CN115294071A

CN115294071A - Tear film detection system and method based on video data

Info

Publication number: CN115294071A
Application number: CN202210959255.5A
Authority: CN
Inventors: 袁进; 谢志; 肖鹏; 周昊; 王耿媛; 何尧; 邓宇晴
Original assignee: Zhongshan Ophthalmic Center
Current assignee: Zhongshan Ophthalmic Center
Priority date: 2022-08-10
Filing date: 2022-08-10
Publication date: 2022-11-04

Abstract

The present disclosure describes a tear film detection system and method based on video data, the system including an acquisition module configured to acquire a plurality of detection frames corresponding to the video data for the tear film and as a tear film data queue; a state analysis module configured to determine an eye state of an intermediate detection frame based on a change in a shape size of a corneal region in three consecutive detection frames in the tear film data queue; a detection segment analysis module configured to acquire a plurality of detection segments corresponding to the video data based on the eye states of the detection frames obtained by the state analysis module, and determine detection results of the respective detection segments based on the detection frames whose eye states are open states; and a result analysis module configured to determine a target detection result based on a plurality of detection results corresponding to the plurality of detection segments. Therefore, the segmentation precision of the tear film detection is high, and the tear film detection support system can support real-time analysis and is convenient.

Description

Tear film detection system and method based on video data

Technical Field

The present disclosure relates generally to intelligent medical systems, and more particularly to a system and method for tear film detection based on video data.

Background

The tear film is a dynamic and complex structure consisting of an anterior lipid layer, a middle aqueous layer, and an inner mucus layer. The tear film structure prevents tear evaporation and provides a smooth optical surface for the cornea. Clinical studies have shown that studies of tear film status (e.g., tear film stability) are of great interest. For example, tear film status may aid in the assessment of some ocular diseases (e.g., dry eye). As another example, tear film status may guide the wearing of some ophthalmic products (e.g., contact lenses).

Currently, two approaches to assessing tear film status are generally used: one is to detect the projected deformation by projecting Placido rings (concentric rings) with the subject's eyes continuously open and analyzing the ring image features to determine the tear film state (e.g., tear film break-up time); another fluorescein tear-break-up time (FTBUT) test is to drop a preset dose of sodium fluorescein into the eye and analyze the time the tear film remains stable without blinking to determine the tear film status. Both of the above approaches can generally be combined with video analysis techniques to obtain the state of the tear film. For example, patent document (CN 111062443A) provides a tear film break-up time detection system based on deep learning, which extracts all frames of a captured tear film break-up time inspection video, inputs the extracted frames into a classification model, and outputs a classification label of each frame of image to form a 0/1 sequence; extracting the longest eye opening segment in the sequence as a detection segment; and (4) respectively inputting each detection section into a dry spot segmentation model for segmentation prediction, and finally obtaining the integral tear film rupture time of the person to be detected.

However, in the detection process based on Placido ring image analysis, the subject needs to continuously open the eyes, the detection result is easily affected by the noise of the eyelashes at the anterior segment of the eyes, the recording of the subject needs to be restarted after the subject blinks or moves the eyes instantly, and the convenience still needs to be improved; the classification model of patent document (CN 111062443A) directly identifies the open-close state, and does not consider the influence of the dynamic change process of the cornea on the classification precision, and further influences the segmentation precision of the detection segment.

Disclosure of Invention

The present disclosure has been made in view of the above circumstances, and an object thereof is to provide a system and a method for tear film detection based on video data, which can improve the accuracy of segmentation of tear film detection, support real-time analysis, and facilitate the detection.

To this end, the present disclosure provides in a first aspect a video data-based tear film detection system comprising: the device comprises an acquisition module, a state analysis module, a detection section analysis module and a result analysis module; the acquisition module is configured to acquire a plurality of detection frames corresponding to the video data for the tear film over time and serve as a tear film data queue; the state analysis module is configured to receive a detection frame to be analyzed, acquire three detection frames in the tear film data queue, in which the detection frame to be analyzed is located in the middle and continuously, and determine an eye state of the detection frame to be analyzed based on a change condition of a shape size of a cornea area in the three detection frames, wherein the eye state includes an open state, an open eye process state and a closed eye process state, and the shape size of the cornea area is represented by a relative size; the detection segment analysis module is configured to read detection frames in the tear film data queue, acquire a plurality of detection segments corresponding to the video data based on the eye states of the detection frames obtained by the state analysis module, and determine detection results of the detection segments based on the detection frames with the eye states being open states, wherein a start frame and an end frame of each detection segment are respectively determined by the detection frames with the eye states being open eye process states and the detection frames with the eye states being closed eye process states in the tear film data queue; and the result analysis module is configured to determine a target detection result based on a plurality of detection results corresponding to the plurality of detection segments.

In the present disclosure, the eye state of the detection frame of the video data of the tear film is determined more accurately based on the dynamic change process of the cornea, a plurality of detection segments are determined based on the eye state of the detection frame, and then the plurality of detection segments are analyzed to obtain the target detection result. In this case, the eye state determined based on the dynamic change process of the cornea enables higher segmentation accuracy of tear film detection, and more accurate segmentation-based support of real-time analysis and interpretation. In addition, the shape and the size of the cornea area are represented by relative sizes, so that the negative influence of possible offset among a plurality of detection frames on the change condition can be reduced, and the method can be compatible with various complex environments for acquiring video data.

Additionally, in the system according to the first aspect of the present disclosure, optionally, the video data includes at least one of a video file and video buffer data, the video data being acquired after staining the tear film with a contrast agent. Thereby, non-real time and/or real time video data can be analyzed. In addition, the contrast agent can improve the contrast of images in video data, and is beneficial to identifying the state of the tear film.

In addition, in the system according to the first aspect of the present disclosure, optionally, the three detection frames sequentially include a previous frame, an intermediate frame, and a next frame, where the intermediate frame is the detection frame to be analyzed, and the state analysis module is configured to: if the change condition is that the shape and size of the cornea region of the previous frame is larger than a preset size, the shape and size of the cornea region of the intermediate frame is larger than the preset size, and the shape and size of the cornea region of the next frame is larger than the preset size, the eye state of the intermediate frame is in an open state; if the change condition is that the shape and size of the cornea region of the previous frame is smaller than the preset size, the shape and size of the cornea region of the intermediate frame is larger than the preset size, and the shape and size of the cornea region of the next frame is larger than the shape and size of the cornea region of the intermediate frame, the eye state of the intermediate frame is the eye opening process state; and if the change condition is that the shape and size of the cornea region of the previous frame is larger than the preset size, the shape and size of the cornea region of the intermediate frame is smaller than the shape and size of the cornea region of the previous frame, and the shape and size of the cornea region of the next frame is smaller than the preset size, the eye state of the intermediate frame is the eye closing process state.

In addition, in the system according to the first aspect of the present disclosure, optionally, the relative size is a ratio of a size of a corneal region of the detection frame to a size of a reference object, and the reference object is at least one of a whole image and a region of interest corresponding to the detection frame. In this case, the negative influence of the possible offset between a plurality of detection frames on the change situation can be reduced, and thus various complex environments for acquiring video data can be compatible.

Additionally, in the system according to the first aspect of the present disclosure, optionally, the tear film data queue is a buffered data queue, and the detection segment analysis module is configured to: reading a detection frame from the buffer data queue according to the time sequence and taking the detection frame as a current frame; determining the eye state of the current frame based on the change situation of the shape and the size of the cornea region between the previous frame of the current frame and the next frame of the current frame; if the eye state of the current frame is the eye opening process state, taking the current frame as the initial frame of the detection section; if the eye state of the current frame is the eye closing process state, taking the current frame as an end frame of the detection section of the determined initial frame, and further determining the detection section; and updating the current frame as the next detection frame in the buffer data queue and continuing to judge based on the current frame until the buffer data queue is read. In this case, a more accurate detection segment is obtained in real time based on the eye state determined by the dynamic change process of the cornea, and real-time analysis can be supported based on the more accurate detection segment.

Additionally, in the system according to the first aspect of the present disclosure, optionally, the detection result includes at least one of a tear film break-up time and a tear film break-up morphology, and the target detection result includes at least one of a target tear film break-up time and a target tear film break-up morphology.

In the system according to the first aspect of the present disclosure, it is preferable that, when the eye state of one detection frame is in an open state, it is determined whether or not a corneal region of the detection frame is ruptured, and when the corneal region of the detection frame is ruptured, a tear film rupture time corresponding to the detection segment to which the detection frame belongs is calculated based on a position of the detection frame in the detection segment to which the detection frame belongs, and/or a tear film rupture state corresponding to the detection segment to which the detection frame belongs is acquired based on the detection frame.

Additionally, in the system according to the first aspect of the present disclosure, optionally, the result analysis module is configured to average a plurality of tear film break up times corresponding to the plurality of detection segments to determine the target tear film break up time.

In addition, in the system according to the first aspect of the present disclosure, optionally, a pre-trained cornea region segmentation model based on deep learning is further included, where the cornea region segmentation model is used to obtain a cornea region of the detection frame; and/or further comprises a pre-trained tear film rupture detection model based on deep learning, wherein the tear film rupture detection model is used for detecting whether the cornea area of the detection frame is ruptured or not. Under the condition, the deep learning-based method can learn the high-level semantic feature information of the corneal region in the video data as much as possible through the marking information of the training data, can be compatible with the video data under different acquisition conditions, and further can obtain a more accurate result.

A second aspect of the present disclosure provides a method for tear film detection based on video data, including obtaining a plurality of detection frames corresponding to the video data for tear film and acting as a tear film data queue; reading detection frames in the tear film data queue, acquiring a plurality of detection segments corresponding to the video data based on eye states of the detection frames, and determining a detection result of each detection segment based on the detection frames of which the eye states are opening states, wherein the eye states comprise opening states, eye opening process states and eye closing process states, determining a starting frame and an ending frame of each detection segment respectively from the detection frames of which the eye states are the opening eye process states and the detection frames of which the eye states are the eye closing process states in the tear film data queue, acquiring three detection frames of which the detection frames to be analyzed are positioned in the middle and are continuous in the eye states of the analysis detection frames in the tear film data queue, determining the eye states of the detection frames to be analyzed based on the change situation of the shape size of a cornea region in the three detection frames, and expressing the shape size of the cornea region by relative size; and determining a target detection result based on a plurality of detection results corresponding to the plurality of detection segments.

In the present disclosure, the eye state of the detection frame of the video data of the tear film is determined more accurately based on the dynamic change process of the cornea, a plurality of detection segments are determined based on the eye state of the detection frame, and then the plurality of detection segments are analyzed to obtain the target detection result. In this case, the eye state determined based on the dynamic change process of the cornea enables higher segmentation accuracy of tear film detection, and more accurate segmentation-based support of real-time analysis and interpretation. In addition, the shape and the size of the cornea area are expressed by relative sizes, so that the negative influence of possible shifts among a plurality of detection frames on the change condition can be reduced, and various complex environments for acquiring video data can be further compatible.

According to the system and the method, the segmentation precision of tear film detection is high, real-time analysis is supported, and the tear film detection system and the method based on the video data are convenient and fast.

Drawings

The disclosure will now be explained in further detail by way of example only with reference to the accompanying drawings, in which:

fig. 1 is an exemplary schematic diagram illustrating a detection environment to which examples of the present disclosure relate.

Fig. 2 is an exemplary block diagram illustrating a detection system to which examples of the present disclosure relate.

Fig. 3 is an exemplary flowchart illustrating determining an eye state for three consecutive detection frames according to an example of the present disclosure.

Fig. 4A is a schematic diagram showing a relevant image of a detection frame in an open state and not broken according to an example of the present disclosure.

Fig. 4B is a schematic diagram illustrating an open state and cracked related image of a detection frame according to an example of the present disclosure.

Fig. 4C is a schematic diagram illustrating a correlation image of a detection frame in which a state of a closing process and a state of rupture cannot be recognized according to an example of the present disclosure.

Fig. 5 is an exemplary flow chart illustrating detection segment identification in accordance with an example of the present disclosure.

Fig. 6 is an exemplary flowchart illustrating a method of training a corneal region segmentation model according to an example of the present disclosure.

Fig. 7 is an exemplary flow chart illustrating a training method of a tear film break-up detection model according to an example of the present disclosure.

Fig. 8 is an exemplary flow chart illustrating a video data-based tear film detection method in accordance with examples of the present disclosure.

Detailed Description

Hereinafter, preferred embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. In the following description, the same components are denoted by the same reference numerals, and redundant description thereof is omitted. The drawings are schematic and the ratio of the dimensions of the components and the shapes of the components may be different from the actual ones. It is noted that the terms "comprises" and "comprising," and any variations thereof, in this disclosure, such that a process, method, system, article, or apparatus that comprises or has a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include or have other steps or elements not expressly listed or inherent to such process, method, article, or apparatus. All methods described in this disclosure can be performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by context.

As mentioned above, there are some problems with the prior art solutions. In addition, the detection process based on Placido ring image analysis is not favorable for acquiring the form of the tear film due to the projection of the ring. The existing method generally simply divides the flow into the judgment of the starting time and the judgment of the fracture time, does not consider the dynamic judgment condition, and cannot support the real-time analysis. Some other methods adopt post-processing, that is, after video data is received, the video data is uniformly preprocessed to preliminarily obtain detection sections, and then the detection sections are analyzed to further screen the detection sections, so that real-time analysis cannot be supported.

Examples of the present disclosure propose a scheme for identifying detection segments and/or tear film detections in order to address all or part of the problems described above. The scheme can enable the segmentation precision of tear film detection to be high, support real-time analysis and be convenient. Specifically, the scheme can accurately determine the eye state of a detection frame of video data of the tear film based on the dynamic change process of the cornea, determine a plurality of detection segments based on the eye state of the detection frame, and further analyze the plurality of detection segments to obtain a target detection result. In this case, the eye state determined based on the dynamic change process of the cornea can make the segmentation accuracy of tear film detection higher, and real-time analysis can be supported based on the more accurate segmentation. In addition, the scheme related to the example of the present disclosure can support real-time analysis, and further can support timely obtaining of the detection result and/or the target detection result of each detection segment.

In addition, the scheme that this disclosed example relates to compares detection process based on Placido ring image analysis more convenient, easily promotes. Examples of the present disclosure relate to an approach that may employ a lightweight deep learning model to operate on video data in real-time in a GPU (graphics processing unit) environment.

It should be noted that, although the scheme related to the examples of the present disclosure is particularly suitable for real-time analysis of video data, the scheme is also suitable for analysis of completely acquired video data.

The tear film detection system based on video data to which examples of the present disclosure relate may be used to identify detection segments and/or tear film detections. Tear film detection systems to which examples of the present disclosure relate may also be referred to as detection systems or tear film detection systems and the like.

Examples of the present disclosure will be described in detail below with reference to the accompanying drawings. Fig. 1 is an exemplary schematic diagram illustrating a detection environment 1 to which an example of the present disclosure relates. As shown in fig. 1, a detection environment 1 may include a detection device 100 and a detection system 200.

The detection device 100 may be used to acquire video data for the tear film 2 of the subject. For example, the detection apparatus 100 may be a slit-lamp microscope. The detection device 100 may transmit the captured video data to the detection system 200. In some examples, an observer 3 (e.g., a doctor) can observe the tear film 2 of the subject via the detection apparatus 100 to adjust an imaging field of view and/or identify a tear film state of the tear film 2.

The detection system 200 may receive video data and identify detected segments in the video data and/or perform tear film detection based on the video data. For example, during the examination of the tear film 2 of the subject using the detection apparatus 100, the detection apparatus 100 may transmit the acquired video data to the detection system 200 in real time, and the detection system 200 may analyze the real-time video data and respond to the detection results of the respective detection segments in real time until the completion of the tear film examination by the subject is confirmed. Additionally, the detection system 200 may be implemented by a computing device. The computing device may be any device with computing capabilities. For example, the computing device may be a cloud server, a personal computer, a mainframe, a distributed computing system, and so on.

Fig. 2 is an exemplary block diagram illustrating a detection system 200 according to examples of the present disclosure.

Referring to fig. 2, the detection system 200 may include an acquisition module 210, a state analysis module 230, a detection segment analysis module 250, and a result analysis module 270. The acquisition module 210 can be configured to determine a tear film data queue based on the video data for tear film 2. The state analysis module 230 may be configured to determine the eye state of the detection frame based on a change in the corneal region between consecutive detection frames. The detection segment analysis module 250 may be configured to acquire a plurality of detection segments corresponding to the video data based on the eye state of the detection frame. The result analysis module 270 may be configured to determine a target detection result based on the plurality of detection segments.

Referring to fig. 2, in this embodiment, the acquisition module 210 can be configured to determine a tear film data queue based on video data for tear film 2. The video data may correspond to a plurality of detection frames (i.e., a plurality of frames of tear film data) that vary over time. In some examples, the acquisition module 210 may be configured to use a plurality of detection frames corresponding to the video data over time as a tear film data queue. For example, for video data having a duration of 1-2 minutes, 60 frames per second, 3600-7200 detection frames may be obtained as a tear film data queue. In addition, the detection frame may be an image or a data stream.

In some examples, the video data may include at least one of a video file and video buffer data (which may also be referred to as a video stream). The video file may be data that has been completely captured. This enables analysis of non-real-time video data. In some examples, the video file may be stored in a storage unit (e.g., a storage server) from which the detection system 200 may read the video file and analyze it. Additionally, the video buffer data may be data collected in real-time. Thus, real-time video data can be analyzed on line. In some examples, the video buffer data may be data generated by the subject in real-time during the tear film examination, which the detection system 200 may receive and analyze in real-time.

In some examples, the video data may be data acquired after staining the tear film 2 with a contrast agent. In this case, the contrast of the image in the video data can be improved, which is advantageous for identifying the state of the tear film (e.g., tear film break morphology). In some examples, the contrast agent may be sodium fluorescein. For example, tear film 2 can be stained with sodium fluorescein, and when tear film 2 breaks, a dry spot appears on the tear film surface. This makes it possible to easily observe and recognize the tear film state. In other examples, the video data may also be acquired in a non-intrusive manner.

In some examples, the acquisition module 210 may be configured to decode video data (e.g., a video file and/or video buffer data) into a plurality of detected frames using video decoding techniques. Thus, a plurality of detection frames varying with time corresponding to the video data can be obtained. In some examples, the video data may be derived from a subject. For example, the video data may be obtained by the detection apparatus 100 during a tear film examination by the subject.

In some examples, a detection frame entering the tear film data queue may wait for detection segment analysis module 250 to acquire for detection segment partitioning. Specifically, the detection segment analysis module 250 may determine the eye state of the detection frame through the state analysis module 230, and then segment the video data based on the eye state of the detection frame to obtain a plurality of detection segments.

In some examples, the tear film data queue may be a buffered data queue. The buffered data queue may be stored in memory. In this case, the manner based on buffering the detection frame can improve the data reading speed and enable quick response, thereby further supporting real-time analysis.

Additionally, the buffered data queue may be flushed. For example, a video file may be loaded in segments by flushing a buffer data queue, and/or real-time video buffer data may be received.

With continued reference to fig. 2, in this embodiment, the state analysis module 230 may be configured to determine the eye state of the detection frame based on the change of the corneal area between consecutive detection frames. A plurality of successive detection frames may constitute a frame data set.

In some examples, the consecutive plurality of detection frames may be three consecutive detection frames (i.e., three adjacent detection frames) in the tear film data queue. That is, the frame data set may consist of three consecutive inspection frames in the tear film data queue. In other words, the frame data set may include consecutive previous, intermediate and next frames in sequence.

Specifically, the state analysis module 230 may be configured to receive a detection frame to be analyzed, obtain a frame data set corresponding to the detection frame to be analyzed, and determine the eye state of the detection frame to be analyzed based on a change in the shape and size of a corneal region between the detection frames in the frame data set, where the frame data set may be composed of three consecutive detection frames in the tear film data queue, and the detection frame to be analyzed is located in the middle of the three consecutive detection frames. That is, the eye state of one detection frame can be determined by the change in the shape and size of the corneal region between the frame preceding the frame, the data of the frame itself (i.e., the intermediate frame), and the frame succeeding the frame. In this case, the eye state of the intermediate frame is determined by analyzing the change in the size of the shape of the corneal region of three consecutive detection frames, taking into account that the tear film 2 is a dynamic structure and the dynamic change process of the cornea, and the eye state of the detection frames can be determined more accurately than in the prior art in which the dynamic change process is ignored by classifying one detection frame alone. In addition, the change condition of the cornea area is used for judging the eye state, and compared with a scheme of predicting the eye state by adopting a machine learning model in the prior art, the method has better interpretability. In addition, the adaptability to various extreme conditions (such as instantaneous blinking) can be more stably met.

In addition, the size and shape of the corneal region can be expressed by relative sizes. That is, the eye state of the detection frame to be analyzed can be determined based on the variation in the relative size of the corneal region between the detection frames in the frame data set. In this case, the negative effect of possible offset between multiple detection frames (i.e., misalignment between multiple detection frames) on the variation situation can be reduced, and thus various complex environments for collecting video data can be compatible.

In addition, the relative size of the corneal region may be the size of the relative reference. In some examples, the relative size may be a ratio of the size of the corneal region of the detection frame to the size of the reference object. The reference object may be an object with a relatively fixed size between a plurality of detection frames. In some examples, the reference object may be at least one of an entire image corresponding to the detection frame (i.e., an entire image corresponding to the detection frame) and a region of interest in the detection frame. For example, the relative size may be a ratio of the size of the corneal region of the detection frame to the size of the entire map. Additionally, the region of interest may include, but is not limited to, a stained area. In other examples, the size of the shape of the corneal region may also be expressed in absolute terms.

In some examples, the shape size of the corneal region may be the area of the corneal region. Accordingly, the relative size may be a relative area and the absolute size may be an absolute area.

Fig. 3 is an exemplary flowchart illustrating determination of an eye state for three consecutive detection frames according to an example of the present disclosure.

In some examples, the eye state may include an open state, an open eye procedure state, and a closed eye procedure state. To this end, examples of the present disclosure also provide an exemplary method of determining an eye state for three consecutive detection frames. The state analysis module 230 may be configured to implement the method of determining eye state. It should be noted that, unless there is a contradiction, the shape and size of the corneal region in the method for determining the state of the eye may be replaced with the relative size of the corneal region or the absolute size of the corneal region without limitation. Referring to fig. 3, for a frame data set including a previous frame, an intermediate frame, and a subsequent frame in this order, a method of determining an eye state of the intermediate frame (i.e., a detection frame to be analyzed) may include:

step S102: it may be determined whether the eye state of the intermediate frame is an open state based on the variation situation. Specifically, if the change is that the shape size of the cornea region of the previous frame is larger than the preset size, the shape size of the cornea region of the intermediate frame is larger than the preset size, and the shape size of the cornea region of the next frame is larger than the preset size, the eye state of the intermediate frame may be the open state. In some examples, if the variation does not satisfy the condition of the open state, it may be continuously determined whether the condition of the other eye state is satisfied or the eye state of the intermediate frame is made the other state.

Step S104: whether the eye state of the intermediate frame is the open eye process state may be determined based on the variation. Specifically, if the change is that the shape size of the cornea region of the previous frame is smaller than the preset size, the shape size of the cornea region of the intermediate frame is larger than the preset size, and the shape size of the cornea region of the next frame is larger than the shape size of the cornea region of the intermediate frame, the eye state of the intermediate frame may be the eye-open process state. In some examples, if the condition of the eye-opening process state is not satisfied by the variation, it may be continuously determined whether the condition of the other eye state is satisfied or the eye state of the intermediate frame is made the other state.

Step S106: whether the eye state of the intermediate frame is the eye-closing process state may be determined based on the change situation. Specifically, if the change is that the shape and size of the cornea region of the previous frame is larger than the preset size, the shape and size of the cornea region of the intermediate frame is smaller than the shape and size of the cornea region of the previous frame, and the shape and size of the cornea region of the next frame is smaller than the preset size, the eye state of the intermediate frame may be the eye closing process state. In some examples, if the variation does not satisfy the condition of the eye-closing process state, it may be continuously determined whether the condition of the other eye state is satisfied or the eye state of the intermediate frame is made the other state.

In some examples, the eye state may also include a closed state. To this end, the above-described method of determining an eye state of an intermediate frame may further include determining whether the eye state of the intermediate frame is a closed state based on the variation. Specifically, if the shape and size of the cornea region of the previous frame is smaller than the preset size, the shape and size of the cornea region of the intermediate frame is smaller than the preset size, and the shape and size of the cornea region of the next frame is smaller than the preset size, the eye state of the intermediate frame may be the closed state.

It should be noted that, the disclosure is not particularly limited to each step in the above method for determining the eye state of the intermediate frame, and the order of the steps may be adjusted or the corresponding step may be deleted according to actual situations. In some examples, the eye state may also include a momentary blink state.

In addition, the preset size may be a fixed value (e.g., an empirical value). For example, if the relative size is a ratio of the size of the cornea region of the detection frame to the size of the reference object, the preset size may be a preset ratio determination threshold (may also be referred to as an occupancy determination threshold). One skilled in the art can experiment and adjust to determine the appropriate preset size according to the scheme exemplified in the present disclosure.

In some examples, the state analysis module 230 may be configured to obtain the corneal region of the detection frame by a segmentation method. Preferably, the corneal region of the detection frame may be acquired by a corneal region segmentation model based on deep learning. The corneal region segmentation model may be pre-trained. Under the circumstance, the segmentation method based on deep learning can learn the high-level semantic feature information of the cornea region in the video data as much as possible through the marking information of the training data, can be compatible with the video data with different acquisition conditions (for example, the video has the influence of different objective conditions such as illumination and dyeing conditions, or the influence of subjective factors such as eyeball movement, focusing, shooting distance angle and eyelash occlusion), and can further segment the cornea region more accurately.

Referring back to fig. 2, in this embodiment, the detection segment analysis module 250 may be configured to acquire a plurality of detection segments corresponding to the video data based on the eye state of the detection frame. In some examples, the eye state of the detection frame may be obtained by state analysis module 230. That is, the eye state of the detection frame obtained by the state analysis module 230 described above may be used to automatically segment the detection segment available for analysis. In this case, the eye state based on the more accurate detection frame can make the segmentation precision of tear film detection higher, reduce the dependency on further processing the segmented detection segment to obtain a better detection segment, and further support real-time analysis. For example, the solution of patent document (CN 111062443A) directly identifies the open-close state, and does not consider the influence of the dynamic change process of the cornea on the classification precision, and further influences the segmentation precision of the detection segment, so that the longest open eye segment needs to be extracted after the detection segment is obtained, and real-time analysis cannot be supported.

In some examples, the detection segment analysis module 250 can be configured to read the detection frames in the tear film data queue and obtain the eye states of the detection frames by the state analysis module 230 (e.g., the detection frames to be analyzed can be used as input to the state analysis module 230 to obtain the eye states of the detection frames).

In some examples, the detection segment analysis module 250 may be configured to read the detection frames in the tear film data queue sequentially in time order. In some examples, the detection segment analysis module 250 can be configured to read the eye state of the detection frame obtained by the state analysis module 230 while reading the detection frame in the tear film data queue. In other examples, detection segment analysis module 250 may be configured to read detection frames in the tear film data queue and read eye states of detection frames that have been obtained by state analysis module 230.

In some examples, detection segment analysis module 250 may be configured to delay reading a detection frame in the tear film data queue for one frame (i.e., may be read starting with the second detection frame) so that state analysis module 230 may determine the eye state of a detection frame based on changes in the corneal region between a previous frame of the detection frame, the detection frame's own data, and a subsequent frame of the detection frame. In other examples, the detection segment analysis module 250 can also be configured to not delay reading the tear film data queue (i.e., can read from the first detection frame), and set the eye state of the first detection frame in the tear film data queue to any state that does not affect the detection segment division. It should be noted that, those skilled in the art can select corresponding means to implement the method according to actual situations, and the disclosure is not particularly limited.

In this embodiment, each detection segment may correspond to a detection frame in a corresponding frame range in the tear film data queue. The frame range may be expressed as a start frame to an end frame. That is, each detection segment may have a start frame and an end frame. In addition, the time corresponding to the start frame may be shorter than the time corresponding to the end frame.

In some examples, the start frame and the end frame of each detection segment may be determined from the detection frame of the eye state as an open-eye process state and the detection frame of the eye state as a closed-eye process state in the tear film data queue, respectively. That is, the start frame of a detection segment may be determined from the detection frame whose eye state is the eye-opening process state, and the end frame of a detection segment may be determined from the detection frame whose eye state is the eye-closing process state. Specifically, the detection segment analysis module 250 may be configured to, upon identifying a detection frame in the tear film data queue whose eye state is an eye-open process state, take the detection frame as a start frame of a detection segment, continue to identify the eye state of the detection frame in the tear film data queue to obtain a detection frame whose eye state is an eye-closed process state as an end frame of the detection segment, and then continue to start identifying a start frame of another new detection segment. Thereby, the detection segment can be specified.

In some examples, the detection segment analysis module 250 may be further configured to determine the detection result of each detection segment based on the detection frame in which the eye state is open. That is, the detection section analysis module 250 may be configured to identify a detection frame whose eye state is an open state, and determine a detection result of a detection section to which the detection frame belongs, based on the detection frame. In some examples, the detection segment analysis module 250 may be further configured to store the detection result of the detection segment after determining the detection result of the detection segment. Therefore, the query can be facilitated.

In some examples, the detection segment analysis module 250 may be configured to detect whether a cornea region of the detection frame is cracked based on the detection frame whose eye state is an open state, and if the cornea region of the detection frame is cracked, determine a detection result of the detection segment to which the detection frame belongs based on the detection frame. In some examples, detecting whether a corneal region of a frame is ruptured may be detecting whether a frame has a dry spot. In some examples, the detection frame for determining the detection result of each detection segment may be a detection frame in which the corneal region rupture first occurs in the detection segment and the eye state is the open state.

Fig. 4A is a schematic diagram showing a relevant image of a detection frame in an open state and not broken according to an example of the present disclosure. Fig. 4B is a schematic diagram illustrating an open state and cracked related image of a detection frame according to an example of the present disclosure. Fig. 4C is a schematic diagram of a correlation image of a detection frame showing a closed process state and failing to identify a rupture state according to an example of the present disclosure.

In addition, fig. 4A shows a cornea segmentation result P11, an open state map P12, and a cornea region map P13 of the frame image P10 as an example of the relevant image of the detection frame in the open state and without fragmentation. As an example of the relevant image of the detection frame in the open state and having broken, fig. 4B shows a cornea segmentation result P21, an open state map P22, and a cornea region map P23 of the frame image P20. As an example of the related images of the detection frame in which the state of the closing process is not recognized and the state of rupture is recognized, fig. 4C shows a cornea segmentation result P31, a closing process state map P32, and a cornea region map P33 of the frame image P30.

In some examples, the detection segment analysis module 250 may be configured to detect whether a corneal region of a detection frame is ruptured by a machine-learned classification method. Preferably, whether or not the corneal region of the detection frame is ruptured can be detected by a tear film rupture detection model based on deep learning. The tear film break up detection model may be pre-trained. Under the condition, the classification method based on deep learning can learn high-level semantic feature information of the corneal region in the video data as much as possible through the marking information of the training data, can be compatible with the video data under different acquisition conditions, and can further perform classification more accurately.

In some examples, the detection results may include at least one of tear film break up time and tear film break up morphology. The tear film break-up time (TBUT) may be the time from the eye being held open after blinking until the first dry spot appears on the tear film 2.

For tear film break up time, in some examples, detection segment analysis module 250 may be configured to calculate tear film break up time for the belonging detection segment based on the location of the detection frame in the belonging detection segment at which the corneal region broke up. In some examples, the tear film break time of the belonging detection segment may be calculated from the frame number of the detection frame in which the corneal region is broken and the frame number of the start frame of the belonging detection segment.

For tear film break up morphology, in some examples, the detection segment analysis module 250 may be configured to obtain tear film break up morphology of the detection segment to which it belongs based on the detection frame of the corneal region breaking up. In some examples, the tear film break up morphology may include at least one of an image corresponding to the detected frame, descriptive information of the tear film state obtained based on the image, and a shape of the tear film 2 obtained based on the image.

In addition, the present disclosure also provides an exemplary method for detecting segment identification, which takes the tear film data queue as the buffer data queue and determines the eye state of the current frame as an open eye process state, a closed eye process state and an open state in order. It should be noted that unless there is a contradiction, the order of judgment of the eye state may be adjusted according to the actual situation without limitation. In addition, the buffered data queue may be replaced with a tear film data queue without limitation.

Referring to fig. 5, the exemplary method of detecting segment identification may include:

step S202: and reading a detection frame from the buffer data queue according to the time sequence and taking the detection frame as a current frame.

Step S204: the eye state of the current frame is determined. Specifically, the eye state of the current frame may be determined based on a change in the shape and size of the corneal region between the previous frame of the current frame, and the next frame of the current frame. For details, reference is made to the above description of an exemplary method of determining the state of the eye.

Step S206: and judging whether the eye state of the current frame is the eye opening process state or not. In some examples, step S208 may be performed if the eye state of the current frame is an eye-open process state. In some examples, step S210 may be performed if the eye state of the current frame is not the open-eye process state.

Step S208: and taking the current frame as the starting frame of the detection section. That is, if the eye state of the current frame is the eye-opening process state, it may indicate that the start frame of a detection segment is identified. In other words, it is possible to wait for the identification of the start frame and then start the identification of a new detection segment. In addition, step S214 may be executed after step S208 is executed.

Step S210: and judging whether the eye state of the current frame is the eye closing process state or not. In some examples, step S212 may be performed if the eye state of the current frame is the closed-eye process state. In some examples, if the eye state of the current frame is not the eye-closing process state, step S214 may be performed or the eye state of the current frame may be continuously determined (e.g., step S213 is performed).

Step S212: the current frame is used as the end frame of the detection segment of the determined start frame, and a detection segment is further determined (i.e., the identification of the detection segment is completed). That is, after the identification of the start frame of a detection segment is confirmed, the detection frame whose eye state is the eye closing process state can be continuously identified as the end frame of the detection segment to complete the identification of the detection segment. In some examples, the status of the detected segment may be set to complete to indicate that the identification of the detected segment is complete. In addition, step S214 may be performed after step S212 is performed.

Step S214: and continuing to read the next detection frame in the buffered data queue as the current frame, and starting to execute from step S204 again until the buffered data queue is completely read. Specifically, the current frame may be updated to be the next detection frame in the buffered data queue and the determination may be continued based on the current frame until the buffered data queue is completely read.

In some examples, with continued reference to fig. 5, the exemplary method of detecting segment identification described above may further include step S213. In step S213, if the eye state of the current frame is open, it is detected whether or not the cornea region of the current frame is ruptured, and if the cornea region of the current frame is ruptured, the detection result of the detection segment to which the current frame belongs is determined based on the current frame. In some examples, in step S213, the detection result may also be recorded. For details, reference is made to the above-mentioned description of determining the detection results of the respective detection segments.

In some examples, in the above exemplary method of detecting segment identification, for the current frame with the eye state being the closed state, the step S214 may be continuously performed.

The disclosed example relates to an exemplary method for detecting segment identification, which obtains a more accurate detecting segment in real time based on the eye state determined by the dynamic change process of the cornea, and can support real-time analysis based on the more accurate detecting segment.

In some examples, the detection segment analysis module 250 may be configured to automatically split the detection segments and display the detection results for each detection segment. In this case, the user can autonomously judge whether to adopt the detection result of the corresponding detection section. For example, the tear film break-up time of a transient blink is generally less useful for assessing tear film status, while the detection segment of a transient blink is often short and easily identifiable, and the user may choose to discard the results of that segment.

Referring back to fig. 2, in the present embodiment, the result analysis module 270 may be configured to determine a target detection result based on a plurality of detection segments. In particular, the result analysis module 270 may be configured to determine a target detection result based on a plurality of detection results corresponding to a plurality of detection segments.

In some examples, the target detection result may include at least one of a target tear film break up time and a target tear film break up morphology. The target tear film break up time can be determined from a plurality of tear film break up times corresponding to the plurality of detection segments. The target tear film breakup morphology may be determined from a plurality of tear film breakup morphologies corresponding to the plurality of detection segments.

In some examples, the result analysis module 270 may be configured to fuse a plurality of tear film break up times corresponding to a plurality of detection segments to obtain a target tear film break up time. In some examples, the fusion method may include, but is not limited to, at least one of randomly taking a value, taking a maximum value, taking a minimum value, taking a median value, and taking an average value. Preferably, the fusion method may be averaging. That is, the result analysis module 270 may be configured to average a plurality of tear film break up times corresponding to a plurality of detection segments to obtain a target tear film break up time.

In some examples, the results analysis module 270 may be configured to perform a comprehensive analysis of a plurality of tear film breakup morphologies corresponding to the plurality of detection segments to obtain a target tear film breakup morphology. In some examples, the target tear film disruption morphology can include a plurality of tear film disruption morphologies corresponding to the plurality of detection segments. In some examples, the result analysis module 270 may be configured to screen or fuse images of detection frames in which corneal regions corresponding to multiple detection segments are ruptured to obtain a target image. Thus, an image with better quality can be obtained. In some examples, the results analysis module 270 may also be configured to obtain descriptive information of the tear film state and/or the shape of the tear film 2 based on the target image. This enables more accurate tear film break-up information to be obtained.

In some examples, the result analysis module 270 may be configured to display a plurality of detection results and/or target detection results corresponding to a plurality of detection segments. In this case, the result can be analyzed conveniently by a user (e.g., a doctor).

In some examples, the result analysis module 270 may be further configured to screen out detection segments of detection frames where no corneal region is ruptured prior to determining the target detection result. In some examples, a user may screen the detection segments used to determine the target detection result via the result analysis module 270.

It should be noted that the result analysis module 270 may not be necessary if the detection system 200 is used to identify a plurality of detection segments.

In some examples, the detection system 200 may also include the pre-trained cornea region segmentation model described above. As described above, the corneal region segmentation model may be used to acquire the corneal region of the detection frame. The corneal region segmentation model may be a lightweight semantic segmentation model. For example, semantic segmentation models may include, but are not limited to, U-Net, deeplab, STDC, or BiSeNet, among others. To this end, examples of the present disclosure also provide a training method of a corneal region segmentation model.

Referring to fig. 6, the training method of the corneal region segmentation model may include:

step S302: a first training sample is constructed. The first training sample may include a plurality of inspection images, each of which may have annotation information. In some examples, a preset number of examination videos for tear film stability may be collected, and a plurality of examination images may be acquired based on the examination videos. For example, an inspection video may be decoded to obtain a plurality of inspection images. In some examples, multiple inspection images of one inspection video may be stored in a chronological order (which may also be referred to as a frame sequential order). Therefore, subsequent labeling of multiple inspection images of the same inspection video can be facilitated.

In some examples, the annotation information for each inspection image may be a corneal profile mask. In some examples, point coordinates may be annotated along the corneal edge of each inspection image from which a corresponding corneal profile mask is generated. This makes it possible to obtain label information for each inspection image. In some examples, for multiple inspection images of one inspection video, the annotations may be sequentially labeled in time order. Therefore, convenience and efficiency of labeling can be improved.

Step S304: a corneal region segmentation model is trained based on the first training sample. In some examples, the first training sample may be divided into a first training set and a first verification set, the corneal region segmentation model is trained by using the first training set, and the trained corneal region segmentation model is verified by using the first verification set to adjust the corneal region segmentation model, so as to obtain an optimal corneal region segmentation model.

Fig. 7 is an exemplary flow chart illustrating a training method of a tear film break detection model according to examples of the present disclosure.

In some examples, the detection system 200 may also include the pre-trained tear film break detection model described above. As described above, the tear film break-up detection model can be used to detect whether a break-up occurs in the corneal region of the detection frame. The tear film break-up detection model may be a lightweight classification network model. For example, the classification network model may include, but is not limited to, resNet, shuffleNet, efficientNet, or MobileNet, among others. To this end, examples of the present disclosure also provide a training method of a tear film break-up detection model. Referring to fig. 7, a training method of a tear film break-up detection model may include:

step S402: a second training sample is constructed. The second training sample may comprise a plurality of inspection images. In some examples, a preset number of examination videos for tear film stability may be collected, and a plurality of examination images may be acquired based on the examination videos. For example, the inspection video may be decoded to obtain a plurality of inspection images. In some examples, multiple inspection images of one inspection video may be stored in chronological order. Thereby, it is possible to facilitate the subsequent determination of the inspection image in which the first fracture occurred at the time of marking.

Step S404: and labeling the inspection images in the second training sample to determine the labeling labels of the inspection images. This enables each inspection image to have a label. In some examples, the above-mentioned scheme related to the detection segment analysis module 250 may be utilized to obtain a plurality of detection segments corresponding to the inspection video, and the inspection images belonging to the same detection segment may be stored in the same storage unit. In this case, the annotator can be facilitated to quickly determine the annotation target, and the convenience of annotation can be improved. Specifically, a plurality of inspection images may be stored in storage units based on the divided inspection segments, and labeled in units of storage units. In some examples, the storage unit that detects storage of segments may be a folder.

In some examples, in labeling multiple inspection images corresponding to one inspection segment, the annotator can mark the inspection image where tear film break-up first exists. In some examples, an annotation tag for an inspection image corresponding to a detection segment may be determined based on the marked inspection image. In some examples, after being marked by a plurality of marking personnel, the marking personnel can determine the final mark after being audited by one expert marking personnel. This can reduce the influence of subjective factors on the accuracy of labeling.

In some examples, the annotation tag can include both unbroken and broken tags. Specifically, a plurality of inspection images corresponding to one inspection segment may be traversed, and after a marked inspection image is obtained, the inspection image is marked as cracked, an image before the inspection image is marked as not cracked, and an image after the inspection image is marked as cracked. Thus, labeling information of a plurality of inspection images corresponding to one detection segment can be obtained.

In some examples, the annotation tag can also include a closed tag. Specifically, an inspection image whose eye state is a closed state may be added to the second training sample, and the annotation label of the inspection image may be set to be closed. In this case, the class proportions of the training samples can be balanced. In addition, the model can distinguish the examination image with the eye state being the closed state, and further can adapt to the condition that the closed eye exists in the video data in the application scene.

In some examples, a label tag may be represented using a three-digit number. The description is intended to be illustrative, and not restrictive. For example, one could have "000" indicating closed, "100" indicating not ruptured, and "111" indicating ruptured.

Step S406: training a tear film break-up detection model based on the second training sample. In some examples, the second training sample may be divided into a second training set and a second validation set, the tear film rupture detection model is trained by the second training set, and the trained tear film rupture detection model is validated by the second validation set to adjust the tear film rupture detection model, thereby obtaining an optimal tear film rupture detection model.

The method for detecting a tear film based on video data according to the present disclosure is described in detail below with reference to the accompanying drawings. The tear film detection method may be implemented by the detection system 200 described above. The description relating to the detection system 200 is equally applicable to the tear film detection method. The tear film detection method according to the examples of the present disclosure may also be referred to as a detection method or a tear film detection method or the like. Fig. 8 is an exemplary flow chart illustrating a method of tear film detection based on video data in accordance with an example of the present disclosure.

In some examples, as shown in fig. 8, the detection method may include determining a tear film data queue based on video data for the tear film 2 (step S502), acquiring a plurality of detection segments corresponding to the video data based on an eye state of a detection frame (step S504), and determining a target detection result based on the plurality of detection segments (step S506).

In this embodiment, in step S502, a tear film data queue may be determined based on the video data for tear film 2. The video data may correspond to a plurality of detection frames that vary over time. In some examples, a plurality of detection frames corresponding to video data over time may be treated as a tear film data queue. For details, refer to the related description of the obtaining module 210.

In this embodiment, in step S504, a plurality of detection segments corresponding to the video data may be acquired based on the eye state of the detection frame. In some examples, in step S504, a detection frame in the tear film data queue may be read, an eye state of the detection frame may be acquired, and a plurality of detection segments corresponding to the video data may be acquired based on the eye state of the detection frame. In some examples, the eye state may include an open state, an open eye process state, and a closed eye process state. In some examples, the start frame and the end frame of each detection segment may be determined from the detection frame of the tear film data queue whose eye state is an open-eye process state and the detection frame of the eye state is a closed-eye process state, respectively. In some examples, in step S504, the detection result of each detection segment may also be determined based on the detection frame in which the eye state is the open state. For details, reference is made to the description relating to the detection segment analysis module 250.

In some examples, the eye state of a detection frame may be determined based on changes in the corneal region between successive detection frames in the tear film data queue. A plurality of successive detection frames may constitute a frame data set. In some examples, the consecutive plurality of detection frames may be three consecutive detection frames in the tear film data queue. Specifically, in analyzing the eye state of the detection frame, a frame data set corresponding to the detection frame to be analyzed may be acquired, and the eye state of the detection frame to be analyzed may be determined based on a change in the shape size of the corneal region between the detection frames in the frame data set, where the frame data set may be composed of three consecutive detection frames in the tear film data queue, and the detection frame to be analyzed is located in the middle of the three consecutive detection frames. In addition, the size of the shape of the corneal region can be expressed in relative size. For details, reference is made to the description of the status analysis module 230.

In the present embodiment, in step S506, the target detection result may be determined based on a plurality of detection segments. Specifically, the target detection result may be determined based on a plurality of detection results corresponding to a plurality of detection segments. For details, see the description related to the result analysis module 270. In some examples, step S506 may not be necessary.

The present disclosure also relates to an electronic device, which may include at least one processing circuit. The at least one processing circuit is configured to perform one or more steps of the above-described method.

The present disclosure also relates to a computer-readable storage medium that may store at least one instruction that, when executed by a processor, performs one or more steps of the method described above.

The detection system 200 and the detection method according to the present disclosure more accurately determine the eye state of the detection frame of the video data of the tear film 2 based on the dynamic change process of the cornea, determine a plurality of detection segments based on the eye state of the detection frame, and further analyze the plurality of detection segments to obtain the target detection result. In this case, the eye state determined based on the dynamic change process of the cornea enables higher segmentation accuracy of tear film detection, and real-time analysis and interpretation can be supported based on the more accurate segmentation. In addition, the shape and the size of the cornea area are represented by relative sizes, so that the negative influence of possible offset among a plurality of detection frames on the change condition can be reduced, and the method can be compatible with various complex environments for acquiring video data.

While the present disclosure has been described in detail above with reference to the drawings and examples, it should be understood that the above description is not intended to limit the disclosure in any way. Those skilled in the art can make modifications and variations to the present disclosure as needed without departing from the true spirit and scope of the disclosure, which fall within the scope of the disclosure.

Claims

1. A tear film detection system based on video data is characterized by comprising an acquisition module, a state analysis module, a detection section analysis module and a result analysis module; the acquisition module is configured to acquire a plurality of detection frames corresponding to the video data for the tear film over time and serve as a tear film data queue; the state analysis module is configured to receive a detection frame to be analyzed, acquire three detection frames in the tear film data queue, wherein the three detection frames are located in the middle and are continuous, and determine the eye state of the detection frame to be analyzed based on the change situation of the shape and the size of a cornea region in the three detection frames, wherein the eye state comprises an open state, an open eye process state and a closed eye process state, and the shape and the size of the cornea region are expressed by relative size; the detection segment analysis module is configured to read detection frames in the tear film data queue, acquire a plurality of detection segments corresponding to the video data based on the eye states of the detection frames acquired by the state analysis module, and determine detection results of the detection segments based on the detection frames with the eye states being open states, wherein a start frame and an end frame of each detection segment are respectively determined by the detection frames with the eye states being open eye process states and the detection frames with the eye states being closed eye process states in the tear film data queue; and the result analysis module is configured to determine a target detection result based on a plurality of detection results corresponding to the plurality of detection segments.

2. The system of claim 1, wherein:

the video data includes at least one of a video file and video buffer data, the video data being data acquired after staining the tear film with a contrast agent.

3. The system of claim 1, wherein:

the three detection frames sequentially include a previous frame, an intermediate frame and a next frame, the intermediate frame is the detection frame to be analyzed, and the state analysis module is configured to:

if the change condition is that the shape and size of the cornea region of the previous frame is larger than a preset size, the shape and size of the cornea region of the intermediate frame is larger than the preset size, and the shape and size of the cornea region of the next frame is larger than the preset size, the eye state of the intermediate frame is in an open state;

if the change condition is that the shape and size of the cornea region of the previous frame is smaller than the preset size, the shape and size of the cornea region of the intermediate frame is larger than the preset size, and the shape and size of the cornea region of the next frame is larger than the shape and size of the cornea region of the intermediate frame, the eye state of the intermediate frame is an eye opening process state;

and if the change condition is that the shape and size of the cornea region of the previous frame is larger than the preset size, the shape and size of the cornea region of the intermediate frame is smaller than the shape and size of the cornea region of the previous frame, and the shape and size of the cornea region of the next frame is smaller than the preset size, the eye state of the intermediate frame is the eye closing process state.

4. The system of claim 1, wherein:

the relative size is a ratio of the size of the cornea region of the detection frame to the size of a reference object, and the reference object is at least one of the whole image and the region of interest corresponding to the detection frame.

5. The system of claim 1, wherein:

the tear film data queue is a buffered data queue, the detection segment analysis module configured to:

reading a detection frame from the buffer data queue according to the time sequence and taking the detection frame as a current frame;

determining the eye state of the current frame based on the change condition of the shape and the size of the cornea area among the previous frame of the current frame, the current frame and the next frame of the current frame;

if the eye state of the current frame is the eye opening process state, taking the current frame as the initial frame of the detection section;

if the eye state of the current frame is the eye closing process state, taking the current frame as an end frame of the detection section of the determined initial frame, and further determining the detection section; and is

And updating the current frame as the next detection frame in the buffer data queue and continuing to judge based on the current frame until the buffer data queue is completely read.

6. The system of claim 1, wherein:

the test results include at least one of a tear film break-up time and a tear film break-up morphology, and the target test results include at least one of a target tear film break-up time and a target tear film break-up morphology.

7. The system of claim 6, wherein:

if the eye state of one detection frame is in an open state, whether the cornea region of the detection frame is cracked or not is judged, if the cornea region of the detection frame is cracked, the tear film cracking time corresponding to the detection section is calculated based on the position of the detection frame in the detection section, and/or the tear film cracking form corresponding to the detection section is obtained based on the detection frame.

8. The system of claim 6, wherein:

the result analysis module is configured to average a plurality of tear film break up times corresponding to the plurality of detection segments to determine the target tear film break up time.

9. The system of claim 7, wherein:

the pre-trained cornea region segmentation model based on deep learning is further included, and the cornea region segmentation model is used for obtaining the cornea region of the detection frame; and/or

The tear film rupture detection system further comprises a pre-trained tear film rupture detection model based on deep learning, wherein the tear film rupture detection model is used for detecting whether a cornea region of the detection frame is ruptured or not.

10. A method for tear film detection based on video data, comprising:

acquiring a plurality of detection frames which change along with time and correspond to the video data of the tear film and using the detection frames as a tear film data queue; reading detection frames in the tear film data queue, acquiring a plurality of detection segments corresponding to the video data based on eye states of the detection frames, and determining a detection result of each detection segment based on the detection frames of which the eye states are opening states, wherein the eye states include an opening state, an eye opening process state and an eye closing process state, determining a start frame and an end frame of each detection segment respectively from the detection frames of which the eye states are the eye opening process state and the eye states are the eye closing process state in the tear film data queue, acquiring three detection frames of which the detection frames to be analyzed in the tear film data queue are positioned in the middle and are continuous in the eye states of the analysis detection frames, determining the eye states of the detection frames to be analyzed based on changes of shape sizes of corneal regions in the three detection frames, and expressing the shape sizes of the corneal regions by relative sizes; and determining a target detection result based on a plurality of detection results corresponding to the plurality of detection segments.