Intelligent decayed tooth identification method and system based on computer vision

Abstract

The invention relates to the field of computer vision, in particular to a decayed tooth intelligent identification method and system based on computer vision, and the method comprises the following steps: image collection, image segmentation, bottleneck feature extraction, reconstruction error optimization, diagnosis result generation, diagnosis result feedback, and decayed tooth development prediction. According to the method, a decoder part of U-Net and an encoder part of ConvNeXt are combined into a segmentation model, and a multi-label recall loss function is used for effective identification; a capsule graph network is introduced to serve as a bottleneck layer, and potential spatial relations and high-level dependence of local features in data are captured through a dynamic routing mechanism; and an MFO-SFR-HWACO algorithm is used to optimize the hyper-parameter configuration of the depth auto-encoder. The system comprises a data acquisition module, an image segmentation module, a bottleneck feature extraction module, a reconstruction error optimization module, a diagnosis result generation module, a decayed tooth development prediction module and an intelligent report generation module.

Description

Technical Field

[0001] The present invention relates to the field of computer vision, and in particular to a method and system for intelligently identifying dental caries based on computer vision. Background Art

[0002] With the rapid development of artificial intelligence technology, computer vision has become an important method for medical image segmentation. However, existing medical image segmentation tasks have problems such as loss of detail information, low precision and inaccurate segmentation. They cannot effectively retain the detailed information of the image, resulting in lesion areas, especially small lesions in the early stage of caries. Traditional convolutional neural networks may not fully extract features in complex images, resulting in missed detection or incorrect segmentation, especially in tooth contact areas or difficult-to-identify areas, resulting in unreliable diagnosis results; traditional loss functions have problems of inaccurate classification and low recall rate. In multi-label problems, only using a single-label loss function cannot accurately optimize the model; ordinary feature extraction methods have the problem of insufficient recognition of complex lesion areas, and often cannot capture the complex spatial relationships and subtle lesions in caries images, resulting in missed diagnosis or misdiagnosis; general hyperparameter optimization algorithms have the problem of low optimization efficiency, resulting in hyperparameter optimization stagnation at the local optimal solution, affecting model performance. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a method and system for intelligent identification of dental caries based on computer vision. In view of the problems of loss of detail information, low precision and inaccurate segmentation in the existing medical image segmentation tasks, two basic architectures, U-Net and ConvNeXt, are selected for image segmentation. The present invention combines the decoder part of U-Net with the encoder part of ConvNeXt as a segmentation model to effectively process features of different scales, especially when processing blurred edges or smaller lesion areas. In addition, the efficient convolution calculation and improved feature representation capabilities make the segmentation task more accurate and improve the accuracy of complex image segmentation. The invention has the ability to identify complex lesions; in view of the problems of inaccurate classification and low recall rate, the present invention uses a multi-label recall loss function in the segmentation model to ensure that caries lesions of different severities can be effectively identified; in view of the inability to capture the complex spatial relationships and subtle lesions in caries images, the present invention introduces a capsule graph network as a bottleneck layer, which can capture the potential spatial relationships and high-level dependencies of local features in the data through a dynamic routing mechanism, and has a good performance for tiny and complex lesions in caries images; in view of the problem of low optimization efficiency of general hyperparameter optimization algorithms, the present invention uses the MFO-SFR-HWACO (Moth-Flame Optimization with Sub-FireflyReplacement and Hybrid Weighted Ant Colony Optimization) algorithm to optimize the hyperparameter configuration of the deep autoencoder, improve the accuracy of the deep autoencoder in the feature extraction process, especially when dealing with complex lesions, it can ensure low reconstruction error, thereby improving the accuracy and robustness of recognition.

[0004] The technical solution adopted by the present invention is as follows: The present invention provides a computer vision-based intelligent caries identification method, which comprises the following steps:

[0005] Step S1: image acquisition, collecting tooth images and enhancing the image quality, constructing a capsule graph network, and performing data amplification on the tooth images through the capsule graph network;

[0006] Step S2: Image segmentation, constructing a segmentation model based on a convolutional neural network, generating a segmentation mask of the tooth image and classifying each potential lesion area, cropping each potential lesion area according to the segmentation mask, and adjusting it to a fixed-size sample;

[0007] Step S3: Bottleneck feature extraction: Use a deep autoencoder to extract the bottleneck features of the sample, introduce the capsule graph network into the deep autoencoder, and generate feature embedding;

[0008] Step S4: Reconstruction error optimization, reconstruct the bottleneck features of the sample through the deep autoencoder to obtain the reconstruction error;

[0009] Step S5: Generate a diagnosis result and determine whether the sample belongs to the caries lesion area based on the reconstruction error;

[0010] Step S6: Feedback of diagnosis results, including lesion area, location, historical diagnosis records and patient metadata;

[0011] Step S7: predicting the development of dental caries, building a prediction model, and predicting the development trend of dental caries lesions based on the diagnosis results;

[0012] Step S8: Generate an intelligent caries report based on the development trend of caries lesions.

[0013] Further, step S2 specifically includes the following steps:

[0014] Step S21: training data collection, collecting historical caries lesion image data from clinical images and annotating each potential lesion area, annotating caries lesion labels, including mild caries, moderate caries and severe caries;

[0015] Step S22: construct a segmentation model. The architecture of the segmentation model includes an encoder part and a decoder part. The decoder part adopts the structure of the U-Net model. The encoder part introduces the ConvNeXt model. The decoder part uses a skip connection to directly access the encoder.

[0016] Step S23: Set the loss function and design a multi-label recall loss function. The formula used is as follows: L=- 1 N ∑ i - 1 N ∑ E - 1 E [ w c · y ic · log y ic ̂ + 1- y ic · log 1- y ic ̂ ] ;

[0017] in, is the multi-label recall loss function value, is the sample size, is the sample index, is the number of caries lesion labels, is the weight coefficient of the caries lesion label, It is The sample in The true label of each category has a value of 1 or 0, where 1 represents the presence of lesions and 0 represents the absence of lesions. The segmentation model is The class probability of the sample is in the range of [0,1];

[0018] Step S24: model training, using the annotated historical caries lesion image data, combined with a multi-label recall loss function and a cross entropy loss function to train the segmentation model;

[0019] Step S25: generating a segmentation mask, inputting the tooth image after data amplification in step S1 into the trained segmentation model to generate a segmentation mask;

[0020] Step S26: cropping the lesion area, using the connected domain analysis technique to extract the boundary coordinates of the lesion area from the segmentation mask, and cropping each potential lesion area from the tooth image after data amplification in step S1 according to the boundary coordinates.

[0021] Further, step S3 specifically includes the following steps:

[0022] Step S31: construct a deep autoencoder, construct a deep autoencoder architecture including a deep encoder, a deep decoder and a bottleneck layer, and introduce a capsule graph network into the bottleneck layer;

[0023] Step S32: Generate pseudo-abnormal data, embed features of samples through capsule graph network to obtain bottleneck features of normal samples, generate pseudo-abnormal features through random perturbations, and use dynamic fusion strategy to combine bottleneck features of normal samples with pseudo-abnormal features to obtain pseudo-abnormal data;

[0024] Step S33: Design a loss function, design a weighted reconstruction error loss function, the formula used is as follows: ;

[0025] in, is the weighted reconstruction error loss function value, For the samples, For the The data output after the samples are reconstructed is For the The bottleneck feature of samples is For the The bottleneck features after the reconstruction of samples, is the bottleneck coefficient;

[0026] Step S34: using the MFO-SFR-HWACO algorithm to optimize the hyperparameters in the deep autoencoder to obtain an optimized deep autoencoder;

[0027] Step S35: Feature embedding generation, using the optimized deep autoencoder to directly extract the bottleneck features of the sample as the final feature embedding.

[0028] Further, step S34 specifically includes the following steps:

[0029] Step S341: Initialize the moth population, where each moth individual represents a set of candidate hyperparameter configurations, and set the maximum number of iterations;

[0030] Step S342: Initialize the ant population and define the path selection rule and pheromone concentration for each individual ant;

[0031] Step S343: define the search space, the hyperparameter configuration includes the bottleneck layer dimension, the number of neurons, the learning rate and the batch size, and set the value range for all the hyperparameters in the hyperparameter configuration;

[0032] Step S344: global search, updating the moth population according to the flame update rule of the moth, and finding potential high-quality hyperparameter configurations in the search space;

[0033] Step S345: local optimization, using potential high-quality hyperparameter configuration to update the ant population, individual ants select paths according to the weighted reconstruction error loss function, and the optimal hyperparameter configuration is obtained by gradually updating the pheromone concentration;

[0034] Step S346: Iterative optimization, iterate step S344 to step S345 until the maximum number of iterations is reached, and output the optimal hyperparameter configuration.

[0035] The computer vision-based caries intelligent recognition system provided by the present invention includes a data acquisition module, an image segmentation module, a bottleneck feature extraction module, a reconstruction error optimization module, a diagnosis result generation module, a caries development prediction module and an intelligent report generation module, which specifically includes the following contents:

[0036] The data acquisition module acquires tooth images and performs image quality enhancement and data amplification;

[0037] The image segmentation module constructs a segmentation model based on a convolutional neural network, generates a segmentation mask and transmits the processed samples to the bottleneck feature extraction module;

[0038] The bottleneck feature extraction module uses a deep autoencoder to extract the bottleneck features of the sample and introduces a capsule graph network to generate feature embedding;

[0039] The reconstruction error optimization module reconstructs the bottleneck features of the sample through a deep autoencoder, and obtains the reconstruction error and transmits it to the diagnosis result generation module;

[0040] The diagnosis result generation module determines whether the sample belongs to the caries lesion area according to the reconstruction error and generates a diagnosis result;

[0041] The dental caries development prediction module constructs a prediction model to predict the development trend of dental caries lesions based on the diagnosis results;

[0042] The intelligent report generation module generates an intelligent caries report, including diagnosis results and predicted development trends of caries lesions.

[0043] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0044] (1) In view of the problems of loss of detail information, low precision and inaccurate segmentation in existing medical image segmentation tasks, U-Net and ConvNeXt are selected for image segmentation. The present invention combines the decoder part of U-Net with the encoder part of ConvNeXt as a segmentation model to effectively process features of different scales, especially when dealing with blurred edges or smaller lesion areas. In addition, efficient convolution calculation and improved feature representation capabilities make the segmentation task more accurate and improve the recognition ability of complex lesions.

[0045] (2) To address the problems of inaccurate classification and low recall rate, the present invention uses a multi-label recall loss function in the segmentation model to ensure that caries lesions of different severities can be effectively identified;

[0046] (3) In order to solve the problem of being unable to capture the complex spatial relationships and subtle lesions in caries images, the present invention introduces a capsule graph network as a bottleneck layer. Through a dynamic routing mechanism, it can capture the potential spatial relationships and high-level dependencies of local features in the data, and has a good performance for small and complex lesions in caries images.

[0047] (4) In view of the problem of low optimization efficiency of general hyperparameter optimization algorithms, the present invention uses the MFO-SFR-HWACO algorithm to optimize the hyperparameter configuration of the deep autoencoder, thereby improving the accuracy of the deep autoencoder in the feature extraction process, especially when dealing with complex lesions, and can ensure low reconstruction errors, thereby improving the accuracy and robustness of recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A schematic diagram of the process of the computer vision-based intelligent caries identification method provided by the present invention;

[0049] Figure 2 Schematic diagram of the image segmentation process;

[0050] Figure 3 Schematic diagram of the flow of the MFO-SFR-HWACO algorithm.

[0051] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0053] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0054] Example 1, see Figure 1 The present invention provides a computer vision-based intelligent dental caries recognition method, which comprises the following steps:

[0055] Step S1: image acquisition, collecting tooth images and enhancing the image quality, constructing a capsule graph network, and performing data amplification on the tooth images through the capsule graph network;

[0056] Step S2: Image segmentation, constructing a segmentation model based on a convolutional neural network, generating a segmentation mask of the tooth image and classifying each potential lesion area, cropping each potential lesion area according to the segmentation mask, and adjusting it to a fixed-size sample;

[0057] Step S3: Bottleneck feature extraction: Use a deep autoencoder to extract the bottleneck features of the sample, introduce the capsule graph network into the deep autoencoder, and generate feature embedding;

[0058] Step S4: Reconstruction error optimization, reconstruct the bottleneck features of the sample through the deep autoencoder to obtain the reconstruction error;

[0059] Step S5: Generate a diagnosis result and determine whether the sample belongs to the caries lesion area based on the reconstruction error;

[0060] Step S6: Feedback of diagnosis results, including lesion area, location, historical diagnosis records and patient metadata;

[0061] Step S7: predicting the development of dental caries, building a prediction model, and predicting the development trend of dental caries lesions based on the diagnosis results;

[0062] Step S8: Generate an intelligent caries report based on the development trend of caries lesions.

[0063] Example 2, see Figure 2 This embodiment is based on the above embodiment, and step S2 specifically includes the following steps:

[0064] Step S21: training data collection, collecting historical caries lesion image data from clinical images and annotating each potential lesion area, annotating caries lesion labels, including mild caries, moderate caries and severe caries;

[0065] Step S22: construct a segmentation model. The architecture of the segmentation model includes an encoder part and a decoder part. The decoder part adopts the structure of the U-Net model. The encoder part introduces the ConvNeXt model. The decoder part uses a skip connection to directly access the encoder.

[0066] Step S23: Set the loss function and design a multi-label recall loss function. The formula used is as follows: L=- 1 N ∑ i - 1 N ∑ E - 1 E [ w c · y ic · log y ic ̂ + 1- y ic · log 1- y ic ̂ ] ;

[0067] in, is the multi-label recall loss function value, is the sample size, is the sample index, is the number of caries lesion labels, is the weight coefficient of the caries lesion label, It is The sample in The true label of each category has a value of 1 or 0, where 1 represents the presence of lesions and 0 represents the absence of lesions. The segmentation model is The class probability of the sample is in the range of [0,1];

[0068] Step S24: model training, using the annotated historical caries lesion image data, combined with a multi-label recall loss function and a cross entropy loss function to train the segmentation model;

[0069] Step S25: generating a segmentation mask, inputting the tooth image after data amplification in step S1 into the trained segmentation model to generate a segmentation mask;

[0070] Step S26: cropping the lesion area, using the connected domain analysis technique to extract the boundary coordinates of the lesion area from the segmentation mask, and cropping each potential lesion area from the tooth image after data amplification in step S1 according to the boundary coordinates.

[0071] In this embodiment, the architecture of the segmentation model is divided into an encoder part, a decoder part and a skip connection;

[0072] Encoder part: Use the ConvNeXt model as the encoder, set the input size of ConvNeXt to 224x224, and use the standard convolution layer and residual block (ResBlock) configuration, the convolution kernel size is 3x3, the stride is 1, and the pooling layer uses 2x2 maximum pooling;

[0073] Decoder: The decoder part of the U-Net architecture is used. The decoder of U-Net restores the spatial resolution of the image through upsampling operations. The decoder part of U-Net contains multiple upsampling layers, each followed by a convolutional layer, which is used to gradually restore the details of the image.

[0074] Skip connection: The decoder directly accesses the feature map output by the encoder through a skip connection, thus ensuring that high-resolution features are preserved during the decoding process;

[0075] The number of output channels of the ConvNeXt encoder part is set to 64, 128, 256, and 512, gradually increasing;

[0076] The upsampling step size of the U-Net decoder part is set to 2, and the spatial size of the feature map is gradually restored through convolution operations.

[0077] In this embodiment, the number of samples in step S23 is set to 1000, the number of lesion categories is set to 3 (corresponding to mild caries, moderate caries and severe caries, respectively), and the category weight coefficient, for different categories, is set to 1, 2, and 1.5 (corresponding to mild caries, moderate caries and severe caries, respectively);

[0078] In this embodiment, in step S24, the annotated historical caries lesion image data is input into the model for training in the model training, and the data set is divided into a training set, a validation set, and a test set, with a ratio of 80%, 10%, and 10%. The Adam optimizer is used, and the learning rate is set to 0.001;

[0079] In this embodiment, in step S25, the amplified tooth image is input into the trained segmentation model to obtain a predicted segmentation mask with a size of 224x224. The value of each pixel indicates whether the position belongs to the lesion area, and the category of each pixel is determined by the category probability.

[0080] In this embodiment, the code used to construct the segmentation model is as follows: import tensorflow as tf from tensorflow.keras import layers, models from tensorflow.keras.applications import ConvNeXt # Build segmentation model def build_segmentation_model(input_shape=(224, 224, 3)): # Encoder part: ConvNeXt model base_model = ConvNeXt(input_shape=input_shape, include_top=False) encoder_output = base_model.output # Decoder part: U-Net structure x = layers.UpSampling2D(size=(2, 2))(encoder_output) x = layers.Conv2D(512, (3, 3), activation='relu', padding='same')(x) x = layers.UpSampling2D(size=(2, 2))(x) x = layers.Conv2D(256, (3, 3), activation='relu', padding='same')(x) x = layers.UpSampling2D(size=(2, 2))(x) x = layers.Conv2D(128, (3, 3), activation='relu', padding='same')(x) x = layers.UpSampling2D(size=(2, 2))(x) x = layers.Conv2D(64, (3, 3), activation='relu', padding='same')(x) # Output layer: segmentation category is 3 (mild, moderate, severe) output = layers.Conv2D(3, (1, 1), activation='softmax')(x) # Create model model = models.Model(inputs=base_model.input, outputs=output) return model # Create model model = build_segmentation_model() model.summary().

[0081] Embodiment 3: This embodiment is based on the above embodiment, and step S3 specifically includes the following steps:

[0082] Step S31: construct a deep autoencoder, construct a deep autoencoder architecture including a deep encoder, a deep decoder and a bottleneck layer, and introduce a capsule graph network into the bottleneck layer;

[0083] Step S32: Generate pseudo-abnormal data, embed features of samples through capsule graph network to obtain bottleneck features of normal samples, generate pseudo-abnormal features through random perturbations, and use dynamic fusion strategy to combine bottleneck features of normal samples with pseudo-abnormal features to obtain pseudo-abnormal data;

[0084] Step S33: Design a loss function, design a weighted reconstruction error loss function, the formula used is as follows: ;

[0085] in, is the weighted reconstruction error loss function value, For the samples, For the The data output after the samples are reconstructed is For the The bottleneck feature of samples is For the The bottleneck features after the reconstruction of samples, is the bottleneck coefficient;

[0086] Step S34: using the MFO-SFR-HWACO algorithm to optimize the hyperparameters in the deep autoencoder to obtain an optimized deep autoencoder;

[0087] Step S35: Feature embedding generation, using the optimized deep autoencoder to directly extract the bottleneck features of the sample as the final feature embedding.

[0088] In this embodiment, the codes used are as follows: import numpy as np import tensorflow as tf from tensorflow.keras import layers, models from tensorflow.keras.optimizers import Adam # Step S31: Build a deep autoencoder # Define the deep autoencoder architecture def build_autoencoder(input_dim, bottleneck_dim): Encoder encoder_input = layers.Input(shape=(input_dim,)) x = layers.Dense(128, activation='relu')(encoder_input) x = layers.Dense(64, activation='relu')(x) bottleneck = layers.Dense(bottleneck_dim, activation='linear',name="bottleneck_layer")(x) # Decoder x = layers.Dense(64, activation='relu')(bottleneck) x = layers.Dense(128, activation='relu')(x) decoder_output = layers.Dense(input_dim, activation='sigmoid')(x) autoencoder = models.Model(encoder_input, decoder_output) encoder = models.Model(encoder_input, bottleneck) return autoencoder, encoder # Build a deep autoencoder input_dim = 20 # Input feature dimension bottleneck_dim = 10 # bottleneck layer feature dimension autoencoder, encoder = build_autoencoder(input_dim, bottleneck_dim) # Introduce capsule graph network as part of bottleneck features (schematic implementation here) # Simple embedding is used in this example, and the full capsule network implementation needs further expansion class CapsuleLayer(layers.Layer): def __init__(self, num_capsules, dim_capsule, **kwargs): super(CapsuleLayer, self).__init__(**kwargs) self.num_capsules = num_capsules self.dim_capsule = dim_capsule def build(self, input_shape): self.W = self.add_weight(shape=(input_shape[-1], self.num_capsules * self.dim_capsule), initializer='glorot_uniform', name='W') def call(self, inputs): u_hat = tf.matmul(inputs, self.W) u_hat = tf.reshape(u_hat, (-1, self.num_capsules, self.dim_capsule)) return tf.nn.sigmoid(u_hat) # Add capsule layer to bottleneck features capsule_layer = CapsuleLayer(num_capsules=5, dim_capsule=2) # Step S32: Generate pseudo-abnormal data def generate_pseudo_anomalies(normal_features, noise_factor=0.1): """Generate pseudo-anomaly features""" noise = np.random.normal(0, noise_factor, normal_features.shape) pseudo_anomalies = normal_features + noise return pseudo_anomalies # Use samples to generate normal and pseudo abnormal data normal_features = np.random.rand(100, bottleneck_dim) # Assume that there are 100 normal samples pseudo_anomalies = generate_pseudo_anomalies(normal_features) # Dynamic Fusion Strategy fused_features = 0.7 * normal_features + 0.3 * pseudo_anomalies # Step S33: Design loss function # Weighted reconstruction error loss function bottleneck_weight = 0.5 def reconstruction_loss(y_true, y_pred, bottleneck_features,bottleneck_recon): reconstruction_loss = tf.reduce_mean(tf.square(y_true - y_pred)) bottleneck_loss = tf.reduce_mean(tf.square(bottleneck_features -bottleneck_recon)) return reconstruction_loss + bottleneck_weight * bottleneck_loss # Compile the autoencoder autoencoder.compile(optimizer=Adam(learning_rate=0.001), loss=lambda y_true, y_pred: reconstruction_loss( y_true, y_pred, encoder(y_true), encoder(y_pred))) # Step S34: Optimize the deep autoencoder # Pseudocode for optimizing hyperparameters using the example MFO-SFR-HWACO algorithm def optimize_hyperparameters(): """Placeholder implementation for optimizing hyperparameters""" best_lr = 0.001 best_batch_size = 32 return best_lr, best_batch_size # Get optimized hyperparameters best_lr, best_batch_size = optimize_hyperparameters() # Update learning rate and recompile model autoencoder.compile(optimizer=Adam(learning_rate=best_lr), loss=lambda y_true, y_pred: reconstruction_loss( y_true, y_pred, encoder(y_true), encoder(y_pred))) # Train the model x_train = np.random.rand(1000, input_dim) # Sample training data y_train = x_train # The target output of the autoencoder is equal to the input autoencoder.fit(x_train, y_train, epochs=50, batch_size=best_batch_size) # Step S35: Feature embedding generation final_features = encoder.predict(x_train).

[0089] Example 4, see Figure 3 , This embodiment is based on the above embodiment, step S34, specifically comprising the following steps;

[0090] Step S341: Initialize the moth population, where each moth individual represents a set of candidate hyperparameter configurations, and set the maximum number of iterations;

[0091] Step S342: Initialize the ant population and define the path selection rule and pheromone concentration for each individual ant;

[0092] Step S343: define the search space, the hyperparameter configuration includes the bottleneck layer dimension, the number of neurons, the learning rate and the batch size, and set the value range for all the hyperparameters in the hyperparameter configuration;

[0093] Step S344: global search, updating the moth population according to the flame update rule of the moth, and finding potential high-quality hyperparameter configurations in the search space;

[0094] Step S345: local optimization, using potential high-quality hyperparameter configuration to update the ant population, individual ants select paths according to the weighted reconstruction error loss function, and the optimal hyperparameter configuration is obtained by gradually updating the pheromone concentration;

[0095] Step S346: Iterative optimization, iterate step S344 to step S345 until the maximum number of iterations is reached, and output the optimal hyperparameter configuration.

[0096] In this example, the moth population is initialized and the hyperparameter setting range is: Bottleneck layer dimension: 8-64; Number of neurons: 64-256; Learning rate: 0.0001-0.01; Batch size: 16-64; The population size is 10 moth individuals, and the initialization parameters of each moth are randomly selected: The first moth individual: 32, 128, 0.001, 32; The second moth individual: 16, 64, 0.005, 16; The maximum number of iterations is 50; Define the ant population size as 10 and initialize the ant population: The first ant individual: 24,96,0.003,32; The second ant individual: 40, 128, 0.002, 48; The initial pheromone concentration was 1.0; Define the search space: the bottleneck layer dimension range is 8-4, the number of neurons range is 64-56, the learning rate range is 0.0001-0.01, and the batch size range is 16-64;

[0097] The moth population update rule is ;

[0098] in, It is The moth in The position at the iteration, It is the best moth in the current population. is the random disturbance factor, ranging from [-0.1, 0.1];

[0099] After 30 global search iterations, the optimal moth individual was found, and its hyperparameter configuration was: 24, 96, 0.003, 32;

[0100] The pheromone update rules are as follows: ;

[0101] in, For path To Path The pheromone concentration, is the weighted reconstruction error loss value of the corresponding configuration;

[0102] Global search and local optimization are performed alternately for a total of 50 iterations, and the final optimal hyperparameter configuration is output: 28, 112, 0.0025, 32.

[0103] Embodiment 5, the computer vision-based caries intelligent recognition system provided by the present invention includes a data acquisition module, an image segmentation module, a bottleneck feature extraction module, a reconstruction error optimization module, a diagnosis result generation module, a caries development prediction module and an intelligent report generation module, specifically including the following contents:

[0104] The data acquisition module acquires tooth images and performs image quality enhancement and data amplification;

[0105] The image segmentation module constructs a segmentation model based on a convolutional neural network, generates a segmentation mask and transmits the processed samples to the bottleneck feature extraction module;

[0106] The bottleneck feature extraction module uses a deep autoencoder to extract the bottleneck features of the sample and introduces a capsule graph network to generate feature embedding;

[0107] The reconstruction error optimization module reconstructs the bottleneck features of the sample through a deep autoencoder, and obtains the reconstruction error and transmits it to the diagnosis result generation module;

[0108] The diagnosis result generation module determines whether the sample belongs to the caries lesion area according to the reconstruction error and generates a diagnosis result;

[0109] The dental caries development prediction module constructs a prediction model to predict the development trend of dental caries lesions based on the diagnosis results;

[0110] The intelligent report generation module generates an intelligent caries report, including diagnosis results and predicted development trends of caries lesions.

[0111] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0112] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0113] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. An intelligent method for identifying dental caries based on computer vision, characterized in that: The method comprises the following steps: Step S1: image acquisition, collecting tooth images and enhancing the image quality, constructing a capsule graph network, and performing data amplification on the tooth images through the capsule graph network; Step S2: Image segmentation: Image segmentation, building a segmentation model based on a convolutional neural network, generating a segmentation mask of the tooth image and classifying each potential lesion area, cropping each potential lesion area according to the segmentation mask, and adjusting it to a fixed-size sample; Step S3: Bottleneck feature extraction: Use a deep autoencoder to extract the bottleneck features of the sample, introduce the capsule graph network into the deep autoencoder, and generate feature embedding; Step S4: Reconstruction error optimization, reconstruct the bottleneck features of the sample through the deep autoencoder to obtain the reconstruction error; Step S5: Generate a diagnosis result and determine whether the sample belongs to the caries lesion area based on the reconstruction error; Step S6: Feedback of diagnosis results, including lesion area, location, historical diagnosis records and patient metadata; Step S7: predicting the development of dental caries, building a prediction model, and predicting the development trend of dental caries lesions based on the diagnosis results; Step S8: Generate an intelligent caries report based on the development trend of caries lesions.

2. The computer vision-based intelligent dental caries identification method according to claim 1, characterized in that: Step S2 specifically includes the following steps: Step S21: training data collection, collecting historical caries lesion image data from clinical images and annotating each potential lesion area, annotating caries lesion labels, including mild caries, moderate caries and severe caries; Step S22: construct a segmentation model. The architecture of the segmentation model includes an encoder part and a decoder part. The decoder part adopts the structure of the U-Net model. The encoder part introduces the ConvNeXt model. The decoder part uses a skip connection to directly access the encoder. Step S23: Set the loss function and design a multi-label recall loss function. The formula used is as follows: ； in, is the multi-label recall loss function value, is the sample size, is the sample index, is the number of caries lesion labels, is the weight coefficient of the caries lesion label, It is The sample in The true label of each category has a value of 1 or 0, where 1 represents the presence of a lesion and 0 represents the absence of a lesion. The segmentation model is The class probability of the sample is in the range of [0,1]; Step S24: model training, using the annotated historical caries lesion image data, combined with a multi-label recall loss function and a cross entropy loss function to train the segmentation model; Step S25: generating a segmentation mask, inputting the tooth image after data amplification in step S1 into the trained segmentation model to generate a segmentation mask; Step S26: cropping the lesion area, using the connected domain analysis technique to extract the boundary coordinates of the lesion area from the segmentation mask, and cropping each potential lesion area from the tooth image after data amplification in step S1 according to the boundary coordinates.

3. The computer vision-based intelligent dental caries identification method according to claim 1, characterized in that: Step S3 specifically includes the following steps: Step S31: construct a deep autoencoder, construct a deep autoencoder architecture including a deep encoder, a deep decoder and a bottleneck layer, and introduce a capsule graph network into the bottleneck layer; Step S32: Generate pseudo-abnormal data, embed features of samples through capsule graph network to obtain bottleneck features of normal samples, generate pseudo-abnormal features through random perturbations, and use dynamic fusion strategy to combine bottleneck features of normal samples with pseudo-abnormal features to obtain pseudo-abnormal data; Step S33: Design a loss function, design a weighted reconstruction error loss function, the formula used is as follows: ； in, is the weighted reconstruction error loss function value, For the samples, For the The data output after the samples are reconstructed is For the The bottleneck feature of samples is For the The bottleneck features after the reconstruction of samples, is the bottleneck coefficient; Step S34: using the MFO-SFR-HWACO algorithm to optimize the hyperparameters in the deep autoencoder to obtain an optimized deep autoencoder; Step S35: Feature embedding generation, using the optimized deep autoencoder to directly extract the bottleneck features of the sample as the final feature embedding.

4. The computer vision-based intelligent dental caries identification method according to claim 3, characterized in that: Step S34 specifically includes the following steps: Step S341: Initialize the moth population, where each moth individual represents a set of candidate hyperparameter configurations, and set the maximum number of iterations; Step S342: Initialize the ant population and define the path selection rule and pheromone concentration for each individual ant; Step S343: define the search space, the hyperparameter configuration includes the bottleneck layer dimension, the number of neurons, the learning rate and the batch size, and set the value range for all the hyperparameters in the hyperparameter configuration; Step S344: global search, updating the moth population according to the flame update rule of the moth, and finding potential high-quality hyperparameter configurations in the search space; Step S345: local optimization, using potential high-quality hyperparameter configuration to update the ant population, individual ants select paths according to the weighted reconstruction error loss function, and the optimal hyperparameter configuration is obtained by gradually updating the pheromone concentration; Step S346: Iterative optimization, iterate step S344 to step S345 until the maximum number of iterations is reached, and output the optimal hyperparameter configuration.

5. The computer vision-based intelligent caries identification system according to claim 4, characterized in that: It includes data acquisition module, image segmentation module, bottleneck feature extraction module, reconstruction error optimization module, diagnosis result generation module, caries development prediction module and intelligent report generation module, specifically including the following contents: The data acquisition module acquires tooth images and performs image quality enhancement and data amplification; The image segmentation module constructs a segmentation model based on a convolutional neural network, generates a segmentation mask and transmits the processed samples to the bottleneck feature extraction module; The bottleneck feature extraction module uses a deep autoencoder to extract the bottleneck features of the sample and introduces a capsule graph network to generate feature embedding; The reconstruction error optimization module reconstructs the bottleneck features of the sample through a deep autoencoder, and obtains the reconstruction error and transmits it to the diagnosis result generation module; The diagnosis result generation module determines whether the sample belongs to the caries lesion area according to the reconstruction error and generates a diagnosis result; The dental caries development prediction module constructs a prediction model to predict the development trend of dental caries lesions based on the diagnosis results; The intelligent report generation module generates an intelligent caries report, including diagnosis results and predicted development trends of caries lesions.

Images