A method and device for intelligent auxiliary diagnosis of skin melanoma

Abstract

The present invention provides a method and device for intelligent auxiliary diagnosis of skin melanoma, and relates to the technical field of medical image analysis. The method for intelligent auxiliary diagnosis of skin melanoma comprises: obtaining a dermoscopic image data set and a feature set; obtaining a pre-diagnosis score set through a trained classification head; calculating the untrustworthy interval of the pre-diagnosis score by an unsupervised clustering algorithm; obtaining an untrustworthy pre-diagnosis score set and a credible pre-diagnosis score set according to the untrustworthy interval of the pre-diagnosis score; obtaining a compensated pre-diagnosis score set according to a trained gating mechanism compensation model; mapping the compensated pre-diagnosis score set using an optimal transmission algorithm to obtain a diagnosis probability; and obtaining a prediction result of skin melanoma according to the diagnosis probability. The present invention can improve the fairness and accuracy of the deep learning diagnosis model.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image analysis, and in particular to an intelligent auxiliary diagnosis method and device for skin melanoma. Background Art

[0002] Melanoma is one of the common types of human skin cancer. Studies have shown that early diagnosis of melanoma can help timely and effective treatment and improve the survival rate of melanoma patients. Deep learning (DL), as a data-driven method of actively learning feature representations from data, has been successfully applied to assist in the diagnosis of skin lesions through skin mirroring. However, while improving the performance of auxiliary diagnosis, the diagnostic unfairness of deep learning algorithms based on different demographic subgroups cannot be ignored. Therefore, it is crucial to study the fairness of DL algorithms in the field of medical image analysis.

[0003] At present, common methods to improve the fairness of DL algorithms can be systematically divided into three categories: preprocessing methods, internal processing methods, and post-processing methods. Among them, the post-processing method modifies the output of the DL classifier, which has the advantages of not requiring access to the internal algorithm and easy deployment compared to the preprocessing method and the internal processing method. However, when fairness is enhanced, the diagnostic performance of the model is usually sacrificed. Therefore, it is necessary to improve the fairness of the DL model while ensuring higher diagnostic performance. Summary of the invention

[0004] In order to solve the problem that when fairness is enhanced in the prior art, the diagnostic performance of the model is usually sacrificed. It is necessary to improve the fairness of the DL model while ensuring higher diagnostic performance. The embodiment of the present invention provides a method and device for intelligent auxiliary diagnosis of skin melanoma. The technical solution is as follows:

[0005] In one aspect, a method for intelligent auxiliary diagnosis of skin melanoma is provided, which is implemented by an intelligent auxiliary diagnosis device for skin melanoma, and comprises:

[0006] S1. Obtain a dermoscopic image dataset; and obtain a feature set of dermoscopic image data according to the dermoscopic image dataset;

[0007] S2, constructing an initial classification head; training the initial classification head according to the feature set to obtain a trained classification head, and obtaining a pre-diagnosis score set through the trained classification head;

[0008] S3, using an unsupervised clustering algorithm to calculate the untrustworthy interval of the pre-diagnosis score; according to the untrustworthy interval of the pre-diagnosis score and the pre-diagnosis score set, obtaining an untrustworthy pre-diagnosis score set and a credible pre-diagnosis score set;

[0009] S4. Construct an initial gating mechanism compensation model;

[0010] S5, training the initial gating mechanism compensation model according to the untrustworthy pre-diagnosis score set and the feature set corresponding to the untrustworthy pre-diagnosis score set to obtain a trained gating mechanism compensation model;

[0011] S6, inputting the untrusted pre-diagnosis score set and the feature set corresponding to the untrusted pre-diagnosis score set into the trained gating mechanism compensation model to obtain a compensated pre-diagnosis score set; merging the compensated pre-diagnosis score set with the trusted pre-diagnosis score set to obtain a final pre-diagnosis score set;

[0012] S7. Use an optimal transmission algorithm to perform mapping processing on the final pre-diagnosis score set to obtain a diagnosis probability; and obtain a prediction result of skin melanoma according to the diagnosis probability.

[0013] Optionally, the step S1 of obtaining a feature set of dermoscopic image data according to the dermoscopic image dataset comprises:

[0014] The dermoscopic image dataset is input into a pre-trained deep learning network model to obtain a feature set of the dermoscopic image data.

[0015] Optionally, the step of training the initial classification head according to the feature set to obtain a trained classification head in S2 includes:

[0016] According to the feature set, the initial classification head is trained using a binary cross entropy loss function to obtain a trained classification head.

[0017] Optionally, the step S3 uses an unsupervised clustering algorithm to calculate an untrustworthy interval of the pre-diagnosis score; and obtains an untrustworthy pre-diagnosis score set and a credible pre-diagnosis score set according to the untrustworthy interval of the pre-diagnosis score and the pre-diagnosis score set, including:

[0018] S31, using an unsupervised clustering algorithm to classify the pre-diagnosis score set into: first-category data, second-category data, and third-category data;

[0019] S32, obtaining the centroid of the first category of data, the centroid of the second category of data, and the centroid of the third category of data;

[0020] S33, sorting the centroids of the first category data, the second category data, and the third category data to obtain a minimum value, a median value, and a maximum value;

[0021] S34. Calculate the unreliable interval of the pre-diagnosis score based on the minimum, median and maximum values;

[0022] S35 . Obtain an unreliable prediagnosis score set and a reliable prediagnosis score set according to the unreliable interval of the prediagnosis score and the prediagnosis score set.

[0023] Optionally, the gating mechanism compensation model of S4 includes 3 fully connected layers, 1 batch normalization layer and a GELU activation function; wherein,

[0024] The fully connected layer is used to standardize the input features;

[0025] The batch normalization layer is used to normalize the input features;

[0026] The GELU activation function is used to perform nonlinear mapping on input features.

[0027] Optionally, the step of S5 training the initial gating mechanism compensation model according to the untrustworthy pre-diagnosis score set and the feature set corresponding to the untrustworthy pre-diagnosis score set to obtain a trained gating mechanism compensation model includes:

[0028] S51, inputting the untrustworthy pre-diagnosis score set and the feature set corresponding to the untrustworthy pre-diagnosis score set into the initial gating mechanism compensation model, and performing normalization processing on the untrustworthy pre-diagnosis score set and the feature set corresponding to the untrustworthy pre-diagnosis score set through the fully connected layer to obtain normalized features;

[0029] S52, fusing the standardized features by an element addition method to obtain fused features;

[0030] S53, inputting the fused features into the batch normalization layer to obtain normalized features;

[0031] S54, inputting the normalized features into the GELU activation function, performing nonlinear mapping processing, and obtaining a nonlinear mapping processing result; inputting the nonlinear mapping processing result into the fully connected layer to obtain a compensated diagnostic score;

[0032] S55. According to the compensated diagnosis score, the initial gating mechanism compensation model is trained using a binary cross entropy loss function to obtain a trained gating mechanism compensation model.

[0033] Optionally, the step of using an optimal transmission algorithm to map the final pre-diagnosis score set to obtain a diagnosis probability; and obtaining a prediction result of skin melanoma according to the diagnosis probability, includes:

[0034] S71, using a Sigmoid function to map the final pre-diagnosis score set to obtain a diagnosis probability;

[0035] S72, mapping the diagnosis probability using the fairness optimal mapping theory to obtain the optimal fairness mapping probability;

[0036] S73. Obtain a prediction result of skin melanoma according to the optimal fairness mapping probability.

[0037] On the other hand, a skin melanoma intelligent auxiliary diagnosis device is provided, which is applied to a skin melanoma intelligent auxiliary diagnosis method, and the device comprises:

[0038] An acquisition unit, configured to acquire a dermoscopic image data set; and obtain a feature set of the dermoscopic image data according to the dermoscopic image data set;

[0039] A first construction unit, used to construct an initial classification head;

[0040] A first training unit, used for training the initial classification head according to the feature set to obtain a trained classification head;

[0041] A first prediction unit, configured to obtain a pre-diagnosis score set through the trained classification head;

[0042] A calculation unit, configured to calculate an untrustworthy interval of the pre-diagnosis score by using an unsupervised clustering algorithm; and obtain an untrustworthy pre-diagnosis score set and a credible pre-diagnosis score set according to the untrustworthy interval of the pre-diagnosis score and the pre-diagnosis score set;

[0043] The second building unit is used to build an initial gating mechanism compensation model;

[0044] A first training unit, configured to train the initial gating mechanism compensation model according to the untrustworthy pre-diagnosis score set and a feature set corresponding to the untrustworthy pre-diagnosis score set, to obtain a trained gating mechanism compensation model;

[0045] The second prediction unit is used to input the unreliable pre-diagnosis score set and the feature set corresponding to the unreliable pre-diagnosis score set into the trained gating mechanism compensation model to obtain a compensated pre-diagnosis score set; and merge the compensated pre-diagnosis score set with the reliable pre-diagnosis score set to obtain a final pre-diagnosis score set.

[0046] The output unit is used to map the final pre-diagnosis score set using an optimal transmission algorithm to obtain a diagnosis probability; and obtain a prediction result of skin melanoma according to the diagnosis probability.

[0047] Optionally, the acquiring unit is used to:

[0048] The dermoscopic image dataset is input into a pre-trained deep learning network model to obtain a feature set of the dermoscopic image data.

[0049] Optionally, the first training unit is used to:

[0050] According to the feature set, the initial classification head is trained using a binary cross entropy loss function to obtain a trained classification head.

[0051] Optionally, the computing unit is used to:

[0052] Using an unsupervised clustering algorithm, the pre-diagnosis score set is classified into: first-category data, second-category data, and third-category data;

[0053] Obtain the centroid of the first type of data, the centroid of the second type of data, and the centroid of the third type of data;

[0054] Sort the centroids of the first category data, the second category data, and the third category data to obtain the minimum value, median value, and maximum value;

[0055] The unconfidence interval of the prediagnostic score was calculated based on the minimum, median, and maximum values;

[0056] According to the untrustworthy interval of the pre-diagnosis score and the pre-diagnosis score set, an untrustworthy pre-diagnosis score set and a trustworthy pre-diagnosis score set are obtained.

[0057] Optionally, the second building unit is used to:

[0058] The fully connected layer is used to standardize the input features;

[0059] The batch normalization layer is used to normalize the input features;

[0060] The GELU activation function is used to perform nonlinear mapping on input features.

[0061] Optionally, the first training unit is used to:

[0062] Inputting the untrustworthy pre-diagnosis score set and the feature set corresponding to the untrustworthy pre-diagnosis score set into the initial gating mechanism compensation model, and performing standardization processing on the untrustworthy pre-diagnosis score set and the feature set corresponding to the untrustworthy pre-diagnosis score set through the fully connected layer to obtain standardized features;

[0063] The standardized features are fused by using an element addition method to obtain fused features;

[0064] Inputting the fused features into the batch normalization layer to obtain normalized features;

[0065] Inputting the normalized features into the GELU activation function, performing nonlinear mapping processing, and obtaining a nonlinear mapping processing result; inputting the nonlinear mapping processing result into the fully connected layer to obtain a compensated diagnostic score;

[0066] According to the compensated diagnosis score, the initial gating mechanism compensation model is trained using a binary cross entropy loss function to obtain a trained gating mechanism compensation model.

[0067] Optionally, the output unit is used to:

[0068] The final pre-diagnosis score set is mapped using a Sigmoid function to obtain a diagnosis probability;

[0069] The diagnosis probability is mapped using the fairness optimal mapping theory to obtain the optimal fairness mapping probability;

[0070] According to the optimal fairness mapping probability, a prediction result of skin melanoma is obtained.

[0071] On the other hand, a smart assisted diagnosis device for skin melanoma is provided, comprising: a processor; a memory, wherein computer-readable instructions are stored on the memory, and when the computer-readable instructions are executed by the processor, any one of the above-mentioned smart assisted diagnosis methods for skin melanoma is implemented.

[0072] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction is stored in the storage medium, and the at least one instruction is loaded and executed by a processor to implement any one of the above-mentioned intelligent auxiliary diagnosis methods for skin melanoma.

[0073] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0074] According to an embodiment of the present invention, a dermoscopic image data set is first acquired; a feature set of dermoscopic image data is obtained based on the dermoscopic image data set; an initial classification head is constructed; the initial classification head is trained based on the feature set to obtain a trained classification head, and a pre-diagnosis score set is obtained through the trained classification head; secondly, an unsupervised clustering algorithm is used to calculate the untrustworthy interval of the pre-diagnosis score; an untrustworthy pre-diagnosis score set and a credible pre-diagnosis score set are obtained based on the untrustworthy interval of the pre-diagnosis score and the pre-diagnosis score set; an initial gating mechanism compensation model is constructed; the initial gating mechanism compensation model is trained based on the untrustworthy pre-diagnosis score set and the feature set corresponding to the untrustworthy pre-diagnosis score set to obtain a trained gating mechanism compensation model; the untrustworthy pre-diagnosis score set and the feature set corresponding to the untrustworthy pre-diagnosis score set are input into the trained gating mechanism compensation model to obtain a compensated pre-diagnosis score set; the compensated pre-diagnosis score set is merged with the credible pre-diagnosis score set to obtain a final pre-diagnosis score set; finally, an optimal transmission algorithm is used to map the final pre-diagnosis score set to obtain a diagnosis probability; and a prediction result of skin melanoma is obtained based on the diagnosis probability.

[0075] The embodiment of the present invention constructs a classification head of a fully connected layer based on a dermoscopic image training set and a feature set corresponding to the dermoscopic image training set, performs pre-diagnosis based on the feature set, and obtains a corresponding pre-diagnosis score, thereby ensuring the input consistency of classification heads across different models and simplifying the fine-tuning process for multiple deep learning classification heads; the embodiment of the present invention constructs a gating mechanism compensation model, identifies unreliable samples based on the results of pre-diagnosis and adjusts unreliable predictions from the pre-diagnosis stage, compensates for the unreliability of the pre-diagnosis score, obtains the compensated diagnosis score, and improves the diagnostic credibility based on the deep learning model; the embodiment of the present invention adopts an optimal transmission algorithm to construct a fairness optimal predictor, maps the fairness of the compensated diagnosis score, and outputs the mapped diagnosis probability, thereby improving the fairness and accuracy of the deep learning diagnosis model. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0077] Figure 1 This is a flow chart of a method for intelligent auxiliary diagnosis of skin melanoma provided by an embodiment of the present invention;

[0078] Figure 2It is a structural schematic diagram of a gating mechanism compensation model of a skin melanoma intelligent auxiliary diagnosis method provided by an embodiment of the present invention;

[0079] Figure 3 It is a schematic diagram of the entire processing framework structure of an intelligent auxiliary diagnosis method for skin melanoma provided by an embodiment of the present invention;

[0080] Figure 4 This is a block diagram of an intelligent auxiliary diagnosis device for skin melanoma provided by an embodiment of the present invention;

[0081] Figure 5 It is a structural schematic diagram of an intelligent auxiliary diagnosis device for skin melanoma provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0082] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0083] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either of the two.

[0084] In the embodiments of the present invention, "image" and "picture" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same. "of", "corresponding, relevant" and "corresponding" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same.

[0085] In the embodiments of the present invention, sometimes the subscripts such as W 1 It may be written in non-subscript form such as W1. When the difference is not emphasized, the meaning is the same.

[0086] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0087] The embodiment of the present invention provides a method for intelligent auxiliary diagnosis of skin melanoma, which can be implemented by an intelligent auxiliary diagnosis device for skin melanoma, and the intelligent auxiliary diagnosis device for skin melanoma can be a terminal or a server. Figure 1 The flowchart of the intelligent auxiliary diagnosis method for skin melanoma is shown in the figure. The processing flow of the method may include the following steps:

[0088] S1. Obtain a dermoscopic image dataset; and obtain a feature set of the dermoscopic image data based on the dermoscopic image dataset.

[0089] Among them, the dermoscopic image dataset includes the ISIC2019 dermoscopic image dataset and the ISIC2020 dermoscopic image dataset.

[0090] In a feasible implementation, the dermoscopic image dataset is divided into a training set and a test set; wherein the ratio of melanoma labels in the training set and the test set is consistent.

[0091] Optionally, S1 obtains a feature set of dermoscopic image data according to the dermoscopic image dataset, including:

[0092] The dermoscopic image dataset is input into the pre-trained deep learning network model to obtain the feature set of the dermoscopic image data.

[0093] Among them, the pre-trained deep learning network model can be selected from any one of a deep learning network model, a Transformer-based model, an LLM model, an LVM model, a Resntet18 model, a Densenet121 model and a Mamba model, and the present invention does not limit this.

[0094] In a feasible implementation, a dermoscopic image dataset is input into a pre-trained deep learning network model, and feature extraction is performed through a fully connected layer to obtain dermoscopic image features. The dermoscopic image features corresponding to each dermoscopic image are scaled to a scale of 1×2048 using a resampling technique to obtain a feature set of the dermoscopic image data.

[0095] S2. Construct an initial classification head; train the initial classification head according to the feature set to obtain a trained classification head, and obtain a pre-diagnosis score set through the trained classification head.

[0096] The classification head includes a fully connected layer; the scale of the input features of the fully connected layer is 2048, and the scale of the output pre-diagnosis score is 1.

[0097] Optionally, S2 trains the initial classification head according to the feature set to obtain a trained classification head, including:

[0098] According to the feature set, the binary cross entropy loss function is used to train the initial classification head to obtain the trained classification head.

[0099] S3. Use an unsupervised clustering algorithm to calculate the unreliable interval of the prediagnosis score; obtain an unreliable prediagnosis score set and a reliable prediagnosis score set based on the unreliable interval of the prediagnosis score and the prediagnosis score set.

[0100] Optionally, specific implementation steps of S3 include S31-S35:

[0101] S31, using an unsupervised clustering algorithm to classify the pre-diagnosis score set into: first-category data, second-category data, and third-category data;

[0102] S32, obtaining the centroid of the first category of data, the centroid of the second category of data, and the centroid of the third category of data;

[0103] S33, sorting the centroids of the first category data, the second category data, and the third category data to obtain a minimum value, a median value, and a maximum value;

[0104] S34. Calculate the unreliable interval of the pre-diagnosis score based on the minimum, median and maximum values;

[0105] The evaluation index of credibility is the distance between the diagnostic score and the preset diagnostic threshold, which can be expressed by the following formula:

[0106] ;

[0107] in, represents the diagnostic score; t represents the preset diagnostic threshold; Indicates distance.

[0108] In one possible implementation, given a trust radius ,when The diagnostic score results are unreliable. The confidence interval for the diagnosis.

[0109] In one feasible implementation, the unconfidence interval of the pre-diagnosis score can be expressed as .

[0110] in, , .

[0111] in, represents the minimum value, which represents the centroid of the untrustworthy sample; represents the median, which represents the centroid of negative samples; Represents the maximum value and represents the centroid of the positive samples.

[0112] S35 . Obtain an unreliable prediagnosis score set and a reliable prediagnosis score set according to the unreliable interval of the prediagnosis score and the prediagnosis score set.

[0113] S4. Construct an initial gating mechanism compensation model.

[0114] Among them, Figure 2 It is a structural schematic diagram of a gating mechanism compensation model of an intelligent auxiliary diagnosis method for skin melanoma provided by an embodiment of the present invention; in a feasible implementation manner, an unreliable pre-diagnosis score set and a feature set corresponding to the unreliable pre-diagnosis score set are input into the fully connected layer of the gating mechanism compensation model, and processed by the first fully connected layer and the fully connected layer to obtain standardized features; the standardized features are fused by the element addition method to obtain fused features; the fused features are input into the batch normalization layer to obtain normalized features; the normalized features are input into the GELU activation function to perform nonlinear mapping processing to obtain nonlinear mapping processing results; the nonlinear mapping processing results are input into the fully connected layer to obtain a compensated diagnosis score; according to the compensated diagnosis score, the model is trained using a binary cross entropy loss function to obtain a trained gating mechanism compensation model.

[0115] Optionally, the gating mechanism compensation model of S4 includes 3 fully connected layers, 1 batch normalization layer and GELU activation function; wherein,

[0116] Fully connected layer, used to standardize input features;

[0117] Batch normalization layer, used to normalize the input features;

[0118] GELU activation function is used to perform nonlinear mapping on input features.

[0119] S5. According to the untrustworthy pre-diagnosis score set and the feature set corresponding to the untrustworthy pre-diagnosis score set, the initial gating mechanism compensation model is trained to obtain a trained gating mechanism compensation model.

[0120] Optionally, specific implementation steps of S5 include S51-S55:

[0121] S51, inputting the untrustworthy pre-diagnosis score set and the feature set corresponding to the untrustworthy pre-diagnosis score set into the initial gating mechanism compensation model, and normalizing the untrustworthy pre-diagnosis score set and the feature set corresponding to the untrustworthy pre-diagnosis score set through a fully connected layer to obtain standardized features;

[0122] S52, fusing the standardized features by using an element addition method to obtain fused features;

[0123] S53, inputting the fused features into a batch normalization layer to obtain normalized features;

[0124] S54, inputting the normalized features into the GELU activation function, performing nonlinear mapping processing, and obtaining a nonlinear mapping processing result; inputting the nonlinear mapping processing result into the fully connected layer to obtain a compensated diagnostic score;

[0125] S55. According to the compensated diagnostic score, the initial gating mechanism compensation model is trained using a binary cross entropy loss function to obtain a trained gating mechanism compensation model.

[0126] Among them, the AdamW optimizer is used to optimize the objective function, and the trained gating mechanism compensation model is obtained by iteratively updating the model parameters.

[0127] S6. Input the untrusted pre-diagnosis score set and the feature set corresponding to the untrusted pre-diagnosis score set into the trained gating mechanism compensation model to obtain the compensated pre-diagnosis score set; merge the compensated pre-diagnosis score set with the trusted pre-diagnosis score set to obtain the final pre-diagnosis score set.

[0128] S7. Use the optimal transmission algorithm to map the final pre-diagnosis score set to obtain the diagnosis probability; according to the diagnosis probability, obtain the prediction result of skin melanoma.

[0129] Optionally, specific implementation steps of S7 include S71-S73:

[0130] S71, using the Sigmoid function to map the final pre-diagnosis score set to obtain the diagnosis probability;

[0131] S72, using the fairness optimal mapping theory to map the diagnosis probability and obtain the optimal fairness mapping probability;

[0132] In a feasible implementation, the diagnosis probability is divided into two different subgroups according to the sensitive attribute of demographics, including: a first subgroup and a second subgroup. The two asynchronous subgroups are divided based on the sensitive attribute of demographics, and the sensitive attribute includes, but is not limited to: male / female, patient age greater than / less than 60 years old. This implementation example uses the sensitive attribute of male / female to obtain two subgroups, one group is male patients, and the other group is female patients.

[0133] In a feasible implementation, based on the data distribution of the first subgroup, the ratio of the number of samples in the first subgroup to the total number of samples is calculated to obtain the prior probability of the first subgroup; based on the data distribution of the second subgroup, the ratio of the number of samples in the second subgroup to the total number of samples is calculated to obtain the prior probability of the second subgroup.

[0134] In a feasible implementation, based on the data distribution of the first subgroup, the cumulative function and the quantile function of the first subgroup are obtained; based on the data distribution of the second subgroup, the cumulative function and the quantile function of the second subgroup are obtained.

[0135] In a feasible implementation, according to each probability value in the diagnosis probability, the fairness optimal mapping theory is adopted to obtain the optimal fairness mapping probability; wherein the fairness optimal mapping theory can be expressed by the following formula:

[0136]

[0137] in, represents the prior probability of the first subgroup; represents the quantile function of the first subgroup; represents the cumulative function of the first subgroup; represents the prior probability of the second subgroup; represents the quantile function of the second subgroup; represents the cumulative function of the second subgroup; represents the optimal fairness mapping probability; Indicates the quantile function value of the second subgroup corresponding to the quantile function value of the first subgroup.

[0138] Among them, the optimal fairness mapping probability combines the diagnosis probability data distribution under different sensitive attributes according to the prior probabilities of the two subgroups, reducing the impact of different sensitive attributes on the diagnosis probability.

[0139] Among them, the test set is used to verify the entire processing framework, such as Figure 3 The figure shows a schematic diagram of the entire processing framework structure of an intelligent auxiliary diagnosis method for skin melanoma provided by an embodiment of the present invention; in a feasible implementation manner, a test set is input into a pre-trained feature extraction model to obtain a feature set of dermoscopic image data; the feature set of dermoscopic image data is input into a trained classification head to obtain a pre-diagnosis score; the pre-diagnosis score is input into a gating mechanism compensation model to obtain a compensated pre-diagnosis score; an optimal transmission algorithm is used to map the final pre-diagnosis score set to obtain a diagnosis probability; and a prediction result of skin melanoma is obtained based on the diagnosis probability.

[0140] S73. According to the optimal fairness mapping probability, a prediction result of skin melanoma is obtained.

[0141] In a feasible implementation, the embodiment of the present invention uses accuracy and group mean as evaluation indicators of diagnostic performance and fairness performance; wherein, a larger accuracy value indicates a more accurate diagnosis, and a smaller group mean indicates a fairer diagnosis.

[0142] According to an embodiment of the present invention, a dermoscopic image data set is first acquired; a feature set of dermoscopic image data is obtained based on the dermoscopic image data set; an initial classification head is constructed; the initial classification head is trained based on the feature set to obtain a trained classification head, and a pre-diagnosis score set is obtained through the trained classification head; secondly, an unsupervised clustering algorithm is used to calculate the untrustworthy interval of the pre-diagnosis score; an untrustworthy pre-diagnosis score set and a credible pre-diagnosis score set are obtained based on the untrustworthy interval of the pre-diagnosis score and the pre-diagnosis score set; an initial gating mechanism compensation model is constructed; the initial gating mechanism compensation model is trained based on the untrustworthy pre-diagnosis score set and the feature set corresponding to the untrustworthy pre-diagnosis score set to obtain a trained gating mechanism compensation model; the untrustworthy pre-diagnosis score set and the feature set corresponding to the untrustworthy pre-diagnosis score set are input into the trained gating mechanism compensation model to obtain a compensated pre-diagnosis score set; the compensated pre-diagnosis score set is merged with the credible pre-diagnosis score set to obtain a final pre-diagnosis score set; finally, an optimal transmission algorithm is used to map the final pre-diagnosis score set to obtain a diagnosis probability; and a prediction result of skin melanoma is obtained based on the diagnosis probability.

[0143] The embodiment of the present invention constructs a classification head of a fully connected layer based on a dermoscopic image training set and a feature set corresponding to the dermoscopic image training set, performs pre-diagnosis based on the feature set, and obtains a corresponding pre-diagnosis score, thereby ensuring the input consistency of classification heads across different models and simplifying the fine-tuning process for multiple deep learning classification heads; the embodiment of the present invention constructs a gating mechanism compensation model, identifies unreliable samples based on the results of pre-diagnosis and adjusts unreliable predictions from the pre-diagnosis stage, compensates for the unreliability of the pre-diagnosis score, obtains the compensated diagnosis score, and improves the diagnostic credibility based on the deep learning model; the embodiment of the present invention adopts an optimal transmission algorithm to construct a fairness optimal predictor, maps the fairness of the compensated diagnosis score, and outputs the mapped diagnosis probability, thereby improving the fairness and accuracy of the deep learning diagnosis model.

[0144] Figure 4 is a block diagram of a skin melanoma intelligent auxiliary diagnosis device according to an exemplary embodiment, and the device is used in a skin melanoma intelligent auxiliary diagnosis method. Figure 4 The device includes an acquisition unit 300, a first construction unit 310, a first training unit 320, a first prediction unit 330, a calculation unit 340, a second construction unit 350, a second training unit 360, a second prediction unit 370 and an output unit 380. Wherein:

[0145] An acquisition unit 300 is used to acquire a dermoscopic image data set; and obtain a feature set of the dermoscopic image data according to the dermoscopic image data set;

[0146] A first construction unit 310 is used to construct an initial classification head;

[0147] A first training unit 320, configured to train the initial classification head according to the feature set to obtain a trained classification head;

[0148] A first prediction unit 330, configured to obtain a pre-diagnosis score set through the trained classification head;

[0149] The calculation unit 340 is used to calculate the untrustworthy interval of the pre-diagnosis score by using an unsupervised clustering algorithm; and obtain an untrustworthy pre-diagnosis score set and a credible pre-diagnosis score set according to the untrustworthy interval of the pre-diagnosis score and the pre-diagnosis score set;

[0150] A second construction unit 350 is used to construct an initial gating mechanism compensation model;

[0151] A second training unit 360 is used to train the initial gating mechanism compensation model according to the untrustworthy pre-diagnosis score set and the feature set corresponding to the untrustworthy pre-diagnosis score set to obtain a trained gating mechanism compensation model;

[0152] The second prediction unit 370 is used to input the untrustworthy pre-diagnosis score set and the feature set corresponding to the untrustworthy pre-diagnosis score set into the trained gating mechanism compensation model to obtain a compensated pre-diagnosis score set; and merge the compensated pre-diagnosis score set with the credible pre-diagnosis score set to obtain a final pre-diagnosis score set;

[0153] The output unit 380 is used to use the optimal transmission algorithm to perform mapping processing on the final pre-diagnosis score set to obtain a diagnosis probability; and obtain a prediction result of skin melanoma according to the diagnosis probability.

[0154] Optionally, the acquiring unit 300 is configured to:

[0155] The dermoscopic image dataset is input into a pre-trained deep learning network model to obtain a feature set of the dermoscopic image data.

[0156] Optionally, the first training unit 320 includes:

[0157] According to the feature set, the initial classification head is trained using a binary cross entropy loss function to obtain a trained classification head.

[0158] Optionally, the computing unit 340 is configured to:

[0159] Using an unsupervised clustering algorithm, the pre-diagnosis score set is classified into: first-category data, second-category data, and third-category data;

[0160] Obtain the centroid of the first type of data, the centroid of the second type of data, and the centroid of the third type of data;

[0161] Sort the centroids of the first category data, the second category data, and the third category data to obtain the minimum value, median value, and maximum value;

[0162] The unconfidence interval of the prediagnostic score was calculated based on the minimum, median, and maximum values;

[0163] According to the untrustworthy interval of the pre-diagnosis score and the pre-diagnosis score set, an untrustworthy pre-diagnosis score set and a trustworthy pre-diagnosis score set are obtained.

[0164] Optionally, the second constructing unit 350 is used to:

[0165] The fully connected layer is used to standardize the input features;

[0166] The batch normalization layer is used to normalize the input features;

[0167] The GELU activation function is used to perform nonlinear mapping on input features.

[0168] Optionally, the first training unit 360 is used to:

[0169] Inputting the untrustworthy pre-diagnosis score set and the feature set corresponding to the untrustworthy pre-diagnosis score set into the initial gating mechanism compensation model, and performing standardization processing on the untrustworthy pre-diagnosis score set and the feature set corresponding to the untrustworthy pre-diagnosis score set through the fully connected layer to obtain standardized features;

[0170] The standardized features are fused by using an element addition method to obtain fused features;

[0171] Inputting the fused features into the batch normalization layer to obtain normalized features;

[0172] Inputting the normalized features into the GELU activation function, performing nonlinear mapping processing, and obtaining a nonlinear mapping processing result; inputting the nonlinear mapping processing result into the fully connected layer to obtain a compensated diagnostic score;

[0173] According to the compensated diagnosis score, the initial gating mechanism compensation model is trained using a binary cross entropy loss function to obtain a trained gating mechanism compensation model.

[0174] Optionally, the output unit 380 is used to:

[0175] The final pre-diagnosis score set is mapped using a Sigmoid function to obtain a diagnosis probability;

[0176] The diagnosis probability is mapped using the fairness optimal mapping theory to obtain the optimal fairness mapping probability;

[0177] According to the optimal fairness mapping probability, a prediction result of skin melanoma is obtained.

[0178] According to an embodiment of the present invention, a dermoscopic image data set is first acquired; a feature set of dermoscopic image data is obtained based on the dermoscopic image data set; an initial classification head is constructed; the initial classification head is trained based on the feature set to obtain a trained classification head, and a pre-diagnosis score set is obtained through the trained classification head; secondly, an unsupervised clustering algorithm is used to calculate the untrustworthy interval of the pre-diagnosis score; an untrustworthy pre-diagnosis score set and a credible pre-diagnosis score set are obtained based on the untrustworthy interval of the pre-diagnosis score and the pre-diagnosis score set; an initial gating mechanism compensation model is constructed; the initial gating mechanism compensation model is trained based on the untrustworthy pre-diagnosis score set and the feature set corresponding to the untrustworthy pre-diagnosis score set to obtain a trained gating mechanism compensation model; the untrustworthy pre-diagnosis score set and the feature set corresponding to the untrustworthy pre-diagnosis score set are input into the trained gating mechanism compensation model to obtain a compensated pre-diagnosis score set; the compensated pre-diagnosis score set is merged with the credible pre-diagnosis score set to obtain a final pre-diagnosis score set; finally, an optimal transmission algorithm is used to map the final pre-diagnosis score set to obtain a diagnosis probability; and a prediction result of skin melanoma is obtained based on the diagnosis probability.

[0179] The embodiment of the present invention constructs a classification head of a fully connected layer based on a dermoscopic image training set and a feature set corresponding to the dermoscopic image training set, performs pre-diagnosis based on the feature set, and obtains a corresponding pre-diagnosis score, thereby ensuring the input consistency of classification heads across different models and simplifying the fine-tuning process for multiple deep learning classification heads; the embodiment of the present invention constructs a gating mechanism compensation model, identifies unreliable samples based on the results of pre-diagnosis and adjusts unreliable predictions from the pre-diagnosis stage, compensates for the unreliability of the pre-diagnosis score, obtains the compensated diagnosis score, and improves the diagnostic credibility based on the deep learning model; the embodiment of the present invention adopts an optimal transmission algorithm to construct a fairness optimal predictor, maps the fairness of the compensated diagnosis score, and outputs the mapped diagnosis probability, thereby improving the fairness and accuracy of the deep learning diagnosis model.

[0180] Figure 5 FIG. 1 is a schematic diagram of a structure of an intelligent auxiliary diagnosis device for skin melanoma provided by an embodiment of the present invention. Figure 5 As shown, the skin melanoma intelligent auxiliary diagnosis device may include the above Figure 4Optionally, the skin melanoma intelligent auxiliary diagnosis device 410 may include a first processor 2001 .

[0181] Optionally, the skin melanoma intelligent auxiliary diagnosis device 410 may further include a memory 2002 and a transceiver 2003 .

[0182] The first processor 2001, the memory 2002 and the transceiver 2003 may be connected via a communication bus.

[0183] Combine the following Figure 5 The components of the skin melanoma intelligent auxiliary diagnosis device 410 are specifically introduced:

[0184] The first processor 2001 is the control center of the skin melanoma intelligent auxiliary diagnosis device 410, which can be a processor or a general term for multiple processing elements. For example, the first processor 2001 is one or more central processing units (CPUs), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention, such as one or more microprocessors (digital signal processors, DSPs), or one or more field programmable gate arrays (field programmable gate arrays, FPGAs).

[0185] Optionally, the first processor 2001 can perform various functions of the skin melanoma intelligent auxiliary diagnosis device 410 by running or executing a software program stored in the memory 2002 and calling data stored in the memory 2002.

[0186] In a specific implementation, as an embodiment, the first processor 2001 may include one or more CPUs, such as Figure 5 CPU0 and CPU1 are shown in FIG.

[0187] In a specific implementation, as an embodiment, the skin melanoma intelligent auxiliary diagnosis device 410 may also include multiple processors, such as Figure 5 The first processor 2001 and the second processor 2004 are shown in FIG. Each of these processors may be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). The processor here may refer to one or more devices, circuits, and / or processing cores for processing data (e.g., computer program instructions).

[0188] The memory 2002 is used to store the software program for executing the solution of the present invention, and is controlled to be executed by the first processor 2001. The specific implementation method can refer to the above method embodiment, which will not be repeated here.

[0189] Optionally, the memory 2002 may be a read-only memory (ROM) or other types of static storage devices that can store static information and instructions, a random access memory (RAM) or other types of dynamic storage devices that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory 2002 may be integrated with the first processor 2001, or may exist independently, and may be accessed through the interface circuit ( Figure 5 (not shown) is coupled to the first processor 2001, which is not specifically limited in this embodiment of the present invention.

[0190] The transceiver 2003 is used to communicate with a network device or a terminal device.

[0191] Optionally, the transceiver 2003 may include a receiver and a transmitter ( Figure 5 The receiver is used to implement a receiving function, and the transmitter is used to implement a sending function.

[0192] Optionally, the transceiver 2003 may be integrated with the first processor 2001, or may exist independently and communicate with the first processor 2001 through the interface circuit ( Figure 5 (not shown) is coupled to the first processor 2001, which is not specifically limited in this embodiment of the present invention.

[0193] It should be noted that Figure 5 The structure of the skin melanoma intelligent auxiliary diagnosis device 410 shown in the figure does not constitute a limitation on the router, and the actual knowledge structure recognition device may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0194] In addition, the technical effects of the skin melanoma intelligent auxiliary diagnosis device 410 can refer to the technical effects of the skin melanoma intelligent auxiliary diagnosis method described in the above method embodiment, and will not be repeated here.

[0195] It should be understood that the first processor 2001 in the embodiment of the present invention may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0196] It should also be understood that the memory in the embodiments of the present invention may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic random access memory (DRAM), synchronous DRAM (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0197] The above embodiments can be implemented in whole or in part by software, hardware (such as circuits), firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0198] It should be understood that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship, but it may also indicate an "and / or" relationship. Please refer to the context for specific understanding.

[0199] In the present invention, "at least one" means one or more, and "plurality" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0200] It should be understood that in various embodiments of the present invention, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0201] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0202] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described equipment, devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0203] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0204] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0205] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0206] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program codes.

[0207] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for intelligent auxiliary diagnosis of skin melanoma, characterized in that: The method comprises: S1. Obtain a dermoscopic image dataset; and obtain a feature set of dermoscopic image data according to the dermoscopic image dataset; S2, constructing an initial classification head; training the initial classification head according to the feature set to obtain a trained classification head, and obtaining a pre-diagnosis score set through the trained classification head; S3, using an unsupervised clustering algorithm to calculate the untrustworthy interval of the pre-diagnosis score; according to the untrustworthy interval of the pre-diagnosis score and the pre-diagnosis score set, obtaining an untrustworthy pre-diagnosis score set and a credible pre-diagnosis score set; The step S3 uses an unsupervised clustering algorithm to calculate the unreliable interval of the pre-diagnosis score; and obtains an unreliable pre-diagnosis score set and a reliable pre-diagnosis score set according to the unreliable interval of the pre-diagnosis score and the pre-diagnosis score set, including: S31, using an unsupervised clustering algorithm to classify the pre-diagnosis score set into: first-category data, second-category data, and third-category data; S32, obtaining the centroid of the first category of data, the centroid of the second category of data, and the centroid of the third category of data; S33, sorting the centroids of the first category data, the second category data, and the third category data to obtain a minimum value, a median value, and a maximum value; S34. Calculate the unreliable interval of the pre-diagnosis score based on the minimum, median and maximum values; S35, obtaining an unreliable prediagnosis score set and a reliable prediagnosis score set according to the unreliable interval of the prediagnosis score and the prediagnosis score set; S4. Construct an initial gating mechanism compensation model; The gating mechanism compensation model includes three fully connected layers, one batch normalization layer and a GELU activation function; The fully connected layer is used to standardize the input features; The batch normalization layer is used to normalize the input features; The GELU activation function is used to perform nonlinear mapping on the input features; S5, training the initial gating mechanism compensation model according to the untrustworthy pre-diagnosis score set and the feature set corresponding to the untrustworthy pre-diagnosis score set to obtain a trained gating mechanism compensation model; The step S5, training the initial gating mechanism compensation model according to the untrustworthy pre-diagnosis score set and the feature set corresponding to the untrustworthy pre-diagnosis score set to obtain a trained gating mechanism compensation model, includes: S51, inputting the untrustworthy pre-diagnosis score set and the feature set corresponding to the untrustworthy pre-diagnosis score set into the initial gating mechanism compensation model, and performing normalization processing on the untrustworthy pre-diagnosis score set and the feature set corresponding to the untrustworthy pre-diagnosis score set through the fully connected layer to obtain normalized features; S52, fusing the standardized features by an element addition method to obtain fused features; S53, inputting the fused features into the batch normalization layer to obtain normalized features; S54, inputting the normalized features into the GELU activation function, performing nonlinear mapping processing, and obtaining a nonlinear mapping processing result; inputting the nonlinear mapping processing result into a fully connected layer to obtain a compensated diagnostic score; S55, training the initial gating mechanism compensation model using a binary cross entropy loss function according to the compensated diagnostic score to obtain a trained gating mechanism compensation model; S6, inputting the untrusted pre-diagnosis score set and the feature set corresponding to the untrusted pre-diagnosis score set into the trained gating mechanism compensation model to obtain a compensated pre-diagnosis score set; merging the compensated pre-diagnosis score set with the trusted pre-diagnosis score set to obtain a final pre-diagnosis score set; S7, performing mapping processing on the final pre-diagnosis score set to obtain a diagnosis probability; and obtaining a prediction result of skin melanoma according to the diagnosis probability; The step S7 performs mapping processing on the final pre-diagnosis score set to obtain a diagnosis probability; and obtains a prediction result of skin melanoma according to the diagnosis probability, including: S71, using a Sigmoid function to map the final pre-diagnosis score set to obtain a diagnosis probability; S72, mapping the diagnosis probability using the fairness optimal mapping theory to obtain the optimal fairness mapping probability; S73. Obtain a prediction result of skin melanoma according to the optimal fairness mapping probability.

2. The intelligent auxiliary diagnosis method for skin melanoma according to claim 1, characterized in that: The step S1 of obtaining a feature set of dermoscopic image data according to the dermoscopic image data set includes: The dermoscopic image dataset is input into a pre-trained deep learning network model to obtain a feature set of the dermoscopic image data.

3. The intelligent auxiliary diagnosis method for skin melanoma according to claim 2, characterized in that: The step S2 of training the initial classification head according to the feature set to obtain a trained classification head includes: According to the feature set, the initial classification head is trained using a binary cross entropy loss function to obtain a trained classification head.

4. An intelligent auxiliary diagnosis device for skin melanoma, the intelligent auxiliary diagnosis device for skin melanoma is used to implement the intelligent auxiliary diagnosis method for skin melanoma according to any one of claims 1 to 3, characterized in that: The device comprises: An acquisition unit, configured to acquire a dermoscopic image data set; and obtain a feature set of the dermoscopic image data according to the dermoscopic image data set; A first construction unit, used to construct an initial classification head; A first training unit, used for training the initial classification head according to the feature set to obtain a trained classification head; A first prediction unit, configured to obtain a pre-diagnosis score set through the trained classification head; A calculation unit, configured to calculate an untrustworthy interval of the pre-diagnosis score by using an unsupervised clustering algorithm; and obtain an untrustworthy pre-diagnosis score set and a credible pre-diagnosis score set according to the untrustworthy interval of the pre-diagnosis score and the pre-diagnosis score set; Wherein, the computing unit is used for: Using an unsupervised clustering algorithm, the pre-diagnosis score set is classified into: first-category data, second-category data, and third-category data; Obtain the centroid of the first type of data, the centroid of the second type of data, and the centroid of the third type of data; Sort the centroids of the first category data, the second category data, and the third category data to obtain the minimum value, median value, and maximum value; The unconfidence interval of the prediagnostic score was calculated based on the minimum, median, and maximum values; According to the untrustworthy interval of the pre-diagnosis score and the pre-diagnosis score set, obtaining an untrustworthy pre-diagnosis score set and a credible pre-diagnosis score set; The second building unit is used to build an initial gating mechanism compensation model; A second training unit is used to train the initial gating mechanism compensation model according to the untrustworthy pre-diagnosis score set and the feature set corresponding to the untrustworthy pre-diagnosis score set to obtain a trained gating mechanism compensation model; Wherein, the second training unit is used for: Inputting the untrustworthy pre-diagnosis score set and the feature set corresponding to the untrustworthy pre-diagnosis score set into the initial gating mechanism compensation model, and normalizing the untrustworthy pre-diagnosis score set and the feature set corresponding to the untrustworthy pre-diagnosis score set through a fully connected layer to obtain standardized features; The standardized features are fused by using an element addition method to obtain fused features; Input the fused features into the batch normalization layer to obtain normalized features; Inputting the normalized features into the GELU activation function, performing nonlinear mapping processing, and obtaining a nonlinear mapping processing result; inputting the nonlinear mapping processing result into the fully connected layer to obtain a compensated diagnostic score; According to the compensated diagnosis score, the initial gating mechanism compensation model is trained using a binary cross entropy loss function to obtain a trained gating mechanism compensation model; A second prediction unit is used to input the untrustworthy pre-diagnosis score set and the feature set corresponding to the untrustworthy pre-diagnosis score set into the trained gating mechanism compensation model to obtain a compensated pre-diagnosis score set; and merge the compensated pre-diagnosis score set with the credible pre-diagnosis score set to obtain a final pre-diagnosis score set; An output unit, used to perform mapping processing on the final pre-diagnosis score set to obtain a diagnosis probability; and obtain a prediction result of skin melanoma according to the diagnosis probability; Wherein, the output unit is used for: The final pre-diagnosis score set is mapped using a Sigmoid function to obtain a diagnosis probability; The diagnosis probability is mapped using the fairness optimal mapping theory to obtain the optimal fairness mapping probability; According to the optimal fairness mapping probability, a prediction result of skin melanoma is obtained.

5. An intelligent auxiliary diagnosis device for skin melanoma, characterized in that: The skin melanoma intelligent auxiliary diagnosis device comprises: processor; A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the method according to any one of claims 1 to 3 is implemented.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program code, and the computer program code can be called by a processor to execute the method according to any one of claims 1 to 3.

Images