Nail-based tumor risk prediction method, device and system, computer equipment and computer readable storage medium

Through the nail-based tumor risk prediction method, Raman detection and scanning electron microscopy technology, the problems of invasiveness, long time and poor sensitivity of existing serum detection are solved, and non-invasive, fast and accurate tumor risk detection is achieved, which is suitable for Large-scale screening and early diagnosis.

CN118983083APending Publication Date: 2024-11-19SHANGHAI TENTH PEOPLES HOSPITAL
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410897352.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-05
Publication Date
2024-11-19

AI Technical Summary

Technical Problem

Existing tumor detection methods such as serum detection have problems such as invasiveness, long detection time, poor sensitivity, inability to reflect tumor burden accumulation and inconvenient large-scale screening, especially in terms of early diagnosis.

Method used

Using a nail-based tumor risk prediction method, chemical information data is obtained through Raman detection and scanning electron microscopy is obtained, pre-trained tumor risk prediction model is input, and multi-layer perceptron is used to obtain tumor risk prediction results. Including tumor classification, accuracy and specificity.

Benefits of technology

It realizes non-invasive, fast and simple tumor risk detection, can provide preliminary judgments within a few minutes, improves the accuracy and coverage of the detection, is suitable for daily health monitoring and large-scale screening, and is especially suitable for promotion in areas with limited resources.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118983083A_ABST
    Figure CN118983083A_ABST
Patent Text Reader

Abstract

The invention relates to a nail-based tumor risk prediction method, device and system, computer equipment and a computer readable storage medium. The nail-based tumor risk prediction method comprises the following steps: acquiring chemical information data and structural image data of a nail coronal plane; and inputting the chemical information data and the structural image data into a pre-trained tumor risk prediction model to obtain a tumor risk prediction result. The method has the advantages that operation is simple and fast; real noninvasive collection is realized; rapid diagnosis and real-time feedback are realized; the accuracy and the reliability are high; periodic changes of metabolic diseases are reflected; popularization and application are easy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of detection and analysis, and in particular to a method, device, system, computer equipment and computer-readable storage medium for predicting tumor risk based on nails. Background Art

[0002] Currently, almost all tumor detection is done through body fluids and imaging. Among them, imaging changes are in a relatively late stage, which means that the tumor entity has been formed. Early tumor diagnostic markers all rely on body fluid diagnosis, and the most commonly used are serum or plasma. For example, markers such as alpha-fetoprotein (AFP) and carbohydrate antigen (CA) are generally significantly elevated in solid tumors. For example, carcinoembryonic antigen (CEA) is a broad-spectrum tumor marker, and its serum concentration is associated with a variety of tumors, especially digestive tract tumors. These are all tumor markers in plasma. However, there are some significant limitations based on body fluid detection: first, only the increase of markers can be seen, but it is impossible to determine its source organ; second, it cannot reflect the accumulation of tumor load in the body; the gold standard for tumor diagnosis is tissue biopsy, but this cannot be performed before the tumor is found.

[0003] Let's take liver cancer (HCC) as an example. At present, elevated serum AFP is an important indicator for the clinical diagnosis of primary liver cancer. In the absence of liver disease activity, and the exclusion of pregnancy and gonadal embryonal carcinoma, ≥400ng / mL for 4 weeks or ≥200ng / mL for 8 weeks, combined with imaging examinations, it can be diagnosed as primary liver cancer. AFP will also increase in patients with cirrhosis and hepatitis, but generally does not exceed 300ng / ml. It is recommended that those with abnormalities undergo dynamic measurements every 1 to 2 months to avoid false negative and false positive results.

[0004] The existing cancer detection technology has the following defects:

[0005] 1. Invasive: Currently, all plasma-based tests must be based on preparation. Drawing blood is traumatic and usually requires fasting, which will cause discomfort to patients due to hunger.

[0006] 2. Insensitivity of diagnostic criteria: AFP detection only reflects the expression level at that time and cannot reflect the damage accumulation of HCC in the body. ≥400ng / mL for 4 weeks means that multiple preparations and tests are required within four weeks for confirmation. If the level is lower than this level once or twice, it cannot be confirmed.

[0007] 3. Long testing time: Different serum marker tests take 15 minutes to 12 hours or even longer. Biopsy surgery and immunohistochemistry of postoperative pathological specimens also take time.

[0008] 4. Unable to reflect the tumor load in the body: Because the plasma metabolites change very quickly, it cannot reflect the changes in metabolic diseases at different stages, and cannot reflect the accumulation of tumors in the body;

[0009] 5. Inconvenience: For elderly groups such as nursing homes or communities, or those with limited mobility, going to the hospital for blood drawing is quite troublesome and time-consuming. In addition, the immunity of this group is relatively low, and going to crowded places in the hospital is prone to secondary infection.

[0010] Currently, no effective solutions have been proposed to address the problems existing in related technologies, such as the need for blood testing, the inability to reflect the accumulation of tumor load in the body, the long detection time, the poor sensitivity of the test results, and the inability to use biopsy diagnosis before the tumor is discovered. Summary of the invention

[0011] The purpose of the present invention is to provide a nail-based tumor risk prediction method, device, system, computer equipment and computer-readable storage medium to address the deficiencies in the prior art, such as the need for blood testing, inability to reflect the accumulation of tumor load in the body, long detection time, poor sensitivity of detection results, and inability to use biopsy diagnosis before the tumor is discovered.

[0012] To achieve the above object, the technical solution adopted by the present invention is:

[0013] In a first aspect, the present invention provides a method for predicting tumor risk based on nails, comprising:

[0014] Acquiring chemical information data and structural image data of the nail coronal surface, wherein the chemical information data is obtained by Raman detection, and the structural image data is obtained by scanning electron microscopy;

[0015] The chemical information data and the structural image data are input into a pre-trained tumor risk prediction model to obtain a tumor risk prediction result, wherein the tumor risk prediction result includes tumor classification, accuracy, sensitivity, and specificity.

[0016] In some embodiments, inputting the chemical information data and the structural image data into a pre-trained tumor risk prediction model to obtain a tumor risk prediction result includes:

[0017] Processing the chemical information data and the structural image data to obtain a 64-dimensional initial feature vector;

[0018] Inputting the initial feature vector into the pre-trained tumor risk prediction model to obtain a final feature vector of 2048 dimensions;

[0019] The final feature vector is input into a multi-layer perceptron for processing to obtain the tumor risk prediction result.

[0020] In some embodiments, the training method of the tumor risk prediction model includes:

[0021] Constructing a tumor risk prediction model, wherein the tumor risk prediction model includes a plurality of residual blocks;

[0022] Constructing a training set and a test set according to the chemical information data and the structural image data, wherein the ratio of the training set to the test set is 3:1;

[0023] Inputting the training set into a tumor risk prediction model to obtain an output result;

[0024] Inputting the output result into a multilayer perceptron for processing to obtain a tumor risk prediction result;

[0025] The tumor risk prediction results are tested to detect the effectiveness of the training, and based on the effectiveness test results, the tumor risk prediction model is iterated until the training is completed.

[0026] In some of the embodiments, the method for training a tumor risk prediction model further includes:

[0027] Calculating the loss value of the tumor risk prediction model;

[0028] The gradient descent method and back propagation algorithm are used to optimize the network parameters and network weights of the tumor risk prediction model.

[0029] In a second aspect, the present invention provides a nail-based tumor risk prediction device, comprising:

[0030] A chemical information acquisition unit, used to acquire chemical information data of the nail coronal surface, wherein the chemical information data is obtained by Raman detection;

[0031] A structural image acquisition unit, used to acquire structural image data of the nail coronal surface, wherein the structural image data is obtained by a scanning electron microscope;

[0032] The risk prediction unit is used to input the chemical information data and the structural image data into a pre-trained tumor risk prediction model to obtain a tumor risk prediction result, wherein the tumor risk prediction result includes tumor classification, accuracy, sensitivity, and specificity.

[0033] In some embodiments, the risk prediction unit includes:

[0034] A processing element, used for processing the chemical information data and the structural image data to obtain a 64-dimensional initial feature vector;

[0035] A first output element, used to input the initial feature vector into the pre-trained tumor risk prediction model to obtain a final feature vector of 2048 dimensions;

[0036] The second output element inputs the final feature vector into a multi-layer perceptron for processing to obtain the tumor risk prediction result.

[0037] In some embodiments, the risk prediction unit includes:

[0038] A construction element, used to construct a tumor risk prediction model, wherein the tumor risk prediction model includes a plurality of residual blocks;

[0039] A data set component, used to construct a training set and a test set according to the chemical information data and the structural image data, wherein the ratio of the training set to the test set is 3:1;

[0040] A first training component, used for inputting the training set into a tumor risk prediction model to obtain an output result;

[0041] A second training element is used to input the output result into a multi-layer perceptron for processing to obtain a tumor risk prediction result;

[0042] The testing element is used to test the tumor risk prediction result to detect the effectiveness of the training, and iterate the tumor risk prediction model according to the effectiveness test result until the training is completed.

[0043] In some of the embodiments, the risk prediction unit further includes:

[0044] A loss calculation element, used to calculate the loss value of the tumor risk prediction model;

[0045] The optimization element is used to optimize the network parameters and network weights of the tumor risk prediction model using a gradient descent method and a back propagation algorithm.

[0046] In a third aspect, a nail-based tumor risk prediction system is provided, comprising:

[0047] A nail processing device for obtaining and processing a nail to obtain a coronal flat-sectioned nail;

[0048] A Raman detection device is used to perform Raman detection on the nail whose coronal surface is cut flat, so as to obtain chemical information data of the nail coronal surface;

[0049] An electron microscope scanning device is used to perform electron microscope scanning on the nail whose coronal surface is cut flat, so as to obtain structural image data of the nail coronal surface;

[0050] The tumor risk prediction device as described in the second aspect is used to obtain chemical information data and structural image data of the nail coronal surface, and input the chemical information data and the structural image data into a pre-trained tumor risk prediction model to obtain a tumor risk prediction result, wherein the tumor risk prediction result includes tumor classification, accuracy, sensitivity, and specificity.

[0051] In a fourth aspect, the present invention provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the tumor risk prediction method as described in the first aspect is implemented.

[0052] In a fifth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the tumor risk prediction method as described in the first aspect.

[0053] The present invention adopts the above technical solution, and has the following technical effects compared with the prior art:

[0054] 1) Easy and quick operation: The nail collection process is simple and does not require complicated preliminary preparation or special environmental conditions. It can be performed in a variety of environments (such as home and community). This convenience makes the test widely used in daily health monitoring and large-scale screening, greatly improving the efficiency and coverage of health management;

[0055] 2) Truly non-invasive collection: The detection method greatly reduces the discomfort and anxiety of patients during the detection process, and is especially suitable for special groups who are sensitive to pain or have difficulty cooperating with traditional sampling;

[0056] 4) Rapid diagnosis and real-time feedback: Combining artificial intelligence technology, especially deep learning algorithms, it can complete the preliminary diagnosis of the disease within a few minutes after obtaining nail data (chemical information data of nails, structural images obtained by scanning electron microscope), and provide patients with immediate health status feedback. This feature is particularly important in clinical environments that require rapid decision-making, which helps to provide early treatment, improve disease prognosis, and scientifically manage the disease process;

[0057] 5) High accuracy and reliability: Raman spectroscopy and scanning electron microscopy are used to obtain information on changes in the chemical composition and structure of nails, as well as clinical information used in previous modeling. Therefore, it is more comprehensive and systematic than single plasma metabolite diagnosis. Through the continuous optimization of big data and machine learning models, the diagnostic accuracy of the system can continue to improve over time and with the accumulation of data;

[0058] 6) Reflecting the stage changes of tumors: Currently, tissue biopsy can be used as the gold standard for diagnosis, but it cannot be obtained before the tumor surgery is clearly diagnosed. Secondly, imaging can reflect the pathological conditions of the body, but it is already in the late stage when imaging is discovered. Conventional serum / plasma tests cannot reflect the accumulation of operations in the body. However, changes in nails can reflect the accumulation of damage to the body. Generally speaking, nails can reflect the overall metabolic status of the past 3 months;

[0059] 7) Easy to promote and apply: Due to its convenience and economy, it is particularly suitable for promotion and use in resource-limited areas and developing countries. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 is a flow chart of a tumor risk prediction method according to an embodiment of the present invention (I);

[0061] Figure 2 is a flow chart of a tumor risk prediction method according to an embodiment of the present invention (II);

[0062] Figure 3 is a flowchart of a tumor risk prediction method according to an embodiment of the present invention (III);

[0063] Figure 4 is a framework diagram of a tumor risk prediction device according to an embodiment of the present invention;

[0064] Figure 5 is a framework diagram of a tumor risk prediction system according to an embodiment of the present invention;

[0065] Figure 6 This is the Raman spectrum data of 10 healthy control samples;

[0066] Figure 7 This is the Raman spectrum data of 7 liver cancer patient samples;

[0067] Figure 8 This is the Raman spectrum data of 8 lung cancer patient samples;

[0068] Fig. 9 This is the Raman spectrum data of 7 gastric cancer patient samples;

[0069] Fig.10 This is the Raman spectrum data of 8 thyroid cancer patient samples;

[0070] Fig.11 It is a comparison chart of the numerical average data of healthy control samples, samples of patients with high uric acid, samples of patients with osteoporosis, samples of patients with stones, and samples of patients with diabetes;

[0071] Fig.12 is the scanning electron micrograph of 6 healthy control samples;

[0072] Fig.13 These are scanning electron micrographs of samples from six patients with liver cancer;

[0073] Fig.14 is a scanning electron micrograph of samples from six lung cancer patients;

[0074] Fig.15 is a scanning electron micrograph of samples from six gastric cancer patients;

[0075] Fig.16 These are scanning electron micrographs of samples from six patients with thyroid cancer;

[0076] Fig.17 It is a schematic diagram of a specific implementation of a tumor risk prediction method according to an embodiment of the present invention.

[0077] The accompanying drawings are marked as follows: 410, chemical information acquisition unit; 420, structural image acquisition unit; 430, risk prediction unit; 510, nail processing device; 520, Raman detection device; 530, electron microscope scanning device; 540, tumor risk prediction device. DETAILED DESCRIPTION

[0078] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is described and illustrated below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. Based on the embodiments provided in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application.

[0079] Obviously, the drawings described below are only some examples or embodiments of the present application. For ordinary technicians in this field, the present application can also be applied to other similar scenarios based on these drawings without creative work. In addition, it can also be understood that although the efforts made in this development process may be complicated and lengthy, for ordinary technicians in this field related to the content disclosed in this application, some changes in design, manufacturing or production based on the technical content disclosed in this application are just conventional technical means, and should not be understood as insufficient content disclosed in this application.

[0080] Reference to "embodiments" in this application means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those of ordinary skill in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0081] Unless otherwise defined, the technical terms or scientific terms involved in this application should be understood by people with ordinary skills in the technical field to which this application belongs. The words "one", "a", "a", "the" and the like involved in this application do not indicate a quantitative limitation, and may represent the singular or plural. The terms "include", "comprise", "have" and any of their variations involved in this application are intended to cover non-exclusive inclusions; for example, a process, method, system, product or device that includes a series of steps or units (units) is not limited to the listed steps or units, but may also include steps or units that are not listed, or may also include other steps or units inherent to these processes, methods, products or devices. The words "connect", "connected", "coupled" and the like involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The "multiple" involved in this application refers to two or more. "And / or" describes the association relationship of associated objects, indicating that there can be three relationships, for example, "A and / or B" can mean: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects before and after are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific ordering of the objects.

[0082] Example 1

[0083] This embodiment relates to the tumor risk prediction method of the present invention.

[0084] Figure 1 FIG. 1 is a flow chart of a tumor risk prediction method according to an embodiment of the present invention (I). Figure 1 As shown, a method for predicting tumor risk based on nails comprises:

[0085] Step S102, obtaining chemical information data and structural image data of the nail coronal surface, wherein the chemical information data is obtained by Raman detection, and the structural image data is obtained by scanning electron microscopy;

[0086] Step S104: input the chemical information data and the structural image data into a pre-trained tumor risk prediction model to obtain a tumor risk prediction result, wherein the tumor risk prediction result includes tumor classification, accuracy, sensitivity, and specificity.

[0087] In the present invention, tumors include but are not limited to prostate cancer, lung cancer, stomach cancer, liver cancer, colon cancer, etc.

[0088] In the present invention, the tumor risk prediction results are only used to indicate the risk of having a certain tumor, and are not used to directly diagnose whether a tumor exists, which requires other direct diagnostic methods to confirm the diagnosis. The tumor risk prediction results of the present invention can be used to assist doctors in disease risk screening before obtaining the results of examinations and tests (such as blood tests, biopsies, etc.).

[0089] In step S102, the cross section of the nail coronal surface is rectangular. Generally, the length of the nail coronal surface is 400-800 um and the width is 100 um.

[0090] In step S102, the Raman detection parameters of the chemical information data are:

[0091] Excitation wavelength: 532nm; power: 7.5-10mw; grating: 1200g / mm; single spectrum integration time: 5-8s; objective lens: 100x / 0.9.

[0092] Number of sample collection points: When modeling, high-quality Raman peak spectra were collected for each sample. The disease group had 360 spectra, and the normal control group had good uniformity, so 240 spectra were collected.

[0093] In sample prediction, it is sufficient to sample 15-20 points from the coronal surface of the nail from top to bottom to reflect the overall information of the nail. The acquisition time of 15-20 points is 5-10 minutes for the chemical information data of the nail and 15 minutes for the SEM sampling.

[0094] In step S102, when performing Raman detection on the nail coronal plane, the points of row A×column B on the nail coronal plane are detected. Generally, A≥5, B≥5.

[0095] Preferably, A = 10, B = 36. Generally, 36 points are collected in the length direction of the nail coronal plane and 10 points are collected in the width direction, that is, Raman detection is performed on 360 points on the nail coronal plane.

[0096] In the present invention, by performing Raman detection on the nail coronal plane, the metabolic status of the patient within a preset time period can be obtained. For example, the metabolic status of the patient for 3 months can be obtained.

[0097] In step S102, the chemical information data is output in the form of an image.

[0098] In step S102, Raman detection includes but is not limited to resonance Raman spectroscopy (RRS), surface-enhanced Raman spectroscopy (SERS), tip-enhanced Raman spectroscopy (TERS), coherent anti-Stokes Raman spectroscopy (CARS), and stimulated Raman scattering (SRS).

[0099] In step S102, taking liver cancer as an example, liver cancer is compared with a normal control group, and the chemical information data includes at least the following characteristic peaks:

[0100]

[0101] In step S102, electron microscope detection includes but is not limited to scanning electron microscope (Scanning Electron Microscope, SEM) and transmission electron microscope (Transmission Electron Microscope, TEM).

[0102] In step S104, in the tumor risk prediction model, the weight of the chemical information data is 50%, and the weight of the structural image data is 50%.

[0103] In step S104, the tumor risk prediction result may include one tumor classification or several tumor classifications.

[0104] Specifically, the tumor risk prediction results can be output in the following format (taking liver cancer as an example):

[0105] (1) Hepatocellular carcinoma;

[0106] (2) Hepatocellular carcinoma (in situ);

[0107] (3) hepatocellular carcinoma (metastasis);

[0108] (4) Hepatocellular carcinoma + high sensitivity + high specificity; (wherein, “+” is only used to indicate the interval and has no other function. In addition, the interval can also be expressed in other ways, such as “,”, “;”, etc.).

[0109] In step S104, the time for obtaining the tumor risk prediction result is 0 to 5 minutes. Preferably, the time for obtaining the tumor risk prediction result is 0 to 3 minutes. More preferably, the time for obtaining the tumor risk prediction result is 0 to 2 minutes. More preferably, the time for obtaining the tumor risk prediction result is 0.5 to 2 minutes. More preferably, the time for obtaining the tumor risk prediction result is 0.5 to 1.5 minutes.

[0110] Furthermore, before step S102, the method further includes:

[0111] A nail slice was collected along the coronal surface of the patient's nail;

[0112] Smooth the two crown surfaces of the nail;

[0113] The free edge of the smoothed nail plate is cut to expose the nail coronal surface structure, which can be placed on a glass slide and fixed with adhesive so that Raman spectroscopy and scanning electron microscopy can be performed later.

[0114] Furthermore, it also includes:

[0115] Obtain clinical information from patients;

[0116] Chemical information data, structural image data, and clinical information are input into a pre-trained tumor risk prediction model to obtain tumor risk prediction results, where the tumor risk prediction results include tumor classification, accuracy, sensitivity, and specificity.

[0117] In some embodiments, the clinical information includes but is not limited to gender, age, disease diagnosis type, and tumor markers.

[0118] In some embodiments, tumor markers include but are not limited to AFP (alpha-fetoprotein), CA199 (carbohydrate antigen 199), CA125 (carbohydrate antigen 125), CEA (carcinoembryonic antigen, DCP (abnormal prothrombin), HBsAg (hepatitis B virus surface antigen), e antibody, core antibody, etc.

[0119] Through steps S102 to S104, using Raman detection and electron microscopy detection, patients can be screened non-invasively. The operation is simple and fast, and real-time feedback can be obtained (after obtaining chemical information data and structural image data, as long as the above two aspects of information are uploaded, the results can generally be obtained within 3-5 minutes). In addition, the tumor risk prediction results can also provide feedback on stage changes. The sensitivity, specificity, and accuracy of the tumor risk prediction results are high, and compared with the traditional single plasma metabolite diagnosis, the prediction results are more comprehensive.

[0120] Figure 2FIG. 2 is a flow chart of a tumor risk prediction method according to an embodiment of the present invention. Figure 2 As shown, step S104 includes:

[0121] Step S202, processing the chemical information data and the structural image data to obtain a 64-dimensional initial feature vector;

[0122] Step S204, inputting the initial feature vector into a pre-trained tumor risk prediction model to obtain a final feature vector of 2048 dimensions;

[0123] Step S206: input the final feature vector into a multi-layer perceptron for processing to obtain a tumor risk prediction result.

[0124] In step S206, the output length of the tumor risk prediction result can be adjusted according to the number of tumor classifications to be detected (accuracy of tumor classification). For example, if the number of tumor classifications to be detected (accuracy of tumor classification) is 3, the output length of the tumor risk prediction result is 3.

[0125] Figure 3 FIG. 3 is a flow chart of a tumor risk prediction method according to an embodiment of the present invention. Figure 3 As shown, the training method of the tumor risk prediction model includes:

[0126] Step S302: constructing a tumor risk prediction model, wherein the tumor risk prediction model includes a plurality of residual blocks;

[0127] Step S304, constructing a training set and a test set according to the chemical information data and the structural image data, wherein the ratio of the training set to the test set is 3:1;

[0128] Step S306, inputting the training set into the tumor risk prediction model to obtain an output result;

[0129] Step S308, inputting the output result into a multi-layer perceptron for processing to obtain a tumor risk prediction result;

[0130] Step S310: Testing the tumor risk prediction results to detect the effectiveness of the training, and iterating the tumor risk prediction model according to the effectiveness test results until the training is completed.

[0131] In step S302, the tumor risk prediction model is a residual-based convolutional neural network model. By introducing the residual learning framework, the gradient vanishing or gradient exploding problems that may occur in the traditional deep network during training can be solved.

[0132] In step S302, the number of residual blocks is at least 3. Preferably, the number of residual blocks is 3 to 6. More preferably, the number of residual blocks is 3 to 5.

[0133] Generally, the dimension of the vector feature output by the next residual block is greater than the dimension of the vector feature output by the previous residual block. Preferably, the ratio of the dimension of the vector feature output by the next residual block to the dimension of the vector feature output by the previous residual block is 2.

[0134] Furthermore, the training method of the tumor risk prediction model also includes:

[0135] Step S312, calculating the loss value of the tumor risk prediction model;

[0136] Step S314: Use the gradient descent method and back propagation algorithm to optimize the network parameters and network weights of the tumor risk prediction model.

[0137] Through steps S302 to S314, a tumor risk prediction model with an accuracy greater than 95%, a sensitivity greater than 93%, and a specificity greater than 97% can be trained, which greatly improves the accuracy, sensitivity, and specificity while reducing the prediction time.

[0138] In addition, the tumor risk prediction method of the embodiment of the present application can be implemented by a computer device. The components of the computer device may include but are not limited to a processor and a memory storing computer program instructions.

[0139] In some embodiments, the processor may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0140] In some embodiments, the memory may include a large capacity memory for data or instructions. By way of example and not limitation, the memory may include a hard disk drive (HDD), a floppy disk drive, a solid state drive (SSD), a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory may include a removable or non-removable (or fixed) medium. Where appropriate, the memory may be inside or outside a data processing device. In a specific embodiment, the memory is a non-volatile memory. In a specific embodiment, the memory includes a read-only memory (ROM) and a random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM (Programmable Read-Only Memory, PROM for short), an erasable PROM (Erasable ProgrammableRead-Only Memory, EPROM for short), an electrically erasable PROM (Electrically Erasable ProgrammableRead-Only Memory, EEPROM for short), an electrically alterable ROM (Electrically Alterable Read-Only Memory, EAROM for short) or a flash memory (FLASH) or a combination of two or more of these. Under appropriate circumstances, the RAM can be a static random access memory (SRAM) or a dynamic random access memory (DRAM), wherein the DRAM can be a fast page mode dynamic random access memory (FPMDRAM), an extended data output dynamic random access memory (EDODRAM), a synchronous dynamic random access memory (SDRAM), etc.

[0141] The memory may be used to store or cache various data files required for processing and / or communication, as well as possible computer program instructions executed by the processor.

[0142] The processor implements any one of the tumor risk prediction methods in the above embodiments by reading and executing computer program instructions stored in the memory.

[0143] In some embodiments, the computer device may further include a communication interface and a bus, wherein the processor, the memory, and the communication interface are connected via the bus and communicate with each other.

[0144] The communication interface is used to realize the communication between each unit, device, unit and / or device in the embodiment of the present application. The communication interface can also realize data communication with other components such as: external devices, image / data acquisition equipment, databases, external storage, and image / data processing workstations.

[0145] The bus includes hardware, software or both, and couples the components of the computer device to each other. The bus includes but is not limited to at least one of the following: data bus, address bus, control bus, expansion bus, and local bus. By way of example and not limitation, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses or a combination of two or more of these. Where appropriate, the bus may include one or more buses. Although embodiments of the present application describe and illustrate a particular bus, the present application contemplates any suitable bus or interconnect.

[0146] The computer device can execute the tumor risk prediction method in the embodiment of the present application.

[0147] In addition, in combination with the tumor risk prediction method in the above embodiment, the embodiment of the present application can provide a computer-readable storage medium for implementation. The computer-readable storage medium stores computer program instructions; when the computer program instructions are executed by the processor, any one of the tumor risk prediction methods in the above embodiment is implemented.

[0148] The technical effects of the present invention are as follows:

[0149] 1) Early prediction of tumor occurrence: Through the structural changes and chemical composition of nails, the accuracy, sensitivity and specificity of tumor occurrence can be detected with a high probability;

[0150] 2) Truly non-invasive testing: This testing method greatly reduces the discomfort and anxiety of patients during the testing process, and is especially suitable for special groups who are sensitive to pain or have difficulty cooperating with traditional sampling, such as children and the elderly:

[0151] 3) Easy and quick operation: The collection process of nails is simple, without complicated preliminary preparation or special environmental conditions, and can be performed in a variety of environments (such as home and community). This convenience makes the test widely used in daily health monitoring and large-scale screening, greatly improving the efficiency and coverage of health management;

[0152] 4) Rapid diagnosis and real-time feedback: Combining artificial intelligence technology, especially deep learning algorithms, it can realize rapid processing and analysis of nail data (chemical information data of nails, structural images obtained by scanning electron microscope), and can complete the preliminary diagnosis of the disease within a few minutes, providing patients with immediate health status feedback. This feature is particularly important in clinical environments that require rapid decision-making, which helps to intervene and treat early and improve disease prognosis;

[0153] 5) High accuracy and reliability: Using advanced image analysis technology, it can accurately identify and classify changes in microstructure and chemical composition in nail images, which are closely related to metabolic diseases. Through the continuous optimization of big data and machine learning models, the diagnostic accuracy of the system will continue to improve over time and with the accumulation of data, so it is more comprehensive and systematic than the diagnosis of a single plasma metabolite;

[0154] 6) Promotion and application: Due to its convenience and economy, it is particularly suitable for promotion and use in resource-limited areas and developing countries.

[0155] Example 2

[0156] This embodiment relates to the tumor risk prediction device of the present invention.

[0157] Figure 4 is a framework diagram of a tumor risk prediction device according to an embodiment of the present invention. Figure 4As shown, a nail-based tumor risk prediction device includes a chemical information acquisition unit 410, a structural image acquisition unit 420, and a risk prediction unit 430. The chemical information acquisition unit 410 is used to acquire chemical information data of the nail coronal surface, wherein the chemical information data is obtained by Raman detection; the structural image acquisition unit 420 is used to acquire structural image data of the nail coronal surface, wherein the structural image data is obtained by scanning electron microscopy; the risk prediction unit 430 is used to input the chemical information data and the structural image data into a pre-trained tumor risk prediction model to obtain a tumor risk prediction result, wherein the tumor risk prediction result includes tumor classification, accuracy, sensitivity, and specificity.

[0158] In some embodiments, the risk prediction unit 430 includes a processing element, a first output element, and a second output element. The processing element is used to process the chemical information data and the structural image data to obtain a 64-dimensional initial feature vector; the first output element is used to input the initial feature vector into a pre-trained tumor risk prediction model to obtain a 2048-dimensional final feature vector; the second output element inputs the final feature vector into a multilayer perceptron for processing to obtain a tumor risk prediction result.

[0159] In some embodiments, the risk prediction unit 430 includes a construction element, a data set element, a first training element, a second training element, and a test element. The construction element is used to construct a tumor risk prediction model, wherein the tumor risk prediction model includes a plurality of residual blocks; the data set element is used to construct a training set and a test set according to chemical information data and structural image data, wherein the ratio of the training set to the test set is 3:1; the first training element is used to input the training set into the tumor risk prediction model to obtain an output result; the second training element is used to input the output result into a multilayer perceptron for processing to obtain a tumor risk prediction result; the test element is used to test the tumor risk prediction result to detect the effectiveness of the training, and according to the effectiveness test result, the tumor risk prediction model is iterated until the training is completed.

[0160] In some embodiments, the risk prediction unit 430 further includes a loss calculation element and an optimization element. The loss calculation element is used to calculate the loss value of the tumor risk prediction model; the optimization element is used to optimize the network parameters and network weights of the tumor risk prediction model using a gradient descent method and a back propagation algorithm.

[0161] In some embodiments, the tumor risk prediction device further includes a clinical information acquisition unit, wherein the clinical information acquisition unit is used to acquire clinical information of the patient; the risk prediction unit 430 is also used to input the clinical information into a pre-trained tumor risk prediction model to obtain a tumor risk prediction result.

[0162] The technical effects of this embodiment are basically the same as those of Embodiment 1 and will not be described in detail here.

[0163] Example 3

[0164] This embodiment relates to the tumor risk prediction system of the present invention.

[0165] Figure 5 is a framework diagram of a tumor risk prediction system according to an embodiment of the present invention. Figure 5 As shown, a nail-based tumor risk prediction system includes a nail processing device 510, a Raman detection device 520, an electron microscope scanning device 530, and a tumor risk prediction device 540 as described in Example 2. The nail processing device 510 is used to obtain and process the nail to obtain a nail with a coronal plane cut flat; the Raman detection device 520 is used to perform Raman detection on the nail with a coronal plane cut flat to obtain chemical information data of the nail coronal plane; the electron microscope scanning device 530 is used to perform electron microscope scanning on the nail with a coronal plane cut flat to obtain structural image data of the nail coronal plane; the tumor risk prediction device 540 is used to obtain chemical information data and structural image data of the nail coronal plane, and input the chemical information data and structural image data into a pre-trained tumor risk prediction model to obtain a tumor risk prediction result, wherein the tumor risk prediction result includes tumor classification, accuracy, sensitivity, and specificity.

[0166] In some embodiments, the nail processing device 510 includes but is not limited to a nail collection unit, a nail fixing unit, a nail smoothing unit, and a nail cutting unit. The nail collection unit is used to obtain a nail from the patient's finger; the nail fixing unit is used to fix the nail; the nail smoothing unit is used to smooth the two surfaces of the nail; and the nail cutting unit is used to cut the free edge of the nail to expose the nail coronal surface structure.

[0167] Generally, the nail collection unit includes but is not limited to nail clippers.

[0168] Generally, the nail fixing unit includes, but is not limited to, a clamp.

[0169] Generally, the nail smoothing unit includes, but is not limited to, a grinder, such as a nail file.

[0170] Generally, the nail cutting unit includes, but is not limited to, a nail cutter.

[0171] In some of the embodiments, the Raman detection device 520 includes but is not limited to a Raman spectrometer.

[0172] In some of the embodiments, the electronic scanning device 530 includes but is not limited to a scanning electron microscope and a transmission electron microscope.

[0173] In some of the embodiments, the tumor risk prediction device 540 includes but is not limited to a mobile phone, a computer, a server, etc.

[0174] The technical effects of this embodiment are basically the same as those of Embodiment 2, and will not be described in detail here.

[0175] Example 4

[0176] This embodiment is a variation of Embodiments 1 to 3.

[0177] In this embodiment, the nail-based tumor risk prediction method includes:

[0178] Step S102, obtaining chemical information data and structural image data of the nail coronal surface, wherein the chemical information data is obtained by Raman detection, and the structural image data is obtained by scanning electron microscopy;

[0179] Step S104, inputting the chemical information data and the structural image data into a pre-trained tumor risk prediction model to obtain a tumor risk prediction result, wherein the tumor risk prediction result includes tumor classification, metabolic disease classification, accuracy, sensitivity, and specificity.

[0180] In the present invention, metabolic diseases include but are not limited to diabetes, hyperuricemia, hyperthyroidism, hypothyroidism, hyperlipidemia, hypertension, fatty liver, Alzheimer's disease, chronic renal failure and tumors from different organs.

[0181] In the present invention, the metabolic disease risk prediction results are only used to indicate the risk of suffering from a certain metabolic disease, and are not used to directly diagnose whether a person suffers from a metabolic disease. Other direct diagnostic methods are required for diagnosis. The metabolic disease risk prediction results of the present invention can be used to assist doctors in performing disease risk screening before obtaining the results of inspections and tests (e.g., blood tests, urine tests, etc.).

[0182] In this embodiment, step S102 is substantially the same as step S102 in embodiment 1, and will not be described in detail herein.

[0183] In this embodiment, the chemical information data of step S102 also includes the following characteristic peaks:

[0184]

[0185]

[0186] In this embodiment, step S104 is substantially the same as step S104 in embodiment 1, and will not be described in detail herein.

[0187] In this embodiment, the tumor risk prediction result can be output in the following format:

[0188] (1) Hepatocellular carcinoma;

[0189] (2) hepatocellular carcinoma + diabetes;

[0190] (3) hepatocellular carcinoma (in situ);

[0191] (4) hepatocellular carcinoma (metastasis) + diabetes (high risk);

[0192] (5) Hepatocellular carcinoma + high sensitivity + high specificity; (wherein, "+" is only used to indicate the interval and does not play any other role. In addition, the interval can also be expressed in other ways, such as "," ";", etc.)

[0193] (6) Hepatocellular carcinoma + high sensitivity + high specificity, diabetes + high sensitivity + high specificity.

[0194] Through steps S102 to S104, tumor classification and metabolic disease classification can be output simultaneously, and tumor risk and metabolic disease risk can be differentiated, thereby obtaining a prediction result of category identification and avoiding confusion of results.

[0195] Further, step S104 includes:

[0196] Step S202, processing the chemical information data and the structural image data to obtain a 64-dimensional initial feature vector;

[0197] Step S204, inputting the initial feature vector into a pre-trained tumor risk prediction model to obtain a final feature vector of 2048 dimensions;

[0198] Step S206: input the final feature vector into a multi-layer perceptron for processing to obtain a tumor risk prediction result.

[0199] In step S206, the output length of the tumor risk prediction result can be adjusted according to the number of tumor classifications to be detected (accuracy of tumor classification). For example, if the number of tumor classifications to be detected (accuracy of tumor classification) is 3 and the number of metabolic diseases is 3, the output length of the tumor risk prediction result is 6.

[0200] Among them, step S202 to step S206 are basically the same as step S202 to step S206 of embodiment 1, and will not be repeated here.

[0201] In this embodiment, the training method of the tumor risk prediction model is basically the same as that in Embodiment 1 and will not be described in detail here.

[0202] In this embodiment, the nail-based tumor risk prediction device includes a chemical information acquisition unit 410, a structural image acquisition unit 420, and a risk prediction unit 430. The chemical information acquisition unit 410 is used to acquire chemical information data of the nail coronal surface, wherein the chemical information data is obtained by Raman detection; the structural image acquisition unit 420 is used to acquire structural image data of the nail coronal surface, wherein the structural image data is obtained by scanning electron microscopy; the risk prediction unit 430 is used to input the chemical information data and the structural image data into a pre-trained tumor risk prediction model to obtain a tumor risk prediction result, wherein the tumor risk prediction result includes tumor classification, metabolic disease classification, accuracy, sensitivity, and specificity.

[0203] In this embodiment, the chemical information acquisition unit 410, the structural image acquisition unit 420, and the risk prediction unit 430 are substantially the same as those in Embodiment 2, and are not described in detail herein.

[0204] In this embodiment, the nail-based tumor risk prediction system includes a nail processing device 510, a Raman detection device 520, an electron microscope scanning device 530, and a tumor risk prediction device 540 as described in Example 2. The nail processing device 510 is used to obtain and process the nail to obtain a nail cut in the coronal plane; the Raman detection device 520 is used to perform Raman detection on the nail cut in the coronal plane to obtain chemical information data of the nail coronal plane; the electron microscope scanning device 530 is used to perform electron microscope scanning on the nail cut in the coronal plane to obtain structural image data of the nail coronal plane; the tumor risk prediction device 540 is used to obtain chemical information data and structural image data of the nail coronal plane, and input the chemical information data and structural image data into a pre-trained tumor risk prediction model to obtain a tumor risk prediction result, wherein the tumor risk prediction result includes tumor classification, metabolic disease classification, accuracy, sensitivity, and specificity.

[0205] In this embodiment, the nail processing device 510, the Raman detection device 520, the electron microscope scanning device 530, and the tumor risk prediction device 540 are substantially the same as those in Embodiment 3, and are not described in detail herein.

[0206] Example 5

[0207] This embodiment is a specific implementation of the present invention.

[0208] In this embodiment, healthy control samples (10 cases), liver cancer patient samples (7 cases), lung cancer patient samples (8 cases), gastric cancer patient samples (7 cases), and thyroid cancer patient samples (8 cases) were collected, and then Raman spectroscopy and scanning electron microscopy were performed respectively.

[0209] Among them, the detection point of Raman spectroscopy detection (cm -1) are as follows: 399.944, 404.510, 409.073, 413.634, 418.192, 422.748, 427.302, 431.852, 436.401, 440.946, 445.490, 450.030, 454.569, 459.104, 463.638, 4 68.168, 472.696, 477.222, 481.745, 486.266, 490.784, 495.300, 499.813, 504.324, 508.833, 513.338, 517.842, 522.343, 526.841, 531.337, 535.83 1. 540.322, 544.810, 549.296, 553.780, 558.261, 562.740, 567.217, 571.690, 576.162, 580.631, 585.098, 589.562, 594.023, 598.483, 602.940, 607 .394、611.846、616.296、620.743、625.188、629.630、634.070、638.508、642.943、647.376、651.806、656.234、660.660、665.083、669.504、673.922、 678.339, 682.752, 687.164, 691.573, 695.979, 700.383, 704.785, 709.185, 713.582, 717.977, 722.369, 726.759, 731.147, 735.532, 739.915, 744.2 96, 748.674, 753.050, 757.424, 761.795, 766.164, 770.530, 774.895, 779.257, 783.616, 787.974, 792.329, 796.681, 801.032, 805.380, 809.725, 81 4.069, 818.410, 822.749, 827.085, 831.419, 835.751, 840.081, 844.408, 848.733, 853.056, 857.376, 861.695, 866.011, 870.324, 874.635, 878.945 、883.251、887.556、891.858、896.158、900.456、904.751、909.044、913.335、917.624、921.910、926.194、930.476、934.756、939.033、943.309、947.581、951.852、956.121、960.387、964.651、968.912、973.172、977.429、981.684、985.937、990.188、994.436、998.682、1002.930、1007.170、1011.410、1015.640、1019.880、1024.110、1028.340、1032.570、1036.800、1041.020、1045.240、1049.460、1053.680、1057.900、1062.110、1066.320、1070.530、1074.740、1078.940、1083.140、1087.340、1091.540、1095.740、1099.930、1104.120、1108.310、1112.500、1116.680、1120.870、1125.050、1129.220、1133.400、1137.580、1141.750、1145.920、1150.080、1154.250、1158.410、1162.570、1166.730、1170.890、1175.050、1179.200、1183.350、1187.500、1191.640、1195.790、1199.930、1204.070、1208.210、1212.340、1216.480、1220.610、1224.740、1228.870、1232.990、1237.110、1241.230、1245.350、1249.470、1253.580、1257.700、1261.810、1265.920、1270.020、1274.130、1278.230、1282.330、1286.430、1290.520、1294.620、1298.710、1302.800、1306.880、1310.970、1315.050、1319.130、1323.210、1327.290、1331.370、1335.440、1339.510、1343.580、1347.640、1351.710、1355.770、1359.830、1363.890、1367.950、1372.000、1376.050、1380.110、1384.150、1388.200、1392.240、1396.290、1400.330、1404.360、1408.400、1412.430、1416.470、1420.500、1424.520、1428.550、1432.570、1436.600、1440.620、1444.630、1448.650、1452.660、1456.680、1460.690、1464.690、1468.700、1472.700、1476.700、1480.700、1484.700、1488.700、1492.690、1496.680、1500.670、1504.660、1508.650、151 2.630、1516.610、1520.590、1524.570、1528.550、1532.520、1536.490、1540.460、1544.430、1548.400、1552.360、1556.320、1560.280、1564.240、1568.200、1572.150、1576.100、1580.050、1584.000、1587.950、1591.890、1595.830、1599.770、1603.710、1607.650、16 11.580、1615.520、1619.450、1623.380、1627.300、1631.230、1635.150、1639.070、1642.990、1646.910、1650.820、1654.740、1658.650、1662.560、1666.460、1670.370、1674.270、1678.170、1682.070、1685.970、1689.870、1693.760、1697.650、1701.540、1705.430、1 709.320, 1713.200, 1717.090, 1720.970, 1724.840, 1728.720, 1732.600, 1736.470, 1740.340, 1744.210, 1748.080, 1751.940, 1755.810, 1759.670, 1763.530, 1767.380, 1771.240, 1775.090, 1778.950, 1782.800, 1786.640, 1790.490, 1794.340, 1798.180, 1802.020. .

[0210] For the Raman spectral data of the above samples, Figure 6 to Figure 10 shown. Fig.11 This is a comparison chart of the numerical mean of Raman spectral data of various samples. Fig.11 As shown, through 750cm -1 、759cm -1 、855cm -1 、1157cm -1 、1240cm -1 、1309cm -1 、1335cm -1 、1398cm -1 、1518cm -1 、1552cm -1 、1583cm -1 、1605cm -1 、1615cm -1 Characteristic peaks such as these can distinguish patient samples with tumors from normal samples. Taking liver cancer and lung cancer as examples, the details are as follows:

[0211] 1) In liver cancer, the Raman peaks of cytochrome C (750 cm-1), tyrosine (644 / 646 cm-1), ethanolamine group (759 cm-1), ν (CC), proline + δ (CCH) proline breathing ring (855 cm-1), and collagen (1319 / 1340 cm-1) were significantly reduced, while tryptophan (749 cm-1), fatty acids (1065 cm-1), carotenoids (1157 cm-1), and lipids (2855 / 2880 cm-1) were significantly increased;

[0212] 2) In lung cancer, the Raman peaks of disulfide bonds (509cm-1), tyrosine (644 / 646cm-1), protein (1242cm-1), and CH2deformation (1398cm-1) were significantly reduced, while carotenoids (1518cm-1), tryptophan (749cm-1), fatty acids (1065cm-1), and lipids (2855 / 2880cm-1) were significantly increased.

[0213] For the scanning electron microscope images of the above samples, Figure 12 to Figure 16 As shown. Figure 12 to Figure 16 It can be seen that the SEM images of the nail coronal plane of the samples of patients with metabolic diseases are different from those of normal samples. These differences can be used to distinguish the samples of patients with metabolic diseases from the normal samples. The details are as follows:

[0214] 1) Normal samples: The surface is relatively smooth and flat, without obvious cracks or protrusions; the texture is dense, the grain is regular, and there is a slight ridge structure; the color is uniform, without obvious color difference;

[0215] 2) Gastric cancer patient samples: Compared with the normal control group, the surface is rough, with obvious protrusions; the texture is irregular; the nails are relatively fragile, showing regular longitudinal jump-like cuts;

[0216] 3) Samples from liver cancer patients: The surface is very rough, with obvious protrusions or depressions; the texture is disordered, with obvious cracks and fissures; the nails are relatively fragile, with regular jumping knife marks;

[0217] 4 Samples from lung cancer patients: The surface is rough, with obvious protrusions; the texture is irregular; the nails are relatively fragile, showing regular jumping and relatively delicate knife marks;

[0218] 5) Samples from thyroid cancer patients: The surface is slightly rough, the texture is slightly disordered, the nails are relatively fragile, and there are regular jumping and large-amplitude cuts caused by a knife.

[0219] Example 6

[0220] This embodiment is a specific implementation of the present invention.

[0221] In this embodiment, hepatocellular carcinoma (HCC) is taken as an example for specific research. Fig.17 As shown, the nail-based tumor risk prediction method includes the following steps:

[0222] 1. Sample collection

[0223] During this study, a total of 297 samples were collected, including 147 samples from patients diagnosed with HCC and 150 samples from healthy controls. Sample collection followed a standardized process and used a specially designed nail collection tool to ensure data consistency and high quality.

[0224] During this study, the clinical information of HCC patients included gender, age, diagnosis, AFP, DCP, HBsAg, e antibody, core antibody, and CA199.

[0225] (II) Raman spectroscopy image and scanning electron microscope image acquisition

[0226] For each sample, Raman spectroscopic images and scanning electron microscope images of the nails were collected. Raman spectroscopic images are used to analyze changes in the chemical composition of the nails, especially the content of disulfide bonds, a chemical marker related to structural stability. Scanning electron microscope images provide microstructural information of the nails, including surface roughness, periodic damage, and changes in microstructure.

[0227] (III) Construction of residual-based convolutional neural network model

[0228] Construct a residual-based convolutional neural network model. Fig.17 As shown, the residual-based convolutional neural network model includes multiple residual blocks, each of which includes multiple convolutional layers and activation layers, which are used to effectively extract key features of image data. Through the deep learning framework, the residual-based convolutional neural network model can automatically learn and identify image features related to hyperuricemia from image data.

[0229] For the residual-based convolutional neural network model of the present invention, the first layer of the network input is the image data x, which will then undergo a convolution operation F(x, {W i}), W i is the trainable convolution kernel parameter. At the same time, the feature map after the convolution operation is added to the original input x to obtain the output feature y of this layer of network. The input feature of the next layer is the output feature y of the previous layer. This process is repeated to finally obtain a 2048-dimensional feature vector k representing the high-dimensional features of the current input image. The network calculation process is as follows:

[0230] .

[0232] In order to further improve the performance of the model, a multi-layer perceptron (MLP) network is designed to integrate the high-dimensional features extracted from the convolutional neural network. MLP consists of multiple fully connected layers, which can learn the complex functional relationship between features, so as to perform effective classification and recognition. Specifically, the input feature of this part is the 2048-dimensional feature vector k of the previous deep network, and then the MLP structure maps the 2048-dimensional feature vector to the number of diseases that need to be classified. For example: If there are currently three diseases that need to be diagnosed, then the output prediction probability p i The length of is 3. This structure helps the model to learn more deeply after feature extraction and improve the accuracy of diagnosis.

[0233] (IV) Training and evaluation of residual-based convolutional neural network models

[0234] 75% of the samples are used for training and 25% of the samples are used for final test evaluation.

[0235] In addition, during the training process, the back propagation algorithm is used to optimize the network weights. By calculating the loss function (such as cross entropy loss) and applying the gradient descent method to gradually adjust the network parameters, the model's prediction results on the training data are as close to the actual labels as possible. Through multiple iterations of training, the model gradually achieves high-precision prediction performance on complex data sets. The loss is calculated as follows:

[0236]

[0237] Where C is the number of categories, y i is the true value of the current sample label (0 or 1), p i is the output prediction probability of the MLP.

[0238] The difference between the true value and the predicted value is calculated by the above formula, and then these differences are back-propagated to W in the deep model i And MLP structure, these parameters are further tuned to achieve the purpose of training.

[0239] Through iterative training, the residual-based convolutional neural network model performed well on the validation set, and then the performance was verified on the test set. The test results showed that the residual-based convolutional neural network model achieved a classification accuracy of 90.25% in the comprehensive test set under two image modalities in 90.45 seconds. This high accuracy proves the effectiveness and reliability of the model in practical applications.

[0240] The above description is only a preferred embodiment of the present invention, and does not limit the implementation mode and protection scope of the present invention. For those skilled in the art, it should be aware that all solutions obtained by equivalent substitutions and obvious changes made using the description and illustrations of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting tumor risk based on nails, characterized in that: include: Acquiring chemical information data and structural image data of the nail coronal surface, wherein the chemical information data is obtained by Raman spectrometer detection, and the structural image data is obtained by scanning electron microscopy; The chemical information data and the structural image data are input into a pre-trained tumor risk prediction model to obtain a tumor risk prediction result, wherein the tumor risk prediction result includes tumor classification, accuracy, sensitivity, and specificity.

2. The tumor risk prediction method according to claim 1, characterized in that: Inputting the chemical information data and the structural image data into a pre-trained tumor risk prediction model to obtain a tumor risk prediction result includes: Processing the chemical information data and the structural image data to obtain a 64-dimensional initial feature vector; Inputting the initial feature vector into the pre-trained tumor risk prediction model to obtain a final feature vector of 2048 dimensions; The final feature vector is input into a multi-layer perceptron for processing to obtain the tumor risk prediction result.

3. The tumor risk prediction method according to claim 1 or 2, characterized in that: The training methods of tumor risk prediction models include: Constructing a tumor risk prediction model, wherein the tumor risk prediction model includes a plurality of residual blocks; Constructing a training set and a test set according to the chemical information data and the structural image data, wherein the ratio of the training set to the test set is 3:1; Inputting the training set into a tumor risk prediction model to obtain an output result; Inputting the output result into a multilayer perceptron for processing to obtain a tumor risk prediction result; The tumor risk prediction results are tested to detect the effectiveness of the training, and based on the effectiveness test results, the tumor risk prediction model is iterated until the training is completed.

4. The tumor risk prediction method according to claim 3, characterized in that: The training method of the tumor risk prediction model also includes: Calculating the loss value of the tumor risk prediction model; The gradient descent method and back propagation algorithm are used to optimize the network parameters and network weights of the tumor risk prediction model.

5. A nail-based tumor risk prediction device, characterized in that: include: A chemical information acquisition unit, used to acquire chemical information data of the nail coronal surface, wherein the chemical information data is obtained by Raman detection; A structural image acquisition unit, used to acquire structural image data of the nail coronal surface, wherein the structural image data is obtained by a scanning electron microscope; The risk prediction unit is used to input the chemical information data and the structural image data into a pre-trained tumor risk prediction model to obtain a tumor risk prediction result, wherein the tumor risk prediction result includes tumor classification, accuracy, sensitivity, and specificity.

6. The tumor risk prediction device according to claim 5, characterized in that: The risk prediction unit comprises: A processing element, used for processing the chemical information data and the structural image data to obtain a 64-dimensional initial feature vector; A first output element, used to input the initial feature vector into the pre-trained tumor risk prediction model to obtain a final feature vector of 2048 dimensions; A second output element inputs the final feature vector into a multilayer perceptron for processing to obtain the tumor risk prediction result; and / or The risk prediction unit comprises: A construction element, used to construct a tumor risk prediction model, wherein the tumor risk prediction model includes a plurality of residual blocks; A data set component, used to construct a training set and a test set according to the chemical information data and the structural image data, wherein the ratio of the training set to the test set is 3:1; A first training component, used for inputting the training set into a tumor risk prediction model to obtain an output result; A second training element is used to input the output result into a multi-layer perceptron for processing to obtain a tumor risk prediction result; The testing element is used to test the tumor risk prediction result to detect the effectiveness of the training, and iterate the tumor risk prediction model according to the effectiveness test result until the training is completed.

7. The tumor risk prediction device according to claim 5 or 6, characterized in that: The risk prediction unit also includes: A loss calculation element, used to calculate the loss value of the tumor risk prediction model; The optimization element is used to optimize the network parameters and network weights of the tumor risk prediction model using a gradient descent method and a back propagation algorithm.

8. A nail-based tumor risk prediction system, characterized in that: include: A nail processing device for obtaining and processing a nail to obtain a coronal flat-sectioned nail; A Raman detection device is used to perform Raman detection on the nail whose coronal surface is cut flat, so as to obtain chemical information data of the nail coronal surface; An electron microscope scanning device is used to perform electron microscope scanning on the nail whose coronal surface is cut flat, so as to obtain structural image data of the nail coronal surface; The tumor risk prediction device as described in any one of claims 5 to 7 is used to obtain chemical information data and structural image data of the nail coronal surface, and input the chemical information data and the structural image data into a pre-trained tumor risk prediction model to obtain a tumor risk prediction result, wherein the tumor risk prediction result includes tumor classification, accuracy, sensitivity, and specificity.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the tumor risk prediction method according to any one of claims 1 to 4 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the tumor risk prediction method as described in any one of claims 1 to 4 is implemented.