pressure centroid, The first in the flexible array pressure sensor The spatial coordinates of each sensing unit For the first The calibration pressure value corresponding to each sensing unit This represents the total number of sensing units in the flexible array pressure sensor.

5. The method for intraoperative intraocular pressure measurement in the prone position based on array pressure sensing and machine learning according to claim 1, characterized in that: In step S3, the equivalent elastic model of the eyelid and periocular soft tissue abstracts the eyelid and periocular soft tissue into an equivalent elastic body region with a preset thickness. The equivalent elastic body region includes an upper surface in contact with the cornea, a lower surface in contact with the flexible array pressure sensor, a side, and an internal enclosed space. In the physical inverse solution process, the corrected pressure matrix is ​​used as a discrete surface load acting on the lower surface of the equivalent elastic body. The intermediate features of the physical inverse solution are obtained by numerical solution, table lookup or flattening approximation model calculation. The intermediate features of the physical inverse solution include the equivalent contact area, equivalent contact radius, equivalent total force, and preliminary intraocular pressure estimate based on the elasticity model.

6. The method for intraoperative intraocular pressure measurement in the prone position based on array pressure sensing and machine learning according to claim 1, characterized in that: In step S4, the individual patient parameters include the patient's age, central corneal thickness, and preoperative baseline intraocular pressure. When constructing the feature vector for intraocular pressure estimation, each feature component is normalized or standardized.

7. The method for intraoperative intraocular pressure measurement in the prone position based on array pressure sensing and machine learning according to claim 1, characterized in that: In step S4, the random forest regression model consists of multiple regression decision trees, and the model is pre-trained offline. During the training phase, a mapping relationship between the input feature vector and the real intraocular pressure value is constructed using multi-source datasets, including finite element simulation datasets, simulated eye experimental datasets, and clinical calibration datasets. During the operation phase, each regression decision tree outputs a local intraocular pressure prediction value for the input feature vector, integrates and calculates all the local intraocular pressure prediction values, and outputs the intraocular pressure estimate value at the corresponding time through nonlinear inversion.

8. The method for intraoperative intraocular pressure measurement in the prone position based on array pressure sensing and machine learning according to claim 1, characterized in that: In step S5, the smoothing process adopts first-order exponential smoothing to obtain a smooth intraocular pressure curve. The process suppresses sensor noise and instantaneous fluctuations while retaining the trend characteristics of slow or sustained high intraocular pressure, ensuring a sensitive response to abnormal changes in intraocular pressure.

9. The method for intraoperative intraocular pressure measurement in the prone position based on array pressure sensing and machine learning according to claim 1, characterized in that: In step S5, when the smoothed intraocular pressure value is consistently lower than the preset lower limit of the individualized intraocular pressure safety range and the duration exceeds the preset threshold, it can be determined as abnormally low intraocular pressure according to clinical needs, triggering a corresponding reminder signal; the preset lower limit is set according to the patient's preoperative baseline intraocular pressure value.

10. The method for intraoperative intraocular pressure measurement in the prone position based on array pressure sensing and machine learning according to claim 9, characterized in that: In step S5, the warning signal is an audible and visual alarm signal, and the reminder signal is an audible and visual reminder signal.

Images