Hereditary cancer auxiliary diagnosis method, device and program product based on adaptive dynamic range loss function

Abstract

The invention relates to the field of intelligent medical treatment, in particular to an adaptive dynamic range loss function-based hereditary cancer auxiliary diagnosis method, equipment and a program product. Comprising the steps of obtaining image data and genetic information data of a to-be-tested person; the image data and the genetic information data are transmitted to a genetic cancer discrimination model to obtain a discrimination result of whether the cancer is a genetic cancer; the hereditary cancer discrimination model is obtained by pre-training image data and genetic information data, and the discrimination model is optimized by calculating an average value of absolute errors of different scale data between a predicted value and a true value through an adaptive dynamic range loss function and adjusting dynamic errors. The method can assist doctors to diagnose and discriminate genetic cancers, and has good clinical value.

Description

Technical Field

[0001] This application relates to the field of intelligent medicine, and specifically relates to a method, device, program product, and computer-readable storage medium for the auxiliary diagnosis of hereditary cancer based on an adaptive dynamic range loss function. Background Art

[0002] Hereditary renal cancer is a kidney tumor with a family genetic tendency, accounting for about 2%-8% of the total renal cancer. Although its incidence is not high, its pathological types are complex, with various disease syndromes and characteristic genetic gene mutations in different clinical manifestations, bringing great difficulties to clinical diagnosis and treatment. Traditional medical imaging diagnosis mainly relies on doctors' experience and professional knowledge, while genetic screening is based on the patient's family history and limited gene test results. These methods have limitations in diagnosis efficiency and accuracy. Summary of the Invention

[0003] In view of the above problems, the present invention provides a method for the auxiliary diagnosis of hereditary cancer based on an adaptive dynamic range loss function, aiming to accurately and quickly identify the characteristics of hereditary cancer from a large amount of medical imaging data and genetic information through deep learning technology, and improve the diagnosis efficiency and accuracy. Specifically, it includes: Obtain the imaging data and genetic information data of the person to be tested; Input the imaging data and genetic information data into the hereditary cancer discrimination model to obtain the discrimination result of whether it is hereditary cancer; the hereditary cancer discrimination model is pre-trained through imaging data and genetic information data. Among them, the discrimination model is optimized by calculating the average value of the absolute error of different-scale data between the predicted value and the true value and dynamic error adjustment through the adaptive dynamic range loss function.

[0004] The calculation formula of the adaptive dynamic range loss function ADRL(y, is expressed as: ) is expressed as: ADRL(y, ) = *exp(- ) where y is the true label, is the output value predicted by the model, N is the total number of samples, is the absolute error between the predicted value and the true value, is the standard deviation of the true label y, is a very small positive number, represents the mean value of the true label y, is a dynamic adjustment term.

[0005] The structure of the hereditary cancer discrimination model includes an input layer, an encoder layer, an attention layer, and a fully connected layer. The image data and genetic information data are input to the input layer, converted into vectors and then fused, and then sequentially pass through the encoder layer, the attention layer, and the fully connected layer to obtain the discrimination result; Optionally, the encoder layer performs vector encoding through a self-attention mechanism to obtain a self-attention feature vector. The attention layer performs feature processing through multi-head attention. After the self-attention feature vector is input to the attention layer, it is divided into N heads, where N is a natural number greater than 1. Each head independently calculates to obtain N features, and the N features are fused to obtain an attention feature. The attention feature is processed through a fully connected layer to obtain the discrimination result; Optionally, the training process of the hereditary discrimination model is as follows: Obtain a patient image data set and a genetic information data set; Input the image data set and the genetic information data set into the hereditary cancer discrimination model to be trained for feature learning, and calculate the loss value through an adaptive dynamic range loss function. Optimize the model based on the loss value; Repeat the above steps until the adaptive dynamic loss function remains unchanged to obtain the hereditary discrimination model.

[0006] The hereditary discrimination model further includes a hereditary cancer gene auxiliary diagnosis module. The feature vector output by the hereditary discrimination model is input to the hereditary cancer gene auxiliary diagnosis module for feature analysis to obtain the diagnosis result of the hereditary cancer type. The hereditary cancer gene auxiliary diagnosis module includes a feature extraction layer and a classification layer. The feature vector is processed through the feature extraction layer and the classification layer to obtain the diagnosis result.

[0007] Optionally, the diagnosis result further includes the genetic mutation type.

[0008] The feature vector is the feature vector output by the encoder layer or the fully connected layer of the genetic discrimination model.

[0009] The training process of the hereditary cancer gene auxiliary diagnosis module is as follows: Obtain a feature vector set; Input the feature vector set into the hereditary cancer gene auxiliary diagnosis module for feature extraction, classification calculation, and loss calculation; Repeat the above steps until the loss remains unchanged to obtain the hereditary cancer gene auxiliary diagnosis module.

[0010] The loss calculation of the hereditary cancer gene auxiliary diagnosis module is through one or more of the following loss functions: cross-entropy loss function, logarithmic loss function, hinge loss function, adaptive dynamic range loss function.

[0011] The object of the present invention is to provide a computer program product, which includes a computer program or instruction, and the computer program or instruction is executed by a processor to implement the above-mentioned hereditary cancer assisted diagnosis method based on an adaptive dynamic range loss function.

[0012] The object of the present invention is to provide a computer device, which includes a memory, a processor, and a computer program or instruction stored on the memory, and the computer program or instruction is executed by the processor to implement the above-mentioned hereditary cancer assisted diagnosis method based on an adaptive dynamic range loss function.

[0013] The object of the present invention is to provide a computer-readable storage medium, which stores a computer program or instruction, and the computer program or instruction is executed by a processor to implement the above-mentioned hereditary cancer assisted diagnosis method based on an adaptive dynamic range loss function.

[0014] Advantages of the present invention: 1. The present invention proposes an adaptive dynamic range loss function ADRL. First, by introducing the standard deviation and a small constant ϵ, ADRL can automatically adapt to data distributions of different scales, reducing the need for data preprocessing; second, dynamic range sensitivity: by introducing an exponential function and a dynamic variance θ, ADRL can dynamically adjust the sensitivity to errors according to the position of the predicted value, which helps the model to better learn in the case of class imbalance or uneven data distribution. Finally, combining the absolute error and uncertainty: ADRL combines the absolute error and the uncertainty of the model prediction, and in the scenario of uncertain or noisy data, it can obtain more information during model training to ensure a certain stability.

[0015] 2. The present invention conducts hereditary cancer assisted diagnosis by training a hereditary cancer discrimination model, which includes a two-stage module. In the first stage, whether it is a hereditary cancer is discriminated through an input layer, an encoder layer, an attention layer, and a fully connected layer; in the second stage, the type of hereditary cancer and the type of genetic genes are discriminated through an input layer, an encoder layer, an attention layer, a fully connected layer, a feature extraction layer, and a classification layer, which helps doctors use this model to discriminate hereditary cancer and reduces the workload of manual discrimination. Description of the Drawings

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those skilled in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0017] Figure 1Schematic flowchart of a method for hereditary cancer assisted diagnosis based on an adaptive dynamic range loss function provided by an embodiment of the present invention;

[0018] Figure 2 Schematic diagram of a hereditary cancer assisted diagnosis system based on an adaptive dynamic range loss function provided by an embodiment of the present invention;

[0019] Figure 3 Schematic diagram of a hereditary cancer assisted diagnosis device based on an adaptive dynamic range loss function provided by an embodiment of the present invention;

[0020] Figure 4 Schematic flowchart of the structure of a hereditary cancer discrimination model provided by an embodiment of the present invention. Detailed implementation manners

[0021] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0022] In some processes described in the specification and claims of the present invention and the above-mentioned drawings, a plurality of operations appear in a specific order. However, it should be clearly understood that these operations may not be executed in the order in which they appear herein or may be executed in parallel. The serial numbers of the operations, such as S101, S102, etc., are only used to distinguish different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions such as "first" and "second" in this article are used to distinguish different messages, devices, modules, etc., and do not represent a sequence, nor do they limit that "first" and "second" are of different types.

[0023] Figure 1 The schematic diagram of the method for hereditary cancer assisted diagnosis based on an adaptive dynamic range loss function provided by an embodiment of the present invention specifically includes:

[0024] S101: Obtain the image data and genetic information data of the person to be tested;

[0025] In one embodiment, the image data includes one or more of the following: CT, MRI, ultrasound, x-ray image.

[0026] In one embodiment, the genetic information includes but is not limited to the following types: family inheritance pattern, family member disease status, gene mutation information (mutation site, mutation type), gene polymorphism, chromosomal abnormality (number abnormality, structure abnormality), pharmacogenetic information.

[0027] S102: Input the image data and genetic information data into the hereditary cancer discrimination model to obtain the discrimination result of whether it is hereditary cancer; the hereditary cancer discrimination model is pre-trained through image data and genetic information data, and among them, the average value of the absolute error of data at different scales between the predicted value and the true value and the dynamic error adjustment are calculated through the adaptive dynamic range loss function to optimize the discrimination model.

[0028] In one embodiment, the calculation formula of the adaptive dynamic range loss function ADRL(y, ) is expressed as: ADRL(y, ) = *exp(- ) where y is the true label, is the output value predicted by the model, N is the total number of samples, is the absolute error between the predicted value and the true value, is the standard deviation of the true label y, is a very small positive number, represents the mean value of the true label y, is a dynamic adjustment term.

[0029] In one embodiment, the hereditary cancer discrimination model structure includes an input layer, an encoder layer, an attention layer, and a fully connected layer. The image data and genetic information data are input to the input layer, converted into vectors and then fused, and then sequentially pass through the encoder layer, the attention layer, and the fully connected layer to obtain the discrimination result.

[0030] In one embodiment, the encoder layer performs vector encoding through the self-attention mechanism to obtain the self-attention feature vector. The attention layer performs feature processing through multi-head attention. After the self-attention feature vector is input to the attention layer, it is divided into N heads, where N is a natural number greater than 1. Each head independently calculates to obtain N features, and the N features are fused to obtain the attention feature. The attention feature is processed through the fully connected layer to obtain the discrimination result.

[0031] In one embodiment, the training process of the hereditary discrimination model is as follows: Obtain the patient image data set and genetic information data set; Input the image data set and genetic information data set into the hereditary cancer discrimination model to be trained for feature learning, and calculate the loss value through the adaptive dynamic range loss function, and optimize the model based on the loss value; Repeat the above steps until the adaptive dynamic loss function remains unchanged to obtain the hereditary discrimination model.

[0032] In one embodiment, the hereditary discrimination model further includes a hereditary cancer gene-assisted diagnosis module. The feature vector output by the hereditary discrimination model is input into the hereditary cancer gene-assisted diagnosis module for feature analysis to obtain a diagnosis result of the hereditary cancer type. The hereditary cancer gene-assisted diagnosis module includes a feature extraction layer and a classification layer. The feature vector is processed through the feature extraction layer and the classification layer to obtain the diagnosis result.

[0033] In one embodiment, the diagnosis result further includes the type of genetic gene mutation.

[0034] In one embodiment, the feature vector is the feature vector output by the encoder layer or the fully connected layer of the genetic discrimination model.

[0035] In one embodiment, the training process of the hereditary cancer gene-assisted diagnosis module is as follows: Obtain a set of feature vectors; Input the set of feature vectors into the hereditary cancer gene-assisted diagnosis module for feature extraction, classification calculation, and loss calculation; Repeat the above steps until the loss remains unchanged to obtain the hereditary cancer gene-assisted diagnosis module.

[0036] In one embodiment, the loss calculation of the hereditary cancer gene-assisted diagnosis module is performed through one or more of the following loss functions: cross-entropy loss function, logarithmic loss function, hinge loss function, adaptive dynamic range loss function.

[0037] In a specific embodiment, the structure of the hereditary cancer discrimination model is as follows: 1 Data acquisition module: used to collect the medical image data (such as CT, MRI, etc.) and genetic information (such as the detection results of germline gene mutations) of the patient.

[0038] 2 Large model construction module: Based on deep learning technology, a large model for image analysis and genetic information analysis is constructed. This model can accurately identify the characteristics of kidney tumors from medical images and combine genetic information to determine whether the tumor is hereditary renal cancer. The specific structure of the large model module is as follows: 2.1 Encoder layer;

[0039] The sequence X after concatenating the medical image feature vector and the genetic information feature vector is calculated through the self-attention mechanism to obtain the attention weights of each element in the sequence to other elements, and an attention matrix is obtained.

[0040] Att = softmax(QK^T*X / d_k)V

[0041] Among them, Q (Query), K (Key), and V (Value) are the results of linear transformations of the input sequence, and d_k is the dimension of Key. Next, the output of the self-attention mechanism is further processed. The calculation formula is:

[0042] F = max(0, Att *W_1 + b_1)W_2 + b_2

[0043] Among them, W_1 and W_2 are weight matrices, and b_1 and b_2 are bias terms.

[0044] Output: The output of the encoder layer serves as the input to the next layer, or as the final feature representation, or as the input to the diagnostic module.

[0045] 2.3 Attention mechanism layer;

[0046] The output of the encoder layer is processed using the multi-head attention mechanism. The input sequence is split into multiple heads, denoted as head_1,..., head_h. Each head independently performs self-attention calculations, and then the results are concatenated. The calculation result is:

[0047] M = Concat(head_1,..., head_h)W_3

[0048] Among them, head_i is the attention result of the i-th head, and W_3 is the weight matrix.

[0049] The output of the attention mechanism serves as the input to the next layer.

[0050] 2.4 Fully connected layer;

[0051] The output of the attention mechanism layer is subjected to linear transformation and activation function processing. The calculation formula is:

[0052] D = σ(W_4x + b_4)

[0053] Among them, W_4 is the weight matrix, b_4 is the bias term, and σ is the ReLU activation function.

[0054] The output of the fully connected layer serves as the final output, or as the input to the diagnostic analysis module.

[0055] In a specific embodiment, the diagnostic analysis module: The output result of the large model module is input into the diagnostic analysis module for further calculation, and the diagnostic result is output. The diagnostic result includes whether the tumor is hereditary renal cancer and the possible types of genetic gene mutations. The model structure of the diagnostic analysis module is:

[0056] Feature extraction layer: The medical image data and genetic information data are used to extract image features through multi-layer convolution and pooling operations. The calculation formula is:

[0057] T = σ(W_5 * D + b_5)

[0058] Where D is the output of the large model module, W_5 is the weight matrix, b_5 is the bias term, and σ is the ReLU activation function.

[0059] Next, genetic information feature extraction is performed, and the calculation formula is:

[0060] U = σ(W_6T + b_6)

[0061] Where W_6 is the weight matrix, b_6 is the bias term, and σ is the ReLU activation function.

[0062] Classification layer: The feature vector after feature extraction is classified through a fully connected layer and a softmax function.

[0063] The calculation formula of the fully connected layer is:

[0064] P(y|x) = W_7U + b_7

[0065] Where W_7 is the weight matrix, b_7 is the bias term, and P(y|x) is the score corresponding to class y in the output of the fully connected layer. The calculated P(y|x) is the probability distribution of each class. The class with the highest probability is selected as the diagnostic result, including the type of hereditary renal cancer and the type of genetic gene mutation.

[0066] In a specific embodiment, the adaptive dynamic range loss function:

[0067] ADRL(y, ) =

[0068] Where: y is the true label, ŷ is the output value predicted by the model, N is the total number of samples, |y_i - ŷ_i| is the absolute error between the predicted value and the true value, STD(y) is the standard deviation of the true label y, which is used to measure the overall fluctuation range of the data, ϵ is a very small positive number to prevent the denominator from being zero. μ(y) represents the mean of the true label y. θ is a dynamically adjustable term.

[0069] In one embodiment, image data and genetic information of a person to be tested are obtained; the image data and genetic information are input into a hereditary cancer discrimination model to obtain a discrimination result; the structure of the hereditary cancer discrimination model includes an input layer, an encoder layer, an attention layer, a fully connected layer, a feature extraction layer, and a classification layer; the image data and genetic information are converted into an image feature vector and a genetic feature vector through the input layer, and after the image feature vector and the genetic feature vector are concatenated, they are successively subjected to feature processing through the encoder layer, the attention layer, and the fully connected layer to obtain processed features, and the processed features are input into the classification layer through feature extraction by the feature extraction layer for classification to obtain a second discrimination result (the type of hereditary cancer, the type of genetic gene mutation). In addition, a first discrimination result (the discrimination result of whether it is hereditary cancer) is obtained by calculating the processed features obtained through the fully connected layer, and the structural process of the model is as Figure 4 shown.

[0070] In one embodiment, the hereditary cancer discrimination model is a pre-trained model, and an adaptive dynamic range loss function is used to calculate the model loss during training. According to requirements, the structure of the hereditary cancer discrimination model includes any one of the following: 1. Input layer + encoder layer + attention layer + fully connected layer;

[0071] 2. Input layer + encoder layer + attention layer + fully connected layer + feature extraction layer + classification layer;

[0072] 3. Input layer + feature extraction layer + classification layer.

[0073] In one embodiment, when the structure of the hereditary cancer discrimination model is an input layer, a feature extraction layer, and a classification layer, a patient medical image data set and a genetic information data set are obtained;

[0074] The medical image data set and genetic information data are input into the input layer for vector conversion to obtain an image vector set and a genetic vector set;

[0075] The image vector set and the genetic vector set are input into the feature extraction layer for feature extraction to obtain feature vectors;

[0076] Then the feature vectors are input into the classification layer for discrimination to obtain the type of hereditary cancer or the type of genetic gene mutation.

[0077] Or feature vectors output by a hereditary cancer discrimination model with a network structure of input layer + encoder layer + attention layer + fully connected layer are obtained, the feature vectors are input into the feature extraction layer for feature extraction to obtain secondary feature vectors, and then the secondary feature vectors are input into the classification layer for discrimination to obtain the type of hereditary cancer or the type of genetic gene mutation.

[0078] Or simultaneously obtain a medical image dataset, a genetic information dataset, and a feature vector output by a hereditary cancer discrimination model with a network structure of an input layer + an encoder layer + an attention layer + a fully connected layer;

[0079] Input the medical image dataset and genetic information data into the input layer for vector conversion to obtain an image vector set and a genetic vector set;

[0080] Fuse the image vector set, the genetic vector set, and the feature vector to obtain a fused vector;

[0081] Input the fused vector into the feature extraction layer for feature extraction to obtain a three - time feature vector;

[0082] Then input the three - time feature vector into the classification layer for discrimination to obtain the type of hereditary cancer or the type of genetic gene mutation.

[0083] The disclosed embodiment of the present invention also provides a computer program product or system, including a computer program, which when executed by a processor implements the steps of the above - mentioned hereditary cancer auxiliary diagnosis method based on an adaptive dynamic range loss function.

[0084] Figure 2 The schematic diagram of the hereditary cancer auxiliary diagnosis system based on the adaptive dynamic range loss function provided by the embodiment of the present invention specifically includes:

[0085] Acquisition module: Acquire the image data and genetic information data of the person to be tested;

[0086] Discrimination module: Input the image data and genetic information data into the hereditary cancer discrimination model to obtain a discrimination result of whether it is hereditary cancer; the hereditary cancer discrimination model is pre - trained through image data and genetic information data, and among them, the average value of the absolute error of different - scale data between the predicted value and the true value and dynamic error adjustment are calculated through an adaptive dynamic range loss function to optimize the discrimination model.

[0087] In one embodiment, the system further includes a diagnosis module: Input the feature vector output by the hereditary discrimination model into the hereditary cancer gene auxiliary diagnosis module for feature analysis to obtain a diagnosis result of the type of hereditary cancer; the hereditary cancer gene auxiliary diagnosis module includes a feature extraction layer and a classification layer, and the feature vector is processed through the feature extraction layer and the classification layer to obtain a diagnosis result.

[0088] In one embodiment, the diagnosis result further includes the type of genetic gene mutation.

[0089] In one embodiment, the system further includes a user interaction module: used to display the diagnosis result, provide treatment suggestions, and receive additional information input by the user for further analysis.

[0090] Figure 3 Schematic diagram of a hereditary cancer assisted diagnosis device based on an adaptive dynamic range loss function provided by an embodiment of the present invention, specifically including:

[0091] A memory and a processor; the memory is used to store program instructions; the processor is used to call the program instructions, and when the program instructions are executed, any one of the above-mentioned hereditary cancer assisted diagnosis methods based on an adaptive dynamic range loss function.

[0092] An embodiment of the present invention also provides a computer-readable storage medium, the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, any one of the above-mentioned hereditary cancer assisted diagnosis methods based on an adaptive dynamic range loss function.

[0093] The verification results of this verification embodiment show that allocating fixed weights for indications can improve the performance of this method compared to the default settings. Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments, which will not be elaborated herein. In several embodiments provided by this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division, and there can be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces, and the indirect couplings or communication connections of devices or units can be in electrical, mechanical, or other forms. The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment. In addition, in each embodiment of the present invention, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated units can be implemented in the form of hardware or in the form of software functional units. Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing relevant hardware through a program, and the program can be stored in a computer-readable storage medium. The storage medium can include: read-only memory (ROM, Read Only Memory), random access memory (RAM, Random Access Memory), magnetic disk, or optical disc, etc.

[0094] Those of ordinary skill in the art can understand that all or part of the steps in implementing the methods of the above embodiments can be completed by instructing relevant hardware through a program, and the program can be stored in a computer-readable storage medium. The above-mentioned medium storage can be read-only memory, magnetic disk, or optical disc, etc.

[0095] The above has introduced in detail a computer device provided by the present invention. For those of ordinary skill in the art, according to the idea of the embodiments of the present invention, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. An auxiliary diagnosis method for hereditary cancer based on an adaptive dynamic range loss function, characterized in that, Including: Obtaining the image data and genetic information data of the person to be tested; Inputting the image data and genetic information data into a hereditary cancer discrimination model to obtain a discrimination result on whether it is hereditary cancer; the hereditary cancer discrimination model is pre-trained through image data and genetic information data, wherein the average value of the absolute error of data at different scales between the predicted value and the true value and dynamic error adjustment are calculated through an adaptive dynamic range loss function to optimize the discrimination model.

2. The hereditary cancer assisted diagnosis method based on the adaptive dynamic range loss function according to claim 1, wherein The calculation formula of the adaptive dynamic range loss function ADRL(y, ) is expressed as: ADRL(y, ) = *exp(- ) where y is the true label, is the output value predicted by the model, N is the total number of samples, is the absolute error between the predicted value and the true value, is the standard deviation of the true label y, is a very small positive number, represents the mean of the true label y, is a dynamic adjustment term.

3. The hereditary cancer assisted diagnosis method based on the adaptive dynamic range loss function according to claim 1, wherein The structure of the hereditary cancer discrimination model includes an input layer, an encoder layer, an attention layer, and a fully connected layer. The image data and genetic information data are input into the input layer, converted into vectors and then fused, and then successively pass through the encoder layer, the attention layer, and the fully connected layer to obtain the discrimination result; Optionally, the encoder layer performs vector encoding through a self-attention mechanism to obtain a self-attention feature vector. The attention layer performs feature processing through multi-head attention. After the self-attention feature vector is input into the attention layer, it is divided into N heads, where N is a natural number greater than 1. Each head independently calculates to obtain N features, and the N features are fused to obtain an attention feature. The attention feature is processed through a fully connected layer to obtain the discrimination result; Optionally, the training process of the hereditary cancer discrimination model is as follows: Obtaining a patient image data set and a genetic information data set; Inputting the image data set and the genetic information data set into the hereditary cancer discrimination model to be trained for feature learning and calculating the loss value through an adaptive dynamic range loss function, and optimizing the model based on the loss value; Repeating the above steps until the adaptive dynamic loss function remains unchanged to obtain a hereditary discrimination model.

4. The hereditary cancer assisted diagnosis method based on the adaptive dynamic range loss function according to claim 1, wherein The hereditary cancer discrimination model further includes a hereditary cancer gene assisted diagnosis module. The feature vector output by the hereditary discrimination model is input into the hereditary cancer gene assisted diagnosis module for feature analysis to obtain a diagnosis result of the hereditary cancer type; the hereditary cancer gene assisted diagnosis module includes a feature extraction layer and a classification layer, and the feature vector is processed through the feature extraction layer and the classification layer to obtain the diagnosis result; Optionally, the diagnosis result further includes the genetic mutation type.

5. The hereditary cancer assisted diagnosis method based on the adaptive dynamic range loss function according to claim 4, characterized in that, The feature vector is the feature vector output by the encoder layer or the fully connected layer of the genetic discrimination model.

6. The hereditary cancer assisted diagnosis method based on the adaptive dynamic range loss function according to claim 4, wherein The training process of the hereditary cancer gene assisted diagnosis module is as follows: Obtaining a feature vector set; Inputting the feature vector set into the hereditary cancer gene assisted diagnosis module for feature extraction, classification calculation, and loss calculation; Repeating the above steps until the loss remains unchanged to obtain the hereditary cancer gene assisted diagnosis module.

7. The method for assisting in the diagnosis of hereditary cancer based on an adaptive dynamic range loss function according to claim 6, wherein The loss calculation of the hereditary cancer gene assisted diagnosis module is through one or more of the following loss functions: cross-entropy loss function, logarithmic loss function, hinge loss function, adaptive dynamic range loss function.

8. A computer program product, which includes a computer program or instructions thereon, characterized in that, The computer program or instruction is executed by a processor to implement the hereditary cancer assisted diagnosis method based on the adaptive dynamic range loss function according to any one of claims 1-7.

9. A computer device, comprising a memory, a processor, and a computer program or instruction stored on the memory, characterized in that, The computer program or instruction, when executed by a processor, implements the method for assisting in the diagnosis of hereditary cancer based on an adaptive dynamic range loss function as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, The computer program or instruction, when executed by a processor, implements the method for assisting in the diagnosis of hereditary cancer based on an adaptive dynamic range loss function as described in any one of claims 1-7.

Images