A lens and software system that can be used for mobile phone to take anterior segment photos

Abstract

The application relates to the field of medical image processing and discloses a lens and a software system which can be used for shooting an anterior segment of an eye by a mobile phone, the lens comprising: a lens barrel which is used as a bearing body of an optical instrument and is used for protecting the precise optical instrument; a clamping plate one which is fixed on one side of the lens barrel and is used for clamping the mobile phone; a clamping plate two which is installed at the bottom of the clamping plate one and is used for assisting the clamping plate one to be installed on a smart terminal; and a rubber strip which is arranged on one side of the clamping plate two and is used for increasing the friction of the clamping plate two on the screen of the mobile phone; and the software system comprises an image acquisition module, an image preprocessing module, an AI disease recognition module, a time sequence progress analysis module, a data and report management module and a man-machine interaction module. Through simplification of the lens, digital image anti-shake processing and illumination normalization correction, the problems of inconvenient carrying of the equipment and unstable imaging of the portable equipment are solved, and the technical effects of improving the imaging quality and consistency of the anterior segment image are achieved.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, specifically to a lens and software system that can be used for mobile phone to capture images of the anterior segment of the eye. Background Technology

[0002] Anterior segment diseases are common in ophthalmology, and continuous monitoring of their progression is crucial for guiding treatment and assessing prognosis. Traditional anterior segment examinations mainly rely on specialized medical equipment such as slit-lamp microscopes. These devices are bulky, expensive, and require operation by specialists, limiting their widespread use in home or community settings and making frequent, continuous monitoring of the condition difficult to achieve.

[0003] With the widespread adoption of smart terminal devices and the improvement of their camera performance, using portable devices such as smartphones for anterior segment health monitoring has become a potential development direction. However, directly using smart terminal devices for shooting faces many challenges. The unavoidable shaking during handheld shooting and the changing ambient lighting conditions often lead to unstable image quality of the acquired anterior segment, resulting in blurriness or uneven exposure, which seriously affects the accuracy of subsequent analysis.

[0004] Secondly, current methods for comparing multiple time-series images rely primarily on doctors' subjective experience, lacking an objective and quantifiable standard for assessing disease progression. This approach struggles to accurately capture subtle changes in lesions, potentially leading to delays or inaccuracies in disease assessment. Furthermore, existing auxiliary applications often fall short in data management and user interaction; users' examination data rarely forms continuous health records, and the operational processes may not be convenient or intuitive for non-professional users. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a lens and software system for capturing anterior segment images using a mobile phone. This solves the problems of inconvenient device portability and unstable imaging quality, subjective assessment of disease progression lacking quantitative standards, and inconvenient health data management when performing anterior segment monitoring on portable smart terminals.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a lens for shooting the front part of a mobile phone, comprising:

[0007] The lens barrel, as the carrier of optical instruments, is used to protect precision optical instruments;

[0008] Clamp 1, which is fixed to one side of the lens barrel, is used to clamp the mobile phone;

[0009] Clamp 2 is installed at the bottom of clamp 1 and is used to cooperate with clamp 1 auxiliary equipment to be installed on the smart terminal;

[0010] A rubber strip is placed on one side of the clamping plate to increase the friction between the clamping plate and the mobile phone screen.

[0011] A fixing post is fixed inside the second clamping plate, and a sliding post slides inside the fixing post. The end of the sliding post away from the fixing post is fixed inside the first clamping plate. A spring is sleeved on the outer wall of the sliding post. One end of the spring is fixed inside the first clamping plate, and the other end of the spring is fixed inside the second clamping plate. An optical component is installed inside the lens barrel.

[0012] Preferably, the optical component includes concave and convex mirrors and a fixed aperture. Multiple concave and convex mirrors are provided and are all installed inside the lens barrel. A spacer is provided between each concave and convex mirror. The fixed aperture is installed inside the lens barrel and at the optical center.

[0013] A software system, comprising:

[0014] The image acquisition module is used to control the camera of the smart terminal device to capture the image in front of the eye and to perform automatic focus, automatic exposure and digital image stabilization.

[0015] An image preprocessing module, connected to the image acquisition module, is used to perform illumination normalization correction, image denoising, and region of interest extraction on the captured raw image.

[0016] The AI ​​disease recognition module is connected to the image preprocessing module and is used to analyze the preprocessed image, identify anterior segment disease features in the image, and generate preliminary recognition results.

[0017] The temporal progression analysis module is used to acquire anterior segment images and related data of the same user at different time points. Through image registration, lesion segmentation, and feature quantification, the disease progression index is calculated to assess changes in the condition.

[0018] The data and report management module is used to perform structured storage and management of images, metadata, recognition results and analysis data, and generate visual reports based on the analysis results;

[0019] The human-computer interaction module provides users with a graphical user interface and a voice command interface to receive user input commands and present the system's processing results and reports.

[0020] Preferably, the digital image stabilization process specifically includes the following steps:

[0021] Motion vector estimation is performed on consecutive image frames using a block matching algorithm;

[0022] The global motion model parameters representing device jitter are calculated from the motion vector using the random sampling consensus algorithm;

[0023] Geometric transformation compensation is performed on the captured original image based on the global motion model parameters to complete the image stabilization process.

[0024] Preferably, the illumination normalization correction specifically includes the following steps:

[0025] An image enhancement algorithm based on Retinex theory is used to decompose the original image into illumination and reflection components;

[0026] By estimating and removing the illumination components, a reflection component image reflecting the inherent properties of the object is obtained, thus completing the illumination normalization correction.

[0027] Preferably, the extraction of the region of interest specifically includes the following steps:

[0028] A pre-trained object detection neural network model is used to process the denoised image;

[0029] After automatically identifying and outputting the bounding box coordinates of the frontal region, the image is cropped based on the bounding box coordinates to extract the region of interest.

[0030] Preferably, the analysis of the preprocessed image specifically includes the following steps:

[0031] By employing a deep convolutional neural network with a residual network structure, multi-level feature extraction is performed on the preprocessed image;

[0032] By integrating features through a fully connected layer, image analysis is completed, and the probability corresponding to different disease categories is output.

[0033] Preferably, the image registration specifically includes the following steps:

[0034] Scale-invariant keypoints are detected in the anterior segment images at different time points, and feature descriptors are generated for them.

[0035] Matching point pairs are determined by matching the feature descriptors of two images;

[0036] A geometric transformation model that describes the spatial mapping relationship between images is estimated using a random sampling consensus algorithm, and the geometric transformation model is applied to align the images to complete image registration.

[0037] Preferably, the lesion segmentation and feature quantification specifically include the following steps:

[0038] A semantic segmentation network model with U-Net architecture is used to perform pixel-level segmentation on the registered image to generate lesion mask images;

[0039] Based on the lesion mask image, morphological features, color features, and texture features are quantitatively extracted from the lesion region;

[0040] Generate multidimensional lesion feature vectors.

[0041] Preferably, calculating the disease progression index specifically includes the following steps:

[0042] Calculate the Euclidean distance between the lesion feature vectors corresponding to two different time points to quantify the overall difference in lesion features;

[0043] The difference magnitude is normalized according to the time interval between the two time points to obtain the disease progression index.

[0044] This invention provides a lens and software system for capturing the front part of a mobile phone camera. It offers the following advantages:

[0045] 1. This invention simplifies the lens, solving the problem of large device size and inaccessibility. At the same time, through digital image stabilization processing in the image acquisition module and illumination normalization correction in the image preprocessing module, it solves the problem of unstable image quality caused by hand shake and changes in ambient light when shooting the anterior segment of the eye with a portable smart terminal device. This achieves the technical effect of improving the imaging quality and consistency of the anterior segment image, providing reliable data input for subsequent disease identification and analysis.

[0046] 2. This invention uses a temporal progression analysis module to register, segment, and quantify images at different time points, and calculates a disease progression index. This solves the problem that traditional disease monitoring relies on subjective judgment and lacks quantitative assessment methods. It achieves the technical effect of automatically tracking and quantitatively assessing the progression of anterior disease, and improves the objectivity and timeliness of disease monitoring.

[0047] 3. This invention uses a data and report management module to store data in a structured manner and generate a visual report containing trend charts. Combined with the voice command and data sharing functions of the human-computer interaction module, it solves the problems of inconvenient management of user health data, unintuitive presentation of analysis results, and complex operation processes. It achieves the technical effect of realizing systematic management of personal health data, improving user operation convenience, and enhancing the value of data application. Attached Figure Description

[0048] Figure 1 This is a perspective view of the lens of the present invention;

[0049] Figure 2 This is a cross-sectional view of clamping plate one and clamping plate two of the present invention;

[0050] Figure 3 This is a diagram showing the internal structure distribution of the lens barrel of the present invention;

[0051] Figure 4 This is a software system architecture diagram of the present invention.

[0052] Among them, 1. Lens tube; 2. Clamping plate one; 3. Clamping plate two; 4. Rubber strip; 5. Sliding column; 6. Spring; 7. Concave and convex mirrors; 8. Spacer ring; 9. Fixed aperture; 10. Fixed column. Detailed Implementation

[0053] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] Please see the appendix Figure 1 -Appendix Figure 3 This invention provides a lens for shooting the front part of a mobile phone, comprising:

[0055] Lens tube 1, which serves as a carrier for optical instruments, is used to protect precision optical instruments;

[0056] Clamp 1 2, which is fixed to one side of lens barrel 1, is used to fit the mobile phone for clamping;

[0057] Clamp 2 3 is installed at the bottom of clamp 1 2 and is used to cooperate with clamp 1 2 auxiliary equipment to be installed on the smart terminal;

[0058] Rubber strip 4 is set on one side of clamp 2 3 to increase the friction of clamp 2 3 on the mobile phone screen;

[0059] A fixing post 10 is fixed inside the clamping plate 2 3. A sliding post 5 slides inside the fixing post 10. One end of the sliding post 5 away from the fixing post 10 is fixed inside the clamping plate 2. A spring 6 is sleeved on the outer wall of the sliding post 5. One end of the spring 6 is fixed inside the clamping plate 2. The other end of the spring 6 is fixed inside the clamping plate 2 3. An optical component is installed inside the lens barrel 1.

[0060] The optical components include concave and convex mirrors 7 and fixed aperture 9. Multiple concave and convex mirrors 7 are provided and are all installed inside the lens barrel 1. Spacers 8 are provided between the concave and convex mirrors 7. The fixed aperture 9 is installed inside the lens barrel 1 and at the optical center.

[0061] Specifically, when the device needs to be used, simply align the position of the lens barrel 1 with the position of the camera on the smart terminal device. At this time, depending on the thickness of the smart terminal, stretch the clamping plate 2 3 so that the sliding column 5 slides inside the fixed column 10 while pulling the spring 6. The reaction force of the spring 6 causes the rubber strip 4 installed at the position of the clamping plate 2 3 to cooperate with the clamping plate 1 2 to clamp the lens barrel 1 onto the smart terminal. At this time, the camera of the smart terminal can capture the front part of the eye through the cooperation of the concave and convex mirrors 7, the spacer 8 and the fixed aperture 9 inside the lens barrel 1.

[0062] Please see the appendix Figure 4 A software system comprising:

[0063] The image acquisition module 110 is used to control the camera of the smart terminal device to capture the frontal image and perform automatic focus, automatic exposure and digital image stabilization processing.

[0064] The image preprocessing module 120, connected to the image acquisition module 110, is used to perform illumination normalization correction, image denoising, and region of interest extraction on the captured raw image.

[0065] The AI ​​disease recognition module 130 is connected to the image preprocessing module 120 and is used to analyze the preprocessed image, identify anterior segment disease features in the image, and generate preliminary recognition results.

[0066] The temporal progression analysis module 140 is used to acquire anterior segment images and related data of the same user at different time points, and calculate the disease progression index to assess changes in the condition through image registration, lesion segmentation, and feature quantification.

[0067] The data and report management module 150 is used to perform structured storage and management of images, metadata, recognition results and analysis data, and generate visual reports based on the analysis results.

[0068] The human-computer interaction module 160 provides users with a graphical user interface and a voice command interface to receive user input commands and present the system's processing results and reports.

[0069] In one specific implementation, the image acquisition module 110 is responsible for interacting with and controlling the imaging hardware and external optical components of the smart terminal device to acquire high-quality anterior segment images. The specific functions of the image acquisition module 110 are implemented through communication with the application programming interface (API) provided by the operating system of the smart terminal device.

[0070] The external optical component that accompanies the image acquisition module 110 is a macro lens that can be detachably mounted in front of the main camera of a smart terminal device. The optical design of this component provides the magnification and working distance required for anterior segment examination, solving the problem that the native lens of a smart terminal device cannot clearly capture the tiny structures of the eye.

[0071] During image acquisition, the image acquisition module 110 performs a series of image acquisition and real-time optimization control operations. The image acquisition module 110 activates the camera of the smart terminal device and sets it to continuous autofocus mode. In this mode, the system uses image sensor data to continuously evaluate and adjust the sharpness of the central area of ​​the preview image through contrast detection autofocus or phase detection autofocus, ensuring that the focus accurately falls on target structures such as the cornea and iris during shooting. Simultaneously, the image acquisition module 110 automatically adjusts exposure parameters, such as shutter speed and ISO, based on the metering results of the central area to obtain appropriate image brightness. Users can also manually compensate for exposure through voice commands issued by the human-computer interaction module 160.

[0072] To mitigate the impact of unavoidable hand-held camera shake on image quality, the image acquisition module 110 integrates a digital image stabilization method. This method is triggered when the user issues a shooting command, and its specific implementation steps may include:

[0073] Step S201: In the brief period before capturing the final image, the image acquisition module 110 continuously buffers a sequence of image frames from the camera preview video stream. The system selects one frame from the sequence as a reference frame and divides it into multiple macroblocks. Subsequently, using a block matching algorithm, it searches for the most similar corresponding region to each macroblock in the reference frame within subsequent frames of the sequence, thereby calculating the displacement of each macroblock between frames and forming a set of initial motion vectors.

[0074] Step S202: Since the initial motion vector contains global motion caused by hand shaking and local motion caused by the movement of objects in the scene, the system needs to extract the global motion representing camera shaking. This embodiment uses the Random Sample Consensus (RANSAC) algorithm to process the initial motion vector set, filtering out outlier data such as local motion, thereby robustly estimating a global motion model parameter that can describe the translation, rotation, or scaling of the entire image frame.

[0075] Step S203: After acquiring the final captured original image, the image acquisition module 110 performs geometric transformation compensation on the original image based on the calculated global motion model parameters. In a simplified implementation, this compensation is manifested as an image translation operation, and the compensated image... It can be expressed by the following formula:

[0076] ;

[0077] In the formula, To compensate the image in coordinates Pixel value at; The original image that was captured; and These are the horizontal and vertical jitter displacements calculated based on the global motion model parameters. This operation reduces image blurring caused by jitter.

[0078] For the specific implementation of image sensor control, autofocus algorithm and block matching algorithm, those skilled in the art can use existing mature technical solutions, which are well known in the field and will not be elaborated here.

[0079] In one specific implementation, the image preprocessing module 120 receives the raw anterior segment image output by the image acquisition module 110 and performs a series of processing operations on it to eliminate the variability of imaging conditions and extract key image regions for subsequent analysis. The output of the image preprocessing module 120 is a standardized, clear image to be analyzed.

[0080] The specific processing flow of the image preprocessing module 120 may include the following steps:

[0081] Step S301: Since images captured by smart terminal devices in different environments are affected by factors such as ambient light intensity and color temperature, this embodiment employs an image enhancement algorithm based on Retinex theory to ensure the consistency and accuracy of subsequent AI analysis and time-series comparison. This algorithm decomposes the image into illumination and reflection components, and obtains the reflection component image that reflects the inherent properties of the object by estimating and removing the illumination component.

[0082] Specifically, this embodiment employs the multi-scale Retinex algorithm with color restoration. This algorithm first performs a logarithmic transformation on each color channel of the original image, then estimates the illumination components by convolving them with a set of Gaussian kernels of different scales, and subtracts this estimate from the logarithm of the original image. Its core calculation can be represented by the following formula:

[0083] ;

[0084] In the formula, In the first Image of the reflected components estimated from each color channel; For the first An input image with multiple color channels; The number of Gaussian kernel sizes used; For the first Weights for each scale; For the first Gaussian kernel function of scale; This represents a two-dimensional convolution operation. To avoid color distortion, the algorithm also includes a color restoration step, which linearly adjusts the calculated reflection component image to ensure the color fidelity of the final output image.

[0085] Step S302: In this step, a median filtering algorithm can be used. The median filter slides a window of a specific size across the image and replaces the value of the center pixel of the window with the median value of all pixels within that window. This method can effectively remove impulse noise that may exist in the image and better preserve the edge details of the image.

[0086] Step S303: To enable the subsequent AI disease recognition module 130 to focus on key eye structures, the anterior segment region needs to be automatically located and cropped from the denoised image. This embodiment uses a lightweight object detection neural network model to achieve this function. This model is pre-trained on a large dataset of images containing eyes, enabling it to accurately identify and output the bounding box coordinates of the anterior segment region. The image preprocessing module 120 crops the image based on the coordinates output by the model, generating the final region of interest image, and then sends it to the next processing module.

[0087] For the specific structure and training method of the target detection neural network model, those skilled in the art can use existing mature technical solutions, which are well-known technologies in the field and will not be elaborated here.

[0088] In one specific implementation, the AI ​​disease recognition module 130 receives a standardized region of interest image output by the image preprocessing module 120. Its core function is to use a pre-trained deep learning model to analyze the image content in order to automatically identify possible anterior segment disease features.

[0089] The AI ​​disease identification module 130 embeds a deep convolutional neural network (CNN) as its core analysis engine. (See attached document.) Figure 2 , Figure 2 This is a schematic diagram of a convolutional neural network structure according to an embodiment of the present invention. The architecture of this network draws on the idea of ​​residual networks, and is constructed by introducing residual learning units. This structure contains multiple cascaded residual blocks, each containing several convolutional layers, batch normalization layers, and activation function layers. Short-circuit connections allow information to propagate across layers, which helps to effectively avoid the gradient vanishing problem while increasing network depth, thereby improving the model's ability to learn complex image features.

[0090] During recognition, a standardized anterior segment image is used as input and sequentially passed through multiple convolutional layers in the network. These convolutional layers perform convolution operations on the input feature map using a series of learnable kernels to extract visual features at different levels, from low-level edges, colors, and textures to high-level abstract features such as lesion morphology. Pooling layers downsample the feature map to reduce its spatial dimensionality, increase the receptive field, and impart a degree of translation invariance to the model.

[0091] After deep feature extraction through multiple residual blocks, the network's output feature map is flattened into a one-dimensional feature vector and fed into one or more fully connected layers. The fully connected layers integrate and weight the extracted high-level features, ultimately outputting a logistic vector. The dimension of this vector is equal to the total number of disease categories predefined by the system.

[0092] To convert this logistic value vector into probabilities corresponding to each disease category, the system uses the Softmax function. The calculation formula is as follows:

[0093] ;

[0094] In the formula, The logical value vector output by a given network Under the condition that the image belongs to the first... Disease categories The posterior probability; Logical value vector The corresponding number in the middle Components of each category; The total number of predefined disease categories.

[0095] The AI ​​disease identification module 130 takes the category with the highest calculated probability value as the preliminary identification result of this analysis, and sends the result along with its corresponding probability value to the data and report management module 150 for further processing.

[0096] The convolutional neural network model was obtained through offline training. The training process used a dataset containing a large number of anterior segment images annotated by professional ophthalmologists. During training, the cross-entropy loss function was used as the objective function, and optimization algorithms such as adaptive moment estimation were employed. The network's weight parameters were iteratively updated through backpropagation until the model's performance on the validation set reached a preset standard. After training, the model was embedded and deployed in application software on smart terminal devices to perform real-time inference tasks. For training methods of convolutional neural networks, those skilled in the art can use existing mature technical solutions, which will not be elaborated upon here.

[0097] In one specific implementation, the temporal progression analysis module 140 first performs temporal image registration before comparing multiple anterior segment images of the same user. The purpose of this step is to eliminate geometric misalignment between images caused by minor changes in factors such as device pose, shooting distance, and patient head position during shooting, ensuring that subsequent difference analysis is performed in spatially corresponding regions.

[0098] This embodiment employs an image registration method based on feature point matching, the specific implementation process of which may include the following steps:

[0099] Step S401: When the system receives at least two anterior segment images to be compared, one of them (e.g., the earliest one in chronological order) is defined as the reference image, and the other is defined as the floating image to be registered.

[0100] Step S402: The system applies the Scale Invariant Feature Transform (SIFT) algorithm to both the reference image and the floating image. This algorithm detects keypoints in both images that are invariant to scale, rotation, and brightness changes, and generates a high-dimensional feature descriptor for each keypoint, which is used to uniquely identify the local image region surrounding the keypoint.

[0101] Step S403: After obtaining the keypoints and their feature descriptor sets of two images, the system finds matching point pairs between the two sets by calculating the Euclidean distance between the descriptors. To improve the accuracy of matching, a ratio test strategy is used to eliminate fuzzy matches, that is, only those point pairs where the best match is better than the second-best match are retained.

[0102] Step S404: Since the feature matching process may generate erroneous matching point pairs (i.e., outliers), this embodiment uses the Random Sample Consensus (RANSAC) algorithm to estimate a robust geometric transformation model. This model is used to describe the spatial mapping relationship from the floating image to the reference image. In one implementation, the geometric transformation model is an affine transformation, which can model complex deformations such as translation, rotation, scaling, and shearing.

[0103] Step S405: Calculate the optimal affine transformation matrix using the RANSAC algorithm. The system then applies this transformation to the entire floating image to generate a registered image. Any pixel in the image... After transformation, its coordinates in the new image It can be represented as:

[0104] ;

[0105] In the formula, It is a 2x3 affine transformation matrix; These are the original pixel coordinates in the floating image; These are the corresponding pixel coordinates in the registered image.

[0106] After the above steps, the registered image is spatially aligned with the reference image, providing a foundation for subsequent lesion segmentation and quantitative comparison. For the specific implementation of scale-invariant feature transformation and random sample consensus algorithms, those skilled in the art can use existing mature technical solutions, which will not be elaborated here.

[0107] After completing the temporal image registration, the temporal progression analysis module 140 then performs lesion segmentation and feature quantification on the aligned anterior segment image. This step aims to accurately separate the lesion region from the image and convert it into a set of numerical data that can be used for quantitative comparison.

[0108] The specific implementation of this process may include the following steps:

[0109] Step S411: For each registered anterior segment image, the system calls a pre-trained semantic segmentation network model to perform pixel-level segmentation of the lesion region. In one specific implementation, the network model adopts the U-Net architecture.

[0110] This architecture comprises an encoder path for extracting image context features and a symmetric decoder path for precise localization. High-resolution features from the encoder path are fused with those from the decoder path via skip connections, achieving accurate segmentation of the target boundary. The model is trained on a large number of anterior segment images with pixel-level lesion annotations and outputs a binary lesion mask image of the same size as the input image. In this mask image, pixels belonging to the lesion region are assigned a specific value (e.g., 1), while pixels in the background region are assigned another specific value (e.g., 0).

[0111] Step S412: Based on the lesion mask image generated in the previous step, the system performs feature quantification on the identified lesion regions and extracts a set of numerical features that can objectively describe their state. These features may include the following categories:

[0112] Morphological features: used to describe the geometry and size of the lesion. This includes the lesion area obtained by counting all pixels with a value of 1 in the masked image. The lesion perimeter is obtained by detecting the outline of the masked region and calculating its length. ; and the shape compactness index used to describe the compactness of a region. Its calculation method can be expressed as:

[0113] ;

[0114] In the formula, The area of ​​the lesion; This represents the perimeter of the lesion area. The value of this index ranges from 0 to 1, with a value closer to 1 indicating a shape closer to a circle.

[0115] Color features: used to describe the color distribution of the lesion area. The system uses the lesion mask to locate the corresponding area in the original color image, and calculates the mean and standard deviation of the intensity of all pixels in the R, G, and B color channels within that area.

[0116] Texture features: These describe the texture patterns on the surface of the lesion. By converting the lesion region into a grayscale image and calculating its gray-level co-occurrence matrix (GLCM), a series of texture descriptors, such as contrast, correlation, energy, and homogeneity, can be further extracted. For texture feature extraction, those skilled in the art can use analysis methods based on the gray-level co-occurrence matrix, which are well-known techniques in the field and will not be elaborated upon here.

[0117] Step S413: Combine all the numerical features calculated above, such as lesion area, perimeter, shape compactness, color mean, color standard deviation, and multiple texture descriptors, into a multidimensional lesion feature vector in a predetermined order. This feature vector constitutes a comprehensive quantitative description of the lesion state at a single time point and is used as input to the subsequent disease progression index calculation stage.

[0118] After obtaining the lesion feature vectors at different time points, the temporal progression analysis module 140 performs the calculation and early warning judgment of the disease progression index to achieve quantitative assessment and automated monitoring of disease changes.

[0119] The specific implementation of this process may include the following steps:

[0120] Step S421: To objectively and quantitatively assess the rate of change of lesions over a period of time, this embodiment defines a disease progression index. This index is obtained by calculating the difference between the lesion feature vectors at two consecutive monitoring time points and normalizing it according to the time interval. The specific calculation formula is as follows:

[0121] ;

[0122] In the formula, From a point in time arrive Disease progression index; and At time points respectively and Extracted lesion feature vector; L2 norm, or Euclidean distance, is used to calculate the overall difference between two multidimensional feature vectors. The time interval between two inspections; It is a preset scalar weighting coefficient used to adjust the index value to a range of values ​​that have clinical interpretability.

[0123] Step S422: The system internally presets corresponding disease progression rate thresholds for different anterior segment disease types. These thresholds are based on statistical analysis of extensive historical clinical data or defined by a medical expert knowledge base, representing the clinically significant rate of change thresholds for a specific disease. The system compares the DPI value calculated in step S521 with the corresponding disease type threshold.

[0124] If the calculated DPI value is greater than the threshold corresponding to the disease type, the system determines that the current disease progression rate exceeds the warning level and generates a warning indicator. Conversely, if the DPI value is not greater than the threshold, no warning indicator is generated.

[0125] Step S423: The time-series progression analysis module 140 transmits the calculated numerical disease progression index (DPI) and the generated warning indicator (if any) to the data and report management module 150 for the generation of the final report and the archiving of the data.

[0126] In one specific implementation, the data and report management module 150 serves as the central data hub of the system, responsible for the unified and structured management of the images, data and analysis results generated by the front-end modules, and generating visual reports that users can view based on these data.

[0127] One of the core functions of the data and report management module 150 is structured case data management. The system establishes a local database on the smart terminal device, for example using the SQLite database engine, to persistently store all user-related data. The data is organized into multiple related tables, such as a user information table and an examination record table. After each examination, the data and report management module 150 creates a new examination record associated with a specific user and includes the following information: a unique record identifier, an examination timestamp, a standardized anterior segment image captured by the image acquisition module 110 and processed by the image preprocessing module 120, preliminary identification results and their confidence levels generated by the AI ​​disease recognition module 130, lesion feature vectors and disease progression indices calculated by the temporal progression analysis module 140, and corresponding warning indicators. To protect user privacy, sensitive personal information stored in the database is encrypted.

[0128] Based on the stored data, the data and report management module 150 is also responsible for generating visual reports. The specific implementation of this process may include the following steps:

[0129] Step S501: When a user requests to view the results of a certain inspection or a historical comparison report, the module retrieves one or more relevant inspection records from the local database according to the request.

[0130] Step S502: The system populates the retrieved data into a preset report template. For a single examination report, the content includes basic information such as the anterior segment image of this examination and the AI ​​recognition results. For a time-series comparison report, the content will display the registered images of the current and previous examinations side by side for easy comparison. The report will also display the calculated Disease Progression Index (DPI) in numerical and graphical form, and will prominently alert the user when a warning indicator is generated by the time-series progression analysis module 140.

[0131] Step S503: To more intuitively display the long-term trend of the disease, the report may include one or more trend graphs. The trend graph uses time as the horizontal axis and one or more key lesion features (such as lesion area) or disease progression index (DPI) as the vertical axis, connecting data points from multiple examinations to visually present the trajectory of the disease's changes over time.

[0132] In addition, the data and report management module 150 also provides data sharing and medical service integration functions. Users can export generated single or comparative reports to common file formats (such as PDF or image files) through the human-computer interaction module 160 and share them via other applications on smart terminal devices (such as email or instant messaging tools). In another implementation, the system can also provide a standardized application programming interface (API). This interface follows secure network communication protocols (such as HTTPS) and uses standardized data exchange formats (such as JSON), allowing, with explicit user authorization, the secure transmission of user health data to third-party electronic medical record systems or telemedicine service platforms, achieving seamless integration with professional medical services.

[0133] In one specific implementation, the human-computer interaction module 160 is the only interface for users to interact with the system of the present invention. It provides users with a graphical interface and voice command interface of the operating system, and is responsible for presenting the processing results of each module inside the system to the user in a visual manner.

[0134] The human-computer interaction module 160 provides users with a graphical user interface (UI). This interface includes a preview and capture interface for image acquisition, a list interface for browsing historical records, and a report details interface for displaying analysis results. Users can perform functions such as launching the program, capturing images, viewing historical records, and sharing reports by using standard gestures such as clicking and swiping on the touchscreen of the smart terminal device.

[0135] To enhance ease of operation, especially in scenarios where users need to hold the device stably with both hands for shooting, the human-computer interaction module 160 also integrates a voice command control function. The specific implementation of this function may include the following steps:

[0136] Step S601: When the user triggers the voice input function through an interface button or a specific gesture, the human-computer interaction module 160 captures the user's voice signal through the microphone of the smart terminal device.

[0137] Step S602: The captured voice signal is transmitted to a voice recognition engine for processing. This engine converts the voice signal into a text string. In one embodiment, the voice recognition engine may be a voice service built into the operating system of the smart terminal device.

[0138] Step S603: After receiving the text string, the human-computer interaction module 160 matches it with a preset set of instruction keywords. This set includes instructions such as "take a picture", "focus", and "view history".

[0139] Step S604: Once a match is successful, the human-computer interaction module 160 generates a corresponding control signal and distributes the signal to the corresponding business logic module for execution. For example, when a "take a picture" command is detected, the human-computer interaction module 160 calls the image acquisition module 110 to perform an image capture operation; when a "view history" command is detected, the data and report management module 150 is notified to prepare and display historical data.

[0140] In terms of result presentation, the human-computer interaction module 160 receives structured report data from the data and report management module 150 and renders the front-end image, analysis conclusions, data charts and other content on the report details interface according to predefined layout rules for users to view.

[0141] For the specific layout design and implementation of the graphical user interface, those skilled in the art can use standard software development kits (SDKs) applicable to the target operating system platform, which are well-known technologies in the field and will not be elaborated here.

Claims

1. A lens that can be used for shooting the front part of a mobile phone, characterized in that, include: The lens tube (1) serves as a carrier for optical instruments and is used to protect precision optical instruments. Clamp 1 (2) is fixed to one side of the lens barrel (1) and is used to clamp the mobile phone; Clamp 2 (3) is installed at the bottom of clamp 1 (2) and is used to cooperate with clamp 1 (2) to install auxiliary equipment on the smart terminal; A rubber strip (4) is provided on one side of the clamping plate (3) to increase the friction of the clamping plate (3) on the mobile phone screen; The clamping plate 2 (3) has a fixed column (10) inside, and a sliding column (5) slides inside the fixed column (10). The end of the sliding column (5) away from the fixed column (10) is fixed inside the clamping plate 1 (2). A spring (6) is sleeved on the outer wall of the sliding column (5). One end of the spring (6) is fixed inside the clamping plate 1 (2), and the other end of the spring (6) is fixed inside the clamping plate 2 (3). An optical component is provided inside the lens barrel (1).

2. The lens and software system for shooting the front part of a mobile phone according to claim 1, characterized in that, The optical components include concave and convex mirrors (7) and fixed apertures (9). Multiple concave and convex mirrors (7) are provided and are all installed inside the lens barrel (1). Spacers (8) are provided between the concave and convex mirrors (7). The fixed apertures (9) are installed inside the lens barrel (1) and at the optical center.

3. A software system applied to a lens as described in any one of claims 1-2, which can be used for shooting the front part of a mobile phone, characterized in that, include: The image acquisition module is used to control the camera of the smart terminal device to capture the image in front of the eye and to perform automatic focus, automatic exposure and digital image stabilization. An image preprocessing module, connected to the image acquisition module, is used to perform illumination normalization correction, image denoising, and region of interest extraction on the captured raw image. The AI ​​disease recognition module is connected to the image preprocessing module and is used to analyze the preprocessed image, identify anterior segment disease features in the image, and generate preliminary recognition results. The temporal progression analysis module is used to acquire anterior segment images and related data of the same user at different time points. Through image registration, lesion segmentation, and feature quantification, the disease progression index is calculated to assess changes in the condition. The data and report management module is used to perform structured storage and management of images, metadata, recognition results and analysis data, and generate visual reports based on the analysis results; The human-computer interaction module provides users with a graphical user interface and a voice command interface to receive user input commands and present the system's processing results and reports.

4. A software system according to claim 3, characterized in that, The digital image stabilization process specifically includes the following steps: Motion vector estimation is performed on consecutive image frames using a block matching algorithm; The global motion model parameters representing device jitter are calculated from the motion vector using the random sampling consensus algorithm; Geometric transformation compensation is performed on the captured original image based on the global motion model parameters to complete the image stabilization process.

5. A software system according to claim 3, characterized in that, The illumination normalization correction specifically includes the following steps: An image enhancement algorithm based on Retinex theory is used to decompose the original image into illumination and reflection components; By estimating and removing the illumination components, a reflection component image reflecting the inherent properties of the object is obtained, thus completing the illumination normalization correction.

6. A software system according to claim 3, characterized in that, The extraction of the region of interest specifically includes the following steps: A pre-trained object detection neural network model is used to process the denoised image; After automatically identifying and outputting the bounding box coordinates of the frontal region, the image is cropped based on the bounding box coordinates to extract the region of interest.

7. A software system according to claim 3, characterized in that, The analysis of the preprocessed image specifically includes the following steps: By employing a deep convolutional neural network with a residual network structure, multi-level feature extraction is performed on the preprocessed image; By integrating features through a fully connected layer, image analysis is completed, and the probability corresponding to different disease categories is output.

8. A software system according to claim 3, characterized in that, The image registration specifically includes the following steps: Scale-invariant keypoints are detected in the anterior segment images at different time points, and feature descriptors are generated for them. Matching point pairs are determined by matching the feature descriptors of two images; A geometric transformation model that describes the spatial mapping relationship between images is estimated using a random sampling consensus algorithm, and the geometric transformation model is applied to align the images to complete image registration.

9. A software system according to claim 3, characterized in that, The lesion segmentation and feature quantification specifically include the following steps: A semantic segmentation network model with U-Net architecture is used to perform pixel-level segmentation on the registered image to generate lesion mask images; Based on the lesion mask image, morphological features, color features, and texture features are quantitatively extracted from the lesion region; Generate multidimensional lesion feature vectors.

10. A software system according to claim 3, characterized in that, The calculation of the disease progression index specifically includes the following steps: Calculate the Euclidean distance between the lesion feature vectors corresponding to two different time points to quantify the overall difference in lesion features; The difference magnitude is normalized according to the time interval between the two time points to obtain the disease progression index.

Images