KR20230158264A - Radiomix-based machine learning model for predictive diagnosis of invasive cancer using medical images - Google Patents

Abstract

The present invention relates to a radiomics-based machine learning model for predicting and diagnosing invasive cancer using medical images, and more specifically, to detect invasive cancer from patients with ductal carcinoma in situ before surgery using magnetic resonance imaging. It relates to a model that can predict or diagnose invasive cancer. The invasive cancer diagnosis model according to the present invention can accurately diagnose invasive cancer in a subject, and can be used in diagnostic surgery and various medical fields related to invasive cancer diagnosis. It can be.

Description

Radiomix-based machine learning model for predictive diagnosis of invasive cancer using medical images}

The present invention relates to a radiomics-based machine learning model for predicting and diagnosing invasive cancer using medical images, and more specifically, to detect invasive cancer from patients with ductal carcinoma in situ before surgery using magnetic resonance imaging. This relates to a model that can predict or diagnose invasive cancer.

Breast ductal carcinoma in situ (DCIS) is a breast malignant tumor that does not invade the basement membrane. In theory, it is a disease that can be cured through surgical treatment because there is no possibility of metastasis, whereas ductal carcinoma in situ (DCIS) is a breast malignant tumor that does not invade the basement membrane. If it invades, it progresses to invasive cancer. When diagnosed with invasive cancer, it is known that there is a possibility of axillary lymph node metastasis in approximately 7% of cases, so additional axillary lymph node biopsy is essential.

Ultrasound-guided central core biopsy is an invasive biopsy and is known to be the standard technique for breast cancer diagnosis. However, since only a portion of the lesion is biopsied, if intraductal carcinoma is diagnosed through ultrasound-guided core biopsy, invasive cancer is diagnosed at the final biopsy after surgery. Underestimation may occur. In addition, mammography, ultrasound, and magnetic resonance imaging (MRI) are used as testing methods to diagnose breast cancer. Although dynamic contrast-enhanced MRI is the most sensitive test for diagnosing breast lesions, it has the disadvantage of low specificity. There is.

To date, many studies have been conducted to predict invasive cancer in cases diagnosed with ductal carcinoma in situ through central core biopsy, and the possibility of diagnosing invasive cancer in preoperative ductal carcinoma in situ is determined by the thickness of the needle, tissue grade, size, mammography findings, and palpability. has been suggested as a significant factor in determining . Meanwhile, the field of precision medicine for personalized treatment and prevention has been emerging in recent years. Recently, radiomics has been used to determine medical treatment for patients. It is attracting attention. Radiomics aims to create a diagnosis and prognosis prediction model by integrating quantitative feature information extracted from medical images to provide personalized medical services. As the use of medical images in patient diagnosis increases, this research field is designed to analyze large amounts of medical image information and predict disease diagnosis, differentiation, and prognosis based on useful diagnostic marker information extracted from it. came into focus. Traditional analyzes of medical images, such as computed tomography scans (CT), MRI, or positron emission tomography (PET), are typically based on visual interpretation of simple features such as tumor size and overall shape, and contrast enhancement. First, in the case of radiomics techniques, these images are analyzed by a computer, converted into quantitative complex data, and then highly efficient quantitative features are extracted, including texture, advanced shape modeling, and heterogeneity. In addition, as resolution characteristics increase, it becomes possible to acquire three-dimensional images containing millions of voxels available for analysis, and accordingly, advances in radiomics are made to efficiently utilize more information from the increased large volume of medical imaging data. It has progressed naturally.

Accordingly, through medical image analysis using radiomics techniques, not only can tumors be monitored over time by regularly acquiring images throughout the treatment process, but also image biomarkers can be used for cancer detection, diagnosis, and selection of treatment strategies. There is an emerging need for models that can be used for prognostic inference, response prediction, and management.

Accordingly, the present inventors created a radiomics-based machine learning model to predict and diagnose invasive cancer using medical images, and confirmed that its diagnostic performance was significantly excellent.

Accordingly, the purpose of the present invention is to provide a method of providing information for diagnosing invasive cancer from medical image data by a computing device.

Another object of the present invention is to provide a computing device for providing information for diagnosing invasive cancer.

Another object of the present invention is to provide a computer program stored in a storage medium for diagnosing invasive cancer.

The present invention relates to a radiomics-based machine learning model for predicting and diagnosing invasive cancer using medical images. The model according to the present invention can diagnose invasive cancer with high accuracy.

Hereinafter, the present invention will be described in more detail.

One aspect of the present invention provides a method of providing information for diagnosing invasive cancer from medical image data by a computing device, comprising: a processing step of generating a segmented image of a region of interest based on the medical image data; An extraction step of extracting one or more features from the segmented image; and a generation step of generating diagnostic information by inputting the extracted features into a pre-learned invasive cancer diagnosis model; A method of providing information for diagnosing invasive cancer, including.

The term "invasive cancer" in this specification refers to cancer that spreads and invades adjacent tissues or cells. Unlike benign tumors, it grows quickly and does not have a sheath surrounding the mass, so it pushes or penetrates surrounding tissues and in this process, It refers to a disease that deprives normal cells of nutrients and weakens or destroys surrounding tissues.

In the present invention, invasive cancer includes squamous cell carcinoma, small-cell lung cancer, non-small cell lung cancer, glioma, gastrointestinal cancer, and renal cell cancer. (renal cancer), ovarian cancer, liver cancer, colorectal cancer, endometrial cancer, kidney cancer, prostate cancer, thyroid cancer, neuroblastoma, brain cancer, central nervous system cancer, pancreatic cancer, glioblastoma multiforme (glioblastoma multiforme), cervical cancer, stomach cancer, bladder cancer, malignant hepatoma, breast ductal carcinoma in situ (DCIS), breast cancer, colon carcinoma, head and neck cancer, stomach cancer, germ cell tumor, pediatric sarcoma ), rhabdomyosarcoma, Ewing's sarcoma, osteosarcoma, soft tissue sarcoma, sinonasal NK/T-cell lymphoma, myeloma, It may be caused by one or more selected from the group consisting of melanoma and multiple myeloma, for example, it may be caused by breast ductal carcinoma in situ (DCIS). , but is not limited to this.

The term “medical image data” in this specification may include multidimensional medical image data composed of discrete image elements (eg, pixels in a two-dimensional image). Specifically, medical data may include a visible object or a digital representation of that object (e.g., a file corresponding to the pixel output of a CT, MRI detector, etc.). Medical imaging data may be collected by computed tomography (CT), magnetic resonance imaging (MRI), fundus imaging, ultrasound, or any other medical imaging system known in the art ( It may include a medical image of a subject).

The term “subject” in this specification may be an object (or subject) that has an invasive cancer disease or is suspected of having an invasive cancer disease. In one embodiment of the invention, the subject may be a mammal, for example, the subject may be one or more selected from the group consisting of humans, monkeys, dogs, cats, mice, rats, cattle, horses, pigs, goats and sheep. You can.

In one embodiment of the present invention, the segmented image includes: a first segmented image including a difference image (subtraction image); a second segmented image including an intratumor image; and a third segmented image including images within the tumor and around the tumor.

And, one or more features are based on the first segment image including the difference image, the second segment image including the intratumor image, and the third segment image including the intratumor and tumor periphery images. can be extracted.

In the present invention, the difference image may be an image obtained by subtracting the pre-contrast image from the post-contrast image.

In one implementation of the present invention, the feature may be a radiomic signature.

In one implementation of the invention, the radiomic signature is calculated based on primary feature measurements including shape features and intensity features, and secondary feature measurements including texture features. It may be.

In one embodiment of the present invention, the radiomic signature may include primary features including shape features and intensity features, and secondary features including texture features.

In one embodiment of the present invention, the extraction step may be performed through the PyRadiomics module.

In one embodiment of the invention, the shape features include Elongation, Flatness, LeastAxisLength, MajorAxisLength, Maximum2DDiameterColumn, Maximum2DDiameterRow, Maximum2DDiameterSlice, Maximum3DDiameter, MeshVolume, MinorAxisLength, Sphericity, SurfaceArea, SurfaceVolumeRatio, and VoxelVolume. to It may include

In one embodiment of the present invention, the intensity features are 10Percentile, 90Percentile, Energy, Entropy, InterquartileRange, Kurtosis, Maximum, MeanAbsoluteDeviation, Mean, Median, Minimum, Range, RobustMeanAbsoluteDeviation, RootMeanSquared, Skewness , may include Total Energy, Uniformity, and Variance.

In one implementation of the invention, the texture features include the intensity co-occurrence matrix (GLCM), intensity dependence matrix (GLDM), intensity interval length matrix (GLRLM), intensity magnitude zone matrix (GLSZM), and neighboring grayscale difference matrix (NGTDM). It may include

In one embodiment of the present invention, the radiomic signature includes Surface Volume Ratio, Sphericity, Energy, Dependence Variance, ZonePercentage, and Skewness. , Large Area Emphasis, Elongation, Large Area High Gray Level Emphasis, Zone Variance, and Short Run High Gray Level Emphasis. It may include one or more features selected from the group consisting of.

In one embodiment of the present invention, the invasive cancer diagnosis model may be learned using features including radiomic signatures.

In one embodiment of the present invention, the radiomic signature may include primary features including shape features and intensity features, and texture features.

In one embodiment of the present invention, the invasive cancer diagnostic model includes Surface Volume Ratio, Sphericity, Energy, Dependence Variance, ZonePercentage, and Skewness. ), Large Area Emphasis, Elongation, Large Area High Gray Level Emphasis, Zone Variance, and Short Run High Gray Level Emphasis It may be learned by a radiomic signature containing one or more features selected from the group consisting of ).

In one embodiment of the present invention, the invasive cancer diagnosis model may be a machine learning model or a deep neural network model.

If the invasive cancer diagnosis model is a machine learning model, the invasive cancer diagnosis model according to the present invention may be a random forest, Naive Bayes decision tree, Bayesian network, or support It may include a vector machine (support vector machine (SVM)) algorithm.

The term “deep neural network” in this specification may refer to a neural network that includes a plurality of hidden layers in addition to an input layer and an output layer. The term “deep neural network” may be used interchangeably with the terms “neural network,” “network function,” and “neural network” throughout this specification. Using deep neural networks, it is possible to identify latent structures in data. In other words, it is possible to identify the potential structure of a photo, text, video, voice, or music (e.g., what object is in the photo, what the content and emotion of the text are, what the content and emotion of the voice are, etc.) . In the present invention, the deep neural network includes a convolutional neural network (CNN), a recurrent neural network (RNN), a restricted Boltzmann machine (RBM), and a deep trust network (deep neural network). It may include, but is not limited to, belief network (DBN), Q network, U network, Siamese network, etc.

Another aspect of the present invention is a computing device for providing information for diagnosing invasive cancer, comprising: a processor including one or more cores; and a memory, wherein the processor generates a segmented image of the region of interest based on medical image data, extracts one or more features from the segmented image, and applies the extracted features to normal and pre-learned infiltrates through diagnostic information. It is a computing device that generates diagnostic information by inputting it into a cancer diagnosis model.

In one embodiment of the present invention, the segmented image includes: a first segmented image including a difference image (subtraction image); a second segmented image including an intratumor image; and a third segmented image including images within the tumor and around the tumor.

In one implementation of the present invention, the feature may be a radiomic signature.

In one embodiment of the present invention, the radiomic signature may include primary features including shape features and intensity features, and secondary features including texture features.

Another aspect of the present invention is a computer program stored in a storage medium, wherein the computer program, when executed on one or more processors, performs the following operations for providing information for diagnosing invasive cancer, the operations being: medical imaging. A processing operation of generating a segmented image of a region of interest based on data; An extraction operation for extracting one or more features from a segmented image; and a generating operation of inputting the extracted features into a pre-learned invasive cancer diagnostic model to generate diagnostic information. It is a computer program stored in a storage medium.

In one embodiment of the present invention, the segmented image includes: a first segmented image including a difference image (subtraction image); a second segmented image including an intratumor image; and a third segmented image including images within the tumor and around the tumor.

In one implementation of the present invention, the feature may be a radiomic signature.

In one embodiment of the present invention, the radiomic signature may include primary features including shape features and intensity features, and secondary features including texture features.

The present invention relates to a radiomics-based machine learning model for predicting and diagnosing invasive cancer using medical images, and more specifically, to detect invasive cancer from patients with ductal carcinoma in situ before surgery using magnetic resonance imaging. It relates to a model that can predict or diagnose invasive cancer. The invasive cancer diagnosis model according to the present invention can accurately diagnose invasive cancer in a subject, and can be used in diagnostic surgery and various medical fields related to invasive cancer diagnosis. It can be.

Figure 1 is a block diagram of a computing device that performs an operation to provide information for diagnosing invasive cancer according to an embodiment of the present invention.
Figure 2 is a block diagram illustrating a process for providing information for diagnosing invasive cancer according to an embodiment of the present invention.
Figure 3 is a photograph showing the results of generating a segmented image for diagnosis of invasive cancer according to an embodiment of the present invention.
Figure 4 is a graph showing the results of analyzing the accuracy of the diagnosis using the invasive cancer diagnosis model according to an embodiment of the present invention through the area under the receiver operating characteristic curve (AUROC). .

Below, with reference to the attached drawings, embodiments of the present invention will be described in detail so that those skilled in the art can easily implement the present invention. However, the present invention may be implemented in various different forms and is not limited to the embodiments described herein. In order to clearly explain the present invention in the drawings, parts that are not related to the description are omitted, and similar parts are given similar reference numerals throughout the specification.

Throughout the specification, when a part "includes" a certain component, this means that it may further include other components rather than excluding other components, unless specifically stated to the contrary. The term “and/or,” “and/or” includes any one and all combinations of the associated listed items.

In addition, the terms “…unit” and “…module” used in the specification refer to a unit that processes at least one function or operation, which may be implemented as hardware, software, or a combination of hardware and software.

The various illustrative logical blocks, components, modules, circuits, means, logic, and algorithm steps described in connection with embodiments herein may be implemented in electronic hardware, computer software, or combinations of both. We must recognize that it can be implemented. To clearly illustrate the interchangeability of hardware and software, various illustrative components, blocks, configurations, means, logics, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented in hardware or software will depend on the specific application and design constraints imposed on the overall system. A skilled technician can implement the described functionality in a variety of ways for each specific application. However, such implementation decisions should not be construed as causing a departure from the scope of the present disclosure.

Figure 1 is a block diagram of a computing device that performs an operation to provide information for diagnosing invasive cancer according to an embodiment of the present invention.

Referring to FIG. 1, a computing device 1000 that performs an operation to provide information for diagnosing invasive cancer according to an embodiment may include a processor 100 and a memory 200.

The processor 100 processes medical image data to generate a segmented image of the region of interest based on the medical image data, extracts one or more features from the segmented image, and uses the extracted features to create a pre-learned invasive cancer diagnostic model. You can perform an operation to generate diagnostic information by entering it in .

When processing to generate a segmented image of a region of interest based on medical image data, the processor 100 may generate one or more segmented images. For example, the processor may generate a first segment image including a subtraction image, a second segment image including an intratumor image, and a third segment image including the intratumoral and tumor periphery images. A segmented image containing an image can be created.

Medical image data may include medical images of invasive cancer and the area around the invasive cancer. At this time, the medical data may include multidimensional medical image data composed of discrete image elements. For example, medical data may be collected by computed tomography (CT), magnetic resonance imaging (MRI), fundus imaging, ultrasound, or any other medical imaging system known in the art. It may be a medical image of a subject.

When extracting one or more features from a segmented image, the processor 100 may extract one or more radiomic signatures from the segmented image. At this time, the radiomic signature may include primary features including shape features and intensity features, and secondary features including texture features. At this time, the shape features may include Elongation, Flatness, LeastAxisLength, MajorAxisLength, Maximum2DDiameterColumn, Maximum2DDiameterRow, Maximum2DDiameterSlice, Maximum3DDiameter, MeshVolume, MinorAxisLength, Sphericity, SurfaceArea, SurfaceVolumeRatio, and VoxelVolume. . The intensity features are 10Percentile, 90Percentile, Energy, Entropy, InterquartileRange, Kurtosis, Maximum, MeanAbsoluteDeviation, Mean, Median, Minimum, Range, RobustMeanAbsoluteDeviation, RootMeanSquared, Skewness, TotalEnergy, Uniformity , may include Variance. The texture features may include a intensity co-occurrence matrix (GLCM), a intensity dependence matrix (GLDM), a intensity interval length matrix (GLRLM), a intensity magnitude zone matrix (GLSZM), and a neighboring grayscale difference matrix (NGTDM). .

When extracting one or more features from a segmented image, the processor 100 extracts Surface Volume Ratio, Sphericity, Energy, Dependence Variance, and Area from the segmented image. ZonePercentage, Skewness, Large Area Emphasis, Elongation, Large Area High Gray Level Emphasis, Zone Variance and Short-Term High Gray One or more radiomic signatures selected from the group consisting of Level Emphasis (Short Run High Gray Level Emphasis) can be extracted.

When extracting one or more features from a segmented image, the processor 100 may extract one or more features using the PyRadiomics module.

The learned invasive cancer diagnosis model may be a machine learning model or a deep neural network. If the learned invasive cancer diagnosis model is a machine learning model, the invasive cancer diagnosis model is Random Forest, Naive Bayes Decision Tree, Bayesian network, or Support Vector Machine. It may include a (support vector machine; SVM) algorithm. When the learned invasive cancer diagnosis model is a deep neural network, the deep neural network may include an input layer and an output layer, or may include a plurality of separate hidden layers other than the input layer and the output layer. Deep neural networks include convolutional neural network (CNN), recurrent neural network (RNN), restricted Boltzmann machine (RBM), and deep belief network (DBN). , may include a Q network, U network, Siamese network, etc., and for example, a deep neural network may be a convolutional neural network.

The invasive cancer diagnosis model may be learned by features including radiomic signatures. The radiomic signature may include primary features including shape features and intensity features, and texture features. For example, the invasive cancer diagnosis model includes Surface Volume Ratio, Sphericity, Energy, Dependence Variance, ZonePercentage, Skewness, and Large Area. A group consisting of Large Area Emphasis, Elongation, Large Area High Gray Level Emphasis, Zone Variance, and Short Run High Gray Level Emphasis. It may be learned by a radiomic signature containing one or more features selected from .

The processor 100 may be comprised of one or more cores, and may include a central processing unit (CPU), a graphics processing unit (GPU), or a tensor processing unit (TPU) of a computing device. It may include a processor for data analysis, machine learning, or deep learning. The processor may read a computer program stored in a memory and perform data processing for machine learning according to an embodiment. According to one embodiment, the processor may perform calculations for learning a neural network. The processor performs calculations for neural network learning, such as processing input data for learning in deep learning (DL), extracting features from input data, calculating errors, and updating the weights of the neural network using backpropagation. It can be done. At least one of the CPU, GPU, and TPU of the processor 100 may process learning of the network function.

The memory 200 includes a flash memory type, hard disk type, multimedia card micro type, card type memory (for example, SD or XD memory, etc.), RAM (Random Access Memory), SRAM (Static Random Access Memory), ROM (Read-Only Memory; ROM), EEPROM (Electrically Erasable Programmable Read-Only Memory), PROM (Programmable Read-Only Memory), magnetic memory, It may include at least one type of storage medium among magnetic disks and optical disks.

A computing device that performs an operation to provide information for diagnosing invasive cancer according to an embodiment includes a network unit for transmitting and receiving brain images and other data as needed, and a network unit for storing and managing data used in other computing processes. Additional databases may be included. Accordingly, a computing device that performs an operation to provide information for diagnosing invasive cancer according to an embodiment may be capable of wired or wireless communication with imaging devices including an MRI imaging device and a PET imaging device. Additionally, data such as weight data, cerebral cortex thickness data, and biomarker accumulation data can be stored in a database, and data stored in the database can be managed, such as deletion and modification, by external input authorized by the user.

Figure 2 is a block diagram illustrating a process for providing information for diagnosing invasive cancer according to an embodiment of the present invention.

Referring to FIG. 2, a computing device that provides information for diagnosing invasive cancer according to an embodiment generates a segmented image of a region of interest based on medical image data, extracts one or more features from the segmented image, and extracts Diagnostic information can be generated by inputting the obtained features into a pre-learned invasive cancer diagnosis model.

At this time, the computing device generates a first segment image including a subtraction image, a second segment image including an intratumor image, and a third segment image including an intratumoral and tumor periphery image. A segmented image containing can be generated.

When extracting one or more features from a segmented image, the computing device may extract one or more radiomic signatures from the segmented image. At this time, the radiomic signature may include primary features including shape features and intensity features, and secondary features including texture features. At this time, the shape features may include Elongation, Flatness, LeastAxisLength, MajorAxisLength, Maximum2DDiameterColumn, Maximum2DDiameterRow, Maximum2DDiameterSlice, Maximum3DDiameter, MeshVolume, MinorAxisLength, Sphericity, SurfaceArea, SurfaceVolumeRatio, and VoxelVolume. . The intensity features are 10Percentile, 90Percentile, Energy, Entropy, InterquartileRange, Kurtosis, Maximum, MeanAbsoluteDeviation, Mean, Median, Minimum, Range, RobustMeanAbsoluteDeviation, RootMeanSquared, Skewness, TotalEnergy, Uniformity , may include Variance. The texture features may include a intensity co-occurrence matrix (GLCM), a intensity dependence matrix (GLDM), a intensity interval length matrix (GLRLM), a intensity magnitude zone matrix (GLSZM), and a neighboring grayscale difference matrix (NGTDM). .

When extracting one or more features from a segmented image, the computing device extracts Surface Volume Ratio, Sphericity, Energy, Dependence Variance, and Area Percentage ( ZonePercentage, Skewness, Large Area Emphasis, Elongation, Large Area High Gray Level Emphasis, Zone Variance, and Short-term High Gray Level Emphasis One or more radiomic signatures selected from the group consisting of (Short Run High Gray Level Emphasis) can be extracted.

When extracting one or more features from a segmented image, the computing device may extract one or more features using the PyRadiomics module.

Hereinafter, the present invention will be described in more detail through the following examples. However, these examples are only for illustrating the present invention, and the scope of the present invention is not limited by these examples.

Example 1: MRI image and region of interest data collection

T1-weighted imaging (T1WI) images and clinical information of patients with ductal carcinoma in situ and invasive carcinoma were collected from Chonnam National University Hwasun Hospital. The data obtained was 359 cases, including the two patient groups. The collected data was divided into training, validation, and test sets as shown in Table 1 below.

Cohort Upstaged DCIS Pure DCIS Total Upstaged : Pure Training + Validation (80%) 121 167 288 1:1.38 Test (20%) 30 41 71 1:1.37

Example 2: Region of Interest Segmentation

In order to perform semi-automatic segmentation of the region of interest, 3D Slicer (ver.4.11, http://download.slicer.org/) was used to check whether the 3D values were well segmented. The accuracy of the semi-automatically segmented images was evaluated through the results confirmed by a radiologist, and is shown in Figure 3. As can be seen in Figure 3, the semi-automatically segmented image used a difference image (subtraction image), which is an image obtained by subtracting the pre-contrast image (pre-contrast) from the post-contrast image (post-contrast). Then, image features were extracted by dividing the regions-of-interest of the intratumor (intratumor) and the tumor periphery (intratumor+peritumor). At this time, since there is a biological possibility that tumor cells exist in areas outside the lesion, segmented images of the tumor periphery were extracted taking this into consideration.

Example 3: Feature extraction

Visual image analysis using the naked eye is influenced by the skill level and subjective experience of the medical staff, and quantitative analysis is difficult, which limits the ability to objectively diagnose lesions. Radiomics can extract objective information that cannot be obtained from existing visual image analysis by extracting lesion characteristics from image data using mathematical and statistical techniques. Therefore, in order to express the characteristics of ductal carcinoma in situ and invasive carcinoma in quantitative numbers, radiomics-based feature extraction was performed on MR images. First order features (first order) and second order features (second order) were extracted using Python's PyRadiomics module. ) were extracted, and the extracted features are shown in Table 2.

From the primary features, shape features and intensity features were extracted. In the secondary features, gray-level co-occurrence matrix (GLCM), gray-level dependence matrix (GLDM), gray-level run length matrix (GLRLM), and grayscale size. By extracting gray-level size zone matrix (GLSZM) and neighboring gray tone difference matrix (NGTDM) features, the total number of features extracted per image was 107.

Feature Category Feature Name Shape (14) Elongation, Flatness, LeastAxisLength, MajorAxisLength, Maximum2DDiameterColumn, Maximum2DDiameterRow, Maximum2DDiameterSlice, Maximum3DDiameter, MeshVolume, MinorAxisLength, Sphericity, SurfaceArea, SurfaceVolumeRatio, VoxelVolume Intensity (18) 10Percentile, 90Percentile, Energy, Entropy, InterquartileRange, Kurtosis, Maximum, MeanAbsoluteDeviation, Mean, Median, Minimum, Range, RobustMeanAbsoluteDeviation, RootMeanSquared, Skewness, TotalEnergy, Uniformity, Variance Texture GLCM (24) Autocorrelation, ClusterProminence, ClusterShade, ClusterTendency, Contrast, Correlation, DifferenceAverage, DifferenceEntropy, DifferenceVariance, Id, Idm, Idmn, Idn, Imc1, Imc2, InverseVariance, JointAverage, JointEnergy, JointEntropy, MCC, MaximumProbability, SumAverage, SumEntropy, SumSquares GLDM (14) DependenceEntropy, DependenceNonUniformity, DependenceNonUniformityNormalized, DependenceVariance, GrayLevelNonUniformity, GrayLevelVariance, HighGrayLevelEmphasis, LargeDependenceEmphasis, LargeDependenceHighGrayLevelEmphasis, LargeDependenceLowGrayLevelEmphasis, LowGrayLevelEmphasis, SmallDependenceE mphasis, SmallDependenceHighGrayLevelEmphasis, SmallDependenceLowGrayLevelEmphasis GLRLM (16) GrayLevelNonUniformity, GrayLevelNonUniformityNormalized, GrayLevelVariance, HighGrayLevelRunEmphasis, LongRunEmphasis, LongRunHighGrayLevelEmphasis, LongRunLowGrayLevelEmphasis, LowGrayLevelRunEmphasis, RunEntropy, RunLengthNonUniformity, RunLengthNonUniformityNormalized, Run Percentage, RunVariance, ShortRunEmphasis, ShortRunHighGrayLevelEmphasis, ShortRunLowGrayLevelEmphasis GLSZM (16) GrayLevelNonUniformity, GrayLevelNonUniformityNormalized, GrayLevelVariance, HighGrayLevelZoneEmphasis, LargeAreaEmphasis, LargeAreaHighGrayLevelEmphasis, LargeAreaLowGrayLevelEmphasis, LowGrayLevelZoneEmphasis, SizeZoneNonUniformity, SizeZoneNonUniformityNormalized, SmallAreaEmphasis, SmallAreaHighGrayLevelEmphasis, SmallAreaLowGrayLevelEmphasis, ZoneEntropy, ZonePercentage, ZoneVariance NGTDM (5) Busyness, Coarseness, Complexity, Contrast, Strength

Example 4: Feature Selection

Because features extracted from images using various methods and filters contain overlapping information that is related to each other, learning a prediction model using all extracted feature information may have a negative impact on the accuracy of the model or cause overfitting of the model. It can react sensitively to noise. At this time, the selected radiomics features must be stable features that can be reproduced, must be related to the value to be predicted, and the correlation between the extracted features must be low.

Accordingly, in order to reduce overfitting, normalization is performed by assigning a penalty corresponding to the absolute value and square of the coefficient size, and the least absolute shrinkage and selection operator (LASSO) of the embedded method with an internal penalty tendency is applied as follows. Eleven features as shown in Table 3 were selected.

Feature
(Category) Mean Median Standard Deviation Upstaged Pure Upstaged Pure Upstaged Pure SurfaceVolumeRatio*
(Shape) 0.7326 0.9497 0.7185 0.9132 0.2303 0.3335 Sphericity* (Shape) 0.4214 0.4703 0.3918 0.4504 0.1487 0.1934 Energy* (Intesity) 109058.9594 90159.7419 53092.4360 23824.9174 143132.1958 169297.3482 DependenceVariance*(GLDM) 28.9512 28.5930 28.7853 29.0125 3.3223 3.3959 ZonePercentage(GLSZM) 0.0007 0.0018 0.0003 0.0007 0.0012 0.00226 Skewness* (Intensity) 0.4235 0.5297 0.4161 0.4780 0.3277 0.3243 LargeAreaEmphasis (GLSZM) 242298922.3590 100603412.9484 22155849 3060702.5 1213173292.1354 343494829.8760 Elongation (Shape) 0.6734 0.6419 0.6993 0.6460 0.1704 0.1583 LargeAreaHighGrayLevelEmphasis* (GLSZM) 268959246.4086 100603412.9484 22155849 3060702.5 1265503190.6186 343494829.8760 ZoneVariance (GLSZM) 53836937.3521 53053781.9475 0 0 216806105.7610 201461345.4732 ShortRunHighGrayLevelEmphasis* (GLRLM) 0.24760 0.27457 0.24341 0.27053 0.06021 0.07393

Example 5: Support Vector Machine (SVM)

Lastly, in order to select and analyze a statistical model or machine learning model suitable for the results to be predicted, when a given data set is expressed as a boundary in the mapped space, the boundary with the largest sum of the distance between the hyperplane and the data is selected. It was analyzed using the support vector machine, a search algorithm. When conducting statistical analysis of radiomics, nonlinear classification was applied and a polynomial kernel was used. As can be seen from the experimental results in Table 4 and Figure 4, the model overfits as the hyperplane degree increases, and the best result was shown when the degree was 2. The AUC was over 0.76, showing high diagnostic usefulness. Confirmed.

Average of Validation 4-fold Independent Test Cohort Sen (%) Spe (%) ACC (%) AUC Sen (%) Spe (%) ACC (%) AUC 61.881 79.094 71.875 0.734 60.000 82.927 73.239 0.764

Claims (15)

In a method of providing information for diagnosing invasive cancer from medical image data by a computing device,
A processing step of generating a segmented image of a region of interest based on medical image data;
An extraction step of extracting one or more features from the segmented image; and
A generation step of generating diagnostic information by inputting the extracted features into a pre-learned invasive cancer diagnosis model;
Method for providing information for diagnosing invasive cancer, including. The method of claim 1, wherein the segmented image is:
A first segmented image including a subtraction image;
a second segmented image including an intratumor image; and
Third segment image including intratumoral and peritumor images;
A method of providing information for diagnosing invasive cancer, comprising: The method of claim 1, wherein the feature is a radiomic signature. The method of claim 3, wherein the radiomic signature is:
A method of providing information for diagnosing invasive cancer, comprising primary features including shape features and intensity features, and secondary features including texture features. The method of claim 1, wherein the extraction step is performed through the PyRadiomics module. The method of claim 4, wherein the radiomic signature is:
Surface Volume Ratio, Sphericity, Energy, Dependence Variance, ZonePercentage, Skewness, Large Area Emphasis, Extension ( A measure of one or more features selected from the group consisting of Elongation, Large Area High Gray Level Emphasis, Zone Variance, and Short Run High Gray Level Emphasis. A method of providing information for diagnosing invasive cancer, which is calculated based on. The method of claim 1, wherein the invasive cancer diagnosis model is:
A method of providing information for diagnosing invasive cancer, which is a support vector machine (SVM). A computing device for providing information for diagnosing invasive cancer, comprising: a processor including one or more cores; and a memory, wherein the processor generates a segmented image of the region of interest based on medical image data, extracts one or more features from the segmented image, and applies the extracted features to normal and pre-learned infiltrates through diagnostic information. A computing device that generates diagnostic information by inputting it into a cancer diagnosis model. The method of claim 8, wherein the segmented image is:
A first segmented image including a subtraction image;
a second segmented image including an intratumor image; and
Third segment image including intratumoral and peritumor images;
A computing device comprising: The computing device of claim 8, wherein the feature is a radiomic signature. The method of claim 10, wherein the radiomic signature is:
A computing device comprising primary features including shape features and intensity features, and secondary features including texture features. A computer program stored in a storage medium, where the computer program, when executed on one or more processors, performs the following operations to provide information for diagnosing invasive cancer, the operations being: Segmentation of a region of interest based on medical image data. a processing operation that generates an image; An extraction operation for extracting one or more features from a segmented image; and a generating operation of generating diagnostic information by inputting the extracted features into a pre-learned invasive cancer diagnostic model. The method of claim 12, wherein the segmented image is:
A first segmented image including a subtraction image;
a second segmented image including an intratumor image; and
Third segment image including intratumoral and peritumor images;
A computer program stored on a storage medium that includes a. 13. The computer program of claim 12, wherein the feature is a radiomic signature. The method of claim 14, wherein the radiomic signature is:
A computer program stored in a storage medium, comprising primary features including shape features and intensity features, and secondary features including texture features.