AI-based eye postoperative care method and system

Abstract

The invention discloses an AI-based eye postoperative nursing method and system, and relates to the technical field of intelligent medical treatment, and the method comprises the steps: obtaining eye postoperative image data of a patient; analyzing an oculentum covering state of the ocular surface by using an artificial intelligence model and identifying an oculentum loss position; according to the loss position of the oculentum, the multi-point oculentum supplementing mechanism is automatically controlled to carry out accurate dispensing; according to the eyelash density, the trichiasis condition and the eyelid shape of a patient, the opening and closing states of the air holes are automatically adjusted, and sodium hyaluronate atomization is started for moisturizing; a patient is supported to preset a moisturizing mode, and a moisturizing enhancement program matched with the current eye environment is automatically executed when it is recognized that the oculentum is lost. Eye ointment loss is accurately recognized through near-infrared imaging and AI segmentation, supplementary application is conducted according to needs, the moisturizing environment is dynamically adjusted in combination with individualized anatomical characteristics, the liquid medicine deposition efficiency is remarkably improved, and dissipation is reduced; and a preset mode and a time-limited enhanced reinspection mechanism are combined, so that the exposure keratitis risk is remarkably reduced.

Description

Technical Field

[0001] This invention relates to the field of intelligent medical technology, and in particular to an AI-based method and system for postoperative eye care. Background Technology

[0002] Following eye surgery, patients often experience incomplete eyelid closure, weakened blinking function, or decreased self-protective ability during sleep, leading to prolonged corneal exposure to air and reduced tear film stability. This can result in corneal epithelial damage, exacerbated dry eye, and in severe cases, exposure keratitis. Traditional methods primarily involve topical application of eye ointment or gel, supplemented by passive protective measures such as sun-blocking eye masks or humidification chambers. However, this method is highly dependent on manual operation and subjective judgment, and the coverage of the eye ointment cannot be continuously monitored, especially at night when it is prone to melting, shifting, or falling off, rendering it ineffective.

[0003] Existing postoperative eye care methods generally have the following limitations: First, the coverage of eye ointment is not continuous enough at night. Due to lack of blinking during sleep, changes in body position, and local temperature, the eye ointment is prone to displacement or loss, making it difficult to achieve stable protection throughout the night. Second, there is a lack of real-time monitoring capabilities for the ocular surface condition. The opening and closing of the vents and the atomization parameters are mostly fixed settings, making it impossible to dynamically identify individualized risk factors such as eye ointment shedding, local corneal exposure, trichiasis, and palpebral fissure morphology. This leads to delayed intervention, making it difficult to balance effective moisturization and comfort, and increasing the risk of postoperative complications. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an AI-based postoperative eye care method and system that solves the problems of insufficient coverage of eye ointment at night and lack of real-time monitoring capability of ocular surface condition.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides an AI-based postoperative eye care method, comprising,

[0008] Obtain postoperative image data of the patient's eye;

[0009] Artificial intelligence models are used to analyze the coverage of eye ointment on the ocular surface and identify the locations where eye ointment has been lost.

[0010] Based on the location where the eye ointment has been lost, the multi-point eye ointment replenishment mechanism is automatically controlled to perform precise application.

[0011] Based on the patient's eyelash density, ingrown eyelashes, and eyelid shape, the system automatically adjusts the opening and closing of the vents and activates sodium hyaluronate atomization for moisturizing.

[0012] It supports patients to preset moisturizing modes and automatically executes a moisturizing enhancement program that matches the current eye environment when it detects that eye ointment has been lost.

[0013] As a preferred embodiment of the AI-based postoperative eye care method of the present invention, the step of acquiring postoperative eye image data of the patient includes:

[0014] By using a near-infrared structured light imaging module integrated inside the smart goggles, a two-second sequence of ocular surface images is continuously acquired while the patient's eyes are closed. The middle frame is selected as the representative frame, and the Laplace variance is calculated to assess the sharpness. If the sharpness reaches a preset threshold, the middle frame is used as the qualified representative frame; otherwise, the image is automatically reacquired until a clear image is obtained.

[0015] As a preferred embodiment of the AI-based postoperative eye care method of the present invention, the step of using an artificial intelligence model to analyze the ointment coverage status on the ocular surface and identify the location of ointment loss includes:

[0016] The qualified representative frames are subjected to grayscale conversion and standardization preprocessing. Based on the eyelid closure contour prior template generated when the patient first wears the device, the normalized cross-correlation matching algorithm is used to align the palpebral fissure area, and the region of interest of a fixed size is cropped with the center of the eyeball as the reference.

[0017] Input the region of interest into the lightweight U-Net++ semantic segmentation model, and output the probability distribution of eye ointment coverage;

[0018] The ointment coverage rate was calculated in three key clinical regions: the central eye area, the inner eye area, and the outer eye area. The results were compared with the clinically effective coverage threshold to construct a set of ointment loss locations.

[0019] If the set of locations where the eye ointment has been lost is empty, then the coverage is considered complete and an "No need to reapply ointment" instruction is output; otherwise, the set of locations where the eye ointment has been lost is output as the basis for ointment reapplying control.

[0020] As a preferred embodiment of the AI-based postoperative eye care method of the present invention, the step of automatically controlling a multi-point eye ointment replenishment mechanism to perform precise dot application based on the location of ointment loss includes:

[0021] For each area in the ointment loss set, the required ointment volume is calculated based on the degree of insufficient ointment coverage, and the volume is limited within a safe upper and lower limit. The corresponding micro piezoelectric pump is driven to perform the dotting operation. After the dotting is completed, there is a delay to wait for the film to stabilize, and a completion signal and operation log are fed back to the main state machine.

[0022] As a preferred embodiment of the AI-based postoperative eye care method of the present invention, the step of automatically adjusting the opening and closing state of the vents and activating sodium hyaluronate atomization for moisturizing according to the patient's eyelash density, ingrown eyelashes, and eyelid shape includes:

[0023] After the dot painting is completed, the candidate eyelash region image within the region of interest is input into the lightweight segmentation model. After lightweight convolution, the eyelash mask is output, and the ingrown eyelash situation is determined based on the calculated eyelash density index and the texture direction.

[0024] The eyelash density index is compared with the density judgment threshold. If the eyelash density is higher than the threshold, it is judged as thick eyelashes; otherwise, it is judged as not thick.

[0025] Simultaneously, the upper and lower eyelid margin curves are fitted, and the degree of incomplete eyelid closure is calculated to classify the exposure risk level, including: low risk, medium risk and high risk;

[0026] The system controls three groups of zonal vents and a central atomizing nozzle: If the eyelashes are thick, all zonal vents are closed; otherwise, if there are ingrown eyelashes, only the zonal vents corresponding to the location of the ingrown eyelashes are closed; if there are neither thick eyelashes nor ingrown eyelashes, the opening and closing of the vents are dynamically controlled according to the exposure risk level: when the exposure risk level is low, all three groups of zonal vents are open; when the exposure risk level is medium or high, only the vents in the eye zone are open.

[0027] The central atomizer is activated synchronously, and the atomization intensity is set according to the exposure risk level. The higher the risk, the greater the atomization intensity. If there are thick eyelashes or ingrown eyelashes, micro-positive pressure is activated to assist in penetrating the eyelash barrier. After atomization, the atomizer settles for a delay and returns to the adaptation completion signal.

[0028] As a preferred embodiment of the AI-based postoperative eye care method described in this invention, the method supporting patient-preset moisturizing modes includes:

[0029] After receiving the signal that the environment adaptation is complete, the current set of locations where the eye cream has been lost is read. When the set of locations where the eye cream has been lost is empty, the preset moisturizing mode is executed.

[0030] The preset moisturizing mode is selected by the user on the mobile terminal when wearing it for the first time, including sleep mode, day mode or sensitive mode, each mode corresponds to different atomization strategies and vent control logic;

[0031] The system operates in the selected mode and monitors the remaining power and medication levels in real time. When the power or medication levels are insufficient, it automatically switches to the sensitive mode and reports a mode downgrade event, while also recording the operation log.

[0032] As a preferred embodiment of the AI-based postoperative eye care method of the present invention, the step of automatically executing a moisturizing enhancement program that matches the current eye environment upon detecting eye ointment loss includes:

[0033] During the operation of the preset moisturizing mode, the image acquisition and eye ointment coverage analysis process is periodically triggered to update the set of eye ointment loss locations;

[0034] When the set of locations where eye ointment has been lost is not empty, the normal execution of the preset moisturizing mode is paused and the enhancement process is triggered. The eyelash density, ingrown eyelash status, and eyelid closure risk level obtained from the most recent environmental adaptation phase are read to generate an enhanced moisturizing strategy:

[0035] When eyelashes are thick or ingrown eyelashes are present, all vents in the zones are closed to form a relatively sealed cavity, and micro-positive pressure is activated to reduce the escape of the atomized liquid and improve droplet arrival. When there are neither thick eyelashes nor ingrown eyelashes, the opening and closing of the vents are dynamically controlled according to the degree of eyelid incomplete closure: when the exposure risk level is low, all three groups of vents in the zones are open; when the exposure risk level is medium or high, only the vents in the center of the eye are open.

[0036] The enhancement process increases atomization intensity and extends atomization duration. After completion, the film is deposited for a delay and a signal indicating that the enhanced moisturizing process is complete is returned to the main state machine. A new round of coverage verification is automatically triggered at a preset time after the enhancement process ends. If the set of locations where the eye ointment has been lost is still not empty, the enhancement process is executed again, and the number of consecutive executions does not exceed the preset limit.

[0037] Secondly, the present invention provides an AI-based postoperative eye care system, comprising,

[0038] The near-infrared imaging and image quality assessment module is used to acquire ocular surface images with eyes closed and to select clear representative frames based on Laplacian variance to ensure the reliability of subsequent analysis.

[0039] The intelligent recognition module for ointment coverage status is used to segment the ointment area using a lightweight semantic model, calculate the coverage rate of each clinical key area, and output a set of ointment loss locations.

[0040] The multi-point precision eye ointment refill module is used to dynamically calculate the ointment volume based on the coverage and drive the corresponding micro piezoelectric pump to complete the precise application of the ointment in different areas.

[0041] The ocular microenvironment adaptive control module is used to simultaneously adjust the opening and closing state of the zoned vents and activate the matching sodium hyaluronate atomization moisturizing based on eyelash density, ingrown eyelashes and eyelid morphology.

[0042] The personalized moisturizing mode management module supports users in setting up preset moisturizing strategies and automatically switches to an enhanced moisturizing program that matches the current eye environment when eye cream loss is detected.

[0043] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the AI-based postoperative eye care method as described in the first aspect of the present invention.

[0044] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the AI-based postoperative eye care method as described in the first aspect of the present invention.

[0045] The beneficial effects of this invention are as follows: This invention ensures high-quality ocular surface image input through near-infrared structured light imaging and a Laplacian variance self-selection mechanism. Based on a lightweight semantic segmentation model, it quantifies the coverage of the central, lateral, and lateral eye regions, constructs a set of ointment loss locations, and drives multi-point micro-piezoelectric pumps for precise reapplication, forming a closed loop of "recognition-decision-execution," effectively solving the problem of nighttime coverage interruption. By integrating individualized anatomical features such as eyelash density index, ingrown eyelash area, and eyelid incomplete closure angle, it dynamically adjusts the zonal venting pores, atomization flow rate, and micro-positive pressure assistance to construct an adaptive moisturizing strategy, improving sodium hyaluronate deposition efficiency and reducing evaporation. Through the synergy of preset modes and enhanced programs, it automatically triggers a timed retest mechanism, significantly reducing the risk of exposure keratitis, especially suitable for high-risk groups such as children and the elderly. Attached Figure Description

[0046] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a flowchart of an AI-based postoperative eye care method in Example 1.

[0048] Figure 2 This is a structural diagram of an AI-based postoperative eye care system in Example 1. Detailed Implementation

[0049] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0050] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0051] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0052] Example 1, referring to Figure 1 and Figure 2 This is the first embodiment of the present invention, which provides an AI-based postoperative eye care method, including the following steps:

[0053] S1. Obtain postoperative image data of the patient's eyes;

[0054] Specifically, acquiring postoperative image data of the patient's eyes includes:

[0055] By using an 850 nm near-infrared structured light imaging module integrated inside the smart goggle, images of the ocular surface region are continuously acquired for 2 seconds at a frame rate of 30 fps while the patient's eyes are closed, resulting in an image sequence containing 60 frames. Among them, the near-infrared structured light adopts a fixed dot matrix projection method that is coaxial with the imaging optical axis;

[0056] Select the 30th frame from the image sequence. As a representative frame, the Laplacian variance was calculated as a sharpness evaluation metric. :

[0057]

[0058] in, Refers to image height. Refers to image width. Frame image , The Laplace operator, Indicates the position of the image The pixel grayscale value at that location, The Laplace response value. The mean of the Laplace response. Refers to the normalization factor;

[0059] Set threshold ,when At that time, Input the data into an artificial intelligence model to identify the application status of the eye ointment;

[0060] Otherwise, control the imaging module to re-execute the image acquisition process until the sharpness threshold condition is met;

[0061] Among them, threshold The value is 150, which can be determined using existing techniques such as ROC curve analysis based on clinical image calibration or correlation experiments between AI segmentation performance and Laplacian variance.

[0062] This invention utilizes 850nm near-infrared structured light imaging and a Laplacian variance self-selection mechanism to automatically acquire high-quality ocular surface images while the patient's eyes are closed, effectively avoiding motion blur and illumination interference. A sharpness threshold is set, and images are only input into the AI ​​model when they meet quality requirements; otherwise, re-import is triggered, ensuring the accuracy and robustness of subsequent ointment application recognition from the source. This method significantly improves image acquisition efficiency and segmentation accuracy, reduces misjudgments and ineffective interventions, and provides a reliable perceptual basis for closed-loop postoperative care, especially suitable for nighttime, non-intrusive monitoring scenarios.

[0063] S2. Use artificial intelligence models to analyze the coverage of eye ointment on the eye surface and identify the locations where eye ointment has been lost;

[0064] Specifically, using artificial intelligence models to analyze the coverage of eye ointment on the ocular surface and identify the locations where eye ointment has been lost includes:

[0065] Output qualified representative frame image Standardized preprocessing was performed to convert the original near-infrared image into a single-channel grayscale image. Based on the eyelid closure contour prior template pre-stored on the inside of the smart goggle, a normalized cross-correlation matching algorithm was used to automatically align the palpebral fissure region in the closed-eye state. A 320×320 pixel region of interest (ROI) was cropped based on the projection point of the eyeball center. The pixel values ​​of the cropped image were normalized to the interval [0,1] to obtain the preprocessed image. :

[0066]

[0067] Will Input a lightweight U-Net++ semantic segmentation model, and the model output is an eye ointment coverage probability mask. ;

[0068] The eyelid closure contour prior template refers to the eyelid edge contour image in the closed state, which is generated by a rapid calibration process when the patient first wears the smart eye mask, and reflects the individualized anatomical characteristics of the patient. The rapid calibration process is an automatic or semi-automatic process for initializing individualized anatomical characteristics.

[0069] The normalized cross-correlation matching algorithm refers to comparing the eyelid closure contour prior template on the target image pixel by pixel, calculating the standardized correlation coefficient of each position. The closer the value is to 1, the more similar the pattern is, thereby achieving accurate positioning of a specific pattern.

[0070] The lightweight U-Net++ semantic segmentation model is a multi-scale nested skip connection encoder-decoder semantic segmentation network. It is a lightweight improvement on the standard U-Net++ structure, including an input layer, encoder, fusion layer, decoder and output layer.

[0071] The input layer is used to receive a single-channel preprocessed ocular surface image with a size of 320×320.

[0072] The encoder is composed of multi-level downsampling convolution modules, used to extract multi-scale eye ointment texture features;

[0073] The fusion layer consists of feature maps output from each downsampling stage of the encoder, which are fused step by step with the upsampling path through lateral convolution. Multiple resolution features at the same semantic level are added together after being subjected to 3×3 depthwise separable convolution to form enhanced features. All fusion paths use lightweight convolution to reduce computational overhead.

[0074] The decoder consists of a multi-level upsampling module. Each level improves the resolution of the feature map through bilinear interpolation and fuses multi-scale features from the corresponding level of the encoder and the horizontal path of the same semantic depth. After lightweight convolution, it outputs an enhanced feature map, thereby restoring spatial details and improving the boundary segmentation accuracy of the eye ointment region.

[0075] The output layer is used to output an eye ointment coverage probability mask of the same size as the input image;

[0076] Model training steps: Construct an offline eye image dataset for model training, perform preprocessing operations on the training images, input the preprocessed training images into the lightweight U-Net++ semantic segmentation model, and output the corresponding eye ointment coverage probability prediction results through the model's encoder, decoder, nested skip connections and feature fusion structure. Compare the prediction results with the corresponding labeled mask, calculate the segmentation loss function, and update the model parameters based on the error backpropagation algorithm. Repeat the above forward calculation and parameter update process, and gradually optimize the model parameters through multiple rounds of iterative training until the model's segmentation performance on the validation dataset meets the preset convergence condition. Solidify the trained model parameters and deploy them to the smart eye mask or its supporting computing unit for real-time recognition of eye ointment coverage status.

[0077] Define three clinically critical regions in the ROI image:

[0078] Eye area A circular area with a radius of 40 pixels centered on the image center corresponds to the high-risk area of ​​central corneal exposure.

[0079] Inner corner of the eye : The rectangular area along the lower edge of the nose, with coordinates ranging from ;

[0080] Outer corner of the eye The rectangular region at the lower temporal border, with coordinates ranging from [value missing]. ;

[0081] In each region Inside, calculate the coverage of the eye ointment. :

[0082]

[0083] in, For the region The total number of pixels, For indicator functions, The confidence threshold for segmentation;

[0084] The lowest ointment coverage threshold used to characterize clinical effectiveness was employed. Construct a set of locations where eye ointment is lost. :

[0085]

[0086] in, This is a set of region indexes used to determine the target area for applying touch-up paste;

[0087] like If the ointment coverage is complete, the command "No need to reapply ointment" will be output.

[0088] Otherwise, the set As a control signal output to the multi-point eye ointment refill mechanism, it drives the precise dotting operation of the corresponding area;

[0089] It should be noted that, The range of values The threshold calibration method of the semantic segmentation model in the existing technology can be used to determine the balance between the precision and recall of pixel classification in the covered area;

[0090] The value range can be set to 0.5 to 0.8, preferably 0.6. It can be determined by existing clinical drug use guidelines, statistical analysis of eye ointment coverage effectiveness experiments, or empirical threshold calibration methods based on historical sample data. 0.6 is an optimal value jointly calibrated by clinical evidence, anatomical patterns, algorithm performance, and empirical feedback, and has sufficient technical rationality and feasibility.

[0091] This invention achieves high-precision regional quantification of ointment coverage in the central, inner, and outer corners of the eye by aligning with a priori template of the eyelid closure contour and using a lightweight U-Net++ semantic segmentation model. This significantly improves the accuracy of segmentation boundaries and computational efficiency. Combined with clinically effective coverage thresholds, it dynamically constructs a set of ointment loss locations to accurately identify areas requiring reapplication. This method avoids the errors of traditional subjective assessments, improves the objectivity and reliability of postoperative care, and provides a stable and quantifiable decision-making basis for closed-loop automated intervention, making it particularly suitable for nighttime, non-intrusive monitoring scenarios.

[0092] S3. Based on the location where the eye ointment has been lost, the multi-point eye ointment replenishment mechanism is automatically controlled to perform precise dot application;

[0093] Specifically, based on the location where the eye ointment has been lost, the automatic control of the multi-point eye ointment replenishment mechanism performs precise spot application, including:

[0094] Collection of locations where eye ointment is lost For each missing region Using the calculated coverage of the eye ointment Replenish eye ointment volume :

[0095]

[0096] in, and These refer to the upper safety limit volume and the lower safety limit volume, respectively. This is the empirical gain coefficient;

[0097] It has three built-in independent micro piezoelectric pumps, each connected to one of the three ointment outlets via a microfluidic channel:

[0098] Pump C ointment outlet: located in the center of the lower eyelid, directly opposite the pupil projection point, 1.55mm from the eyelid margin;

[0099] Pump N ointment outlet at the inner corner of the eye: located 2 mm lateral to the lower eyelid margin on the nasal side;

[0100] Pump T ointment outlet at the outer corner of the eye: located 2 mm lateral to the lower eyelid margin on the temporal side;

[0101] For each missing region, activate the corresponding pump. Apply a drive signal and calculate the control duration. :

[0102]

[0103] in, This represents the pump's measured average flow rate.

[0104] After the activated pump completes the dotting process, it delays for 3 seconds to allow the eye ointment to spread and stabilize into a film before returning a "dotting complete" signal to the main state machine. It also records the replenishment area index and the volume of ointment replenished in each area. With driving time Timestamp and operation version number;

[0105] It should be noted that, and The determination is based on the minimum effective coverage dose and the maximum safe volume of the ocular surface, respectively, and can be obtained through existing technologies such as in vitro simulation, clinical dose response experiments, or microfluidic calibration.

[0106] The range of values ​​is It can be determined through existing techniques such as dose-response experiments based on animal models or clinical subjects, combined with regression analysis of the effect of improved ointment coverage and the degree of reduction in corneal exposure risk.

[0107] This invention precisely calculates the required ointment volume for each area based on the location of ointment loss, and uses three independent micro-piezoelectric pumps to achieve on-demand, quantitative application. The amount of ointment applied is strictly limited within a safe and effective range to avoid over-stimulation or insufficient coverage. The application location is precisely matched to the anatomical features of the central, incisional, and coccygeal regions of the eye, and the driving duration is dynamically determined by measured flow rate to ensure accurate dosage. After application, a delayed film formation process is implemented, and the execution log is logged to form a closed-loop control system. This approach significantly improves drug delivery accuracy, safety, and automation, effectively maintaining the postoperative corneal protective barrier.

[0108] S4. Based on the patient's eyelash density, ingrown eyelashes, and eyelid shape, automatically adjust the opening and closing of the vents and activate sodium hyaluronate atomization for moisturizing;

[0109] Specifically, based on the patient's eyelash density, ingrown eyelashes, and eyelid shape, the system automatically adjusts the opening and closing of the vents and activates sodium hyaluronate atomization for moisturizing, including:

[0110] After receiving the returned "dot painting complete" signal, the image represents the closed-eye frame. Within the ROI region (320×320), a lightweight segmentation model was used to extract the eyelash regions 5 mm above and below the lower eyelid margin to obtain the eyelash mask. :

[0111]

[0112] in, Represents pixels This area belongs to the eyelash region. Represents pixels Non-eyelash area;

[0113] The lightweight segmentation model includes: an input layer, an encoder, a decoder, a fusion layer, and an output layer;

[0114] The input layer is used to receive representative frame images of closed-eye images. The input image is a single-channel grayscale image with dimensions aligned to the ROI (Region of Interest).

[0115] The encoder performs multi-scale feature extraction on the input image and consists of at least three levels of downsampling convolutional modules, each of which includes depthwise separable convolutions. with point convolution The combination of these elements is used to connect batch normalization and nonlinear activation functions in sequence, and downsampling is achieved through convolution or pooling with a stride of 2.

[0116] The decoder is a module that upsamples and reconstructs the deep features output by the encoder step by step. It consists of at least three upsampling modules corresponding to the encoder. Each upsampling module includes an interpolation upsampling module and a convolution thinning module.

[0117] The fusion layer is as follows: it fuses features of the same scale as the decoder in the encoder. The fusion method is feature concatenation or element-wise addition. It also integrates channel and spatial information through fusion convolution to enhance the segmentation ability of eyelash filament boundaries.

[0118] The output layer is used to output an eyelash probability map of the same size as the input image, and obtains pixel confidence through Sigmoid activation, followed by binarization thresholding. Convert the probability map into an eyelash mask. ;

[0119] Model training steps: Construct an offline training dataset, perform preprocessing on the training images in the offline training dataset, including at least grayscale conversion, size normalization, eyelash candidate region cropping and pixel normalization, and perform data augmentation on the training images to improve robustness to different eyelash shapes and lighting conditions.

[0120] The preprocessed training image is input into the lightweight segmentation model, which outputs a predicted eyelash probability map.

[0121] The predicted probability map is compared with the labeled mask, and the segmentation loss function is calculated.

[0122] Based on the segmentation loss function, backpropagation of error is used to update the model parameters, and the process of "input training image - output prediction result - calculate loss - parameter update" is repeatedly executed for multiple rounds of iterative training until the performance of the validation set meets the preset convergence condition. The trained model parameters are then solidified and deployed to the smart eye mask or its supporting computing unit.

[0123] Eyelash masking area Internally, Canny edge detection is performed, the number of edge response pixels is counted, and the eyelash density index is calculated. :

[0124]

[0125] in, Refers to mask The number of Canny edge response pixels within the area. refer to The total number of pixels in the mask;

[0126] To increase the eyelash density index The signal is converted into an executable control signal, and the threshold is determined through calibration experiments before leaving the factory. Collect a sample set of eyelash images with manually labeled closed eyes, and calculate the value of each sample. Based on receiver operating characteristic (ROC) curve analysis, the Youden index was selected. The largest value was used as the optimal classification threshold, which was determined after validation with a large number of clinical samples. :

[0127] in, This represents the proportion of all real, thick eyelashes that are correctly judged as "thick." This represents the proportion of real, non-thick eyelashes that are misclassified as "thick" in all images.

[0128] like If so, it is judged as "thick eyelashes". If so, it is judged as "not dense";

[0129] right Orientation field estimation is performed on local textures within the eyelashes. If the main axis of the eyelashes points towards the cornea (angle < 30°), the ingrown eyelash area is marked. ;

[0130] Calculate the angle of incomplete eyelid closure and maximum gap width Exposure risk level is defined as: low risk Medium risk and high risk ;

[0131] The angle of incomplete eyelid closure and maximum gap width The results were obtained by fitting the quadratic curves of the upper and lower eyelid margins and calculating the tangent angle and the maximum vertical gap at the center of the pupil.

[0132] The smart eye mask features three groups of ventilation holes (C / N / T) and a central atomizing nozzle:

[0133] like Then close all three groups of vents.

[0134] Otherwise, if there is an ingrown eyelash area Then only close Corresponding ventilation holes for different zones;

[0135] If there are neither thick eyelashes nor ingrown eyelashes, the opening is dynamically adjusted according to the exposure risk level, including full opening in low risk, and opening only the central aperture group C in medium and high risk.

[0136] At the same time, immediately start the central ultrasonic nebulizer, and according to Dynamically set atomization flow rate :

[0137] If the risk level is low, then set the atomization flow rate to [value missing]. If the risk level is medium, then set the atomization flow rate to [value missing]. If it is high risk, then set the atomization flow rate. ;

[0138] If you have thick eyelashes or ingrown eyelashes, also use [the product / treatment]. A slightly positive pressure-assisted fan pushes droplets to penetrate the eyelash barrier;

[0139] Atomization remains constant The process takes 60 seconds, with the vent state switching 100ms before atomization begins. After atomization ends, there is a 10-second delay to allow droplets to settle and form a film. Then, a "Environment adaptation complete" signal is returned to the main state machine, and a log is recorded, including: , , , Opening and closing status of each orifice group, atomization flow rate Whether positive pressure is enabled, duration, timestamp, and operation version number;

[0140] After the environment adaptation is completed, it enters a low-power detection state and re-executes the image acquisition and eye ointment coverage analysis process according to a preset cycle or event trigger, continuously updating the set of eye ointment loss locations. .

[0141] This invention uses a lightweight model to accurately identify eyelash density and ingrown eyelash areas, and dynamically classifies exposure risk levels based on the angle of incomplete eyelid closure. It then adaptively closes corresponding vents, adjusts the sodium hyaluronate atomization flow rate, and, when necessary, activates micro-positive pressure to assist in penetrating the eyelash barrier. This strategy significantly improves the deposition efficiency of the medication on the corneal surface, reduces leakage, and simultaneously creates a personalized moisturizing microenvironment, effectively reducing the risk of exposure keratitis and achieving safe, efficient, and painless postoperative care.

[0142] S5. Supports patients to preset moisturizing modes and automatically executes a moisturizing enhancement program that matches the current eye environment when eye ointment loss is detected;

[0143] Specifically, the preset moisturizing modes supported by patients include:

[0144] After receiving the "Environment adaptation complete" signal from the main state machine, the current set of locations where the eye ointment has been lost is read. :

[0145] when If the preset moisturizing mode is not entered, the process will proceed; otherwise, the eye cream application state machine will continue running and trigger the next round of image acquisition and eye cream coverage analysis to update the process. Until Then it enters the preset mode;

[0146] When patients first wear the smart eye mask, a moisturizing mode option is provided via the accompanying mobile terminal, including:

[0147] Sleep mode: Low flow nebulization With all ventilation holes fully open, it is suitable for keeping your eyes closed for extended periods at night.

[0148] Daytime mode: Intermittent nebulization (starts for 30 seconds every 30 minutes, flow rate) The ventilation holes are controlled according to a medium-risk strategy.

[0149] Sensitive mode: Nebulization is turned off, and only basic ventilation through the vents is maintained. This mode is suitable for patients with a history of sodium hyaluronate allergy.

[0150] After user selection, pattern identifier and corresponding moisturizing parameter configuration table It is encrypted and stored in the eye mask's local non-volatile memory, and synchronized to the cloud account;

[0151] The moisturizing parameter configuration table Includes a set of atomization control parameters (in For atomization flow rate, This is an intermittent atomization cycle. (Single atomization duration), vent control strategy parameters Mode priority identifier ;

[0152] After loading the parameters, the preset moisturizing mode will be executed according to the parameter configuration table. The vent control module and the atomization module work together to drive the operation of the ventilation port control module.

[0153] when At that time, the atomization module is controlled to operate at a continuous low flow rate. It operates with all zone vents fully open to maintain a stable, humid environment over a long period. Indicates sleep mode;

[0154] when At that time, according to the atomization interval set in the parameter configuration table. The atomization process is activated every 30 minutes, with each atomization session lasting 30 seconds and a flow rate of [missing information]. Meanwhile, the ventilation control module dynamically adjusts according to the medium-risk ventilation strategy, among which... Indicates daytime mode;

[0155] when At this time, keep the atomizing module closed, and only control the vent to basic ventilation mode to avoid drug irritation. Indicates a sensitive mode;

[0156] During the process, the battery power and the remaining sodium hyaluronate solution were continuously monitored, and the monitoring results were compared with the safety thresholds in real time.

[0157] When the battery level is detected to be below 10% or the remaining liquid volume is below 2 mL, the current preset mode of atomization will be immediately stopped and the system will automatically switch to the sensitive mode while maintaining basic ventilation through the vents.

[0158] Push resource shortage warning messages to user terminals, report mode degradation events to the main state machine, and record the following logs: user-selected mode, actual execution parameters (hole status, atomization flow / frequency), resource status (battery level, liquid), timestamp, and version number.

[0159] This invention supports users to preset three moisturizing modes: sleep, daytime, and sensitive, to match different care scenarios as needed: the sleep mode provides continuous low-flow atomization and fully open vents to ensure nighttime comfort; the daytime mode uses intermittent atomization to balance efficiency and breathability; and the sensitive mode disables atomization but only provides ventilation, suitable for allergy sufferers. Once the eye ointment is fully applied, the selected mode is automatically activated, and the battery level and remaining medication are monitored in real time. When resources are insufficient, it intelligently downgrades to the sensitive mode to ensure safe operation. This solution balances personalization, comfort, and safety, significantly improving patient compliance and postoperative care experience.

[0160] Furthermore, upon detecting eye cream loss, it automatically executes a moisturizing enhancement program tailored to the current eye environment, including:

[0161] During the execution of the preset moisturizing mode, the intelligent recognition module for eye cream coverage status is periodically run to output the set of eye cream loss locations in real time, and continuously receive the set of locations.

[0162] When detected When the current eye surface is determined to be in a state of incomplete eye ointment coverage, a "eye ointment loss triggered" event signal is sent to the main state machine, and the normal execution process of the current preset moisturizing mode is paused, and the moisturizing enhancement determination stage is entered.

[0163] Read the eyelash density index calculated in the latest environment adaptation process. Ingrown eyelashes area Incomplete eyelid closure angle and maximum gap width Generate an enhanced moisturizing strategy:

[0164] like If ingrown eyelashes are present, close all three-part ventilation holes (C / N / T) to create a completely sealed cavity and prevent the evaporation of the new eye ointment or the escape of the atomizing liquid.

[0165] If you do not have thick eyelashes and do not have ingrown eyelashes:

[0166] like Only open the eye zone hole group C, if All three groups of ventilation holes are open;

[0167] In the moisturizing enhancement program, extend the atomization duration to 90 seconds and increase the atomization flow rate. The corresponding value is adjusted to an enhanced value:

[0168] when , ;

[0169] when hour, ;

[0170] when hour, ;

[0171] like If ingrown eyelashes are present, activate simultaneously. A slightly positive pressure auxiliary fan propels droplets through the eyelash barrier, employing a single high-intensity continuous atomization.

[0172] After the air vent status is switched (with a 100 ms allowance), immediately start the ultrasonic nebulizer;

[0173] The atomization lasts for 90 seconds, during which time any mode switching requests are prohibited;

[0174] After atomization is complete, delay for 10 seconds to allow the droplets to settle and form a film.

[0175] Return a "Enhanced moisturizing complete" signal to the main state machine;

[0176] Ten minutes after the enhancement program is completed, a new round of image acquisition is automatically triggered to verify whether the eye ointment coverage has been restored. If there is still loss, the enhancement program is restarted again, and it can be executed up to three times consecutively.

[0177] This invention automatically triggers a moisturizing enhancement program tailored to the current eye environment upon detecting eye ointment loss: based on eyelash density, ingrown eyelashes, and degree of eyelid incomplete closure, it dynamically closes vents, activates micro-positive pressure, extends atomization time to 90 seconds, and increases flow rate to 0.6–1.5 mL / min, forming a high-intensity, closed-loop moisturizing intervention. After the enhancement, the coverage status is automatically rechecked, and this process can be repeated up to three times consecutively to ensure effective repair of the protective barrier. This mechanism achieves a closed loop of "sensing-response-verification," significantly reducing the risk of corneal exposure while maintaining comfortable basic care, making it particularly suitable for children and elderly patients with poor compliance.

[0178] This embodiment also provides an AI-based postoperative eye care system, including:

[0179] The near-infrared imaging and image quality assessment module is used to acquire ocular surface images with eyes closed and to select clear representative frames based on Laplacian variance to ensure the reliability of subsequent analysis.

[0180] The intelligent recognition module for ointment coverage status is used to segment the ointment area using a lightweight U-Net++ model, calculate the coverage rate of each clinical key area, and output the set of ointment loss locations.

[0181] The multi-point precision eye ointment refill module is used to adjust the amount of ointment based on the number of cisplatins and coverage. Dynamically calculate the paste volume and drive the corresponding micro piezoelectric pump to complete precise spot application in designated areas;

[0182] The ocular microenvironment adaptive control module is used to simultaneously adjust the opening and closing state of the zoned vents and activate the matching sodium hyaluronate atomization moisturizing based on eyelash density, ingrown eyelashes and eyelid morphology.

[0183] The personalized moisturizing mode management module supports users in setting up preset moisturizing strategies and automatically switches to an enhanced moisturizing program that matches the current eye environment when eye cream loss is detected.

[0184] This embodiment also provides a computer device applicable to an AI-based postoperative eye care method, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize an AI-based postoperative eye care method as proposed in the above embodiment.

[0185] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0186] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements an AI-based postoperative eye care method as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0187] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An AI-based postoperative eye care method, characterized in that: include, Obtain postoperative image data of the patient's eye; Artificial intelligence models are used to analyze the coverage of eye ointment on the ocular surface and identify the locations where eye ointment has been lost. Based on the location where the eye ointment has been lost, the multi-point eye ointment replenishment mechanism is automatically controlled to perform precise application. Based on the patient's eyelash density, ingrown eyelashes, and eyelid shape, the system automatically adjusts the opening and closing of the vents and activates sodium hyaluronate atomization for moisturizing. It supports patients to preset moisturizing modes and automatically executes a moisturizing enhancement program that matches the current eye environment when it detects that eye ointment has been lost.

2. The AI-based postoperative eye care method as described in claim 1, characterized in that: The acquisition of postoperative image data of the patient's eye includes: By using a near-infrared structured light imaging module integrated inside the smart goggles, a two-second sequence of ocular surface images is continuously acquired while the patient's eyes are closed. The middle frame is selected as the representative frame, and the Laplace variance is calculated to assess the sharpness. If the sharpness reaches a preset threshold, the middle frame is used as the qualified representative frame; otherwise, the image is automatically reacquired until a clear image is obtained.

3. The AI-based postoperative eye care method as described in claim 2, characterized in that: The method of using artificial intelligence models to analyze the coverage of eye ointment on the ocular surface and identify the locations where eye ointment has been lost includes: The qualified representative frames are subjected to grayscale conversion and standardization preprocessing. Based on the eyelid closure contour prior template generated when the patient first wears the device, the normalized cross-correlation matching algorithm is used to align the palpebral fissure area, and the region of interest of a fixed size is cropped with the center of the eyeball as the reference. Input the region of interest into the lightweight U-Net++ semantic segmentation model, and output the probability distribution of eye ointment coverage; The ointment coverage rate was calculated in three key clinical regions: the central eye area, the inner eye area, and the outer eye area. The results were compared with the clinically effective coverage threshold to construct a set of ointment loss locations. If the set of locations where the eye ointment has been lost is empty, then the coverage is considered complete and an "No need to reapply ointment" instruction is output; otherwise, the set of locations where the eye ointment has been lost is output as the basis for ointment reapplying control.

4. The AI-based postoperative eye care method as described in claim 3, characterized in that: The automatic control of the multi-point eye ointment replenishment mechanism for precise application based on the location of ointment loss includes: For each area in the ointment loss set, the required ointment volume is calculated based on the degree of insufficient ointment coverage, and the volume is limited within a safe upper and lower limit. The corresponding micro piezoelectric pump is driven to perform the dotting operation. After the dotting is completed, there is a delay to wait for the film to stabilize, and a completion signal and operation log are fed back to the main state machine.

5. The AI-based postoperative eye care method as described in claim 4, characterized in that: The automatic adjustment of the opening and closing of the vents and activation of sodium hyaluronate atomization for moisturizing based on the patient's eyelash density, ingrown eyelashes, and eyelid shape includes: After the dot painting is completed, the candidate eyelash region image within the region of interest is input into the lightweight segmentation model. After lightweight convolution, the eyelash mask is output, and the ingrown eyelash situation is determined based on the calculated eyelash density index and the texture direction. The eyelash density index is compared with the density judgment threshold. If the eyelash density is higher than the threshold, it is judged as thick eyelashes; otherwise, it is judged as not thick. Simultaneously, the upper and lower eyelid margin curves are fitted, and the degree of incomplete eyelid closure is calculated to classify the exposure risk level, including: low risk, medium risk and high risk; The system controls three groups of zonal vents and a central atomizing nozzle: If the eyelashes are thick, all zonal vents are closed; otherwise, if there are ingrown eyelashes, only the zonal vents corresponding to the location of the ingrown eyelashes are closed; if there are neither thick eyelashes nor ingrown eyelashes, the opening and closing of the vents are dynamically controlled according to the exposure risk level: when the exposure risk level is low, all three groups of zonal vents are open; when the exposure risk level is medium or high, only the vents in the eye zone are open. The central atomizer is activated synchronously, and the atomization intensity is set according to the exposure risk level. The higher the risk, the greater the atomization intensity. If there are thick eyelashes or ingrown eyelashes, micro-positive pressure is activated to assist in penetrating the eyelash barrier. After atomization, the atomizer settles for a delay and returns to the adaptation completion signal.

6. The AI-based postoperative eye care method as described in claim 5, characterized in that: The patient-preset moisturizing modes include: After receiving the signal that the environment adaptation is complete, the current set of locations where the eye cream has been lost is read. When the set of locations where the eye cream has been lost is empty, the preset moisturizing mode is executed. The preset moisturizing mode is selected by the user on the mobile terminal when wearing it for the first time, including sleep mode, day mode or sensitive mode, each mode corresponds to different atomization strategies and vent control logic; The system operates in the selected mode and monitors the remaining power and medication levels in real time. When the power or medication levels are insufficient, it automatically switches to the sensitive mode and reports a mode downgrade event, while also recording the operation log.

7. The AI-based postoperative eye care method as described in claim 6, characterized in that: The process of automatically executing a moisturizing enhancement procedure that matches the current eye environment upon detecting eye cream loss includes: During the operation of the preset moisturizing mode, the image acquisition and eye ointment coverage analysis process is periodically triggered to update the set of eye ointment loss locations; When the set of locations where eye ointment has been lost is not empty, the normal execution of the preset moisturizing mode is paused and the enhancement process is triggered. The eyelash density, ingrown eyelash status, and eyelid closure risk level obtained from the most recent environmental adaptation phase are read to generate an enhanced moisturizing strategy: When eyelashes are thick or ingrown eyelashes are present, all vents in the zones are closed to form a relatively sealed cavity, and micro-positive pressure is activated to reduce the escape of the atomized liquid and improve droplet arrival. When there are neither thick eyelashes nor ingrown eyelashes, the opening and closing of the vents are dynamically controlled according to the degree of eyelid incomplete closure: when the exposure risk level is low, all three groups of vents in the zones are open; when the exposure risk level is medium or high, only the vents in the center of the eye are open. The enhancement process increases atomization intensity and extends atomization duration. After completion, the film is deposited for a delay and a signal indicating that the enhanced moisturizing process is complete is returned to the main state machine. A new round of coverage verification is automatically triggered at a preset time after the enhancement process ends. If the set of locations where the eye ointment has been lost is still not empty, the enhancement process is executed again, and the number of consecutive executions does not exceed the preset limit.

8. An AI-based postoperative eye care system, based on the AI-based postoperative eye care method according to any one of claims 1 to 7, characterized in that: include, The near-infrared imaging and image quality assessment module is used to acquire ocular surface images with eyes closed and to select clear representative frames based on Laplacian variance to ensure the reliability of subsequent analysis. The intelligent recognition module for eye ointment coverage status is used to segment the eye ointment area using a lightweight semantic segmentation model, calculate the coverage rate of key clinical areas, and output a set of eye ointment loss locations. The multi-point precision eye ointment replenishment module is used to dynamically calculate the replenishment volume based on the set of eye ointment loss locations and coverage, and drive the corresponding micro piezoelectric pump to complete the precise application of the ointment in different areas. The ocular microenvironment adaptive control module is used to simultaneously adjust the opening and closing state of the zoned vents and activate the matching sodium hyaluronate atomization moisturizing based on eyelash density, ingrown eyelashes and eyelid morphology. The personalized moisturizing mode management module supports users in setting up preset moisturizing strategies and automatically switches to an enhanced moisturizing program that matches the current eye environment when eye cream loss is detected.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the AI-based postoperative eye care method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the AI-based postoperative eye care method according to any one of claims 1 to 7.

Images