Method, device and system for predicting epidermal growth factor receptor gene mutation probability

Abstract

The invention provides a method, a device and a system for predicting the gene mutation probability of an epidermal growth factor receptor. The method comprises the following steps: acquiring first computed tomography image data; determining a lesion area in the first computed tomography image data, and performing segmentation processing on the lesion area of the first computed tomography image data by adopting a semantic segmentation algorithm to determine iconographic features of the lesion; extracting blood supply features of the first computed tomography image data by adopting a three-dimensional segmentation model; constructing a mutation probability prediction model according to the iconography features, the blood supply features and the clinical features of the focus; and determining the epidermal growth factor receptor gene mutation probability of the to-be-detected object according to the mutation probability prediction model. According to the embodiment of the invention, more comprehensive lesion information and influence factors of occurrence of EGFR mutation can be captured, so that the accuracy of EGFR mutation probability prediction is remarkably improved.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a method, device and system for predicting the probability of epidermal growth factor receptor gene mutation. Background Art

[0002] At present, lung cancer is divided into two major categories according to histopathology: non-small cell lung cancer (NSCLC) and small cell lung cancer (SCLC), with the former accounting for about 85%. Epidermal Growth Factor Receptor (EGFR) is currently considered to be one of the target genes highly associated with non-small cell lung cancer (NSCLC). Clearly identifying the epidermal growth factor receptor (EGFR) gene mutation status of patients with non-small cell lung cancer (NSCLC) is a prerequisite for determining whether the patient is suitable for subsequent related targeted therapy. Epidermal Growth Factor Receptor (EGFR) gene mutation is correlated with age, gender, smoking history, etc., but traditional epidermal growth factor receptor (EGFR) gene mutation detection is performed by sampling tissues, such as blood, human-related tissues, etc. Although the detection accuracy is high, the cost is high and there is a certain risk of infection. Therefore, it is necessary to propose a method for predicting the probability of epidermal growth factor receptor gene mutation to at least partially solve the problems existing in the prior art. Summary of the invention

[0003] In view of at least one of the above technical problems existing in the prior art, the present application is proposed. According to one aspect of the present application, a method for predicting the probability of epidermal growth factor receptor gene mutation is provided, the method comprising:

[0004] acquiring first computed tomography image data;

[0005] Determine a lesion region in the first computed tomography image data, and segment the lesion region in the first computed tomography image data using a semantic segmentation algorithm to determine imaging features of the lesion;

[0006] extracting blood supply characteristics of the first computed tomography image data using a three-dimensional segmentation model;

[0007] constructing a mutation probability prediction model according to the imaging characteristics of the lesion, the blood supply characteristics and the clinical characteristics;

[0008] The mutation probability of the epidermal growth factor receptor gene of the object to be detected is determined according to the mutation probability prediction model.

[0009] In some embodiments, before acquiring the first computed tomography image data, the method further includes:

[0010] obtaining an initial computed tomography image data sample;

[0011] The initial computed tomography image data samples are preprocessed to form first computed tomography image data.

[0012] In some embodiments, the preprocessing includes at least: rotating, translating and scaling the initial computed tomography image data.

[0013] In some embodiments, determining the lesion area in the first computed tomography image data comprises:

[0014] The first computed tomography image data is detected using a target detection algorithm to determine the lesion area on the first computed tomography image and determine the three-dimensional coordinates of the lesion area.

[0015] In some embodiments, segmenting the lesion area of ​​the first computed tomography image data using a semantic segmentation algorithm includes:

[0016] Using a semantic segmentation algorithm to crop the lesion area from the first computed tomography image data to form a three-dimensional image of the lesion area;

[0017] The imaging features corresponding to the three-dimensional image of the lesion area are obtained through a preset imaging feature open source library.

[0018] In some embodiments, extracting blood supply characteristics of the first computed tomography image data using a three-dimensional segmentation model includes:

[0019] Using a three-dimensional segmentation model to segment the first computer tomography image to obtain a three-dimensional image of the bronchi and arteries and veins;

[0020] The shortest distance between the lesion area and the bronchus and / or the artery and vein is calculated by using the three-dimensional image of the lesion area and the three-dimensional image of the bronchus and / or the artery and vein to determine whether the lesion area intersects with the bronchus and / or the artery and vein.

[0021] In some embodiments, when the lesion area intersects with the bronchus and / or the artery and vein, extracting the blood supply characteristics of the first computer tomography image data using a three-dimensional segmentation model further includes:

[0022] The volume of the bronchus and / or the artery and vein covered by the lesion area and the number of branches of the bronchus and / or the artery and vein covered by the lesion area are calculated through the three-dimensional matrix of the lesion area and the three-dimensional matrix of the bronchus and / or the artery and vein.

[0023] In some embodiments, a mutation probability prediction model is constructed based on the imaging characteristics of the lesion, the blood supply characteristics and the clinical characteristics, including:

[0024] acquiring corresponding clinical features from a pathology report of the first computed tomography image data;

[0025] Based on a preset algorithm, training multiple initial mutation probability prediction models according to the imaging features, the blood supply features, and the clinical features;

[0026] Using a test data set to verify the prediction effects of the multiple initial mutation probability prediction models, and selecting the model with the best prediction effect as the mutation probability prediction model;

[0027] Among them, the clinical characteristics at least include the patient's basic characteristics and pathological characteristics; the patient's basic characteristics at least include the patient's age, gender, smoking history and medical history; the pathological characteristics at least include tumor cell staging, pathological classification, lesion size and lesion location.

[0028] In some embodiments, determining the probability of epidermal growth factor receptor gene mutation in the subject to be detected according to the mutation probability prediction model includes:

[0029] Acquiring a second computer tomography image sequence of the object to be detected;

[0030] inputting the second computer tomography image sequence of the object to be detected into the mutation probability prediction model;

[0031] An output result of a mutation probability prediction model is received, wherein the output result represents the mutation probability of the epidermal growth factor receptor gene of the subject to be detected.

[0032] In another aspect, an embodiment of the present application provides a device for predicting the probability of epidermal growth factor receptor gene mutation, the device comprising:

[0033] A sample acquisition module, used for acquiring first computer tomography image data;

[0034] a lesion detection module, for determining a lesion region in the first computed tomography image data;

[0035] a lesion segmentation module, which uses a semantic segmentation algorithm to segment the lesion area of ​​the first computer tomography image data to determine the imaging features of the lesion;

[0036] a bronchus and arteriovenous segmentation module, which extracts blood supply characteristics of the first computed tomography image data using a three-dimensional segmentation model;

[0037] The epidermal growth factor receptor gene mutation prediction module constructs a mutation probability prediction model according to the imaging characteristics, blood supply characteristics and clinical characteristics of the lesion, and determines the epidermal growth factor receptor gene mutation probability of the object to be detected according to the mutation probability prediction model.

[0038] Another aspect of the present application embodiment provides a system for predicting the probability of epidermal growth factor receptor gene mutation, the system comprising:

[0039] A memory and a processor, wherein the memory stores a computer program executed by the processor, and when the computer program is executed by the processor, the processor executes the method for predicting the probability of epidermal growth factor receptor gene mutation as described above.

[0040] On another aspect, an embodiment of the present application provides a storage medium having a computer program stored thereon. When the computer program is executed by a processor, the processor executes the method for predicting the probability of epidermal growth factor receptor gene mutation as described above.

[0041] The method for predicting the probability of epidermal growth factor receptor gene mutation in the embodiment of the present application uses the imaging characteristics, blood supply characteristics, and clinical characteristics of the lesions in the computed tomography image data to construct a mutation probability prediction model, and then uses the mutation probability prediction model to determine the probability of epidermal growth factor receptor gene mutation of the object to be detected. By extracting features of multimodal computed tomography image features and clinical characteristics, more comprehensive lesion information and influencing factors of epidermal growth factor receptor (EGFR) mutation can be captured, thereby significantly improving the accuracy of epidermal growth factor receptor (EGFR) mutation probability prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0043] Figure 1 A schematic flow chart showing a method for predicting the probability of epidermal growth factor receptor gene mutation according to an embodiment of the present application;

[0044] Figure 2 A schematic flow chart showing preprocessing of initial computed tomography image data samples according to an embodiment of the present application is shown;

[0045] Figure 3 A schematic flowchart of step S102 according to an embodiment of the present application is shown;

[0046] Figure 4 A schematic flow chart showing step S102 according to another embodiment of the present application is shown;

[0047] Figure 5 A schematic flowchart of step S103 according to an embodiment of the present application is shown;

[0048] Figure 6 A schematic flowchart of step S103 according to another embodiment of the present application is shown;

[0049] Figure 7 A schematic flowchart of step S104 according to an embodiment of the present application is shown;

[0050] FIG8( a) shows a schematic diagram of a random forest model according to an embodiment of the present application;

[0051] FIG8( b ) shows a schematic diagram of an XGBOOST model according to an embodiment of the present application;

[0052] FIG8( c ) shows a schematic diagram of a deep learning model according to an embodiment of the present application;

[0053] Fig. 9 A schematic flowchart of step S105 according to an embodiment of the present application is shown;

[0054] Fig.10 A schematic diagram showing a device for predicting the probability of epidermal growth factor receptor gene mutation according to an embodiment of the present application;

[0055] Fig.11 A schematic block diagram of a system for predicting the probability of epidermal growth factor receptor gene mutation according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0056] In order to enable those skilled in the art to better understand the technical solutions of the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0057] Based on at least one of the aforementioned technical problems, the present application provides a method for predicting the probability of epidermal growth factor receptor gene mutation, the method comprising: acquiring first computed tomography image data; determining a lesion area in the first computed tomography image data, and performing segmentation processing on the lesion area of ​​the first computed tomography image data using a semantic segmentation algorithm to determine the imaging features of the lesion; extracting the blood supply features of the first computed tomography image data using a three-dimensional segmentation model; constructing a mutation probability prediction model based on the imaging features of the lesion, the blood supply features and clinical features; and determining the probability of epidermal growth factor receptor gene mutation of the object to be detected based on the mutation probability prediction model. The method for predicting the probability of epidermal growth factor receptor gene mutation in the embodiment of the present application uses the imaging characteristics, blood supply characteristics, and clinical characteristics of the lesions in the computed tomography (CT) image data to construct a mutation probability prediction model, and then uses the mutation probability prediction model to determine the probability of epidermal growth factor receptor gene mutation of the object to be detected. By extracting features of multimodal computed tomography (CT) image features and clinical characteristics, it is possible to capture more comprehensive lesion information and factors affecting the occurrence of epidermal growth factor receptor (EGFR) mutations, thereby significantly improving the accuracy of epidermal growth factor receptor (EGFR) mutation probability prediction.

[0058] Figure 1 A schematic flow chart of a method for predicting the probability of epidermal growth factor receptor gene mutation according to an embodiment of the present application is shown; Figure 1 As shown, the method 100 for predicting the probability of epidermal growth factor receptor gene mutation according to an embodiment of the present application may include the following steps S101, S102, S103, S104 and S105:

[0059] In step S101, a first computed tomography (CT) image data is acquired.

[0060] The computed tomography (CT) image data here is a computed tomography (CT) image sequence, which can completely represent a three-dimensional image of the lungs.

[0061] In one embodiment of the present application, Figure 2 As shown, before acquiring the first computer tomography (CT) image data, the method further includes step A1 and step A2:

[0062] In step A1, an initial computed tomography (CT) image data sample is obtained;

[0063] In step A2, the initial computed tomography (CT) image data samples are preprocessed to form first computed tomography (CT) image data.

[0064] Wherein, the preprocessing at least includes: performing rotation, translation and scaling operations on the initial computed tomography (CT) image data.

[0065] When acquiring a computed tomography (CT) image, if the subject has a large movement, the computed tomography (CT) image will have problems such as ghosting. Therefore, when preprocessing the computed tomography (CT) image, it is necessary to perform image preprocessing such as denoising on the computed tomography (CT) image.

[0066] The initial computed tomography (CT) image data samples are preprocessed in a three-dimensional space, which may include image rotation operations, translation operations, and scaling operations, etc. For example, a three-dimensional image with pixels of 100x512x512 is reduced to a three-dimensional image with pixels of 50x256x256.

[0067] In step S102, a lesion region in the first computed tomography (CT) image data is determined, and a semantic segmentation algorithm is used to segment the lesion region in the first computed tomography (CT) image data to determine imaging features of the lesion.

[0068] Wherein, the lesions include pulmonary nodules (PN). In medicine, pulmonary nodules (PN) refer to round or irregular lesions with a diameter less than or equal to 3 cm in the lungs. Imaging shows shadows with increased density, which may be single or multiple, with clear or unclear boundaries. Pulmonary nodules of different densities have different probabilities of malignancy. Early detection of pulmonary nodules, rapid diagnosis of benign or malignant nature, and timely removal of malignant nodules are of great significance for early screening of lung cancer and reducing lung cancer mortality. In the embodiment of the present application, the lesion area, namely the pulmonary nodule area, may also be referred to as a region of interest (ROI).

[0069] In one embodiment of the present application, Figure 3 As shown, step S102 of determining the lesion area in the first computed tomography (CT) image data includes step B1:

[0070] In step B1, a target detection algorithm is used to detect the first computed tomography (CT) image data to determine the lesion area on the first computed tomography (CT) image and determine the three-dimensional coordinates of the lesion area.

[0071] The embodiment of the present application can use a target detection algorithm to detect the lesion area. In one example, a three-dimensional convolutional neural network (3D-CNN) is constructed using a deep learning algorithm, and then the region candidate network (Region Proposal Network, RPN) of the regional convolutional neural network (Faster-RCNN) in the target detection algorithm is used to detect the first computed tomography (CT) image data to screen out the region of interest (Region of Interest, ROI), that is, the candidate pulmonary nodule area (lesion area). At the same time, the three-dimensional coordinates, diameter and other information of the lesion area can also be determined, and the accuracy is high.

[0072] In one embodiment of the present application, Figure 4 As shown, step S102 uses a semantic segmentation algorithm to segment the lesion area of ​​the first computed tomography (CT) image data, including step C1 and step C2:

[0073] In step C1, a semantic segmentation algorithm is used to crop the lesion area from the first computed tomography (CT) image data to form a three-dimensional image of the lesion area.

[0074] After determining the lesion area in the first computer tomography (CT) image data and the three-dimensional coordinates of the lesion area, the embodiment of the present application can use the semantic segmentation technology in the deep learning algorithm to cut out the lesion area. The model used can be a nodule segmentation model, and the network structure used is DenseUNet. This network structure is a supervised training method, and the supervisory label (label) is a single-channel binary image. In a single-channel binary image, 0 represents the background and 1 represents the nodule tissue area. According to the lesion area (ROI area), the lesion area is cut out from the first computer tomography (CT) image data as the input of the nodule segmentation model. The nodule segmentation model segments the area of ​​the lesion according to the input, and can be specifically accurate to the pixel level.

[0075] In step C2, imaging features corresponding to the three-dimensional image of the lesion area are obtained through a preset imaging feature open source library.

[0076] For example, for a program developed in Python, the imaging features corresponding to the lesion can be obtained through the open source library of imaging features in Python.

[0077] In step S103, a three-dimensional segmentation model is used to extract blood supply characteristics of the first computed tomography (CT) image data.

[0078] The blood supply information in the embodiments of the present application refers to the distribution of bronchi and arteries and veins around the lesion, as well as whether the lesion penetrates bronchi and arteries and veins, and the volume and depth of the site of penetration of bronchi and arteries and veins.

[0079] In one embodiment of the present application, Figure 5 As shown, step S103 of extracting the blood supply characteristics of the first computed tomography (CT) image data using a three-dimensional segmentation model includes step D1, step D2 and step D2:

[0080] In step D1, the first computed tomography (CT) image is segmented using a three-dimensional segmentation model to obtain a three-dimensional image of the bronchi and arteries and veins.

[0081] The embodiment of the present application uses a three-dimensional segmentation model (3D-UNETS model) to segment the first computer tomography (CT) image to obtain the corresponding three-dimensional images of the bronchi and arteries and veins. When training the 3D-UNETS model, a supervised model training is adopted, and the supervised label (LABEL) is the current channel image. In the current channel image, 0 represents the background, 1 represents the bronchus, 2 represents the artery, and 3 represents the vein. When training the 3D-UNETS model, data enhancement, attention mechanism and other technologies are also added to optimize the model segmentation capability. The first computer tomography (CT) image data is used as the input of the 3D-UNETS model. The 3D-UNETS model identifies the regions of the bronchi and arteries and veins based on the input first computer tomography (CT) image data, which can be accurate to the pixel level.

[0082] Because different patients have different lung sizes. For example, there are relatively large differences between the lungs of children and adults. Moreover, the shapes of blood vessels and bronchi of different patients will also be different. For example, some patients will have some variant blood vessels and bronchial branches. The embodiment of the present application adopts the 3DU-NET model. The embodiment of the present application adopts a three-dimensional segmentation model (3D-UNETS model) to segment the first computed tomography (CT) image. The model is a U-shaped architecture as a whole, which combines shallow local features and deep semantic features, and can better process computed tomography (CT) images of bronchi and arteries and veins of different sizes and shapes.

[0083] In step D2, the shortest distance between the lesion area and the bronchus and / or the artery and vein is calculated through the three-dimensional image of the lesion area and the three-dimensional image of the bronchus and / or the artery and vein to determine whether the lesion area intersects with the bronchus and / or the artery and vein.

[0084] Specifically, the shortest distance between the lesion and the bronchus and / or artery and vein can be calculated by the three-dimensional coordinates of the lesion and the three-dimensional coordinates of the bronchus and / or artery and vein. For example, by calculating the Euclidean distance between the three-dimensional coordinates of each point on the bronchus, artery and vein and the three-dimensional coordinates of each point on the lesion, the positional relationship between the lesion and the bronchus, artery and vein can be obtained. If the distance is equal to 0, it means that the lesion has an intersection with the bronchus, artery and vein and other tissues, that is, blood supply information.

[0085] In step D3, if they intersect, it is determined that there is a lesion in the bronchus and / or the artery and vein.

[0086] In one embodiment of the present application, Figure 6 As shown, in the case where the lesion area intersects with the bronchus and / or the artery and vein, step S103 of extracting the blood supply characteristics of the first computer tomography image data using a three-dimensional segmentation model also includes step E1:

[0087] In step E1, the volume of the bronchus and / or the artery and vein covered by the lesion area and the number of branches of the bronchus and / or the artery and vein covered by the lesion area are calculated using the three-dimensional matrix of the lesion area and the three-dimensional matrix of the bronchus and / or the artery and vein.

[0088] In the embodiment of the present application, the three-dimensional images of the lesion area coverage, bronchi and arteries and veins can also be represented in the form of a three-dimensional array. In the three-dimensional array, 0 can be used to represent the background, and 1 can be used to represent the presence of a bronchus, artery, vein or lesion at that location.

[0089] A computed tomography (CT) image can be regarded as a two-dimensional array (or two-dimensional matrix) consisting of N rows and M columns of pixels, and a patient's lung computed tomography (CT) image sequence often contains hundreds of computed tomography (CT) images, so these computed tomography (CT) images can be regarded as a three-dimensional array (or three-dimensional matrix). For example, if there are K lung CT images, then the complete lung CT image is equivalent to a three-dimensional array (or three-dimensional matrix) consisting of K×N×M pixels. In such a three-dimensional array, an array value of 1 indicates a lesion, and an array value of 0 indicates that it is not a lesion. The segmentation result of the lesion indicates whether these pixels are the location of the lesion, so that the three-dimensional lesion area corresponding to the complete lung computed tomography (CT) image can be obtained. This segmentation method can also obtain the three-dimensional areas of bronchi and arteries and veins. For the three-dimensional pixels of lesions, arteries, and veins, the expression formula is as follows;

[0090] Lesions x,y,z =lesion x,y,z , where x∈[1,K],y∈[1,N],z∈[1,M] (1)

[0091] artery x,y,z =artery x,y,z , where x∈[1,K],y∈[1,N],z∈[1,M] (2)

[0092] vein x,y,z =vein x,y,z , where x∈[1,K],y∈[1,N],z∈[1,M] (3)

[0093] Bronchial tube x,y,z =bronchi x,y,z , where x∈[1,K],y∈[1,N],z∈[1,M] (4)

[0094] Among them, lesion represents the three-dimensional matrix of the lesion; artery represents the three-dimensional matrix of the artery; vein represents the three-dimensional matrix of the vein; bronchi represents the three-dimensional matrix of the bronchus, and x, y, and z represent the pixel positions corresponding to the three-dimensional matrix. x,y,z Indicates whether the pixel at the yth row and zth column of the xth CT image in the CT image sequence is a lesion. If the pixel is a lesion, then lesion x,y,z =1, otherwise 0. At the same time, the pixels on the bronchi and arteries and veins are similar.

[0095] For example, taking lesions and bronchus as an example, when the three-dimensional matrix lesion and the three-dimensional matrix bronchi are multiplied, the result is as follows:

[0096] Result x,y,z =lesion x,y,z ×bronchi x,y,z

[0097] Among them, lesion x,y,z Indicates whether the pixel at the yth row and zth column of the xth computed tomography (CT) image in the CT image sequence is a lesion; bronze x,y,z Indicates whether the pixel at the yth row and zth column of the xth CT image in the computed tomography (CT) image sequence is a bronchus; result x,y,z It indicates whether there is a lesion intersecting the bronchus at the pixel point at the yth row and the zth column of the xth computed tomography (CT) image in the computed tomography (CT) image sequence.

[0098] Those skilled in the art will know that for any pixel (x, y, z), where x∈[1,K], y∈[1,N], z∈[1,M], if result x,y,z =1, it means lesion x,y,z =1 and bronze x,y,z =1, then the pixel is both a pixel of the lesion and a pixel of the bronchus, which means that the lesion is interspersed with the bronchus (or that the lesion and the bronchus have an intersection). In this way, by counting all the positions where the result is 1 in the result three-dimensional matrix, the number of pixels covering the bronchus by the lesion can be obtained. Moreover, it can be known that the layer thickness and spatial position (spacing) information of the lesion in the K, N, and M directions can be known through the number of pixels and the position of each pixel. In this way, the layer thickness and spatial position information can be used to obtain the volume of the bronchus covered by the lesion, and / or the volume of the artery and vein covered by the lesion. Similarly, the number of branches of the bronchi and / or the arteries and veins can also be counted.

[0099] The embodiment of the present application can use matrix operations, etc. to calculate a three-dimensional representation of the skeleton of the bronchus and artery and vein. For example, the pixel value of the skeleton position corresponding to the first branch will be represented by 1, and the pixel values ​​of all skeleton positions corresponding to the second branch will be represented by 2, and so on. By counting the number of different numbers (except 0), the number of bronchus and / or arteriovenous branches can be obtained.

[0100] In step S104, a mutation probability prediction model is constructed according to the imaging characteristics, blood supply characteristics and clinical characteristics of the lesion.

[0101] In one embodiment of the present application, Figure 7 As shown, step S104 constructs a mutation probability prediction model according to the imaging characteristics, blood supply characteristics and clinical characteristics of the lesion, including step F1, step F2 and step F3:

[0102] In step F1, corresponding clinical features are obtained from a pathology report of the first computed tomography image data;

[0103] In step F2, based on a preset algorithm, multiple initial mutation probability prediction models are trained according to the imaging features, the blood supply features, and the clinical features;

[0104] In step F3, the prediction effects of the multiple initial mutation probability prediction models are verified using a test data set, and the model with the best prediction effect is selected as the mutation probability prediction model.

[0105] Among them, the clinical characteristics at least include the patient's basic characteristics and pathological characteristics; the patient's basic characteristics at least include the patient's age, gender, smoking history and medical history; the pathological characteristics at least include tumor cell (Tumor, Node and Metastasis, abbreviated as TNM) staging, pathological classification, lesion size and lesion location.

[0106] For example, the preset algorithm may include an XGBOOST algorithm, a random forest algorithm, and a deep learning algorithm. In this way, the multiple initial mutation probability prediction models trained according to the imaging features, the blood supply features, and the clinical features are an XGBOOST model, a random forest model, and a deep learning model. Then, the model with the best effect is selected as the final mutation probability prediction model.

[0107] As shown in Figure 8(a), it is a schematic diagram of the random forest model. The random forest model is a classifier that contains many decision trees. It can be used to deal with classification and regression problems, as well as dimensionality reduction problems. It is also very tolerant to outliers and noise, and has better prediction and classification performance than decision trees.

[0108] As shown in Figure 8(b), it is a schematic diagram of the XGBOOST model. In the figure, the XGBOOST model is constructed by taking the statistics of males whose age (Age) is less than 15 years old and the statistics of people who use computers daily (Use Computer Daily) as examples. The XGBOOST model is an integrated learning algorithm and belongs to the category of gradient boosting tree (GBDT) model. It can be implemented by building an additive model. When building the XGBOOST model, you can choose a tree model, but you can also choose other types of models such as logistic regression.

[0109] As shown in Figure 8(c), it is a schematic diagram of a deep learning model. The basic structure of a deep learning model is a complex network system formed by a large number of processing units (called neurons) widely connected to each other. It is a highly complex nonlinear dynamic learning system. Neural networks have large-scale parallelism, distributed storage and processing, self-organization, self-adaptation and self-learning capabilities, and are particularly suitable for processing imprecise and fuzzy information processing problems that require considering many factors and conditions at the same time.

[0110] This embodiment first extracts the imaging features of the lesion from the CT image. At the same time, by segmenting the bronchi and arteries and veins and determining whether the lesion has an intersection with the bronchi and arteries and veins, the blood supply characteristics of the lesion can be obtained. In addition, clinical features can also be extracted from the pathology report. Clinical features may include features such as basic patient information and medical history, pathological features of the tumor, and the like. Then, the above-mentioned imaging features, blood supply features, and clinical features may be fused by residual connections and the like to construct a mutation probability prediction model for predicting epidermal growth factor receptor (EGFR) gene mutations. This process can be divided into two stages: training and testing:

[0111] In the training phase, a variety of algorithms can be used to train the mutation probability prediction model. For example, the mutation probability prediction model has been trained and optimized multiple times using XGBoost, neural network model, and random forest model. Similarly, the models mentioned above can also be trained and optimized multiple times. For example, the number of XGBoost network models and random forest decision trees and the depth of decision trees, the number of layers of the neural network model, the type of activation function, etc. are adjusted to ensure that the model can fully learn the features and patterns in the data. When optimizing the mutation probability prediction model, the Adam algorithm can be used as the optimizer and the cross entropy as the loss function to improve the training efficiency and accuracy of the model.

[0112] In the testing phase, the trained mutation probability prediction model was tested on the test set to evaluate its accuracy. By comparing the performance of different models on the test set, we selected the best model as the final epidermal growth factor receptor (EGFR) mutation probability prediction model.

[0113] Specifically, when collecting the initial computed tomography image data samples, they can be randomly divided into a training set and a test set in a ratio of 8:2, wherein the training set is used to train the mutation probability prediction model, and the test set is used to verify and evaluate the performance of the mutation probability prediction model.

[0114] In step S105, the mutation probability of the epidermal growth factor receptor gene of the subject to be detected is determined according to the mutation probability prediction model.

[0115] In one embodiment of the present application, Fig. 9 As shown, step S105 of determining the mutation probability of the epidermal growth factor receptor gene of the object to be detected according to the mutation probability prediction model includes step G1, step G2 and step G3:

[0116] Step G1, acquiring a second computer tomography image sequence of the object to be detected;

[0117] Step G2, inputting the second computer tomography image data of the object to be detected into the mutation probability prediction model;

[0118] Step G3, receiving the output result of the mutation probability prediction model, wherein the output result represents the mutation probability of the epidermal growth factor receptor gene of the subject to be detected.

[0119] In an embodiment of the present application, after completing the construction of the mutation probability prediction model, when there is a need for a patient CT image to be detected, a second computer tomography image sequence of the object to be detected can be collected, and then the imaging features and blood supply features are extracted from the second computer tomography image sequence of the object to be detected. The imaging features and blood supply features, as well as the clinical features, are input into the mutation probability prediction model of the epidermal growth factor receptor (EGFR) for prediction, and the probability of epidermal growth factor receptor gene mutation in the patient to be detected is obtained. The value of the probability is a decimal between 0 and 1. This probability value can provide a reference for clinicians, helping them to better judge the probability of epidermal growth factor receptor (EGFR) mutation in the patient to be detected, so as to formulate a more personalized treatment plan.

[0120] The embodiments of the present application jointly predict the probability of epidermal growth factor receptor (EGFR) gene mutation through blood supply characteristics (whether bronchi, arteries and veins are interspersed with lesions, and when interspersed with lesions, the closest distance between bronchi, arteries and veins and lesions, etc.), as well as imaging characteristics and clinical characteristics. In traditional technology, although computed tomography (CT) can qualitatively determine whether bronchi and blood vessels are interspersed with lesions, it requires a relatively high cost to determine information such as the interspersed volume, depth, and interspersed position, and it is impossible to determine the three-dimensional spatial characteristics of the lesion and bronchi / blood vessels when no interspersal occurs. The embodiments of the present application can quantitatively determine the blood supply information of the lesion in three dimensions by reconstructing the three-dimensional images of the bronchi / arteries and veins, and then combine the imaging characteristics and clinical characteristics of the lesions to predict the probability of epidermal growth factor receptor (EGFR) mutation.

[0121] The method for predicting the probability of epidermal growth factor receptor gene mutation in the embodiment of the present application uses the imaging characteristics, blood supply characteristics, and clinical characteristics of the lesions in the computed tomography image data to construct a mutation probability prediction model, and then uses the mutation probability prediction model to determine the predicted epidermal growth factor receptor gene mutation probability of the object to be tested. By extracting features from multimodal computed tomography image features and clinical characteristics, it is possible to capture more comprehensive lesion information and influencing factors of EGFR mutation, thereby significantly improving the accuracy of epidermal growth factor receptor (EGFR) mutation probability prediction.

[0122] Combine the following Fig.10The device for predicting the probability of epidermal growth factor receptor gene mutation of the present application is described. Fig.10 A schematic diagram of a device 1000 for predicting the probability of epidermal growth factor receptor gene mutation according to an embodiment of the present application is shown. The device 1000 comprises:

[0123] The sample acquisition module 1001 is used to acquire first computer tomography image data;

[0124] A lesion detection module 1002 is configured to determine a lesion region in the first computed tomography image data;

[0125] A lesion segmentation module 1003, which uses a semantic segmentation algorithm to segment the lesion area of ​​the first computer tomography image data to determine the imaging features of the lesion;

[0126] a bronchus and arteriovenous segmentation module 1004, which extracts blood supply characteristics of the first computed tomography image data using a three-dimensional segmentation model;

[0127] The epidermal growth factor receptor gene mutation prediction module 1005 constructs a mutation probability prediction model according to the imaging characteristics of the lesion, the blood supply characteristics and the clinical characteristics, and determines the epidermal growth factor receptor gene mutation probability of the object to be detected according to the mutation probability prediction model.

[0128] First, the sample acquisition module 1001 acquires an initial computed tomography (CT) image sequence. Then the computed tomography (CT) image sequence is preprocessed. During the preprocessing, the computed tomography (CT) image can be filtered, rotated, translated, and scaled, and can also be cropped to obtain the preprocessed first computed tomography image data.

[0129] Secondly, the lesion detection module performs lesion detection on the preprocessed first computer tomography image data to determine the lesion area therein. The computer tomography (CT) image with the determined lesion area is sent to the lesion segmentation module for processing. The lesion segmentation module cuts out a three-dimensional image of the lesion area therefrom, and uses a semantic segmentation algorithm to segment the lesion area of ​​the first computer tomography image data to determine the imaging characteristics of the lesion.

[0130] Thirdly, the bronchus and arteriovenous segmentation module uses a three-dimensional segmentation model to extract the blood supply characteristics of the first computed tomography image data. The blood supply information refers to the distribution of bronchi and arteries and veins around the lesion, whether the lesion intersects the bronchi and arteries and veins, and the volume and depth of the interspersed bronchi and arteriovenous sites.

[0131] In addition, clinical characteristics can be extracted from pathology reports.

[0132] Then, the above imaging features, blood supply features and clinical features are integrated through residual connection and other algorithms to form an epidermal growth factor receptor (EGFR) mutation probability prediction model. Through the mutation probability prediction model of the embodiment of the present application, more comprehensive lesion information and influencing factors of epidermal growth factor receptor (EGFR) mutation can be captured, thereby significantly improving the accuracy of epidermal growth factor receptor (EGFR) mutation probability prediction.

[0133] Combine the following Fig.11 The system for predicting the probability of epidermal growth factor receptor gene mutation of the present application is described, wherein: Fig.11 A schematic block diagram of a system for predicting the probability of epidermal growth factor receptor gene mutation according to an embodiment of the present application is shown.

[0134] like Fig.11 As shown, the system 1100 for predicting the probability of epidermal growth factor receptor gene mutation includes: one or more memories 1101 and one or more processors 1102, wherein the memory 1101 stores a computer program executed by the processor 1102, and when the computer program is executed by the processor 1102, the processor 1102 executes the method for predicting the probability of epidermal growth factor receptor gene mutation described above.

[0135] The system 1100 for predicting the probability of epidermal growth factor receptor gene mutation may be part or all of a computer device that can implement a method for predicting the probability of epidermal growth factor receptor gene mutation by means of software, hardware, or a combination of software and hardware.

[0136] like Fig.11 As shown, the system 1100 for predicting the probability of epidermal growth factor receptor gene mutation includes one or more memories 1101, one or more processors 1102, a display (not shown), and a communication interface, etc. These components are interconnected via a bus system and / or other forms of connection mechanisms (not shown). It should be noted that Fig.11 The components and structure of the system 1100 for predicting the probability of epidermal growth factor receptor gene mutation are merely exemplary and non-limiting. The system 1100 for predicting the probability of epidermal growth factor receptor gene mutation may also have other components and structures as needed.

[0137] The memory 1101 is used to store various data and executable program instructions generated during the operation of the related method, such as for storing various application programs or algorithms for implementing various specific functions. It can include one or more computer program products, and the computer program product can include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory can include, for example, random access memory (RAM) and / or cache memory (cache), etc. The non-volatile memory can include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0138] Processor 1102 can be a central processing unit (CPU), a graphics processing unit (GPU), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other forms of processing units with data processing capabilities and / or instruction execution capabilities, and can be other components in the system 1100 for predicting the probability of epidermal growth factor receptor gene mutation to perform the desired functions.

[0139] In one example, the system 1100 for predicting the probability of epidermal growth factor receptor gene mutation also includes an output device that can output various information (such as images or sounds) to the outside (such as a user), and can include one or more of a display device, a speaker, etc.

[0140] The communication interface can be an interface of any currently known communication protocol, such as a wired interface or a wireless interface, wherein the communication interface may include one or more serial ports, USB interfaces, Ethernet ports, WiFi, wired networks, DVI interfaces, device integrated interconnect modules or other suitable ports, interfaces, or connections.

[0141] In addition, according to an embodiment of the present application, a storage medium is also provided, on which program instructions are stored, and when the program instructions are run by a computer or a processor, the corresponding steps of the method for predicting the probability of epidermal growth factor receptor gene mutation in the embodiment of the present application are executed. The storage medium may include, for example, a memory card of a smart phone, a storage component of a tablet computer, a hard disk of a personal computer, a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a portable compact disk read-only memory (CD-ROM), a USB memory, or any combination of the above storage media.

[0142] The system and storage medium for predicting the probability of epidermal growth factor receptor gene mutation according to the embodiments of the present application can implement the aforementioned method for predicting the probability of epidermal growth factor receptor gene mutation, and therefore have the same advantages as the aforementioned method for predicting the probability of epidermal growth factor receptor gene mutation.

[0143] Although example embodiments have been described herein with reference to the accompanying drawings, it should be understood that the above example embodiments are merely exemplary and are not intended to limit the scope of the present application to this. Those of ordinary skill in the art may make various changes and modifications therein without departing from the scope and spirit of the present application. All these changes and modifications are intended to be included within the scope of the present application as required by the appended claims.

[0144] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0145] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed.

[0146] In the description provided herein, a large number of specific details are described. However, it is understood that the embodiments of the present application can be practiced without these specific details. In some instances, well-known methods, structures and techniques are not shown in detail so as not to obscure the understanding of this description.

[0147] Similarly, it should be understood that in order to streamline the present application and help understand one or more of the various inventive aspects, in the description of the exemplary embodiments of the present application, the various features of the present application are sometimes grouped together into a single embodiment, figure, or description thereof. However, the method of the present application should not be interpreted as reflecting the following intention: the claimed application requires more features than the features clearly stated in each claim. More specifically, as reflected in the corresponding claims, the inventive point is that the corresponding technical problem can be solved with features less than all the features of a single disclosed embodiment. Therefore, the claims following the specific embodiment are hereby explicitly incorporated into the specific embodiment, wherein each claim itself serves as a separate embodiment of the present application.

[0148] It will be understood by those skilled in the art that, except for mutually exclusive features, all features disclosed in this specification (including the accompanying claims, abstracts and drawings) and all processes or units of any method or device disclosed in this specification may be combined in any combination. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstracts and drawings) may be replaced by an alternative feature that provides the same, equivalent or similar purpose.

[0149] In addition, those skilled in the art will appreciate that, although some embodiments described herein include certain features included in other embodiments but not other features, the combination of features of different embodiments is meant to be within the scope of the present application and form different embodiments. For example, in the claims, any one of the claimed embodiments can be used in any combination.

[0150] The various component embodiments of the present application can be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. It should be understood by those skilled in the art that a microprocessor or digital signal processor (DSP) can be used in practice to implement some or all functions of some modules according to the embodiments of the present application. The application can also be implemented as a device program (e.g., computer program and computer program product) for executing a part or all of the methods described herein. Such a program implementing the present application can be stored on a computer-readable medium, or can have the form of one or more signals. Such a signal can be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

[0151] It should be noted that the above embodiments illustrate the present application rather than limit the present application, and that those skilled in the art may design alternative embodiments without departing from the scope of the appended claims. In the claims, any reference symbol between brackets should not be constructed as a limitation to the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "one" or "an" preceding an element does not exclude the presence of multiple such elements. The present application may be implemented by means of hardware including several different elements and by means of a suitably programmed computer. In a unit claim that lists several devices, several of these devices may be embodied by the same hardware item. The use of the words first, second, and third, etc. does not indicate any order. These words may be interpreted as names.

[0152] The above is only a specific implementation or description of a specific implementation of the present application, and the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. The protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A method for predicting the probability of epidermal growth factor receptor gene mutation, characterized in that: The method comprises: acquiring first computed tomography image data; Determine a lesion region in the first computed tomography image data, and segment the lesion region in the first computed tomography image data using a semantic segmentation algorithm to determine imaging features of the lesion; extracting blood supply characteristics of the first computed tomography image data using a three-dimensional segmentation model; constructing a mutation probability prediction model according to the imaging characteristics of the lesion, the blood supply characteristics and the clinical characteristics; The mutation probability of the epidermal growth factor receptor gene of the object to be detected is determined according to the mutation probability prediction model.

2. The method according to claim 1, characterized in that Before acquiring the first computed tomography image data, the method further includes: obtaining an initial computed tomography image data sample; The initial computed tomography image data samples are preprocessed to form first computed tomography image data.

3. The method according to claim 2, characterized in that The preprocessing at least includes: performing rotation, translation and scaling operations on the initial computed tomography image data.

4. The method according to claim 1, characterized in that: Determining a lesion area in the first computer tomography image data includes: The first computed tomography image data is detected using a target detection algorithm to determine the lesion area on the first computed tomography image and determine the three-dimensional coordinates of the lesion area.

5. The method according to claim 3, characterized in that: The lesion area of ​​the first computer tomography image data is segmented using a semantic segmentation algorithm, including: Using a semantic segmentation algorithm to crop the lesion area from the first computed tomography image data to form a three-dimensional image of the lesion area; The imaging features corresponding to the three-dimensional image of the lesion area are obtained through a preset imaging feature open source library.

6. The method according to claim 5, characterized in that Extracting blood supply characteristics of the first computer tomography image data using a three-dimensional segmentation model includes: Using a three-dimensional segmentation model to segment the first computer tomography image data to obtain a three-dimensional image of the bronchi and arteries and veins; Calculating the shortest distance between the lesion area and the bronchus and / or the artery and vein through the three-dimensional image of the lesion area and the three-dimensional image of the bronchus and / or the artery and vein to determine whether the lesion area intersects with the bronchus and / or the artery and vein; If they intersect, it is determined that a lesion exists in the bronchus and / or the artery and vein.

7. The method according to claim 6, characterized in that In the case where the lesion area intersects with the bronchus and / or the artery and vein, extracting blood supply characteristics of the first computer tomography image data using a three-dimensional segmentation model also includes: The volume of the bronchus and / or the artery and vein covered by the lesion area and the number of branches of the bronchus and / or the artery and vein covered by the lesion area are calculated through the three-dimensional matrix of the lesion area and the three-dimensional matrix of the bronchus and / or the artery and vein.

8. The method according to claim 1, characterized in that A mutation probability prediction model is constructed according to the imaging characteristics of the lesion, the blood supply characteristics and the clinical characteristics, including: acquiring corresponding clinical features from a pathology report of the first computed tomography image data; Based on a preset algorithm, training multiple initial mutation probability prediction models according to the imaging features, the blood supply features, and the clinical features; Using a test data set to verify the prediction effects of the multiple initial mutation probability prediction models, and selecting the model with the best prediction effect as the mutation probability prediction model; Among them, the clinical characteristics at least include the patient's basic characteristics and pathological characteristics; the patient's basic characteristics at least include the patient's age, gender, smoking history and medical history; the pathological characteristics at least include tumor cell staging, pathological classification, lesion size and lesion location.

9. The method according to claim 1, characterized in that: Determining the epidermal growth factor receptor gene mutation probability of the subject to be detected according to the mutation probability prediction model includes: Acquiring a second computer tomography image sequence of the object to be detected; inputting the second computer tomography image sequence of the object to be detected into the mutation probability prediction model; An output result of a mutation probability prediction model is received, wherein the output result represents the mutation probability of the epidermal growth factor receptor gene of the subject to be detected.

10. A device for predicting the probability of epidermal growth factor receptor gene mutation, characterized in that: The device comprises: A sample acquisition module, used for acquiring first computer tomography image data; a lesion detection module, for determining a lesion region in the first computed tomography image data; a lesion segmentation module, which uses a semantic segmentation algorithm to segment the lesion area of ​​the first computer tomography image data to determine the imaging features of the lesion; a bronchus and arteriovenous segmentation module, which extracts blood supply characteristics of the first computed tomography image data using a three-dimensional segmentation model; The epidermal growth factor receptor gene mutation prediction module constructs a mutation probability prediction model according to the imaging characteristics, blood supply characteristics and clinical characteristics of the lesion, and determines the epidermal growth factor receptor gene mutation probability of the object to be detected according to the mutation probability prediction model.

11. A system for predicting the probability of epidermal growth factor receptor gene mutation, characterized in that: The system comprises: A memory and a processor, wherein the memory stores a computer program executed by the processor, and when the computer program is executed by the processor, the processor executes the method for predicting the probability of epidermal growth factor receptor gene mutation as described in any one of claims 1 to 9.

12. A storage medium, characterized in that: The storage medium stores a computer program, which, when executed by a processor, enables the processor to execute the method for predicting the probability of epidermal growth factor receptor gene mutation according to any one of claims 1 to 9.

Images