Artificial intelligence model for predicting BRCA gene mutations by using magnetic resonance imaging, and use thereof

Abstract

The present invention relates to an artificial intelligence model and a method for implementing same, the artificial intelligence model being capable of extracting significant radiomics features from magnetic resonance imaging (MRI) of a prostate cancer patient and non-invasively predicting gene mutations. The present invention selects radiomics features having a close correlation with BRCA mutations and predicts BRCA mutations with high accuracy by applying various binary classification algorithms on the basis of the radiomics features. In addition, by combining various feature selection methods with a binary classifier, the present invention establishes that a BRCA mutation prediction model composed of a combination of forward feature selection (FFS) and a random forest (RF) classifier learns characteristics of BRCA gene mutations more effectively than other combinations and also exhibits high performance in actual prediction, thereby implementing a learning model for predicting BRCA mutations with high prediction reliability.

Description

Artificial intelligence model for predicting BRCA gene mutations using magnetic resonance imaging and its use

[0001] The present invention relates to an artificial intelligence model that extracts radiomics features from magnetic resonance imaging (MRI) and predicts whether or not a BRCA gene mutation is present based on the extracted features, and to a use thereof.

[0002]

[0003] Prostate cancer is one of the most common male cancers worldwide. While early detection offers a relatively good prognosis, metastatic prostate cancer is difficult to treat and has a sharply reduced survival rate. Consequently, the need for technologies that not only detect prostate cancer early but also proactively identify high-risk patients at high risk for metastasis is growing.

[0004] The BRCA1 gene is located on chromosome 17 and consists of 23 exons, while the BRCA2 gene is located on chromosome 13 and consists of 26 exons. While these genes were initially associated with an increased risk of breast and ovarian cancer, subsequent research has revealed that they are also associated with various cancers, including male prostate, pancreatic, and colon cancers. Recently, it has been reported that mutations in the BRCA genes significantly increase the risk of developing metastatic cancer due to impaired DNA repair function. In particular, BRCA2 mutations are closely associated with a poor prognosis in prostate cancer patients, and patients with these mutations often exhibit more aggressive cancer progression and lower responsiveness to existing treatments. Therefore, identifying BRCA gene mutations in prostate cancer patients is essential for precise prognosis prediction and treatment response assessment, and can serve as an important criterion for establishing personalized treatment strategies.

[0005] Currently, genetic sequencing-based analysis methods using blood or tissue samples are used to determine whether or not a BRCA gene mutation exists. However, these methods have several problems, including expensive equipment, long analysis times, limited accessibility, and patient reluctance due to invasiveness. Therefore, demand for simpler and non-invasive alternative technologies is increasing.

[0006] Magnetic resonance imaging (MRI), one of these noninvasive approaches, is a widely used imaging technique for diagnosing, staging, and assessing treatment response to prostate cancer. It can assess not only tumor size, margins, and extent of invasion, but also functional changes in the tissue. MRI, in particular, offers the advantage of providing quantitative imaging information that can be used to predict patient prognosis and assess risk.

[0007] Prostate cancer patients harboring BRCA gene mutations exhibit specific imaging characteristics on MRI, and research into imaging-based genetic mutation prediction techniques utilizing these characteristics is actively underway. However, technology that can reliably predict BRCA gene mutations in prostate cancer patients based on MRI remains insufficiently established. Therefore, there is a pressing need to develop a predictive model that can link imaging and genetic information for precise diagnosis and treatment planning.

[0008]

[0009] Numerous papers and patents are referenced and cited throughout this specification. The disclosures of these cited papers and patents are incorporated herein by reference in their entirety to provide a clearer understanding of the state of the art and the scope of the present invention.

[0010]

[0011] The present inventors have made extensive research efforts to develop a method for rapidly and reliably predicting the presence or absence of mutations in the BRCA gene, which is closely correlated with a poor prognosis in prostate cancer patients. As a result, we extracted image information significant for determining BRCA gene mutations from magnetic resonance images (MRIs) of prostate cancer patients, and selected features important for mutation prediction among the information to develop a mutation prediction model that can rapidly predict the presence or absence of BRCA mutations. The present invention identifies key image features significant for predicting BRCA gene mutations through machine learning-based analysis and utilizes them, thereby implementing image features that can function as markers for predicting BRCA gene mutations and a mutation prediction model utilizing the same.

[0012] Accordingly, the purpose of the present invention is to provide a method for providing information necessary for predicting BRCA gene mutations in cancer patients.

[0013] Another object of the present invention is to provide a device for predicting BRCA gene mutations in cancer patients.

[0014] Another object of the present invention is to provide a learning method of an artificial intelligence model for predicting BRCA gene mutations in cancer patients.

[0015]

[0016] Other objects and advantages of the present invention will become more apparent from the detailed description, claims and drawings below.

[0017]

[0018] According to one aspect of the present invention, the present invention provides a method for providing information necessary for predicting a BRCA gene mutation in a cancer patient, comprising the following steps.

[0019] (a) A step of extracting digitized image information from a magnetic resonance image (MRI) of a cancer patient with segmented lesions;

[0020] (b) a step of performing consistency analysis on the extracted image information;

[0021] (c) a step of performing a feature selection process on the image information for which the above consistency has been analyzed;

[0022] (d) A step of applying a machine learning model that predicts the presence or absence of a BRCA gene mutation using the above-mentioned selected features as input values.

[0023]

[0024] To accurately predict the presence of mutations in the BRCA gene, which are closely correlated with poor cancer prognosis, the present inventors selected significant features for BRCA mutation detection from magnetic resonance imaging (MRI) of cancer patients and developed an artificial intelligence model that can rapidly predict the presence or absence of BRCA mutations based on these features. This prediction model performs a concordance analysis on the numerical features extracted from the images to identify features highly correlated with BRCA mutations and utilizes various classification algorithms to compare and evaluate the predictive performance to produce highly reliable results.

[0025] Accordingly, the present invention can be utilized as an efficient alternative technology that can replace or complement next-generation sequencing (NGS), which requires invasive and expensive equipment. Furthermore, the predictive model of the present invention can accurately predict the presence or absence of BRCA gene mutations based on patient images, thereby contributing substantially to individual patient genetic risk assessment, treatment strategy determination, prognosis estimation, and the implementation of precision medicine.

[0026]

[0027] The term "BRCA gene (breast cancer gene)" used herein is a tumor suppressor gene that performs the role of repairing damaged DNA that can develop into cancer and regulating cell division, etc., and the BRCA gene in the present invention includes both BRCA 1 and BRCA 2. The BRCA 1 gene is located on chromosome 17 and consists of 22 exons, and the BRCA 2 gene is located on chromosome 13 and consists of 26 exons. Furthermore, the BRCA gene is closely related to the development and prognosis of prostate cancer, and if there is a mutation in the gene, the DNA repair function is reduced, causing genomic instability, and thus the risk of developing cancer can significantly increase. In addition, the BRCA gene is also closely correlated with the poor prognosis of prostate cancer patients. It has been reported that prostate cancer patients with pathological mutations in the BRCA gene have a faster disease progression rate, a high metastasis rate, and a shorter survival period compared to general prostate cancer patients. In particular, patients with mutations in the BRCA 2 gene have a lower responsiveness to hormone therapy and a higher risk of progression to metastatic prostate cancer.

[0028] The term "mutation" as used herein refers to a case where a permanent or temporary change occurs in the base sequence of an existing gene, and includes both a case where there is an abnormality in the structure or number of chromosomes (chromosomal abnormality) depending on the effect on the structure of the genetic material, and a case where a small-scale change occurs in the base sequence of the gene. Cases where there is an abnormality in the structure of the chromosome include deletion, duplication, inversion, and translocation, and a deletion mutation in which a part of the chromosome is lost, and a point mutation in which the nucleotide sequence constituting the gene locus is deleted over one or several thousand pairs are also included in the mutation. More specifically, the BRCA gene mutation of the present invention refers to a permanent or temporary change in a gene that directly or indirectly significantly affects the prognosis or treatment responsiveness of a cancer patient.

[0029] The term "metastasis" as used herein refers to the process by which cancer cells break off from a primary tumor (the cancer that first occurred) and migrate to other parts of the body, forming new tumors at the site of migration. Metastasis is a critical stage that increases the lethality of cancer, and it is known that more than 90% of cancer deaths are due to metastasis (Nature Reviews Cancer, 2006, 6:449-458). The present invention can more accurately diagnose the metastasis of cancer, particularly prostate cancer. The mechanism by which cancer cells acquire motility during the metastatic process is explained by the epithelial to mesenchymal transition (EMT), in which tumor epithelial cells acquire the characteristics of mesenchymal cells through genetic mutations, and the opposite process, mesenchymal to epithelial transition (MET) (J Clin Invest. 2009, 119:1417-1419). That is, epithelial cells that have acquired mesenchymal characteristics have weakened cell-to-cell bonds, leave their original location, and migrate into blood vessels. The cells that migrated through the blood vessels regain their original epithelial characteristics (MET) and settle in secondary sites far from the primary site, thereby causing tumor growth.

[0030]

[0031] The term “magnetic resonance imaging (MRI)” used in this specification refers to a non-invasive imaging diagnostic technology that uses a high-frequency magnetic field to image the tissue structure inside the human body at high resolution, and in the present invention, it is used as basic image data for visualizing lesions in cancer patients and extracting image-based features.

[0032] The term “lesion segmentation” as used herein refers to a series of processes for identifying tumor tissue or a lesion region of interest on an MRI and independently identifying or extracting the lesion region from the entire image by defining a spatial boundary or mask of the region. Lesion segmentation can be performed through an automated artificial intelligence-based image segmentation algorithm, or can be manually segmented based on morphological features including boundary clarity and tissue density differences. In addition, it includes a series of processes in which segmentation is performed based on functional or quantitative image information, including signal intensity and temporal changes in the image, as well as morphological features. Therefore, “lesion segmentation” as used herein includes all procedures for recognizing a lesion or region of interest from an MRI and defining the region.

[0033] Therefore, in this specification, “magnetic resonance image of a cancer patient with segmented lesions” means image data in which tumor tissue or a lesion area of ​​interest is identified and segmented manually, semi-automatically or automatically from an magnetic resonance image, and boundary information or a mask of the lesion is defined, and the image includes a state in which it can be used as basic data for quantitative image feature extraction, analysis or artificial intelligence model learning in a later stage.

[0034]

[0035] The term “quantified image information” as used herein refers to various quantitative feature values ​​extracted from lesions or regions of interest within medical images such as magnetic resonance imaging (MRI). This information includes all information that can be extracted and quantified from magnetic resonance images, as well as the intensity, texture, shape, gradient, statistical distribution, and dynamic change of the image. More specifically, the quantified image information refers to features that reflect patterns, morphological changes, or functional changes within the image that may be correlated with the presence or absence of a BRCA gene mutation. The quantified image information can be quantified using specific image analysis software such as PyRadiomics.

[0036] The term "consistency analysis," as used herein, refers to a series of quantitative analysis processes for assessing the reproducibility and reliability of image information extracted from medical images. Specifically, it refers to the process of numerically assessing the consistency between information extracted from different viewpoints, readers, or imaging conditions, including calculating correlation coefficients, relative standard deviations, or intraclass correlation coefficients.

[0037] The term “feature selection process” used herein refers to a series of procedures for selecting key features that significantly influence the prediction of BRCA gene mutations from among digitized image information whose reliability and reproducibility have been confirmed through consistency analysis. The feature selection process is performed to significantly improve the accuracy of the prediction model and prevent overfitting, and includes not only all statistical methods including statistical testing and correlation analysis, but also various feature selection methods based on machine learning. Specifically, the feature selection process includes all of Corr (Correlation-based Selection), LASSO (Least Absolute Shrinkage and Selection Operator), FFS (Forward Feature Selection), RFE (Recursive Feature Elimination), PCA (Principal Component Analysis), and Mutual Information, but is not limited thereto, and various feature selection methods commonly used in the art may be included therein.

[0038]

[0039] The term “machine learning” as used herein refers to a statistical-based learning algorithm for predicting metastasis based on gene expression information collected from cancer patients, and includes, but is not limited to, supervised learning, unsupervised learning, and reinforcement learning. Specifically, the machine learning provided by the present invention performs a consistency analysis and feature selection process on digitized image information extracted from magnetic resonance imaging (MRI) of cancer patients, thereby predicting the presence or absence of a BRCA gene mutation with high reliability using image features that can function as markers of BRCA mutations as input values.

[0040]

[0041] According to a specific embodiment of the present invention, the cancer of the present invention is prostate cancer.

[0042] The term "prostate cancer" as used herein refers to cancer that develops in the prostate gland. The prostate gland is a gland in the male reproductive system that surrounds the urethra, just below the bladder. Cancer cells can spread to other parts of the body, particularly the bones and lymph nodes. While there may be no symptoms in the early stages, later stages can cause symptoms such as pain, difficulty urinating, blood in the urine, and pelvic or back pain. Benign prostatic hyperplasia (BPH) can cause similar symptoms, including fatigue due to low red blood cell counts. Factors that increase the risk of prostate cancer include older age, family history, and race, with approximately 99% of cases occurring after age 50.

[0043]

[0044] According to a specific embodiment of the present invention, the above-described consistency analysis of the method for providing information necessary for predicting a BRCA gene mutation in a cancer patient provided by the present invention is performed by extracting image information having an ICC (Intraclass Correlation Coefficient) value of 0.8 or higher.

[0045] The term "ICC (Intraclass Correlation Coefficient)" used herein refers to a statistical index value that quantitatively evaluates the consistency between data measured repeatedly on the same subject. A higher ICC value indicates less variability between measurements and higher reliability. The ICC value is determined by comparing the variance between images and the variance within the measurement error based on data measured repeatedly by multiple observers or at multiple time points on magnetic resonance images of cancer patients. Specifically, the ICC value is calculated using the following equation (1).

[0046]

[0047] [Formula 1]

[0048] ICC = σ between 2 / (σ between 2 +σ within 2 )

[0049] The above σ between 2 refers to the variance between data measured by multiple observers on magnetic resonance images of cancer patients,

[0050] The above σ within 2 refers to the variance corresponding to the measurement error that occurs between observers for images of the same cancer patient.

[0051] As described above, the present invention performed a concordance analysis to evaluate the reproducibility of digitized image information extracted from medical images, and predicted the presence or absence of a BRCA mutation based on highly reliable image information by extracting only image information having an ICC value of 0.8 or higher. More specifically, the concordance analysis of the present invention can be performed by extracting image information having an ICC value of 0.85 or higher, and even more specifically by extracting image information having an ICC value of 0.9 or higher, even more specifically by extracting image information having an ICC value of 0.93 or higher, image information having an ICC value of 0.94 or higher, and most specifically by extracting image information having an ICC value of 0.95 or higher.

[0052]

[0053] According to a specific embodiment of the present invention, the above-described match analysis of the method for providing information necessary for predicting a BRCA gene mutation in a cancer patient provided by the present invention additionally performs positive sample amplification.

[0054] In order to implement higher reliability and improve imbalance between data, the present invention can additionally perform positive sample amplification on image information having an ICC value of 0.8 or higher.

[0055] The term “positive sample amplification” as used herein refers to a series of procedures for correcting imbalances between data and improving the sensitivity and prediction accuracy of the model for a patient group with a BRCA gene mutation by augmenting or repeatedly using digitized image information of patients with a BRCA gene mutation during the training process of a machine learning model.

[0056] More specifically, the positive sample amplification in the present invention includes, but is not limited to, oversampling, data augmentation, and weight adjustment for magnetic resonance images of cancer patients with BRCA gene mutations, and includes all procedures commonly used in the art to artificially augment or repeat the digitized image information of patients with BRCA gene mutations.

[0057] According to a specific embodiment of the present invention, the feature selection process of the method for providing information necessary for predicting a BRCA gene mutation in a cancer patient provided by the present invention is performed by No FS (NO Feature Selection), Corr (Correlation-based Selection), LASSO (Least Absolute Shrinkage and Selection Operator) or FFS (Forward Feature Selection).

[0058] The term “No FS (NO Feature Selection)” used in this specification refers to a method of using all digitized information as input variables of a machine learning model, and feature loss can be prevented by utilizing all extracted image features for learning.

[0059] The term “Corr (Correlation-based Selection)” used in this specification refers to a method of evaluating the significance of features based on the correlation coefficient between each feature and the target variable to be predicted, and selecting only features above a certain threshold or with a high rank, which is effective in removing duplication between features.

[0060] The term "Least Absolute Shrinkage and Selection Operator (LASSO)" used herein refers to a regression analysis-based feature selection method that automatically selects features by adding an L1 regularization term to the objective function to reduce the regression coefficients of unnecessary features to zero. This method simultaneously performs variable selection and regularization, preventing overfitting in high-dimensional data and improving the model's generalization ability.

[0061] The term “Forward Feature Selection (FFS)” used in this specification is a sequential feature selection technique that determines the optimal feature combination by sequentially adding features that contribute to improving classification performance. It does not initially include any features, but evaluates the performance of the model while repeatedly adding features one by one.

[0062]

[0063] According to a specific embodiment of the present invention, the feature selection process of the method for providing information necessary for predicting BRCA gene mutations in cancer patients provided by the present invention is performed by FFS (Forward Feature Selection).

[0064] As described below, the inventors experimentally demonstrated that the combination of Forward Feature Selection (FFS) feature selection and a Random Forest classifier dramatically improved BRCA gene mutation prediction performance compared to other feature selection methods and classifiers. This combination demonstrated superior performance compared to other combinations in various performance indicators, including sensitivity, specificity, precision, and AUC (Area Under the Curve), for BRCA gene mutation prediction, demonstrating its potential for practical clinical application.

[0065]

[0066] According to a specific embodiment of the present invention, the step of applying a machine learning model of the method for providing information necessary for predicting a BRCA gene mutation in a cancer patient provided by the present invention is performed by Adab (Adaptive Boosting), RF (Random Forest), SGD (Stochastic Gradient Descent), or LDA (Linear Discriminant Analysis).

[0067] The term “Adaptive Boosting” used in this specification refers to an ensemble learning method that repeatedly trains a weak classifier and sequentially combines multiple classifiers to create a strong classifier with excellent predictive performance, and at each iteration step, a higher weight is given to samples that were misclassified in the previous step to improve the accuracy of the classifier.

[0068] The term "RF (Random Forest)" used in this specification refers to an ensemble-based classification technique that trains multiple decision trees and then makes final predictions using majority voting or averaging. Each tree is trained based on randomly selected data samples and feature sets, ensuring model diversity and predictive stability.

[0069] The term "Stochastic Gradient Descent (SGD)" used in this specification refers to a learning method that iteratively updates weights to minimize a model's loss function using small, randomly selected data batches (mini-batches) rather than the entire dataset. Its high computational efficiency allows it to be effectively applied to large-scale data and high-dimensional feature spaces.

[0070] The term “Linear Discriminant Analysis (LDA)” as used herein refers to a linear classification technique that maximizes the separability between classes by linearly transforming data in a direction that maximizes between-class variance and minimizes within-class variance.

[0071]

[0072] According to a specific embodiment of the present invention, the step of applying a machine learning model of the method of providing information necessary for predicting a BRCA gene mutation in a cancer patient provided by the present invention is performed by RF (Random Forest).

[0073] The present invention developed an artificial intelligence model that can quickly and accurately predict the presence or absence of a BRCA mutation by combining feature selection using the Forward Feature Selection (FFS) method and a Random Forest classifier. The artificial intelligence model extracted 32 features with an ICC value of 0.9 or higher from a total of 1,422 image information and applied them to a training set. As a result, the model showed high prediction performance with an AUROC (Area Under the Receiver Operating Characteristic Curve) value exceeding 0.8 for the training data set (Test set), experimentally proving that the model of the present invention can classify the presence or absence of a BRCA mutation with significantly higher reliability compared to cases using other prediction models.

[0074]

[0075] According to another aspect of the present invention, the present invention provides a device for predicting BRCA gene mutation in a cancer patient, comprising the following configuration.

[0076] (a) A receiving unit for receiving a magnetic resonance image (MRI) of a cancer patient with segmented lesions;

[0077] (b) an extraction unit that extracts digitized image information from the segmented lesion;

[0078] (c) an analysis unit that performs a consistency analysis process on the extracted image information;

[0079] (d) a feature selection unit that performs a feature selection process on the image information for which the above consistency has been analyzed;

[0080] (e) A prediction unit that applies a machine learning model to predict the presence or absence of a BRCA gene mutation using the above-selected features as input values.

[0081] BRCA, mutations, lesion segmentation, magnetic resonance imaging, digitized image information, concordance analysis, feature selection process, and the meaning of machine learning have already been described above, so their description is omitted to avoid excessive duplication.

[0082] The present inventors have developed an artificial intelligence model that can noninvasively and rapidly predict the presence or absence of BRCA gene mutations by implementing a prediction device consisting of a receiving unit, an extraction unit, an analysis unit, a feature selection unit, and a prediction unit. The prediction device can predict the presence or absence of BRCA mutations in cancer patients using only magnetic resonance imaging without expensive genetic testing, and thus serves as an efficient alternative that can complement or replace existing diagnostic technologies. In addition, the present invention implements the entire process from image analysis preprocessing to the prediction stage as a single integrated system, making it easy to automate and apply clinically.

[0083]

[0084] The present invention selects image features with an ICC (Intraclass Correlation Coefficient) value of 0.9 or higher from a total of 1,422 image information, and applies a machine learning classifier based on the selected image features to accurately predict the presence or absence of a BRCA gene mutation. In addition, as a result of applying various feature selection methods to high-confidence image information with an ICC value of 0.9 or higher, it was confirmed that the number of extracted image features varied depending on the method used. Specifically, the LASSO (Least Absolute Shrinkage and Selection Operator) method extracted 39 features, the FFS (Forward Feature Selection) method extracted 32 features, the Corr (Correlation-based Selection) method extracted 65 features, and the No FS (NO Feature Selection) method extracted 250 features.

[0085] Accordingly, the inventors of the present invention identified key features that contribute significantly to BRCA gene mutation prediction based on the characteristic that the number and composition of image features ultimately utilized vary depending on the feature selection method, and utilized these features for model interpretation and performance analysis.

[0086]

[0087] According to another aspect of the present invention, the present invention provides a method for learning an artificial intelligence model for predicting BRCA gene mutation in a cancer patient, comprising the following steps.

[0088] (a) receiving a magnetic resonance image (MRI) of a cancer patient with segmented lesions;

[0089] (b) a step of extracting digitized image information from the segmented lesion;

[0090] (c) a step of performing a consistency analysis process on the extracted image information;

[0091] (d) a step of performing a feature selection process on the image information for which the above consistency has been analyzed;

[0092] (e) A step of training a machine learning model that predicts the presence or absence of a BRCA gene mutation using the above-mentioned selected features as input values.

[0093] The BRCA gene, mutations, lesion segmentation, magnetic resonance imaging, digitized image information, concordance analysis, feature selection process, and machine learning have already been described above, so their description is omitted to avoid excessive duplication.

[0094]

[0095] The features and advantages of the present invention are summarized as follows:

[0096] (a) The present invention extracts image features having a specific correlation with BRCA gene mutations from magnetic resonance images of cancer patients and trains a machine learning model with the image features to quickly and accurately predict whether a patient has a BRCA gene mutation that is closely correlated with poor prognosis and treatment responsiveness.

[0097] (b) The present invention can be utilized as an efficient alternative technology that can replace or supplement next-generation sequencing (NGS), which requires invasive and expensive equipment, by rapidly predicting the presence or absence of a BRCA gene mutation with high reliability based on magnetic resonance imaging.

[0098]

[0099] Figure 1 is a schematic diagram of the overall learning and prediction process of the BRCA mutation prediction model.

[0100] Figure 2 is a diagram showing the overall performance of a BRCA mutation prediction model consisting of a combination of the FFS feature selection method and the Adab classifier. Figure 2a shows the AUROC (Area Under the Receiver Operating Characteristic Curve) value of the training data set, and Figure 2b shows the AUROC of the validation data set. Figure 2c shows the image features that contributed to the prediction decision of the training data set, ranked by rank, and Figure 2d shows the image features that contributed to the prediction decision of the validation data set, ranked by rank.

[0101] Figure 3 is a diagram showing the overall performance of a BRCA mutation prediction model consisting of a combination of the LASSO feature selection method and the Adab classifier. Figure 3a shows the AUROC (Area Under the Receiver Operating Characteristic Curve) value of the training data set, and Figure 3b shows the AUROC of the validation data set. Figure 3c shows the image features that contributed to the prediction decision of the training data set, ranked by rank, and Figure 23 shows the image features that contributed to the prediction decision of the validation data set, ranked by rank.

[0102] Figure 4 is a diagram showing the overall performance of a BRCA mutation prediction model consisting of a combination of a LASSO feature selection method and an RF classifier. Figure 4a shows the AUROC (Area Under the Receiver Operating Characteristic Curve) value of the training data set, and Figure 4b shows the AUROC of the validation data set. Figure 4c shows the image features that contributed to the prediction decision of the training data set, ranked by rank, and Figure 4d shows the image features that contributed to the prediction decision of the validation data set, ranked by rank.

[0103] Figure 5 is a diagram showing the overall performance of a BRCA mutation prediction model consisting of a combination of the LASSO feature selection method and the SGD classifier. Figure 5a shows the AUROC (Area Under the Receiver Operating Characteristic Curve) value of the training data set, and Figure 5b shows the AUROC of the validation data set. Figure 5c shows the image features that contributed to the prediction decision of the training data set, ranked by rank, and Figure 5d shows the image features that contributed to the prediction decision of the validation data set, ranked by rank.

[0104] Figure 6 is a diagram showing the overall performance of a BRCA mutation prediction model composed of a combination of the LASSO feature selection method and the LDA classifier. Figure 6a shows the AUROC (Area Under the Receiver Operating Characteristic Curve) value of the training data set, and Figure 6b shows the AUROC of the validation data set. Figure 6c shows the image features that contributed to the prediction decision of the training data set, ranked by rank, and Figure 6d shows the image features that contributed to the prediction decision of the validation data set, ranked by rank.

[0105] Figure 7 is a diagram showing the overall performance of a BRCA mutation prediction model consisting of a combination of the Corr feature selection method and the SGD classifier. Figure 7a shows the AUROC (Area Under the Receiver Operating Characteristic Curve) value of the training data set, and Figure 7b shows the AUROC of the validation data set. Figure 7c shows the image features that contributed to the prediction decision of the training data set, ranked by rank, and Figure 7d shows the image features that contributed to the prediction decision of the validation data set, ranked by rank.

[0106] Figure 8 is a diagram showing the overall performance of a BRCA mutation prediction model consisting of a combination of a LASSO feature selection method and an RF classifier. Figure 8a shows the AUROC (Area Under the Receiver Operating Characteristic Curve) value of the training data set, and Figure 8b shows the AUROC of the validation data set. Figure 8c shows the image features that contributed to the prediction decision of the training data set in order of rank, and Figure 8d shows the image features that contributed to the prediction decision of the validation data set in order of rank.

[0107] Figure 9 is a diagram showing the overall performance of a BRCA mutation prediction model consisting of a combination of the FFS feature selection method and the RF classifier. Figure 9a shows the AUROC (Area Under the Receiver Operating Characteristic Curve) value of the training data set, and Figure 9b shows the AUROC of the validation data set. Figure 9c shows the image features that contributed to the prediction decision of the training data set, ranked by rank, and Figure 9d shows the image features that contributed to the prediction decision of the validation data set, ranked by rank.

[0108]

[0109] Hereinafter, the present invention will be described in more detail through examples. These examples are intended solely to illustrate the present invention more specifically, and it will be apparent to those skilled in the art that the scope of the present invention is not limited by these examples, in accordance with the gist of the present invention.

[0110]

[0111] Example

[0112] Training method for BRCA gene mutation prediction model

[0113] Data collection and image information collection

[0114] The data for this study consisted of MRI images of prostate cancer patients collected from two independent institutions. The training dataset consisted of 199 MRI images of prostate cancer patients obtained from Severance Hospital in Sinchon, and the test dataset consisted of 158 MRI images of prostate cancer patients obtained from Seoul National University Bundang Hospital (Table 1). For the prostate cancer lesions included in the training dataset from Severance Hospital in Sinchon, the latest version of ITK-SNAP (version 4.0.1, release date: December 22, 2022) was used to segment the lesion area.

[0115]

[0116] Severance Hospital in Sinchon, Seoul National University Hospital in Bundang, prostate cancer patient 199158

[0117]

[0118] Image information extraction and selection

[0119] To quantitatively extract image information about lesion sites within images, the latest version of PyRadiomics (version 3.0.1, released on June 13, 2022) was utilized, and a total of 1,422 2D and 3D-based image data were calculated. The calculated image data are as follows:

[0120] diagnostics_Image-original_Mean

[0121] diagnostics_Image-original_Minimum

[0122] diagnostics_Image-original_Maximum

[0123] diagnostics_Image-interpolated_Mean

[0124] diagnostics_Image-interpolated_Minimum

[0125] diagnostics_Image-interpolated_Maximum

[0126] diagnostics_Mask-interpolated_Mean

[0127] diagnostics_Mask-interpolated_Minimum

[0128] diagnostics_Mask-interpolated_Maximum

[0129] original_shape_Elongation

[0130] original_shape_Flatness

[0131] original_shape_LeastAxisLength

[0132] original_shape_MajorAxisLength

[0133] original_shape_Maximum2DDiameterColumn

[0134] original_shape_Maximum2DDiameterRow

[0135] original_shape_Maximum2DDiameterSlice

[0136] original_shape_Maximum3DDiameter

[0137] original_shape_MeshVolume

[0138] original_shape_MinorAxisLength

[0139] original_shape_Sphericity

[0140] original_shape_SurfaceArea

[0141] original_shape_SurfaceVolumeRatio

[0142] original_shape_VoxelVolume

[0143] original_firstorder_10Percentile

[0144] original_firstorder_90Percentile

[0145] original_firstorder_Energy

[0146] original_firstorder_Entropy

[0147] original_firstorder_InterquartileRange

[0148] original_firstorder_Kurtosis

[0149] original_firstorder_Maximum

[0150] original_firstorder_MeanAbsoluteDeviation

[0151] original_firstorder_Mean

[0152] original_firstorder_Median

[0153] original_firstorder_Minimum

[0154] original_firstorder_Range

[0155] original_firstorder_RobustMeanAbsoluteDeviation

[0156] original_firstorder_RootMeanSquared

[0157] original_firstorder_Skewness

[0158] original_firstorder_TotalEnergy

[0159] original_firstorder_Uniformity

[0160] original_firstorder_Variance

[0161] original_glcm_Autocorrelation

[0162] original_glcm_ClusterProminence

[0163] original_glcm_ClusterShade

[0164] original_glcm_ClusterTendency

[0165] original_glcm_Contrast

[0166] original_glcm_Correlation

[0167] original_glcm_DifferenceAverage

[0168] original_glcm_DifferenceEntropy

[0169] original_glcm_DifferenceVariance

[0170] original_glcm_Id

[0171] original_glcm_Idm

[0172] original_glcm_Idmn

[0173] original_glcm_Idn

[0174] original_glcm_Imc1

[0175] original_glcm_Imc2

[0176] original_glcm_InverseVariance

[0177] original_glcm_JointAverage

[0178] original_glcm_JointEnergy

[0179] original_glcm_JointEntropy

[0180] original_glcm_MCC

[0181] original_glcm_MaximumProbability

[0182] original_glcm_SumAverage

[0183] original_glcm_SumEntropy

[0184] original_glcm_SumSquares

[0185] original_gldm_DependenceEntropy

[0186] original_gldm_DependenceNonUniformity

[0187] original_gldm_DependenceNonUniformityNormalized

[0188] original_gldm_DependenceVariance

[0189] original_gldm_GrayLevelNonUniformity

[0190] original_gldm_GrayLevelVariance

[0191] original_gldm_HighGrayLevelEmphasis

[0192] original_gldm_LargeDependenceEmphasis

[0193] original_gldm_LargeDependenceHighGrayLevelEmphasis

[0194] original_gldm_LargeDependenceLowGrayLevelEmphasis

[0195] original_gldm_LowGrayLevelEmphasis

[0196] original_gldm_SmallDependenceEmphasis

[0197] original_gldm_SmallDependenceHighGrayLevelEmphasis

[0198] original_gldm_SmallDependenceLowGrayLevelEmphasis

[0199] original_glrlm_GrayLevelNonUniformity

[0200] original_glrlm_GrayLevelNonUniformityNormalized

[0201] original_glrlm_GrayLevelVariance

[0202] original_glrlm_HighGrayLevelRunEmphasis

[0203] original_glrlm_LongRunEmphasis

[0204] original_glrlm_LongRunHighGrayLevelEmphasis

[0205] original_glrlm_LongRunLowGrayLevelEmphasis

[0206] original_glrlm_LowGrayLevelRunEmphasis

[0207] original_glrlm_RunEntropy

[0208] original_glrlm_RunLengthNonUniformity

[0209] original_glrlm_RunLengthNonUniformityNormalized

[0210] original_glrlm_RunPercentage

[0211] original_glrlm_RunVariance

[0212] original_glrlm_ShortRunEmphasis

[0213] original_glrlm_ShortRunHighGrayLevelEmphasis

[0214] original_glrlm_ShortRunLowGrayLevelEmphasis

[0215] original_glszm_GrayLevelNonUniformity

[0216] original_glszm_GrayLevelNonUniformityNormalized

[0217] original_glszm_GrayLevelVariance

[0218] original_glszm_HighGrayLevelZoneEmphasis

[0219] original_glszm_LargeAreaEmphasis

[0220] original_glszm_LargeAreaHighGrayLevelEmphasis

[0221] original_glszm_LargeAreaLowGrayLevelEmphasis

[0222] original_glszm_LowGrayLevelZoneEmphasis

[0223] original_glszm_SizeZoneNonUniformity

[0224] original_glszm_SizeZoneNonUniformityNormalized

[0225] original_glszm_SmallAreaEmphasis

[0226] original_glszm_SmallAreaHighGrayLevelEmphasis

[0227] original_glszm_SmallAreaLowGrayLevelEmphasis

[0228] original_glszm_ZoneEntropy

[0229] original_glszm_ZonePercentage

[0230] original_glszm_ZoneVariance

[0231] original_ngtdm_Busyness

[0232] original_ngtdm_Coarseness

[0233] original_ngtdm_Complexity

[0234] original_ngtdm_Contrast

[0235] original_ngtdm_Strength

[0236] wavelet-LLH_firstorder_10Percentile

[0237] wavelet-LLH_firstorder_90Percentile

[0238] wavelet-LLH_firstorder_Energy

[0239] wavelet-LLH_firstorder_Entropy

[0240] wavelet-LLH_firstorder_InterquartileRange

[0241] wavelet-LLH_firstorder_Kurtosis

[0242] wavelet-LLH_firstorder_Maximum

[0243] wavelet-LLH_firstorder_MeanAbsoluteDeviation

[0244] wavelet-LLH_firstorder_Mean

[0245] wavelet-LLH_firstorder_Median

[0246] wavelet-LLH_firstorder_Minimum

[0247] wavelet-LLH_firstorder_Range

[0248] wavelet-LLH_firstorder_RobustMeanAbsoluteDeviation

[0249] wavelet-LLH_firstorder_RootMeanSquared

[0250] wavelet-LLH_firstorder_Skewness

[0251] wavelet-LLH_firstorder_TotalEnergy

[0252] wavelet-LLH_firstorder_Uniformity

[0253] wavelet-LLH_firstorder_Variance

[0254] wavelet - LLH_glcm_Autocorrelation

[0255] wavelet-LLH_glcm_ClusterProminence

[0256] wavelet-LLH_glcm_ClusterShade

[0257] wavelet-LLH_glcm_ClusterTendency

[0258] wavelet-LLH_glcm_Contrast

[0259] wavelet-LLH_glcm_Correlation

[0260] wavelet - LLH_glcm_DifferenceAverage

[0261] wavelet - LLH_glcm_DifferenceEntropy

[0262] wavelet - LLH_glcm_DifferenceVariance

[0263] wavelet-LLH_glcm_Id

[0264] wavelet-LLH_glcm_Idm

[0265] wavelet-LLH_glcm_Idmn

[0266] wavelet-LLH_glcm_Idn

[0267] wavelet-LLH_glcm_Imc1

[0268] wavelet-LLH_glcm_Imc2

[0269] wavelet-LLH_glcm_InverseVariance

[0270] wavelet-LLH_glcm_JointAverage

[0271] wavelet-LLH_glcm_JointEnergy

[0272] wavelet-LLH_glcm_JointEntropy

[0273] wavelet-LLH_glcm_MCC

[0274] wavelet - LLH_glcm_MaximumProbability

[0275] wavelet-LLH_glcm_SumAverage

[0276] wavelet-LLH_glcm_SumEntropy

[0277] wavelet-LLH_glcm_SumSquares

[0278] wavelet - LLH_gldm_DependenceEntropy

[0279] wavelet-LLH_gldm_DependenceNonUniformity

[0280] wavelet-LLH_gldm_DependenceNonUniformityNormalized

[0281] wavelet-LLH_gldm_DependenceVariance

[0282] wavelet-LLH_gldm_GrayLevelNonUniformity

[0283] wavelet-LLH_gldm_GrayLevelVariance

[0284] wavelet-LLH_gldm_HighGrayLevelEmphasis

[0285] wavelet-LLH_gldm_LargeDependenceEmphasis

[0286] wavelet-LLH_gldm_LargeDependenceHighGrayLevelEmphasis

[0287] wavelet-LLH_gldm_LargeDependenceLowGrayLevelEmphasis

[0288] wavelet-LLH_gldm_LowGrayLevelEmphasis

[0289] wavelet-LLH_gldm_SmallDependenceEmphasis

[0290] wavelet-LLH_gldm_SmallDependenceHighGrayLevelEmphasis

[0291] wavelet-LLH_gldm_SmallDependenceLowGrayLevelEmphasis

[0292] wavelet-LLH_glrlm_GrayLevelNonUniformity

[0293] wavelet-LLH_glrlm_GrayLevelNonUniformityNormalized

[0294] wavelet-LLH_glrlm_GrayLevelVariance

[0295] wavelet-LLH_glrlm_HighGrayLevelRunEmphasis

[0296] wavelet-LLH_glrlm_LongRunEmphasis

[0297] wavelet-LLH_glrlm_LongRunHighGrayLevelEmphasis

[0298] wavelet-LLH_glrlm_LongRunLowGrayLevelEmphasis

[0299] wavelet-LLH_glrlm_LowGrayLevelRunEmphasis

[0300] wavelet-LLH_glrlm_RunEntropy

[0301] wavelet-LLH_glrlm_RunLengthNonUniformity

[0302] wavelet-LLH_glrlm_RunLengthNonUniformityNormalized

[0303] wavelet-LLH_glrlm_RunPercentage

[0304] wavelet-LLH_glrlm_RunVariance

[0305] wavelet-LLH_glrlm_ShortRunEmphasis

[0306] wavelet-LLH_glrlm_ShortRunHighGrayLevelEmphasis

[0307] wavelet-LLH_glrlm_ShortRunLowGrayLevelEmphasis

[0308] wavelet-LLH_glszm_GrayLevelNonUniformity

[0309] wavelet-LLH_glszm_GrayLevelNonUniformityNormalized

[0310] wavelet-LLH_glszm_GrayLevelVariance

[0311] wavelet-LLH_glszm_HighGrayLevelZoneEmphasis

[0312] wavelet-LLH_glszm_LargeAreaEmphasis

[0313] wavelet-LLH_glszm_LargeAreaHighGrayLevelEmphasis

[0314] wavelet-LLH_glszm_LargeAreaLowGrayLevelEmphasis

[0315] wavelet-LLH_glszm_LowGrayLevelZoneEmphasis

[0316] wavelet-LLH_glszm_SizeZoneNonUniformity

[0317] wavelet-LLH_glszm_SizeZoneNonUniformityNormalized

[0318] wavelet-LLH_glszm_SmallAreaEmphasis

[0319] wavelet-LLH_glszm_SmallAreaHighGrayLevelEmphasis

[0320] wavelet-LLH_glszm_SmallAreaLowGrayLevelEmphasis

[0321] wavelet-LLH_glszm_ZoneEntropy

[0322] wavelet-LLH_glszm_ZonePercentage

[0323] wavelet-LLH_glszm_ZoneVariance

[0324] wavelet-LLH_ngtdm_Busyness

[0325] wavelet-LLH_ngtdm_Coarseness

[0326] wavelet-LLH_ngtdm_Complexity

[0327] wavelet-LLH_ngtdm_Contrast

[0328] wavelet-LLH_ngtdm_Strength

[0329] wavelet-LHL_firstorder_10Percentile

[0330] wavelet-LHL_firstorder_90Percentile

[0331] wavelet-LHL_firstorder_Energy

[0332] wavelet - LHL_firstorder_Entropy

[0333] wavelet-LHL_firstorder_InterquartileRange

[0334] wavelet-LHL_firstorder_Kurtosis

[0335] wavelet-LHL_firstorder_Maximum

[0336] wavelet-LHL_firstorder_MeanAbsoluteDeviation

[0337] wavelet-LHL_firstorder_Mean

[0338] wavelet-LHL_firstorder_Median

[0339] wavelet-LHL_firstorder_Minimum

[0340] wavelet-LHL_firstorder_Range

[0341] wavelet-LHL_firstorder_RobustMeanAbsoluteDeviation

[0342] wavelet-LHL_firstorder_RootMeanSquared

[0343] wavelet-LHL_firstorder_Skewness

[0344] wavelet-LHL_firstorder_TotalEnergy

[0345] wavelet-LHL_firstorder_Uniformity

[0346] wavelet-LHL_firstorder_Variance

[0347] wavelet-LHL_glcm_Autocorrelation

[0348] wavelet-LHL_glcm_ClusterProminence

[0349] wavelet-LHL_glcm_ClusterShade

[0350] wavelet-LHL_glcm_ClusterTendency

[0351] wavelet-LHL_glcm_Contrast

[0352] wavelet-LHL_glcm_Correlation

[0353] wavelet-LHL_glcm_DifferenceAverage

[0354] wavelet-LHL_glcm_DifferenceEntropy

[0355] wavelet-LHL_glcm_DifferenceVariance

[0356] wavelet-LHL_glcm_Id

[0357] wavelet-LHL_glcm_Idm

[0358] wavelet-LHL_glcm_Idmn

[0359] wavelet-LHL_glcm_Idn

[0360] wavelet-LHL_glcm_Imc1

[0361] wavelet-LHL_glcm_Imc2

[0362] wavelet - LHL_glcm_InverseVariance

[0363] wavelet-LHL_glcm_JointAverage

[0364] wavelet-LHL_glcm_JointEnergy

[0365] wavelet-LHL_glcm_JointEntropy

[0366] wavelet-LHL_glcm_MCC

[0367] wavelet - LHL_glcm_MaximumProbability

[0368] wavelet-LHL_glcm_SumAverage

[0369] wavelet-LHL_glcm_SumEntropy

[0370] wavelet-LHL_glcm_SumSquares

[0371] wavelet - LHL_gldm_DependenceEntropy

[0372] wavelet-LHL_gldm_DependenceNonUniformity

[0373] wavelet-LHL_gldm_DependenceNonUniformityNormalized

[0374] wavelet - LHL_gldm_DependenceVariance

[0375] wavelet-LHL_gldm_GrayLevelNonUniformity

[0376] wavelet-LHL_gldm_GrayLevelVariance

[0377] wavelet-LHL_gldm_HighGrayLevelEmphasis

[0378] wavelet-LHL_gldm_LargeDependenceEmphasis

[0379] wavelet-LHL_gldm_LargeDependenceHighGrayLevelEmphasis

[0380] wavelet-LHL_gldm_LargeDependenceLowGrayLevelEmphasis

[0381] wavelet-LHL_gldm_LowGrayLevelEmphasis

[0382] wavelet-LHL_gldm_SmallDependenceEmphasis

[0383] wavelet-LHL_gldm_SmallDependenceHighGrayLevelEmphasis

[0384] wavelet-LHL_gldm_SmallDependenceLowGrayLevelEmphasis

[0385] wavelet-LHL_glrlm_GrayLevelNonUniformity

[0386] wavelet-LHL_glrlm_GrayLevelNonUniformityNormalized

[0387] wavelet-LHL_glrlm_GrayLevelVariance

[0388] wavelet-LHL_glrlm_HighGrayLevelRunEmphasis

[0389] wavelet-LHL_glrlm_LongRunEmphasis

[0390] wavelet-LHL_glrlm_LongRunHighGrayLevelEmphasis

[0391] wavelet-LHL_glrlm_LongRunLowGrayLevelEmphasis

[0392] wavelet-LHL_glrlm_LowGrayLevelRunEmphasis

[0393] wavelet-LHL_glrlm_RunEntropy

[0394] wavelet-LHL_glrlm_RunLengthNonUniformity

[0395] wavelet-LHL_glrlm_RunLengthNonUniformityNormalized

[0396] wavelet-LHL_glrlm_RunPercentage

[0397] wavelet-LHL_glrlm_RunVariance

[0398] wavelet-LHL_glrlm_ShortRunEmphasis

[0399] wavelet-LHL_glrlm_ShortRunHighGrayLevelEmphasis

[0400] wavelet-LHL_glrlm_ShortRunLowGrayLevelEmphasis

[0401] wavelet-LHL_glszm_GrayLevelNonUniformity

[0402] wavelet-LHL_glszm_GrayLevelNonUniformityNormalized

[0403] wavelet-LHL_glszm_GrayLevelVariance

[0404] wavelet-LHL_glszm_HighGrayLevelZoneEmphasis

[0405] wavelet-LHL_glszm_LargeAreaEmphasis

[0406] wavelet-LHL_glszm_LargeAreaHighGrayLevelEmphasis

[0407] wavelet-LHL_glszm_LargeAreaLowGrayLevelEmphasis

[0408] wavelet-LHL_glszm_LowGrayLevelZoneEmphasis

[0409] wavelet-LHL_glszm_SizeZoneNonUniformity

[0410] wavelet-LHL_glszm_SizeZoneNonUniformityNormalized

[0411] wavelet-LHL_glszm_SmallAreaEmphasis

[0412] wavelet-LHL_glszm_SmallAreaHighGrayLevelEmphasis

[0413] wavelet-LHL_glszm_SmallAreaLowGrayLevelEmphasis

[0414] wavelet-LHL_glszm_ZoneEntropy

[0415] wavelet-LHL_glszm_ZonePercentage

[0416] wavelet-LHL_glszm_ZoneVariance

[0417] wavelet-LHL_ngtdm_Busyness

[0418] wavelet-LHL_ngtdm_Coarseness

[0419] wavelet-LHL_ngtdm_Complexity

[0420] wavelet-LHL_ngtdm_Contrast

[0421] wavelet-LHL_ngtdm_Strength

[0422] wavelet-LHH_firstorder_10Percentile

[0423] wavelet-LHH_firstorder_90Percentile

[0424] wavelet-LHH_firstorder_Energy

[0425] wavelet-LHH_firstorder_Entropy

[0426] wavelet-LHH_firstorder_InterquartileRange

[0427] wavelet-LHH_firstorder_Kurtosis

[0428] wavelet-LHH_firstorder_Maximum

[0429] wavelet-LHH_firstorder_MeanAbsoluteDeviation

[0430] wavelet-LHH_firstorder_Mean

[0431] wavelet-LHH_firstorder_Median

[0432] wavelet-LHH_firstorder_Minimum

[0433] wavelet-LHH_firstorder_Range

[0434] wavelet-LHH_firstorder_RobustMeanAbsoluteDeviation

[0435] wavelet-LHH_firstorder_RootMeanSquared

[0436] wavelet-LHH_firstorder_Skewness

[0437] wavelet-LHH_firstorder_TotalEnergy

[0438] wavelet-LHH_firstorder_Uniformity

[0439] wavelet-LHH_firstorder_Variance

[0440] wavelet-LHH_glcm_Autocorrelation

[0441] wavelet-LHH_glcm_ClusterProminence

[0442] wavelet-LHH_glcm_ClusterShade

[0443] wavelet-LHH_glcm_ClusterTendency

[0444] wavelet-LHH_glcm_Contrast

[0445] wavelet-LHH_glcm_Correlation

[0446] wavelet-LHH_glcm_DifferenceAverage

[0447] wavelet-LHH_glcm_DifferenceEntropy

[0448] wavelet-LHH_glcm_DifferenceVariance

[0449] wavelet-LHH_glcm_Id

[0450] wavelet-LHH_glcm_Idm

[0451] wavelet-LHH_glcm_Idmn

[0452] wavelet-LHH_glcm_Idn

[0453] wavelet-LHH_glcm_Imc1

[0454] wavelet-LHH_glcm_Imc2

[0455] wavelet-LHH_glcm_InverseVariance

[0456] wavelet-LHH_glcm_JointAverage

[0457] wavelet-LHH_glcm_JointEnergy

[0458] wavelet-LHH_glcm_JointEntropy

[0459] wavelet-LHH_glcm_MCC

[0460] wavelet-LHH_glcm_MaximumProbability

[0461] wavelet-LHH_glcm_SumAverage

[0462] wavelet-LHH_glcm_SumEntropy

[0463] wavelet-LHH_glcm_SumSquares

[0464] wavelet-LHH_gldm_DependenceEntropy

[0465] wavelet-LHH_gldm_DependenceNonUniformity

[0466] wavelet-LHH_gldm_DependenceNonUniformityNormalized

[0467] wavelet-LHH_gldm_DependenceVariance

[0468] wavelet-LHH_gldm_GrayLevelNonUniformity

[0469] wavelet-LHH_gldm_GrayLevelVariance

[0470] wavelet-LHH_gldm_HighGrayLevelEmphasis

[0471] wavelet-LHH_gldm_LargeDependenceEmphasis

[0472] wavelet-LHH_gldm_LargeDependenceHighGrayLevelEmphasis

[0473] wavelet-LHH_gldm_LargeDependenceLowGrayLevelEmphasis

[0474] wavelet-LHH_gldm_LowGrayLevelEmphasis

[0475] wavelet-LHH_gldm_SmallDependenceEmphasis

[0476] wavelet-LHH_gldm_SmallDependenceHighGrayLevelEmphasis

[0477] wavelet-LHH_gldm_SmallDependenceLowGrayLevelEmphasis

[0478] wavelet-LHH_glrlm_GrayLevelNonUniformity

[0479] wavelet-LHH_glrlm_GrayLevelNonUniformityNormalized

[0480] wavelet-LHH_glrlm_GrayLevelVariance

[0481] wavelet-LHH_glrlm_HighGrayLevelRunEmphasis

[0482] wavelet-LHH_glrlm_LongRunEmphasis

[0483] wavelet-LHH_glrlm_LongRunHighGrayLevelEmphasis

[0484] wavelet-LHH_glrlm_LongRunLowGrayLevelEmphasis

[0485] wavelet-LHH_glrlm_LowGrayLevelRunEmphasis

[0486] wavelet-LHH_glrlm_RunEntropy

[0487] wavelet-LHH_glrlm_RunLengthNonUniformity

[0488] wavelet-LHH_glrlm_RunLengthNonUniformityNormalized

[0489] wavelet-LHH_glrlm_RunPercentage

[0490] wavelet-LHH_glrlm_RunVariance

[0491] wavelet-LHH_glrlm_ShortRunEmphasis

[0492] wavelet-LHH_glrlm_ShortRunHighGrayLevelEmphasis

[0493] wavelet-LHH_glrlm_ShortRunLowGrayLevelEmphasis

[0494] wavelet-LHH_glszm_GrayLevelNonUniformity

[0495] wavelet-LHH_glszm_GrayLevelNonUniformityNormalized

[0496] wavelet-LHH_glszm_GrayLevelVariance

[0497] wavelet-LHH_glszm_HighGrayLevelZoneEmphasis

[0498] wavelet-LHH_glszm_LargeAreaEmphasis

[0499] wavelet-LHH_glszm_LargeAreaHighGrayLevelEmphasis

[0500] wavelet-LHH_glszm_LargeAreaLowGrayLevelEmphasis

[0501] wavelet-LHH_glszm_LowGrayLevelZoneEmphasis

[0502] wavelet-LHH_glszm_SizeZoneNonUniformity

[0503] wavelet-LHH_glszm_SizeZoneNonUniformityNormalized

[0504] wavelet-LHH_glszm_SmallAreaEmphasis

[0505] wavelet-LHH_glszm_SmallAreaHighGrayLevelEmphasis

[0506] wavelet-LHH_glszm_SmallAreaLowGrayLevelEmphasis

[0507] wavelet-LHH_glszm_ZoneEntropy

[0508] wavelet-LHH_glszm_ZonePercentage

[0509] wavelet-LHH_glszm_ZoneVariance

[0510] wavelet-LHH_ngtdm_Busyness

[0511] wavelet-LHH_ngtdm_Coarseness

[0512] wavelet-LHH_ngtdm_Complexity

[0513] wavelet-LHH_ngtdm_Contrast

[0514] wavelet-LHH_ngtdm_Strength

[0515] wavelet-HLL_firstorder_10Percentile

[0516] wavelet-HLL_firstorder_90Percentile

[0517] wavelet-HLL_firstorder_Energy

[0518] wavelet-HLL_firstorder_Entropy

[0519] wavelet-HLL_firstorder_InterquartileRange

[0520] wavelet-HLL_firstorder_Kurtosis

[0521] wavelet-HLL_firstorder_Maximum

[0522] wavelet-HLL_firstorder_MeanAbsoluteDeviation

[0523] wavelet-HLL_firstorder_Mean

[0524] wavelet-HLL_firstorder_Median

[0525] wavelet-HLL_firstorder_Minimum

[0526] wavelet-HLL_firstorder_Range

[0527] wavelet-HLL_firstorder_RobustMeanAbsoluteDeviation

[0528] wavelet-HLL_firstorder_RootMeanSquared

[0529] wavelet-HLL_firstorder_Skewness

[0530] wavelet-HLL_firstorder_TotalEnergy

[0531] wavelet-HLL_firstorder_Uniformity

[0532] wavelet-HLL_firstorder_Variance

[0533] wavelet-HLL_glcm_Autocorrelation

[0534] wavelet-HLL_glcm_ClusterProminence

[0535] wavelet-HLL_glcm_ClusterShade

[0536] wavelet-HLL_glcm_ClusterTendency

[0537] wavelet-HLL_glcm_Contrast

[0538] wavelet-HLL_glcm_Correlation

[0539] wavelet-HLL_glcm_DifferenceAverage

[0540] wavelet-HLL_glcm_DifferenceEntropy

[0541] wavelet-HLL_glcm_DifferenceVariance

[0542] wavelet-HLL_glcm_Id

[0543] wavelet-HLL_glcm_Idm

[0544] wavelet-HLL_glcm_Idmn

[0545] wavelet-HLL_glcm_Idn

[0546] wavelet-HLL_glcm_Imc1

[0547] wavelet-HLL_glcm_Imc2

[0548] wavelet-HLL_glcm_InverseVariance

[0549] wavelet-HLL_glcm_JointAverage

[0550] wavelet-HLL_glcm_JointEnergy

[0551] wavelet-HLL_glcm_JointEntropy

[0552] wavelet-HLL_glcm_MCC

[0553] wavelet-HLL_glcm_MaximumProbability

[0554] wavelet-HLL_glcm_SumAverage

[0555] wavelet-HLL_glcm_SumEntropy

[0556] wavelet-HLL_glcm_SumSquares

[0557] wavelet-HLL_gldm_DependenceEntropy

[0558] wavelet-HLL_gldm_DependenceNonUniformity

[0559] wavelet-HLL_gldm_DependenceNonUniformityNormalized

[0560] wavelet-HLL_gldm_DependenceVariance

[0561] wavelet-HLL_gldm_GrayLevelNonUniformity

[0562] wavelet-HLL_gldm_GrayLevelVariance

[0563] wavelet-HLL_gldm_HighGrayLevelEmphasis

[0564] wavelet-HLL_gldm_LargeDependenceEmphasis

[0565] wavelet-HLL_gldm_LargeDependenceHighGrayLevelEmphasis

[0566] wavelet-HLL_gldm_LargeDependenceLowGrayLevelEmphasis

[0567] wavelet-HLL_gldm_LowGrayLevelEmphasis

[0568] wavelet-HLL_gldm_SmallDependenceEmphasis

[0569] wavelet-HLL_gldm_SmallDependenceHighGrayLevelEmphasis

[0570] wavelet-HLL_gldm_SmallDependenceLowGrayLevelEmphasis

[0571] wavelet-HLL_glrlm_GrayLevelNonUniformity

[0572] wavelet-HLL_glrlm_GrayLevelNonUniformityNormalized

[0573] wavelet-HLL_glrlm_GrayLevelVariance

[0574] wavelet-HLL_glrlm_HighGrayLevelRunEmphasis

[0575] wavelet-HLL_glrlm_LongRunEmphasis

[0576] wavelet-HLL_glrlm_LongRunHighGrayLevelEmphasis

[0577] wavelet-HLL_glrlm_LongRunLowGrayLevelEmphasis

[0578] wavelet-HLL_glrlm_LowGrayLevelRunEmphasis

[0579] wavelet-HLL_glrlm_RunEntropy

[0580] wavelet-HLL_glrlm_RunLengthNonUniformity

[0581] wavelet-HLL_glrlm_RunLengthNonUniformityNormalized

[0582] wavelet-HLL_glrlm_RunPercentage

[0583] wavelet-HLL_glrlm_RunVariance

[0584] wavelet-HLL_glrlm_ShortRunEmphasis

[0585] wavelet-HLL_glrlm_ShortRunHighGrayLevelEmphasis

[0586] wavelet-HLL_glrlm_ShortRunLowGrayLevelEmphasis

[0587] wavelet-HLL_glszm_GrayLevelNonUniformity

[0588] wavelet-HLL_glszm_GrayLevelNonUniformityNormalized

[0589] wavelet-HLL_glszm_GrayLevelVariance

[0590] wavelet-HLL_glszm_HighGrayLevelZoneEmphasis

[0591] wavelet-HLL_glszm_LargeAreaEmphasis

[0592] wavelet-HLL_glszm_LargeAreaHighGrayLevelEmphasis

[0593] wavelet-HLL_glszm_LargeAreaLowGrayLevelEmphasis

[0594] wavelet-HLL_glszm_LowGrayLevelZoneEmphasis

[0595] wavelet-HLL_glszm_SizeZoneNonUniformity

[0596] wavelet-HLL_glszm_SizeZoneNonUniformityNormalized

[0597] wavelet-HLL_glszm_SmallAreaEmphasis

[0598] wavelet-HLL_glszm_SmallAreaHighGrayLevelEmphasis

[0599] wavelet-HLL_glszm_SmallAreaLowGrayLevelEmphasis

[0600] wavelet-HLL_glszm_ZoneEntropy

[0601] wavelet-HLL_glszm_ZonePercentage

[0602] wavelet-HLL_glszm_ZoneVariance

[0603] wavelet-HLL_ngtdm_Busyness

[0604] wavelet-HLL_ngtdm_Coarseness

[0605] wavelet-HLL_ngtdm_Complexity

[0606] wavelet-HLL_ngtdm_Contrast

[0607] wavelet-HLL_ngtdm_Strength

[0608] wavelet-HLH_firstorder_10Percentile

[0609] wavelet-HLH_firstorder_90Percentile

[0610] wavelet-HLH_firstorder_Energy

[0611] wavelet-HLH_firstorder_Entropy

[0612] wavelet-HLH_firstorder_InterquartileRange

[0613] wavelet-HLH_firstorder_Kurtosis

[0614] wavelet-HLH_firstorder_Maximum

[0615] wavelet-HLH_firstorder_MeanAbsoluteDeviation

[0616] wavelet-HLH_firstorder_Mean

[0617] wavelet-HLH_firstorder_Median

[0618] wavelet-HLH_firstorder_Minimum

[0619] wavelet-HLH_firstorder_Range

[0620] wavelet-HLH_firstorder_RobustMeanAbsoluteDeviation

[0621] wavelet-HLH_firstorder_RootMeanSquared

[0622] wavelet-HLH_firstorder_Skewness

[0623] wavelet-HLH_firstorder_TotalEnergy

[0624] wavelet-HLH_firstorder_Uniformity

[0625] wavelet-HLH_firstorder_Variance

[0626] wavelet-HLH_glcm_Autocorrelation

[0627] wavelet-HLH_glcm_ClusterProminence

[0628] wavelet-HLH_glcm_ClusterShade

[0629] wavelet-HLH_glcm_ClusterTendency

[0630] wavelet-HLH_glcm_Contrast

[0631] wavelet-HLH_glcm_Correlation

[0632] wavelet-HLH_glcm_DifferenceAverage

[0633] wavelet-HLH_glcm_DifferenceEntropy

[0634] wavelet-HLH_glcm_DifferenceVariance

[0635] wavelet-HLH_glcm_Id

[0636] wavelet-HLH_glcm_Idm

[0637] wavelet-HLH_glcm_Idmn

[0638] wavelet-HLH_glcm_Idn

[0639] wavelet-HLH_glcm_Imc1

[0640] wavelet-HLH_glcm_Imc2

[0641] wavelet-HLH_glcm_InverseVariance

[0642] wavelet-HLH_glcm_JointAverage

[0643] wavelet-HLH_glcm_JointEnergy

[0644] wavelet-HLH_glcm_JointEntropy

[0645] wavelet-HLH_glcm_MCC

[0646] wavelet-HLH_glcm_MaximumProbability

[0647] wavelet-HLH_glcm_SumAverage

[0648] wavelet-HLH_glcm_SumEntropy

[0649] wavelet-HLH_glcm_SumSquares

[0650] wavelet-HLH_gldm_DependenceEntropy

[0651] wavelet-HLH_gldm_DependenceNonUniformity

[0652] wavelet-HLH_gldm_DependenceNonUniformityNormalized

[0653] wavelet-HLH_gldm_DependenceVariance

[0654] wavelet-HLH_gldm_GrayLevelNonUniformity

[0655] wavelet-HLH_gldm_GrayLevelVariance

[0656] wavelet-HLH_gldm_HighGrayLevelEmphasis

[0657] wavelet-HLH_gldm_LargeDependenceEmphasis

[0658] wavelet-HLH_gldm_LargeDependenceHighGrayLevelEmphasis

[0659] wavelet-HLH_gldm_LargeDependenceLowGrayLevelEmphasis

[0660] wavelet-HLH_gldm_LowGrayLevelEmphasis

[0661] wavelet-HLH_gldm_SmallDependenceEmphasis

[0662] wavelet-HLH_gldm_SmallDependenceHighGrayLevelEmphasis

[0663] wavelet-HLH_gldm_SmallDependenceLowGrayLevelEmphasis

[0664] wavelet-HLH_glrlm_GrayLevelNonUniformity

[0665] wavelet-HLH_glrlm_GrayLevelNonUniformityNormalized

[0666] wavelet-HLH_glrlm_GrayLevelVariance

[0667] wavelet-HLH_glrlm_HighGrayLevelRunEmphasis

[0668] wavelet-HLH_glrlm_LongRunEmphasis

[0669] wavelet-HLH_glrlm_LongRunHighGrayLevelEmphasis

[0670] wavelet-HLH_glrlm_LongRunLowGrayLevelEmphasis

[0671] wavelet-HLH_glrlm_LowGrayLevelRunEmphasis

[0672] wavelet-HLH_glrlm_RunEntropy

[0673] wavelet-HLH_glrlm_RunLengthNonUniformity

[0674] wavelet-HLH_glrlm_RunLengthNonUniformityNormalized

[0675] wavelet-HLH_glrlm_RunPercentage

[0676] wavelet-HLH_glrlm_RunVariance

[0677] wavelet-HLH_glrlm_ShortRunEmphasis

[0678] wavelet-HLH_glrlm_ShortRunHighGrayLevelEmphasis

[0679] wavelet-HLH_glrlm_ShortRunLowGrayLevelEmphasis

[0680] wavelet-HLH_glszm_GrayLevelNonUniformity

[0681] wavelet-HLH_glszm_GrayLevelNonUniformityNormalized

[0682] wavelet-HLH_glszm_GrayLevelVariance

[0683] wavelet-HLH_glszm_HighGrayLevelZoneEmphasis

[0684] wavelet-HLH_glszm_LargeAreaEmphasis

[0685] wavelet-HLH_glszm_LargeAreaHighGrayLevelEmphasis

[0686] wavelet-HLH_glszm_LargeAreaLowGrayLevelEmphasis

[0687] wavelet-HLH_glszm_LowGrayLevelZoneEmphasis

[0688] wavelet-HLH_glszm_SizeZoneNonUniformity

[0689] wavelet-HLH_glszm_SizeZoneNonUniformityNormalized

[0690] wavelet-HLH_glszm_SmallAreaEmphasis

[0691] wavelet-HLH_glszm_SmallAreaHighGrayLevelEmphasis

[0692] wavelet-HLH_glszm_SmallAreaLowGrayLevelEmphasis

[0693] wavelet-HLH_glszm_ZoneEntropy

[0694] wavelet-HLH_glszm_ZonePercentage

[0695] wavelet-HLH_glszm_ZoneVariance

[0696] wavelet-HLH_ngtdm_Busyness

[0697] wavelet-HLH_ngtdm_Coarseness

[0698] wavelet-HLH_ngtdm_Complexity

[0699] wavelet-HLH_ngtdm_Contrast

[0700] wavelet-HLH_ngtdm_Strength

[0701] wavelet-HHL_firstorder_10Percentile

[0702] wavelet-HHL_firstorder_90Percentile

[0703] wavelet-HHL_firstorder_Energy

[0704] wavelet-HHL_firstorder_Entropy

[0705] wavelet-HHL_firstorder_InterquartileRange

[0706] wavelet-HHL_firstorder_Kurtosis

[0707] wavelet-HHL_firstorder_Maximum

[0708] wavelet-HHL_firstorder_MeanAbsoluteDeviation

[0709] wavelet-HHL_firstorder_Mean

[0710] wavelet-HHL_firstorder_Median

[0711] wavelet-HHL_firstorder_Minimum

[0712] wavelet-HHL_firstorder_Range

[0713] wavelet-HHL_firstorder_RobustMeanAbsoluteDeviation

[0714] wavelet-HHL_firstorder_RootMeanSquared

[0715] wavelet-HHL_firstorder_Skewness

[0716] wavelet-HHL_firstorder_TotalEnergy

[0717] wavelet-HHL_firstorder_Uniformity

[0718] wavelet-HHL_firstorder_Variance

[0719] wavelet-HHL_glcm_Autocorrelation

[0720] wavelet-HHL_glcm_ClusterProminence

[0721] wavelet-HHL_glcm_ClusterShade

[0722] wavelet-HHL_glcm_ClusterTendency

[0723] wavelet-HHL_glcm_Contrast

[0724] wavelet-HHL_glcm_Correlation

[0725] wavelet-HHL_glcm_DifferenceAverage

[0726] wavelet-HHL_glcm_DifferenceEntropy

[0727] wavelet-HHL_glcm_DifferenceVariance

[0728] wavelet-HHL_glcm_Id

[0729] wavelet-HHL_glcm_Idm

[0730] wavelet-HHL_glcm_Idmn

[0731] wavelet-HHL_glcm_Idn

[0732] wavelet-HHL_glcm_Imc1

[0733] wavelet-HHL_glcm_Imc2

[0734] wavelet-HHL_glcm_InverseVariance

[0735] wavelet-HHL_glcm_JointAverage

[0736] wavelet-HHL_glcm_JointEnergy

[0737] wavelet-HHL_glcm_JointEntropy

[0738] wavelet-HHL_glcm_MCC

[0739] wavelet-HHL_glcm_MaximumProbability

[0740] wavelet-HHL_glcm_SumAverage

[0741] wavelet-HHL_glcm_SumEntropy

[0742] wavelet-HHL_glcm_SumSquares

[0743] wavelet-HHL_gldm_DependenceEntropy

[0744] wavelet-HHL_gldm_DependenceNonUniformity

[0745] wavelet-HHL_gldm_DependenceNonUniformityNormalized

[0746] wavelet-HHL_gldm_DependenceVariance

[0747] wavelet-HHL_gldm_GrayLevelNonUniformity

[0748] wavelet-HHL_gldm_GrayLevelVariance

[0749] wavelet-HHL_gldm_HighGrayLevelEmphasis

[0750] wavelet-HHL_gldm_LargeDependenceEmphasis

[0751] wavelet-HHL_gldm_LargeDependenceHighGrayLevelEmphasis

[0752] wavelet-HHL_gldm_LargeDependenceLowGrayLevelEmphasis

[0753] wavelet-HHL_gldm_LowGrayLevelEmphasis

[0754] wavelet-HHL_gldm_SmallDependenceEmphasis

[0755] wavelet-HHL_gldm_SmallDependenceHighGrayLevelEmphasis

[0756] wavelet-HHL_gldm_SmallDependenceLowGrayLevelEmphasis

[0757] wavelet-HHL_glrlm_GrayLevelNonUniformity

[0758] wavelet-HHL_glrlm_GrayLevelNonUniformityNormalized

[0759] wavelet-HHL_glrlm_GrayLevelVariance

[0760] wavelet-HHL_glrlm_HighGrayLevelRunEmphasis

[0761] wavelet-HHL_glrlm_LongRunEmphasis

[0762] wavelet-HHL_glrlm_LongRunHighGrayLevelEmphasis

[0763] wavelet-HHL_glrlm_LongRunLowGrayLevelEmphasis

[0764] wavelet-HHL_glrlm_LowGrayLevelRunEmphasis

[0765] wavelet-HHL_glrlm_RunEntropy

[0766] wavelet-HHL_glrlm_RunLengthNonUniformity

[0767] wavelet-HHL_glrlm_RunLengthNonUniformityNormalized

[0768] wavelet-HHL_glrlm_RunPercentage

[0769] wavelet-HHL_glrlm_RunVariance

[0770] wavelet-HHL_glrlm_ShortRunEmphasis

[0771] wavelet-HHL_glrlm_ShortRunHighGrayLevelEmphasis

[0772] wavelet-HHL_glrlm_ShortRunLowGrayLevelEmphasis

[0773] wavelet-HHL_glszm_GrayLevelNonUniformity

[0774] wavelet-HHL_glszm_GrayLevelNonUniformityNormalized

[0775] wavelet-HHL_glszm_GrayLevelVariance

[0776] wavelet-HHL_glszm_HighGrayLevelZoneEmphasis

[0777] wavelet-HHL_glszm_LargeAreaEmphasis

[0778] wavelet-HHL_glszm_LargeAreaHighGrayLevelEmphasis

[0779] wavelet-HHL_glszm_LargeAreaLowGrayLevelEmphasis

[0780] wavelet-HHL_glszm_LowGrayLevelZoneEmphasis

[0781] wavelet-HHL_glszm_SizeZoneNonUniformity

[0782] wavelet-HHL_glszm_SizeZoneNonUniformityNormalized

[0783] wavelet-HHL_glszm_SmallAreaEmphasis

[0784] wavelet-HHL_glszm_SmallAreaHighGrayLevelEmphasis

[0785] wavelet-HHL_glszm_SmallAreaLowGrayLevelEmphasis

[0786] wavelet-HHL_glszm_ZoneEntropy

[0787] wavelet-HHL_glszm_ZonePercentage

[0788] wavelet-HHL_glszm_ZoneVariance

[0789] wavelet-HHL_ngtdm_Busyness

[0790] wavelet-HHL_ngtdm_Coarseness

[0791] wavelet-HHL_ngtdm_Complexity

[0792] wavelet-HHL_ngtdm_Contrast

[0793] wavelet-HHL_ngtdm_Strength

[0794] wavelet-HHH_firstorder_10Percentile

[0795] wavelet-HHH_firstorder_90Percentile

[0796] wavelet-HHH_firstorder_Energy

[0797] wavelet-HHH_firstorder_Entropy

[0798] wavelet-HHH_firstorder_InterquartileRange

[0799] wavelet-HHH_firstorder_Kurtosis

[0800] wavelet-HHH_firstorder_Maximum

[0801] wavelet-HHH_firstorder_MeanAbsoluteDeviation

[0802] wavelet-HHH_firstorder_Mean

[0803] wavelet-HHH_firstorder_Median

[0804] wavelet-HHH_firstorder_Minimum

[0805] wavelet-HHH_firstorder_Range

[0806] wavelet-HHH_firstorder_RobustMeanAbsoluteDeviation

[0807] wavelet-HHH_firstorder_RootMeanSquared

[0808] wavelet-HHH_firstorder_Skewness

[0809] wavelet-HHH_firstorder_TotalEnergy

[0810] wavelet-HHH_firstorder_Uniformity

[0811] wavelet-HHH_firstorder_Variance

[0812] wavelet-HHH_glcm_Autocorrelation

[0813] wavelet-HHH_glcm_ClusterProminence

[0814] wavelet-HHH_glcm_ClusterShade

[0815] wavelet-HHH_glcm_ClusterTendency

[0816] wavelet-HHH_glcm_Contrast

[0817] wavelet-HHH_glcm_Correlation

[0818] wavelet-HHH_glcm_DifferenceAverage

[0819] wavelet-HHH_glcm_DifferenceEntropy

[0820] wavelet-HHH_glcm_DifferenceVariance

[0821] wavelet-HHH_glcm_Id

[0822] wavelet-HHH_glcm_Idm

[0823] wavelet-HHH_glcm_Idmn

[0824] wavelet-HHH_glcm_Idn

[0825] wavelet-HHH_glcm_Imc1

[0826] wavelet-HHH_glcm_Imc2

[0827] wavelet-HHH_glcm_InverseVariance

[0828] wavelet-HHH_glcm_JointAverage

[0829] wavelet-HHH_glcm_JointEnergy

[0830] wavelet-HHH_glcm_JointEntropy

[0831] wavelet-HHH_glcm_MCC

[0832] wavelet-HHH_glcm_MaximumProbability

[0833] wavelet-HHH_glcm_SumAverage

[0834] wavelet-HHH_glcm_SumEntropy

[0835] wavelet-HHH_glcm_SumSquares

[0836] wavelet-HHH_gldm_DependenceEntropy

[0837] wavelet-HHH_gldm_DependenceNonUniformity

[0838] wavelet-HHH_gldm_DependenceNonUniformityNormalized

[0839] wavelet-HHH_gldm_DependenceVariance

[0840] wavelet-HHH_gldm_GrayLevelNonUniformity

[0841] wavelet-HHH_gldm_GrayLevelVariance

[0842] wavelet-HHH_gldm_HighGrayLevelEmphasis

[0843] wavelet-HHH_gldm_LargeDependenceEmphasis

[0844] wavelet-HHH_gldm_LargeDependenceHighGrayLevelEmphasis

[0845] wavelet-HHH_gldm_LargeDependenceLowGrayLevelEmphasis

[0846] wavelet-HHH_gldm_LowGrayLevelEmphasis

[0847] wavelet-HHH_gldm_SmallDependenceEmphasis

[0848] wavelet-HHH_gldm_SmallDependenceHighGrayLevelEmphasis

[0849] wavelet-HHH_gldm_SmallDependenceLowGrayLevelEmphasis

[0850] wavelet-HHH_glrlm_GrayLevelNonUniformity

[0851] wavelet-HHH_glrlm_GrayLevelNonUniformityNormalized

[0852] wavelet-HHH_glrlm_GrayLevelVariance

[0853] wavelet-HHH_glrlm_HighGrayLevelRunEmphasis

[0854] wavelet-HHH_glrlm_LongRunEmphasis

[0855] wavelet-HHH_glrlm_LongRunHighGrayLevelEmphasis

[0856] wavelet-HHH_glrlm_LongRunLowGrayLevelEmphasis

[0857] wavelet-HHH_glrlm_LowGrayLevelRunEmphasis

[0858] wavelet-HHH_glrlm_RunEntropy

[0859] wavelet-HHH_glrlm_RunLengthNonUniformity

[0860] wavelet-HHH_glrlm_RunLengthNonUniformityNormalized

[0861] wavelet-HHH_glrlm_RunPercentage

[0862] wavelet-HHH_glrlm_RunVariance

[0863] wavelet-HHH_glrlm_ShortRunEmphasis

[0864] wavelet-HHH_glrlm_ShortRunHighGrayLevelEmphasis

[0865] wavelet-HHH_glrlm_ShortRunLowGrayLevelEmphasis

[0866] wavelet-HHH_glszm_GrayLevelNonUniformity

[0867] wavelet-HHH_glszm_GrayLevelNonUniformityNormalized

[0868] wavelet-HHH_glszm_GrayLevelVariance

[0869] wavelet-HHH_glszm_HighGrayLevelZoneEmphasis

[0870] wavelet-HHH_glszm_LargeAreaEmphasis

[0871] wavelet-HHH_glszm_LargeAreaHighGrayLevelEmphasis

[0872] wavelet-HHH_glszm_LargeAreaLowGrayLevelEmphasis

[0873] wavelet-HHH_glszm_LowGrayLevelZoneEmphasis

[0874] wavelet-HHH_glszm_SizeZoneNonUniformity

[0875] wavelet-HHH_glszm_SizeZoneNonUniformityNormalized

[0876] wavelet-HHH_glszm_SmallAreaEmphasis

[0877] wavelet-HHH_glszm_SmallAreaHighGrayLevelEmphasis

[0878] wavelet-HHH_glszm_SmallAreaLowGrayLevelEmphasis

[0879] wavelet-HHH_glszm_ZoneEntropy

[0880] wavelet-HHH_glszm_ZonePercentage

[0881] wavelet-HHH_glszm_ZoneVariance

[0882] wavelet-HHH_ngtdm_Busyness

[0883] wavelet-HHH_ngtdm_Coarseness

[0884] wavelet-HHH_ngtdm_Complexity

[0885] wavelet-HHH_ngtdm_Contrast

[0886] wavelet-HHH_ngtdm_Strength

[0887] wavelet-LLL_firstorder_10Percentile

[0888] wavelet-LLL_firstorder_90Percentile

[0889] wavelet-LLL_firstorder_Energy

[0890] wavelet-LLL_firstorder_Entropy

[0891] wavelet-LLL_firstorder_InterquartileRange

[0892] wavelet-LLL_firstorder_Kurtosis

[0893] wavelet-LLL_firstorder_Maximum

[0894] wavelet-LLL_firstorder_MeanAbsoluteDeviation

[0895] wavelet-LLL_firstorder_Mean

[0896] wavelet-LLL_firstorder_Median

[0897] wavelet-LLL_firstorder_Minimum

[0898] wavelet-LLL_firstorder_Range

[0899] wavelet-LLL_firstorder_RobustMeanAbsoluteDeviation

[0900] wavelet-LLL_firstorder_RootMeanSquared

[0901] wavelet-LLL_firstorder_Skewness

[0902] wavelet-LLL_firstorder_TotalEnergy

[0903] wavelet-LLL_firstorder_Uniformity

[0904] wavelet-LLL_firstorder_Variance

[0905] wavelet-LLL_glcm_Autocorrelation

[0906] wavelet-LLL_glcm_ClusterProminence

[0907] wavelet-LLL_glcm_ClusterShade

[0908] wavelet-LLL_glcm_ClusterTendency

[0909] wavelet-LLL_glcm_Contrast

[0910] wavelet-LLL_glcm_Correlation

[0911] wavelet-LLL_glcm_DifferenceAverage

[0912] wavelet-LLL_glcm_DifferenceEntropy

[0913] wavelet-LLL_glcm_DifferenceVariance

[0914] wavelet-LLL_glcm_Id

[0915] wavelet-LLL_glcm_Idm

[0916] wavelet-LLL_glcm_Idmn

[0917] wavelet-LLL_glcm_Idn

[0918] wavelet-LLL_glcm_Imc1

[0919] wavelet-LLL_glcm_Imc2

[0920] wavelet-LLL_glcm_InverseVariance

[0921] wavelet-LLL_glcm_JointAverage

[0922] wavelet-LLL_glcm_JointEnergy

[0923] wavelet-LLL_glcm_JointEntropy

[0924] wavelet-LLL_glcm_MCC

[0925] wavelet-LLL_glcm_MaximumProbability

[0926] wavelet-LLL_glcm_SumAverage

[0927] wavelet-LLL_glcm_SumEntropy

[0928] wavelet-LLL_glcm_SumSquares

[0929] wavelet-LLL_gldm_DependenceEntropy

[0930] wavelet-LLL_gldm_DependenceNonUniformity

[0931] wavelet-LLL_gldm_DependenceNonUniformityNormalized

[0932] wavelet-LLL_gldm_DependenceVariance

[0933] wavelet-LLL_gldm_GrayLevelNonUniformity

[0934] wavelet-LLL_gldm_GrayLevelVariance

[0935] wavelet-LLL_gldm_HighGrayLevelEmphasis

[0936] wavelet-LLL_gldm_LargeDependenceEmphasis

[0937] wavelet-LLL_gldm_LargeDependenceHighGrayLevelEmphasis

[0938] wavelet-LLL_gldm_LargeDependenceLowGrayLevelEmphasis

[0939] wavelet-LLL_gldm_LowGrayLevelEmphasis

[0940] wavelet-LLL_gldm_SmallDependenceEmphasis

[0941] wavelet-LLL_gldm_SmallDependenceHighGrayLevelEmphasis

[0942] wavelet-LLL_gldm_SmallDependenceLowGrayLevelEmphasis

[0943] wavelet-LLL_glrlm_GrayLevelNonUniformity

[0944] wavelet-LLL_glrlm_GrayLevelNonUniformityNormalized

[0945] wavelet-LLL_glrlm_GrayLevelVariance

[0946] wavelet-LLL_glrlm_HighGrayLevelRunEmphasis

[0947] wavelet-LLL_glrlm_LongRunEmphasis

[0948] wavelet-LLL_glrlm_LongRunHighGrayLevelEmphasis

[0949] wavelet-LLL_glrlm_LongRunLowGrayLevelEmphasis

[0950] wavelet-LLL_glrlm_LowGrayLevelRunEmphasis

[0951] wavelet-LLL_glrlm_RunEntropy

[0952] wavelet-LLL_glrlm_RunLengthNonUniformity

[0953] wavelet-LLL_glrlm_RunLengthNonUniformityNormalized

[0954] wavelet-LLL_glrlm_RunPercentage

[0955] wavelet-LLL_glrlm_RunVariance

[0956] wavelet-LLL_glrlm_ShortRunEmphasis

[0957] wavelet-LLL_glrlm_ShortRunHighGrayLevelEmphasis

[0958] wavelet-LLL_glrlm_ShortRunLowGrayLevelEmphasis

[0959] wavelet-LLL_glszm_GrayLevelNonUniformity

[0960] wavelet-LLL_glszm_GrayLevelNonUniformityNormalized

[0961] wavelet-LLL_glszm_GrayLevelVariance

[0962] wavelet-LLL_glszm_HighGrayLevelZoneEmphasis

[0963] wavelet-LLL_glszm_LargeAreaEmphasis

[0964] wavelet-LLL_glszm_LargeAreaHighGrayLevelEmphasis

[0965] wavelet-LLL_glszm_LargeAreaLowGrayLevelEmphasis

[0966] wavelet-LLL_glszm_LowGrayLevelZoneEmphasis

[0967] wavelet-LLL_glszm_SizeZoneNonUniformity

[0968] wavelet-LLL_glszm_SizeZoneNonUniformityNormalized

[0969] wavelet-LLL_glszm_SmallAreaEmphasis

[0970] wavelet-LLL_glszm_SmallAreaHighGrayLevelEmphasis

[0971] wavelet-LLL_glszm_SmallAreaLowGrayLevelEmphasis

[0972] wavelet-LLL_glszm_ZoneEntropy

[0973] wavelet-LLL_glszm_ZonePercentage

[0974] wavelet-LLL_glszm_ZoneVariance

[0975] wavelet-LLL_ngtdm_Busyness

[0976] wavelet-LLL_ngtdm_Coarseness

[0977] wavelet-LLL_ngtdm_Complexity

[0978] wavelet-LLL_ngtdm_Contrast

[0979] wavelet-LLL_ngtdm_Strength

[0980] log-sigma-3-0-mm-3D_firstorder_10Percentile

[0981] log-sigma-3-0-mm-3D_firstorder_90Percentile

[0982] log-sigma-3-0-mm-3D_firstorder_Energy

[0983] log-sigma-3-0-mm-3D_firstorder_Entropy

[0984] log-sigma-3-0-mm-3D_firstorder_InterquartileRange

[0985] log-sigma-3-0-mm-3D_firstorder_Kurtosis

[0986] log-sigma-3-0-mm-3D_firstorder_Maximum

[0987] log-sigma-3-0-mm-3D_firstorder_MeanAbsoluteDeviation

[0988] log-sigma-3-0-mm-3D_firstorder_Mean

[0989] log-sigma-3-0-mm-3D_firstorder_Median

[0990] log-sigma-3-0-mm-3D_firstorder_Minimum

[0991] log-sigma-3-0-mm-3D_firstorder_Range

[0992] log-sigma-3-0-mm-3D_firstorder_RobustMeanAbsoluteDeviation

[0993] log-sigma-3-0-mm-3D_firstorder_RootMeanSquared

[0994] log-sigma-3-0-mm-3D_firstorder_Skewness

[0995] log-sigma-3-0-mm-3D_firstorder_TotalEnergy

[0996] log-sigma-3-0-mm-3D_firstorder_Uniformity

[0997] log-sigma-3-0-mm-3D_firstorder_Variance

[0998] log-sigma-3-0-mm-3D_glcm_Autocorrelation

[0999] log-sigma-3-0-mm-3D_glcm_ClusterProminence

[1000] log-sigma-3-0-mm-3D_glcm_ClusterShade

[1001] log-sigma-3-0-mm-3D_glcm_ClusterTendency

[1002] log-sigma-3-0-mm-3D_glcm_Contrast

[1003] log-sigma-3-0-mm-3D_glcm_Correlation

[1004] log-sigma-3-0-mm-3D_glcm_DifferenceAverage

[1005] log-sigma-3-0-mm-3D_glcm_DifferenceEntropy

[1006] log-sigma-3-0-mm-3D_glcm_DifferenceVariance

[1007] log-sigma-3-0-mm-3D_glcm_Id

[1008] log-sigma-3-0-mm-3D_glcm_Idm

[1009] log-sigma-3-0-mm-3D_glcm_Idmn

[1010] log-sigma-3-0-mm-3D_glcm_Idn

[1011] log-sigma-3-0-mm-3D_glcm_Imc1

[1012] log-sigma-3-0-mm-3D_glcm_Imc2

[1013] log-sigma-3-0-mm-3D_glcm_InverseVariance

[1014] log-sigma-3-0-mm-3D_glcm_JointAverage

[1015] log-sigma-3-0-mm-3D_glcm_JointEnergy

[1016] log-sigma-3-0-mm-3D_glcm_JointEntropy

[1017] log-sigma-3-0-mm-3D_glcm_MCC

[1018] log-sigma-3-0-mm-3D_glcm_MaximumProbability

[1019] log-sigma-3-0-mm-3D_glcm_SumAverage

[1020] log-sigma-3-0-mm-3D_glcm_SumEntropy

[1021] log-sigma-3-0-mm-3D_glcm_SumSquares

[1022] log-sigma-3-0-mm-3D_gldm_DependenceEntropy

[1023] log-sigma-3-0-mm-3D_gldm_DependenceNonUniformity

[1024] log-sigma-3-0-mm-3D_gldm_DependenceNonUniformityNormalized

[1025] log-sigma-3-0-mm-3D_gldm_DependenceVariance

[1026] log-sigma-3-0-mm-3D_gldm_GrayLevelNonUniformity

[1027] log-sigma-3-0-mm-3D_gldm_GrayLevelVariance

[1028] log-sigma-3-0-mm-3D_gldm_HighGrayLevelEmphasis

[1029] log-sigma-3-0-mm-3D_gldm_LargeDependenceEmphasis

[1030] log-sigma-3-0-mm-3D_gldm_LargeDependenceHighGrayLevelEmphasis

[1031] log-sigma-3-0-mm-3D_gldm_LargeDependenceLowGrayLevelEmphasis

[1032] log-sigma-3-0-mm-3D_gldm_LowGrayLevelEmphasis

[1033] log-sigma-3-0-mm-3D_gldm_SmallDependenceEmphasis

[1034] log-sigma-3-0-mm-3D_gldm_SmallDependenceHighGrayLevelEmphasis

[1035] log-sigma-3-0-mm-3D_gldm_SmallDependenceLowGrayLevelEmphasis

[1036] log-sigma-3-0-mm-3D_glrlm_GrayLevelNonUniformity

[1037] log-sigma-3-0-mm-3D_glrlm_GrayLevelNonUniformityNormalized

[1038] log-sigma-3-0-mm-3D_glrlm_GrayLevelVariance

[1039] log-sigma-3-0-mm-3D_glrlm_HighGrayLevelRunEmphasis

[1040] log-sigma-3-0-mm-3D_glrlm_LongRunEmphasis

[1041] log-sigma-3-0-mm-3D_glrlm_LongRunHighGrayLevelEmphasis

[1042] log-sigma-3-0-mm-3D_glrlm_LongRunLowGrayLevelEmphasis

[1043] log-sigma-3-0-mm-3D_glrlm_LowGrayLevelRunEmphasis

[1044] log-sigma-3-0-mm-3D_glrlm_RunEntropy

[1045] log-sigma-3-0-mm-3D_glrlm_RunLengthNonUniformity

[1046] log-sigma-3-0-mm-3D_glrlm_RunLengthNonUniformityNormalized

[1047] log-sigma-3-0-mm-3D_glrlm_RunPercentage

[1048] log-sigma-3-0-mm-3D_glrlm_RunVariance

[1049] log-sigma-3-0-mm-3D_glrlm_ShortRunEmphasis

[1050] log-sigma-3-0-mm-3D_glrlm_ShortRunHighGrayLevelEmphasis

[1051] log-sigma-3-0-mm-3D_glrlm_ShortRunLowGrayLevelEmphasis

[1052] log-sigma-3-0-mm-3D_glszm_GrayLevelNonUniformity

[1053] log-sigma-3-0-mm-3D_glszm_GrayLevelNonUniformityNormalized

[1054] log-sigma-3-0-mm-3D_glszm_GrayLevelVariance

[1055] log-sigma-3-0-mm-3D_glszm_HighGrayLevelZoneEmphasis

[1056] log-sigma-3-0-mm-3D_glszm_LargeAreaEmphasis

[1057] log-sigma-3-0-mm-3D_glszm_LargeAreaHighGrayLevelEmphasis

[1058] log-sigma-3-0-mm-3D_glszm_LargeAreaLowGrayLevelEmphasis

[1059] log-sigma-3-0-mm-3D_glszm_LowGrayLevelZoneEmphasis

[1060] log-sigma-3-0-mm-3D_glszm_SizeZoneNonUniformity

[1061] log-sigma-3-0-mm-3D_glszm_SizeZoneNonUniformityNormalized

[1062] log-sigma-3-0-mm-3D_glszm_SmallAreaEmphasis

[1063] log-sigma-3-0-mm-3D_glszm_SmallAreaHighGrayLevelEmphasis

[1064] log-sigma-3-0-mm-3D_glszm_SmallAreaLowGrayLevelEmphasis

[1065] log-sigma-3-0-mm-3D_glszm_ZoneEntropy

[1066] log-sigma-3-0-mm-3D_glszm_ZonePercentage

[1067] log-sigma-3-0-mm-3D_glszm_ZoneVariance

[1068] log-sigma-3-0-mm-3D_ngtdm_Busyness

[1069] log-sigma-3-0-mm-3D_ngtdm_Coarseness

[1070] log-sigma-3-0-mm-3D_ngtdm_Complexity

[1071] log-sigma-3-0-mm-3D_ngtdm_Contrast

[1072] log-sigma-3-0-mm-3D_ngtdm_Strength

[1073] square_firstorder_10Percentile

[1074] square_firstorder_90Percentile

[1075] square_firstorder_Energy

[1076] square_firstorder_Entropy

[1077] square_firstorder_InterquartileRange

[1078] square_firstorder_Kurtosis

[1079] square_firstorder_Maximum

[1080] square_firstorder_MeanAbsoluteDeviation

[1081] square_firstorder_Mean

[1082] square_firstorder_Median

[1083] square_firstorder_Minimum

[1084] square_firstorder_Range

[1085] square_firstorder_RobustMeanAbsoluteDeviation

[1086] square_firstorder_RootMeanSquared

[1087] square_firstorder_Skewness

[1088] square_firstorder_TotalEnergy

[1089] square_firstorder_Variance

[1090] square_glcm_DifferenceEntropy

[1091] square_glcm_JointEntropy

[1092] square_glcm_SumEntropy

[1093] square_gldm_DependenceEntropy

[1094] square_gldm_DependenceNonUniformity

[1095] square_gldm_DependenceNonUniformityNormalized

[1096] square_gldm_DependenceVariance

[1097] square_gldm_LargeDependenceEmphasis

[1098] square_gldm_LargeDependenceHighGrayLevelEmphasis

[1099] square_gldm_LargeDependenceLowGrayLevelEmphasis

[1100] square_gldm_SmallDependenceEmphasis

[1101] square_gldm_SmallDependenceHighGrayLevelEmphasis

[1102] square_gldm_SmallDependenceLowGrayLevelEmphasis

[1103] square_glrlm_GrayLevelNonUniformity

[1104] square_glrlm_LongRunEmphasis

[1105] square_glrlm_LongRunHighGrayLevelEmphasis

[1106] square_glrlm_LongRunLowGrayLevelEmphasis

[1107] square_glrlm_RunEntropy

[1108] square_glrlm_RunLengthNonUniformity

[1109] square_glrlm_RunLengthNonUniformityNormalized

[1110] square_glrlm_RunPercentage

[1111] square_glrlm_RunVariance

[1112] square_glrlm_ShortRunEmphasis

[1113] square_glrlm_ShortRunHighGrayLevelEmphasis

[1114] square_glrlm_ShortRunLowGrayLevelEmphasis

[1115] square_glszm_LargeAreaEmphasis

[1116] square_glszm_LargeAreaHighGrayLevelEmphasis

[1117] square_glszm_LargeAreaLowGrayLevelEmphasis

[1118] square_glszm_SizeZoneNonUniformityNormalized

[1119] square_glszm_SmallAreaEmphasis

[1120] square_glszm_SmallAreaHighGrayLevelEmphasis

[1121] square_glszm_SmallAreaLowGrayLevelEmphasis

[1122] square_glszm_ZoneEntropy

[1123] square_glszm_ZonePercentage

[1124] square_glszm_ZoneVariance

[1125] squareroot_firstorder_10Percentile

[1126] squareroot_first_order_90Percentile

[1127] squareroot_first_order_Energy

[1128] squareroot_first_order_Entropy

[1129] squareroot_first_order_interquartileRange

[1130] squareroot_firstorder_kurtosis

[1131] squareroot_firstorder_Maximum

[1132] squareroot_first_order_MeanAbsoluteDeviation

[1133] squareroot_firstorder_Mean

[1134] squareroot_firstorder_Median

[1135] squareroot_first_order_Minimum

[1136] squareroot_first_order_Range

[1137] squareroot_first_order_RobustMeanAbsoluteDeviation

[1138] squareroot_first_order_RootMeanSquared

[1139] squareroot_firstorder_skewness

[1140] squareroot_first_order_TotalEnergy

[1141] squareroot_firstorder_uniformity

[1142] squareroot_firstorder_variance

[1143] squareroot_glcm_Autocorrelation

[1144] squareroot_glcm_ClusterProminence

[1145] squareroot_glcm_ClusterShade

[1146] squareroot_glcm_ClusterTendency

[1147] squareroot_glcm_Contrast

[1148] squareroot_glcm_Correlation

[1149] squareroot_glcm_DifferenceAverage

[1150] squareroot_glcm_DifferenceEntropy

[1151] squareroot_glcm_DifferenceVariance

[1152] squareroot_glcm_Id

[1153] squareroot_glcm_Idm

[1154] squareroot_glcm_Idmn

[1155] squareroot_glcm_Idn

[1156] squareroot_glcm_Imc1

[1157] squareroot_glcm_Imc2

[1158] squareroot_glcm_InverseVariance

[1159] squareroot_glcm_JointAverage

[1160] squareroot_glcm_JointEnergy

[1161] squareroot_glcm_JointEntropy

[1162] squareroot_glcm_MCC

[1163] squareroot_glcm_MaximumProbability

[1164] squareroot_glcm_SumAverage

[1165] squareroot_glcm_SumEntropy

[1166] squareroot_glcm_SumSquares

[1167] squareroot_gldm_DependenceEntropy

[1168] squareroot_gldm_DependenceNonUniformity

[1169] squareroot_gldm_DependenceNonUniformityNormalized

[1170] squareroot_gldm_DependenceVariance

[1171] squareroot_gldm_GrayLevelNonUniformity

[1172] squareroot_gldm_GrayLevelVariance

[1173] squareroot_gldm_HighGrayLevelEmphasis

[1174] squareroot_gldm_LargeDependenceEmphasis

[1175] squareroot_gldm_LargeDependenceHighGrayLevelEmphasis

[1176] squareroot_gldm_LargeDependenceLowGrayLevelEmphasis

[1177] squareroot_gldm_LowGrayLevelEmphasis

[1178] squareroot_gldm_SmallDependenceEmphasis

[1179] squareroot_gldm_SmallDependenceHighGrayLevelEmphasis

[1180] squareroot_gldm_SmallDependenceLowGrayLevelEmphasis

[1181] squareroot_glrlm_GrayLevelNonUniformity

[1182] squareroot_glrlm_GrayLevelNonUniformityNormalized

[1183] squareroot_glrlm_GrayLevelVariance

[1184] squareroot_glrlm_HighGrayLevelRunEmphasis

[1185] squareroot_glrlm_LongRunEmphasis

[1186] squareroot_glrlm_LongRunHighGrayLevelEmphasis

[1187] squareroot_glrlm_LongRunLowGrayLevelEmphasis

[1188] squareroot_glrlm_LowGrayLevelRunEmphasis

[1189] squareroot_glrlm_RunEntropy

[1190] squareroot_glrlm_RunLengthNonUniformity

[1191] squareroot_glrlm_RunLengthNonUniformityNormalized

[1192] squareroot_glrlm_RunPercentage

[1193] squareroot_glrlm_RunVariance

[1194] squareroot_glrlm_ShortRunEmphasis

[1195] squareroot_glrlm_ShortRunHighGrayLevelEmphasis

[1196] squareroot_glrlm_ShortRunLowGrayLevelEmphasis

[1197] squareroot_glszm_GrayLevelNonUniformity

[1198] squareroot_glszm_GrayLevelNonUniformityNormalized

[1199] squareroot_glszm_GrayLevelVariance

[1200] squareroot_glszm_HighGrayLevelZoneEmphasis

[1201] squareroot_glszm_LargeAreaEmphasis

[1202] squareroot_glszm_LargeAreaHighGrayLevelEmphasis

[1203] squareroot_glszm_LargeAreaLowGrayLevelEmphasis

[1204] squareroot_glszm_LowGrayLevelZoneEmphasis

[1205] squareroot_glszm_SizeZoneNonUniformity

[1206] squareroot_glszm_SizeZoneNonUniformityNormalized

[1207] squareroot_glszm_SmallAreaEmphasis

[1208] squareroot_glszm_SmallAreaHighGrayLevelEmphasis

[1209] squareroot_glszm_SmallAreaLowGrayLevelEmphasis

[1210] squareroot_glszm_ZoneEntropy

[1211] squareroot_glszm_ZonePercentage

[1212] squareroot_glszm_ZoneVariance

[1213] squareroot_ngtdm_Busyness

[1214] squareroot_ngtdm_Coarseness

[1215] squareroot_ngtdm_Complexity

[1216] squareroot_ngtdm_Contrast

[1217] squareroot_ngtdm_Strength

[1218] exponential_first_order_10Percentile

[1219] exponential_firstorder_90Percentile

[1220] exponential_firstorder_Energy

[1221] exponential_firstorder_Entropy

[1222] exponential_firstorder_InterquartileRange

[1223] exponential_firstorder_Kurtosis

[1224] exponential_firstorder_Maximum

[1225] exponential_firstorder_MeanAbsoluteDeviation

[1226] exponential_firstorder_Mean

[1227] exponential_firstorder_Median

[1228] exponential_firstorder_Minimum

[1229] exponential_firstorder_Range

[1230] exponential_firstorder_RobustMeanAbsoluteDeviation

[1231] exponential_firstorder_RootMeanSquared

[1232] exponential_firstorder_Skewness

[1233] exponential_firstorder_TotalEnergy

[1234] exponential_firstorder_Variance

[1235] exponential_glcm_DifferenceEntropy

[1236] exponential_glcm_JointEntropy

[1237] exponential_glcm_SumEntropy

[1238] exponential_gldm_DependenceEntropy

[1239] exponential_gldm_DependenceNonUniformity

[1240] exponential_gldm_DependenceNonUniformityNormalized

[1241] exponential_gldm_DependenceVariance

[1242] exponential_gldm_LargeDependenceEmphasis

[1243] exponential_gldm_LargeDependenceHighGrayLevelEmphasis

[1244] exponential_gldm_LargeDependenceLowGrayLevelEmphasis

[1245] exponential_gldm_SmallDependenceEmphasis

[1246] exponential_gldm_SmallDependenceHighGrayLevelEmphasis

[1247] exponential_gldm_SmallDependenceLowGrayLevelEmphasis

[1248] exponential_glrlm_GrayLevelNonUniformity

[1249] exponential_glrlm_LongRunEmphasis

[1250] exponential_glrlm_LongRunHighGrayLevelEmphasis

[1251] exponential_glrlm_LongRunLowGrayLevelEmphasis

[1252] exponential_glrlm_RunEntropy

[1253] exponential_glrlm_RunLengthNonUniformity

[1254] exponential_glrlm_RunLengthNonUniformityNormalized

[1255] exponential_glrlm_RunPercentage

[1256] exponential_glrlm_RunVariance

[1257] exponential_glrlm_ShortRunEmphasis

[1258] exponential_glrlm_ShortRunHighGrayLevelEmphasis

[1259] exponential_glrlm_ShortRunLowGrayLevelEmphasis

[1260] exponential_glszm_LargeAreaEmphasis

[1261] exponential_glszm_LargeAreaHighGrayLevelEmphasis

[1262] exponential_glszm_LargeAreaLowGrayLevelEmphasis

[1263] exponential_glszm_SizeZoneNonUniformityNormalized

[1264] exponential_glszm_SmallAreaEmphasis

[1265] exponential_glszm_SmallAreaHighGrayLevelEmphasis

[1266] exponential_glszm_SmallAreaLowGrayLevelEmphasis

[1267] exponential_glszm_ZoneEntropy

[1268] exponential_glszm_ZonePercentage

[1269] exponential_glszm_ZoneVariance

[1270] logarithm_firstorder_10Percentile

[1271] logarithm_firstorder_90Percentile

[1272] logarithm_firstorder_Energy

[1273] logarithm_firstorder_Entropy

[1274] logarithm_firstorder_InterquartileRange

[1275] logarithm_firstorder_Kurtosis

[1276] logarithm_firstorder_Maximum

[1277] logarithm_firstorder_MeanAbsoluteDeviation

[1278] logarithm_firstorder_Mean

[1279] logarithm_firstorder_Median

[1280] logarithm_firstorder_Minimum

[1281] logarithm_firstorder_Range

[1282] logarithm_firstorder_RobustMeanAbsoluteDeviation

[1283] logarithm_firstorder_RootMeanSquared

[1284] logarithm_firstorder_Skewness

[1285] logarithm_firstorder_TotalEnergy

[1286] logarithm_firstorder_Uniformity

[1287] logarithm_firstorder_Variance

[1288] logarithm_glcm_Autocorrelation

[1289] logarithm_glcm_ClusterProminence

[1290] logarithm_glcm_ClusterShade

[1291] logarithm_glcm_ClusterTendency

[1292] logarithm_glcm_Contrast

[1293] logarithm_glcm_Correlation

[1294] logarithm_glcm_DifferenceAverage

[1295] logarithm_glcm_DifferenceEntropy

[1296] logarithm_glcm_DifferenceVariance

[1297] logarithm_glcm_Id

[1298] logarithm_glcm_Idm

[1299] logarithm_glcm_Idmn

[1300] logarithm_glcm_Idn

[1301] logarithm_glcm_Imc1

[1302] logarithm_glcm_Imc2

[1303] logarithm_glcm_InverseVariance

[1304] logarithm_glcm_JointAverage

[1305] logarithm_glcm_JointEnergy

[1306] logarithm_glcm_JointEntropy

[1307] logarithm_glcm_MCC

[1308] logarithm_glcm_MaximumProbability

[1309] logarithm_glcm_SumAverage

[1310] logarithm_glcm_SumEntropy

[1311] logarithm_glcm_SumSquares

[1312] logarithm_gldm_DependenceEntropy

[1313] logarithm_gldm_DependenceNonUniformity

[1314] logarithm_gldm_DependenceNonUniformityNormalized

[1315] logarithm_gldm_DependenceVariance

[1316] logarithm_gldm_GrayLevelNonUniformity

[1317] logarithm_gldm_GrayLevelVariance

[1318] logarithm_gldm_HighGrayLevelEmphasis

[1319] logarithm_gldm_LargeDependenceEmphasis

[1320] logarithm_gldm_LargeDependenceHighGrayLevelEmphasis

[1321] logarithm_gldm_LargeDependenceLowGrayLevelEmphasis

[1322] logarithm_gldm_LowGrayLevelEmphasis

[1323] logarithm_gldm_SmallDependenceEmphasis

[1324] logarithm_gldm_SmallDependenceHighGrayLevelEmphasis

[1325] logarithm_gldm_SmallDependenceLowGrayLevelEmphasis

[1326] logarithm_glrlm_GrayLevelNonUniformity

[1327] logarithm_glrlm_GrayLevelNonUniformityNormalized

[1328] logarithm_glrlm_GrayLevelVariance

[1329] logarithm_glrlm_HighGrayLevelRunEmphasis

[1330] logarithm_glrlm_LongRunEmphasis

[1331] logarithm_glrlm_LongRunHighGrayLevelEmphasis

[1332] logarithm_glrlm_LongRunLowGrayLevelEmphasis

[1333] logarithm_glrlm_LowGrayLevelRunEmphasis

[1334] logarithm_glrlm_RunEntropy

[1335] logarithm_glrlm_RunLengthNonUniformity

[1336] logarithm_glrlm_RunLengthNonUniformityNormalized

[1337] logarithm_glrlm_RunPercentage

[1338] logarithm_glrlm_RunVariance

[1339] logarithm_glrlm_ShortRunEmphasis

[1340] logarithm_glrlm_ShortRunHighGrayLevelEmphasis

[1341] logarithm_glrlm_ShortRunLowGrayLevelEmphasis

[1342] logarithm_glszm_GrayLevelNonUniformity

[1343] logarithm_glszm_GrayLevelNonUniformityNormalized

[1344] logarithm_glszm_GrayLevelVariance

[1345] logarithm_glszm_HighGrayLevelZoneEmphasis

[1346] logarithm_glszm_LargeAreaEmphasis

[1347] logarithm_glszm_LargeAreaHighGrayLevelEmphasis

[1348] logarithm_glszm_LargeAreaLowGrayLevelEmphasis

[1349] logarithm_glszm_LowGrayLevelZoneEmphasis

[1350] logarithm_glszm_SizeZoneNonUniformity

[1351] logarithm_glszm_SizeZoneNonUniformityNormalized

[1352] logarithm_glszm_SmallAreaEmphasis

[1353] logarithm_glszm_SmallAreaHighGrayLevelEmphasis

[1354] logarithm_glszm_SmallAreaLowGrayLevelEmphasis

[1355] logarithm_glszm_ZoneEntropy

[1356] logarithm_glszm_ZonePercentage

[1357] logarithm_glszm_ZoneVariance

[1358] logarithm_ngtdm_Busyness

[1359] logarithm_ngtdm_Coarseness

[1360] logarithm_ngtdm_Complexity

[1361] logarithm_ngtdm_Contrast

[1362] logarithm_ngtdm_Strength

[1363] gradient_firstorder_10Percentile

[1364] gradient_firstorder_90Percentile

[1365] gradient_firstorder_Energy

[1366] gradient_firstorder_Entropy

[1367] gradient_firstorder_InterquartileRange

[1368] gradient_firstorder_Kurtosis

[1369] gradient_firstorder_Maximum

[1370] gradient_firstorder_MeanAbsoluteDeviation

[1371] gradient_firstorder_Mean

[1372] gradient_firstorder_Median

[1373] gradient_firstorder_Minimum

[1374] gradient_firstorder_Range

[1375] gradient_firstorder_RobustMeanAbsoluteDeviation

[1376] gradient_firstorder_RootMeanSquared

[1377] gradient_firstorder_Skewness

[1378] gradient_firstorder_TotalEnergy

[1379] gradient_firstorder_Variance

[1380] gradient_glcm_DifferenceEntropy

[1381] gradient_glcm_JointEntropy

[1382] gradient_glcm_SumEntropy

[1383] gradient_gldm_DependenceEntropy

[1384] gradient_gldm_DependenceNonUniformity

[1385] gradient_gldm_DependenceNonUniformityNormalized

[1386] gradient_gldm_DependenceVariance

[1387] gradient_gldm_LargeDependenceEmphasis

[1388] gradient_gldm_LargeDependenceHighGrayLevelEmphasis

[1389] gradient_gldm_LargeDependenceLowGrayLevelEmphasis

[1390] gradient_gldm_SmallDependenceEmphasis

[1391] gradient_gldm_SmallDependenceHighGrayLevelEmphasis

[1392] gradient_gldm_SmallDependenceLowGrayLevelEmphasis

[1393] gradient_glrlm_GrayLevelNonUniformity

[1394] gradient_glrlm_LongRunEmphasis

[1395] gradient_glrlm_LongRunHighGrayLevelEmphasis

[1396] gradient_glrlm_LongRunLowGrayLevelEmphasis

[1397] gradient_glrlm_RunEntropy

[1398] gradient_glrlm_RunLengthNonUniformity

[1399] gradient_glrlm_RunLengthNonUniformityNormalized

[1400] gradient_glrlm_RunPercentage

[1401] gradient_glrlm_RunVariance

[1402] gradient_glrlm_ShortRunEmphasis

[1403] gradient_glrlm_ShortRunHighGrayLevelEmphasis

[1404] gradient_glrlm_ShortRunLowGrayLevelEmphasis

[1405] gradient_glszm_LargeAreaEmphasis

[1406] gradient_glszm_LargeAreaHighGrayLevelEmphasis

[1407] gradient_glszm_LargeAreaLowGrayLevelEmphasis

[1408] gradient_glszm_SizeZoneNonUniformityNormalized

[1409] gradient_glszm_SmallAreaEmphasis

[1410] gradient_glszm_SmallAreaHighGrayLevelEmphasis

[1411] gradient_glszm_SmallAreaLowGrayLevelEmphasis

[1412] gradient_glszm_ZoneEntropy

[1413] gradient_glszm_ZonePercentage

[1414] gradient_glszm_ZoneVariance

[1415] diagnostics_Mask-original_VoxelNum

[1416] diagnostics_Mask-original_VolumeNum

[1417] diagnostics_Mask-interpolated_VoxelNum

[1418] diagnostics_Mask-interpolated_VolumeNum

[1419] square_firstorder_Uniformity

[1420] square_glcm_Autocorrelation

[1421] square_glcm_ClusterProminence

[1422] square_glcm_ClusterShade

[1423] square_glcm_ClusterTendency

[1424] square_glcm_Contrast

[1425] square_glcm_Correlation

[1426] square_glcm_DifferenceAverage

[1427] square_glcm_DifferenceVariance

[1428] square_glcm_Id

[1429] square_glcm_Idm

[1430] square_glcm_Idmn

[1431] square_glcm_Idn

[1432] square_glcm_Imc1

[1433] square_glcm_Imc2

[1434] square_glcm_InverseVariance

[1435] square_glcm_JointAverage

[1436] square_glcm_JointEnergy

[1437] square_glcm_MCC

[1438] square_glcm_MaximumProbability

[1439] square_glcm_SumAverage

[1440] square_glcm_SumSquares

[1441] square_gldm_GrayLevelNonUniformity

[1442] square_gldm_GrayLevelVariance

[1443] square_gldm_HighGrayLevelEmphasis

[1444] square_gldm_LowGrayLevelEmphasis

[1445] square_glrlm_GrayLevelNonUniformityNormalized

[1446] square_glrlm_GrayLevelVariance

[1447] square_glrlm_HighGrayLevelRunEmphasis

[1448] square_glrlm_LowGrayLevelRunEmphasis

[1449] square_glszm_GrayLevelNonUniformity

[1450] square_glszm_GrayLevelNonUniformityNormalized

[1451] square_glszm_GrayLevelVariance

[1452] square_glszm_HighGrayLevelZoneEmphasis

[1453] square_glszm_LowGrayLevelZoneEmphasis

[1454] square_glszm_SizeZoneNonUniformity

[1455] square_ngtdm_Busyness

[1456] square_ngtdm_Coarseness

[1457] square_ngtdm_Complexity

[1458] square_ngtdm_Contrast

[1459] square_ngtdm_Strength

[1460] exponential_firstorder_Uniformity

[1461] exponential_glcm_Autocorrelation

[1462] exponential_glcm_ClusterProminence

[1463] exponential_glcm_ClusterShade

[1464] exponential_glcm_ClusterTendency

[1465] exponential_glcm_Contrast

[1466] exponential_glcm_Correlation

[1467] exponential_glcm_DifferenceAverage

[1468] exponential_glcm_DifferenceVariance

[1469] exponential_glcm_Id

[1470] exponential_glcm_Idm

[1471] exponential_glcm_Idmn

[1472] exponential_glcm_Idn

[1473] exponential_glcm_Imc1

[1474] exponential_glcm_Imc2

[1475] exponential_glcm_InverseVariance

[1476] exponential_glcm_JointAverage

[1477] exponential_glcm_JointEnergy

[1478] exponential_glcm_MCC

[1479] exponential_glcm_MaximumProbability

[1480] exponential_glcm_SumAverage

[1481] exponential_glcm_SumSquares

[1482] exponential_gldm_GrayLevelNonUniformity

[1483] exponential_gldm_GrayLevelVariance

[1484] exponential_gldm_HighGrayLevelEmphasis

[1485] exponential_gldm_LowGrayLevelEmphasis

[1486] exponential_glrlm_GrayLevelNonUniformityNormalized

[1487] exponential_glrlm_GrayLevelVariance

[1488] exponential_glrlm_HighGrayLevelRunEmphasis

[1489] exponential_glrlm_LowGrayLevelRunEmphasis

[1490] exponential_glszm_GrayLevelNonUniformity

[1491] exponential_glszm_GrayLevelNonUniformityNormalized

[1492] exponential_glszm_GrayLevelVariance

[1493] exponential_glszm_HighGrayLevelZoneEmphasis

[1494] exponential_glszm_LowGrayLevelZoneEmphasis

[1495] exponential_glszm_SizeZoneNonUniformity

[1496] exponential_ngtdm_Busyness

[1497] exponential_ngtdm_Coarseness

[1498] exponential_ngtdm_Complexity

[1499] exponential_ngtdm_Contrast

[1500] exponential_ngtdm_Strength

[1501] gradient_firstorder_Uniformity

[1502] gradient_glcm_Autocorrelation

[1503] gradient_glcm_ClusterProminence

[1504] gradient_glcm_ClusterShade

[1505] gradient_glcm_ClusterTendency

[1506] gradient_glcm_Contrast

[1507] gradient_glcm_Correlation

[1508] gradient_glcm_DifferenceAverage

[1509] gradient_glcm_DifferenceVariance

[1510] gradient_glcm_Id

[1511] gradient_glcm_Idm

[1512] gradient_glcm_Idmn

[1513] gradient_glcm_Idn

[1514] gradient_glcm_Imc1

[1515] gradient_glcm_Imc2

[1516] gradient_glcm_InverseVariance

[1517] gradient_glcm_JointAverage

[1518] gradient_glcm_JointEnergy

[1519] gradient_glcm_MCC

[1520] gradient_glcm_MaximumProbability

[1521] gradient_glcm_SumAverage

[1522] gradient_glcm_SumSquares

[1523] gradient_gldm_GrayLevelNonUniformity

[1524] gradient_gldm_GrayLevelVariance

[1525] gradient_gldm_HighGrayLevelEmphasis

[1526] gradient_gldm_LowGrayLevelEmphasis

[1527] gradient_glrlm_GrayLevelNonUniformityNormalized

[1528] gradient_glrlm_GrayLevelVariance

[1529] gradient_glrlm_HighGrayLevelRunEmphasis

[1530] gradient_glrlm_LowGrayLevelRunEmphasis

[1531] gradient_glszm_GrayLevelNonUniformity

[1532] gradient_glszm_GrayLevelNonUniformityNormalized

[1533] gradient_glszm_GrayLevelVariance

[1534] gradient_glszm_HighGrayLevelZoneEmphasis

[1535] gradient_glszm_LowGrayLevelZoneEmphasis

[1536] gradient_glszm_SizeZoneNonUniformity

[1537] gradient_ngtdm_Busyness

[1538] gradient_ngtdm_Coarseness

[1539] gradient_ngtdm_Complexity

[1540] gradient_ngtdm_Contrast

[1541] gradient_ngtdm_Strength.

[1542] gradient_ngtdm_Complexity

[1543] gradient_ngtdm_Contrast

[1544] gradient_ngtdm_Strength.

[1545]

[1546] Afterwards, in order to include only features with high reproducibility between two different readers in the analysis, the ICC (Intraclass Correlation Coefficient) was calculated for each feature, and only image information with an ICC value of 0.95 or higher was selected as the final analysis target.

[1547]

[1548] Handling data imbalances

[1549] To solve the class imbalance problem in the training data set, positive sample amplification was performed using the Synthetic Minority Oversampling Technique (SMOTE) technique (k_neighbors=5).

[1550]

[1551] Feature Selection Method

[1552] The following four feature selection methods were applied to image information with an ICC value of 0.95 or higher selected as the final analysis target: No FS (NO Feature Selection), Corr (Correlation-based Selection), LASSO (Least Absolute Shrinkage and Selection Operator), and FFS (Forward Feature Selection).

[1553]

[1554] Classifier and model development

[1555] A prediction model was developed to classify the presence or absence of BRCA gene mutation (BRCA mutation vs. wild-type) by combining the above four feature selection methods and the following four binary classifiers: Adab (Adaptive Boosting), RF (Random Forest), SGD (Stochastic Gradient Descent), and LDA (Linear Discriminant Analysis).

[1556]

[1557] Software and Libraries

[1558] To train and validate the BRCA gene mutation prediction model according to the present invention, Python (version 3.11.8), scikit-learn (version 1.5.0), PyRadiomics (version 3.0.1), and ITK-SNAP (version 4.0.1) were used, and NumPy, Pandas, Matplotlib, and Seaborn were used as analysis libraries. In addition, SHAP (version 0.45.1) was used as an analysis tool.

[1559]

[1560] Reliability verification of each BRCA gene mutation prediction model

[1561] We evaluated the performance of BRCA gene mutation prediction models using four feature selection methods and four binary classifier combinations. The AUROC (Area Under the Receiver Operating Characteristic Curve), TPR (sensitivity), TNR (specificity), PPV (precision), NPV (negative predictive value), FPR (false positive rate), FNR (false negative rate), FDR (false discovery rate), ACC (accuracy), and confusion matrix for a total of eight models were calculated.

[1562] AUROC (Area Under the Receiver Operating Characteristic Curve) is an indicator of the overall classification performance of the model; TPR (True Positive Rate, sensitivity) is the proportion of actual positives that the model correctly predicts as positive; TNR (True Negative Rate, specificity) is the proportion of actual negatives that the model correctly predicts as negative; PPV (Positive Predictive Value, precision) is the proportion of actual positives predicted as positive; NPV (Negative Predictive Value, negative predictive value) is the proportion of actual negatives predicted as negative; FPR (False Positive Rate, false positive rate) is the proportion of actual negatives incorrectly predicted as positive; FNR (False Negative Rate, false negative rate) is the proportion of actual positives incorrectly predicted as negative; FDR (False Discovery Rate, false discovery rate) is the proportion of actual negatives predicted as positives; ACC (Accuracy, accuracy) is the proportion of all samples that are correctly predicted.

[1563]

[1564] 1. Combination of FFS feature selection method and Adab classifier

[1565] (1) Information on the prediction model for BRCA gene mutation

[1566] Forward Feature Selection (FFS) was adopted as a feature selection method, and the Adaptive Boosting (Adab) classifier was used to predict BRCA gene mutations. To ensure high reliability, only image information with an ICC value of 0.95 or higher was extracted. As a result, 32 image information with an ICC value of 0.95 were extracted from a total of 1,422 image information.

[1567] (2) Data set composition

[1568] The training dataset consisted of 199 MRI images of prostate cancer patients obtained from Severance Hospital in Sinchon, including 33 samples with BRCA mutations and 166 wild-type samples without BRCA mutations. The validation dataset consisted of 158 MRI images of prostate cancer patients obtained from Seoul National University Bundang Hospital, including 21 samples with BRCA mutations and 137 wild-type samples without BRCA mutations.

[1569] (3) Summary of overall performance indicators

[1570] Tables 2 through 5 below show the overall performance of the combined model of the FFS feature selection method and the Adab classifier. Table 2 also shows the overall performance of the training data set. The AUROC (Area Under the Receiver Operating Characteristic Curve) value was measured at 0.852, demonstrating that the model effectively distinguishes between classes (Fig. 2a).

[1571] Index measurement value 95% confidence interval AUROC 0.852 - TPR (sensitivity) 0.825 (0.760, 0.876) TNR (specificity) 0.723 (0.650, 0.785) PPV (precision) 0.749 (0.681, 0.806) NPV (negative predictive value) 0.805 (0.734, 0.861) FPR (false positive rate) 0.277 (0.215, 0.350) FNR (false negative rate) 0.175 (0.124, 0.240) FDR (false discovery rate) 0.251 (0.194, 0.319) ACC (accuracy) 0.774 (0.726, 0.816)

[1572] Table 3 below shows the overall performance of the test set. The AUROC (Area Under the Receiver Operating Characteristic Curve) value was measured at 0.794, demonstrating that the model effectively distinguishes between classes (Fig. 2b).

[1573] Indicator 95% confidence interval AUROC 0.794 - TPR (sensitivity) 0.476 (0.283, 0.676) TNR (specificity) 0.898 (0.836, 0.938) PPV (precision) 0.417 (0.245, 0.612) NPV (negative predictive value) 0.918 (0.859, 0.954) FPR (false positive rate) 0.102 (0.062, 0.164) FNR (false negative rate) 0.524 (0.324, 0.717) FDR (false discovery rate) 0.583 (0.388, 0.755) ACC (accuracy) 0.842 (0.777, 0.890)

[1574] Table 4 below shows the results of calculating the confusion matrix of the training data set, and Table 5 shows the results of calculating the confusion matrix of the validation data set. TP below is the number of cases in which the model correctly predicted a subject with an actual BRCA mutation as a subject with a BRCA mutation; TN below is the number of cases in which the model correctly predicted a subject without an actual BRCA mutation as a subject without a BRCA mutation; FP below is the number of cases in which the model incorrectly predicted a subject without an actual BRCA mutation as a subject with a BRCA mutation; and FN below is the number of cases in which the model incorrectly predicted a subject with an actual BRCA mutation as a subject without a BRCA mutation.

[1575]

[1576] Predicted: BRCA mutation present Predicted: BRCA mutation present Total Actual: BRCA mutation absent 120 (TN) 46 (FP) 166 Actual: BRCA mutation present 29 (FN) 137 (TP) 166 Total 149 183 332

[1577] Predicted: BRCA mutation present Predicted: BRCA mutation present Total Actual: BRCA mutation absent 123 (TN) 14 (FP) 137 Actual: BRCA mutation present 11 (FN) 10 (TP) 21 Total 13424158

[1578] (4) Feature Importance Analysis

[1579] For the trained model, we performed feature importance analysis based on permutation importance using the Shapley Additive Explanations (SHAP) library. The analysis was performed on the training dataset, and the results showed that the top features contributing to the prediction decision were, in order, squareroot_firstorder_InterquartileRange, exponential_firstorder_MeanAbsoluteDeviation, and wavelet-HHH_glszm_ZoneVariance. The mean importance of each feature was calculated to be 0.150, 0.111, and 0.055, respectively (Fig. 2c). In addition, feature importance analysis based on permutation importance was also performed on the test dataset. According to the results derived from the above analysis, the main features with high contribution to the prediction decision were wavelet-HHH_glszm_ZoneVariance, original_glrlm_RunLengthNonUniformity, squareroot_firstorder_InterquartileRange, and exponential_firstorder_MeanAbsoluteDeviation, in that order. The respective importance averages were calculated as 0.165, 0.058, 0.035, and -0.003, respectively, and it can be confirmed that the results differ from the importance ranking compared to the training data (Fig. 2d).

[1580]

[1581] 2. Combination of LASSO feature selection method and Adab classifier

[1582] (1) Information on the prediction model for BRCA gene mutation

[1583] The Least Absolute Shrinkage and Selection Operator (LASSO) method was adopted as a feature selection method, and the Adaptive Boosting (Adab) classifier was used to predict BRCA gene mutations. To ensure high reliability, only image information with an ICC value of 0.95 or higher was extracted. As a result, 39 images with an ICC value of 0.95 were extracted from a total of 1,422 images.

[1584] (2) Data set composition

[1585] The training dataset consisted of 199 MRI images of prostate cancer patients obtained from Severance Hospital in Sinchon, including 33 samples with BRCA mutations and 166 wild-type samples without BRCA mutations. The validation dataset consisted of 158 MRI images of prostate cancer patients obtained from Seoul National University Bundang Hospital, including 21 samples with BRCA mutations and 137 wild-type samples without BRCA mutations.

[1586] (3) Summary of overall performance indicators

[1587] Tables 6 through 9 below show the overall performance of the combined model using the LASSO feature selection method and the Adab classifier. Table 6 also shows the overall performance of the training data set. The AUROC (Area Under the Receiver Operating Characteristic Curve) value was measured at 0.852, demonstrating that the model effectively distinguishes between classes (Fig. 3a).

[1588] Index measurement value 95% confidence interval AUROC 0.852 - TPR (sensitivity) 0.825 (0.760, 0.876) TNR (specificity) 0.723 (0.650, 0.785) PPV (precision) 0.749 (0.681, 0.806) NPV (negative predictive value) 0.805 (0.734, 0.861) FPR (false positive rate) 0.277 (0.215, 0.350) FNR (false negative rate) 0.175 (0.124, 0.240) FDR (false discovery rate) 0.251 (0.194, 0.319) ACC (accuracy) 0.774 (0.726, 0.816)

[1589] Table 7 below shows the overall performance of the test set. The AUROC (Area Under the Receiver Operating Characteristic Curve) value was measured at 0.802, demonstrating that the model effectively distinguishes between classes (Fig. 3b).

[1590] Indicator 95% confidence interval AUROC 0.802 - TPR (sensitivity) 0.667 (0.454, 0.828) TNR (specificity) 0.854 (0.785, 0.903) PPV (precision) 0.412 (0.264, 0.578) NPV (negative predictive value) 0.944 (0.888, 0.972) FPR (false positive rate) 0.146 (0.097, 0.215) FNR (false negative rate) 0.333 (0.172, 0.546) FDR (false discovery rate) 0.588 (0.422, 0.736) ACC (accuracy) 0.829 (0.763, 0.880)

[1591] Table 8 below shows the results of calculating the confusion matrix of the training data set, and Table 9 below shows the results of calculating the confusion matrix of the validation data set. TP below is the number of cases in which the model correctly predicted a subject with an actual BRCA mutation as a subject with a BRCA mutation; TN below is the number of cases in which the model correctly predicted a subject without an actual BRCA mutation as a subject without a BRCA mutation; FP below is the number of cases in which the model incorrectly predicted a subject without an actual BRCA mutation as a subject with a BRCA mutation; and FN below is the number of cases in which the model incorrectly predicted a subject with an actual BRCA mutation as a subject without a BRCA mutation.

[1592] Predicted: BRCA mutation present Predicted: BRCA mutation present Total Actual: BRCA mutation absent 120 (TN) 46 (FP) 166 Actual: BRCA mutation present 29 (FN) 137 (TP) 166 Total 149 183 332

[1593] Predicted: BRCA mutation present Predicted: BRCA mutation present Total Actual: BRCA mutation absent 117 (TN) 20 (FP) 137 Actual: BRCA mutation present 7 (FN) 14 (TP) 21 Total 12434158

[1594] (4) Feature Importance Analysis

[1595] For the trained model, we performed feature importance analysis based on permutation importance using the Shapley Additive Explanations (SHAP) library. The analysis was performed on the training dataset, and the results showed that the top features contributing to the prediction decision were, in order, squareroot_firstorder_InterquartileRange, exponential_firstorder_MeanAbsoluteDeviation, and wavelet-HHH_glszm_ZoneVariance. The average importance of each feature was calculated to be 0.150, 0.111, and 0.055, respectively (Fig. 3c). In addition, feature importance analysis based on permutation importance was also performed on the test dataset. According to the results derived from the above analysis, the main features with high contribution to the prediction decision were wavelet-HHH_glszm_ZoneVariance, wavelet-LLL_glrlm_RunLengthNonUniformity, and squareroot_firstorder_InterquartileRange, in that order. The average importance of each was calculated to be 0.155, 0.060, and 0.035, respectively, and it can be confirmed that the results differ from the importance ranking compared to the training data (Fig. 3d).

[1596]

[1597] 3. Combination of LASSO feature selection method and RF classifier

[1598] (1) Information on the prediction model for BRCA gene mutation

[1599] The Least Absolute Shrinkage and Selection Operator (LASSO) method was adopted as a feature selection method, and the presence of BRCA gene mutations was predicted using the Random Forest (RF) classifier. To ensure high reliability, only image information with an ICC value of 0.95 or higher was extracted. As a result, 39 image information with an ICC value of 0.95 were extracted from a total of 1,422 image information.

[1600] (2) Data set composition

[1601] The training dataset consisted of 199 MRI images of prostate cancer patients obtained from Severance Hospital in Sinchon, including 33 samples with BRCA mutations and 166 wild-type samples without BRCA mutations. The validation dataset consisted of 158 MRI images of prostate cancer patients obtained from Seoul National University Bundang Hospital, including 21 samples with BRCA mutations and 137 wild-type samples without BRCA mutations.

[1602] (3) Summary of overall performance indicators

[1603] Tables 10 to 13 below show the overall performance of the combined model of the LASSO feature selection method and the RF classifier. Table 10 also shows the overall performance of the training data set. The AUROC (Area Under the Receiver Operating Characteristic Curve) value was measured to be 0.817, demonstrating that the model effectively distinguishes between classes (Fig. 4a).

[1604] Index measurement value 95% confidence interval AUROC 0.817 - TPR (sensitivity) 0.861 (0.801, 0.906) TNR (specificity) 0.542 (0.466, 0.616) PPV (precision) 0.653 (0.588, 0.713) NPV (negative predictive value) 0.796 (0.713, 0.860) FPR (false positive rate) 0.458 (0.384, 0.534) FNR (false negative rate) 0.139 (0.094, 0.199) FDR (false discovery rate) 0.347 (0.287, 0.412) ACC (accuracy) 0.702 (0.651, 0.748)

[1605] Table 11 below shows the overall performance of the test set. The AUROC (Area Under the Receiver Operating Characteristic Curve) value was measured at 0.756, demonstrating that the model effectively distinguishes between classes (Figure 4b).

[1606] Indicator 95% confidence interval AUROC 0.756 - TPR (sensitivity) 0.619 (0.409, 0.792), TNR (specificity) 0.847 (0.777, 0.898), PPV (precision) 0.382 (0.239, 0.550), NPV (negative predictive value) 0.935 (0.878, 0.967), FPR (false positive rate) 0.153 (0.102, 0.223), FNR (false negative rate) 0.381 (0.208, 0.591), FDR (false discovery rate) 0.618 (0.450, 0.761), ACC (accuracy) 0.816 (0.749, 0.869)

[1607] Table 12 below shows the results of calculating the confusion matrix of the training data set, and Table 13 below shows the results of calculating the confusion matrix of the validation data set. TP below is the number of cases in which the model correctly predicted a subject with an actual BRCA mutation as a subject with a BRCA mutation; TN below is the number of cases in which the model correctly predicted a subject without an actual BRCA mutation as a subject without a BRCA mutation; FP below is the number of cases in which the model incorrectly predicted a subject without an actual BRCA mutation as a subject with a BRCA mutation; and FN below is the number of cases in which the model incorrectly predicted a subject with an actual BRCA mutation as a subject without a BRCA mutation.

[1608]

[1609] Predicted: BRCA mutation present Predicted: BRCA mutation present Total Actual: BRCA mutation absent 90 (TN) 76 (FP) 166 Actual: BRCA mutation present 23 (FN) 143 (TP) 166 Total 113219332

[1610] Predicted: BRCA mutation present Predicted: BRCA mutation present Total Actual: BRCA mutation absent 116 (TN) 21 (FP) 137 Actual: BRCA mutation present 8 (FN) 13 (TP) 21 Total 12434158

[1611] (4) Feature Importance Analysis

[1612] For the trained model, we performed feature importance analysis based on permutation importance using the Shapley Additive Explanations (SHAP) library. The analysis was performed based on the training dataset, and the results showed that the top features contributing to the prediction decision were wavelet-HHH_glszm_ZoneVariance, exponential_firstorder_90Percentile, and squareroot_firstorder_10Percentile, in that order. The average importance of each feature was calculated to be 0.119, 0.056, and 0.029, respectively (Fig. 4c). In addition, feature importance analysis based on permutation importance was also performed on the test dataset. According to the results derived from the above analysis, the main features with high contribution to the prediction decision were wavelet-HHH_glszm_ZoneVariance, diagnostics_Image-interpolated_Maximum, squareroot_firstorder_10Percentile, and wavelet-LLL_glrlm_RunLengthNonUniformity, in that order. The average importance of each was calculated to be 0.243, 0.022, 0.014, and 0.010, respectively, and it can be confirmed that the results are different from the importance ranking compared to the training data (Fig. 4d).

[1613]

[1614] 4. Combination of LASSO feature selection method and SGD classifier

[1615] (1) Information on the prediction model for BRCA gene mutation

[1616] The Least Absolute Shrinkage and Selection Operator (LASSO) method was adopted as a feature selection method, and the presence or absence of BRCA gene mutations was predicted using the Stochastic Gradient Descent (SGD) classifier. To ensure high reliability, only image information with an ICC value of 0.95 or higher was extracted, and as a result, 39 image information with an ICC value of 0.95 were extracted from a total of 1,422 image information.

[1617] (2) Data set composition

[1618] The training dataset consisted of 332 MRI images of prostate cancer patients obtained from Severance Hospital in Sinchon, including 166 samples with BRCA mutations and 166 samples with non-BRCA wild-type mutations. The validation dataset consisted of 158 MRI images of prostate cancer patients obtained from Seoul National University Bundang Hospital, including 21 samples with BRCA mutations and 137 samples with non-BRCA wild-type mutations.

[1619] (3) Summary of overall performance indicators

[1620] Tables 14 to 17 below show the overall performance of the combined model using the LASSO feature selection method and the SGD classifier. Table 14 also shows the overall performance of the training data set. The AUROC (Area Under the Receiver Operating Characteristic Curve) value was measured to be 0.775, demonstrating that the model effectively distinguishes between classes (Fig. 5a).

[1621] Index measurement value 95% confidence interval AUROC 0.775 - TPR (sensitivity) 0.627 (0.551, 0.696) TNR (specificity) 0.789 (0.721, 0.844) PPV (precision) 0.748 (0.670, 0.813) NPV (negative predictive value) 0.679 (0.610, 0.741) FPR (false positive rate) 0.211 (0.156, 0.279) FNR (false negative rate) 0.373 (0.304, 0.449) FDR (false discovery rate) 0.252 (0.187, 0.330) ACC (accuracy) 0.708 (0.657, 0.754)

[1622] Table 15 below shows the overall performance of the test set. The AUROC (Area Under the Receiver Operating Characteristic Curve) value was measured at 0.778, demonstrating that the model effectively distinguishes between classes (Figure 5b).

[1623] Indicator 95% confidence interval AUROC 0.778 - TPR (sensitivity) 0.714 (0.500, 0.862) TNR (specificity) 0.752 (0.673, 0.817) PPV (precision) 0.306 (0.195, 0.445) NPV (negative predictive value) 0.945 (0.885, 0.975) FPR (false positive rate) 0.248 (0.183, 0.327) FNR (false negative rate) 0.286 (0.138, 0.500) FDR (false discovery rate) 0.694 (0.555, 0.805) ACC (accuracy) 0.747 (0.674, 0.808)

[1624] Table 16 below shows the results of calculating the confusion matrix of the training data set, and Table 17 below shows the results of calculating the confusion matrix of the validation data set. TP below is the number of cases in which the model correctly predicted a subject with an actual BRCA mutation as a subject with a BRCA mutation; TN below is the number of cases in which the model correctly predicted a subject without an actual BRCA mutation as a subject without a BRCA mutation; FP below is the number of cases in which the model incorrectly predicted a subject without an actual BRCA mutation as a subject with a BRCA mutation; and FN below is the number of cases in which the model incorrectly predicted a subject with an actual BRCA mutation as a subject without a BRCA mutation.

[1625]

[1626] Predicted: BRCA mutation present Predicted: BRCA mutation present Total Actual: BRCA mutation absent 131 (TN) 35 (FP) 166 Actual: BRCA mutation present 62 (FN) 104 (TP) 166 Total 193139332

[1627] Predicted: BRCA mutation status Predicted: BRCA mutation status Total Actual: BRCA mutation status None 103 (TN) 34 (FP) 137 Actual: BRCA mutation status 6 (FN) 15 (TP) 21 Total 10949158

[1628] (4) Feature Importance Analysis

[1629] For the trained model, we performed a feature importance analysis based on permutation importance using the SHAP (Shapley Additive Explanations) library. The analysis was performed based on the training dataset, and the results showed that the top features that contributed to the prediction decision were, in order, diagnostics_Image-interpolated_Maximum, wavelet-HLL_firstorder_Energy, and wavelet-LHL_firstorder_Energy. The average importance of each feature was calculated as 0.142, 0.0858, and 0.056, respectively (Fig. 5c). In addition, we performed a feature importance analysis based on permutation importance on the test dataset. According to the results derived from the above analysis, the top features with high contributions to the prediction decision were, in order, wavelet-HLL_firstorder_Energy, squareroot_firstorder_Skewness, diagnostics_Image-interpolated_Maximum, and wavelet-LHL_firstorder_Energy. The respective importance averages were calculated as 0.043, 0.037, 0.036, and 0.024, respectively, and it can be confirmed that the results are different from the importance rankings compared to the training data (Fig. 5d).

[1630]

[1631] 5. Combination of LASSO feature selection method and LDA classifier

[1632] (1) Information on the prediction model for BRCA gene mutation

[1633] The Least Absolute Shrinkage and Selection Operator (LASSO) method was adopted as a feature selection method, and the presence of BRCA gene mutations was predicted using the Linear Discriminant Analysis (LDA) classifier. To ensure high reliability, only image information with an ICC value of 0.95 or higher was extracted, and as a result, 39 image information with an ICC value of 0.95 were extracted from a total of 1,422 image information.

[1634] (2) Data set composition

[1635] The training dataset consisted of 119 MRI images of prostate cancer patients obtained from Severance Hospital in Sinchon, including 166 samples with BRCA mutations and 33 samples with non-BRCA wild-type mutations. The validation dataset consisted of 158 MRI images of prostate cancer patients obtained from Seoul National University Bundang Hospital, including 21 samples with BRCA mutations and 137 samples with non-BRCA wild-type mutations.

[1636] (3) Summary of overall performance indicators

[1637] Tables 18 to 21 below show the overall performance of the combined model of the LASSO feature selection method and the LDA classifier. Table 18 also shows the overall performance of the training data set. The AUROC (Area Under the Receiver Operating Characteristic Curve) value was measured at 0.857, demonstrating that the model effectively distinguishes between classes (Figure 6a).

[1638] Index measurement value 95% confidence interval AUROC 0.857 - TPR (sensitivity) 0.813 (0.747, 0.865) TNR (specificity) 0.735 (0.664, 0.794) PPV (precision) 0.754 (0.691, 0.809) NPV (negative predictive value) 0.797 (0.730, 0.851) FPR (false positive rate) 0.265 (0.206, 0.336) FNR (false negative rate) 0.187 (0.135, 0.253) FDR (false discovery rate) 0.246 (0.191, 0.309) ACC (accuracy) 0.774 (0.725, 0.818)

[1639] Table 19 below shows the overall performance of the test set. The AUROC (Area Under the Receiver Operating Characteristic Curve) value was measured at 0.759, demonstrating that the model effectively distinguishes between classes (Figure 6b).

[1640] Indicator 95% confidence interval AUROC 0.759 - TPR (sensitivity) 0.714 (0.500, 0.862) TNR (specificity) 0.701 (0.620, 0.772) PPV (precision) 0.268 (0.162, 0.404) NPV (negative predictive value) 0.941 (0.878, 0.973) FPR (false positive rate) 0.299 (0.228, 0.380) FNR (false negative rate) 0.286 (0.138, 0.500) FDR (false discovery rate) 0.732 (0.596, 0.838) ACC (accuracy) 0.703 (0.626, 0.772)

[1641] Table 20 below shows the results of calculating the confusion matrix of the training data set, and Table 21 shows the results of calculating the confusion matrix of the validation data set. TP below is the number of cases in which the model correctly predicted a subject with an actual BRCA mutation as a subject with a BRCA mutation; TN below is the number of cases in which the model correctly predicted a subject without an actual BRCA mutation as a subject without a BRCA mutation; FP below is the number of cases in which the model incorrectly predicted a subject without an actual BRCA mutation as a subject with a BRCA mutation; and FN below is the number of cases in which the model incorrectly predicted a subject with an actual BRCA mutation as a subject without a BRCA mutation.

[1642]

[1643] Predicted: BRCA mutation present Predicted: BRCA mutation present Total Actual: BRCA mutation absent 122 (TN) 44 (FP) 166 Actual: BRCA mutation present 31 (FN) 135 (TP) 166 Total 153179332

[1644] Predicted: BRCA mutation status Predicted: BRCA mutation status Total Actual: BRCA mutation status None 96 (TN) 41 (FP) 137 Actual: BRCA mutation status 6 (FN) 15 (TP) 21 Total 10256158

[1645] (4) Feature Importance Analysis

[1646] For the trained model, we performed feature importance analysis based on permutation importance using the Shapley Additive Explanations (SHAP) library. The analysis was performed on the training dataset, and the results showed that the top features contributing to the prediction decision were, in order, diagnostics_Image-interpolated_Maximum, wavelet-HLL_firstorder_Energy, and squareroot_firstorder_Skewness. The average importance of each feature was calculated to be 0.155, 0.129, and 0.050, respectively (Fig. 6c). In addition, feature importance analysis based on permutation importance was also performed on the test dataset. According to the results derived from the above analysis, the main features with high contribution to the prediction decision were wavelet-HLL_firstorder_Energy, log-sigma-3-0-mm-3D_glszm_GrayLevelNonUniformity, diagnostics_Image-interpolated_Maximum, and squareroot_firstorder_Skewness, in that order. The average importance of each was calculated to be 0.134, 0.054, 0.050, and 0.042, respectively, and it can be confirmed that the results are different from the importance ranking compared to the training data (Fig. 6d).

[1647]

[1648] 6. Combination of Corr feature selection method and SGD classifier

[1649] (1) Information on the prediction model for BRCA gene mutation

[1650] Correlation-based Selection (Corr) was adopted as a feature selection method, and the presence of BRCA gene mutations was predicted using the Stochastic Gradient Descent (SGD) classifier. To ensure high reliability, only image information with an ICC value of 0.95 or higher was extracted, and as a result, 65 image information with an ICC value of 0.95 were extracted from a total of 1,422 image information.

[1651] (2) Data set composition

[1652] The training dataset consisted of 332 MRI images of prostate cancer patients obtained from Severance Hospital in Sinchon, including 166 samples with BRCA mutations and 166 samples with non-BRCA wild-type mutations. The validation dataset consisted of 158 MRI images of prostate cancer patients obtained from Seoul National University Bundang Hospital, including 21 samples with BRCA mutations and 137 samples with non-BRCA wild-type mutations.

[1653] (3) Summary of overall performance indicators

[1654] Tables 22 to 25 below show the overall performance of the combined model of the Corr feature selection method and the SGD classifier. Table 22 also shows the overall performance of the training data set. The AUROC (Area Under the Receiver Operating Characteristic Curve) value was measured at 0.856, demonstrating that the model effectively distinguishes between classes (Figure 7a).

[1655] Index measurement value 95% confidence interval AUROC 0.856 - TPR (sensitivity) 0.723 (0.650, 0.785) TNR (specificity) 0.801 (0.735, 0.856) PPV (precision) 0.784 (0.720, 0.838) NPV (negative predictive value) 0.743 (0.680, 0.798) FPR (false positive rate) 0.199 (0.144, 0.265) FNR (false negative rate) 0.277 (0.215, 0.350) FDR (false discovery rate) 0.216 (0.162, 0.280) ACC (accuracy) 0.762 (0.713, 0.807)

[1656] Table 23 below shows the overall performance of the test set. The AUROC (Area Under the Receiver Operating Characteristic Curve) value was measured at 0.758, demonstrating that the model effectively distinguishes between classes (Figure 7b).

[1657] Indicator 95% confidence interval AUROC 0.758 - TPR (sensitivity) 0.714 (0.500, 0.862), TNR (specificity) 0.715 (0.620, 0.772), PPV (precision) 0.278 (0.162, 0.404), NPV (negative predictive value) 0.942 (0.878, 0.973), FPR (false positive rate) 0.285 (0.228, 0.380), FNR (false negative rate) 0.286 (0.138, 0.500), FDR (false discovery rate) 0.722 (0.596, 0.838), ACC (accuracy) 0.715 (0.626, 0.772)

[1658] Table 24 below shows the results of calculating the confusion matrix of the training data set, and Table 25 shows the results of calculating the confusion matrix of the validation data set. TP below is the number of cases in which the model correctly predicted a subject with an actual BRCA mutation as a subject with a BRCA mutation; TN below is the number of cases in which the model correctly predicted a subject without an actual BRCA mutation as a subject without a BRCA mutation; FP below is the number of cases in which the model incorrectly predicted a subject without an actual BRCA mutation as a subject with a BRCA mutation; and FN below is the number of cases in which the model incorrectly predicted a subject with an actual BRCA mutation as a subject without a BRCA mutation.

[1659]

[1660] Predicted: BRCA mutation present Predicted: BRCA mutation present Total Actual: BRCA mutation absent 133 (TN) 33 (FP) 166 Actual: BRCA mutation present 46 (FN) 120 (TP) 166 Total 179 153 332

[1661] Predicted: BRCA mutation status Predicted: BRCA mutation status Total Actual: BRCA mutation status None 98 (TN) 39 (FP) 137 Actual: BRCA mutation status 6 (FN) 15 (TP) 21 Total 10,454,158

[1662] (4) Feature Importance Analysis

[1663] For the trained model, we performed feature importance analysis based on permutation importance using the Shapley Additive Explanations (SHAP) library. The analysis was performed on the training dataset, and the results showed that the top features contributing to the prediction decision were, in order, original_glrlm_RunLengthNonUniformity, squareroot_firstorder_10Percentile, and original_shape_MeshVolume. The average importance of each feature was calculated to be 0.188, 0.130, and 0.121, respectively (Fig. 7c). In addition, feature importance analysis based on permutation importance was also performed on the test dataset. According to the results derived from the above analysis, the main features with high contribution to the prediction decision were original_glrlm_RunLengthNonUniformity, wavelet-LHL_firstorder_Energy, squareroot_firstorder_Skewness, and square_glrlm_LongRunEmphasis, in that order. The average importance of each was calculated to be 0.173, 0.113, 0.059, and 0.041, respectively, and it can be confirmed that the results are different from the importance ranking compared to the training data (Fig. 7d).

[1664]

[1665] 7. No FS feature selection method and combination of Adab classifiers

[1666] (1) Information on the prediction model for BRCA gene mutation

[1667] We adopted the No FS (No Feature Selection) method for feature selection, and used the Adab (Adaptive Boosting) classifier to predict BRCA gene mutations. To ensure high reliability, only image data with an ICC value of 0.95 or higher were extracted. As a result, 250 images with an ICC value of 0.95 were extracted from a total of 1,422 images.

[1668] (2) Data set composition

[1669] The training dataset consisted of 199 MRI images of prostate cancer patients obtained from Severance Hospital in Sinchon, Korea. 33 samples harbored BRCA mutations and 166 wild-type samples were non-BRCA mutations. The validation dataset consisted of 158 MRI images of prostate cancer patients obtained from Seoul National University Bundang Hospital, 21 samples harbored BRCA mutations and 137 wild-type samples were non-BRCA mutations.

[1670] (3) Summary of overall performance indicators

[1671] Tables 26 to 29 below show the overall performance of the model combining the NoFS feature selection method and the Adab classifier. Table 26 shows the overall performance of the training data set. The AUROC (Area Under the Receiver Operating Characteristic Curve) value was measured at 0.880, demonstrating that the model effectively distinguishes between classes (Figure 8a).

[1672] Index measurement value 95% confidence interval AUROC 0.880 - TPR (sensitivity) 0.849 (0.787, 0.896) TNR (specificity) 0.831 (0.767, 0.882) PPV (precision) 0.834 (0.775, 0.881) NPV (negative predictive value) 0.847 (0.784, 0.895) FPR (false positive rate) 0.169 (0.118, 0.233) FNR (false negative rate) 0.151 (0.104, 0.213) FDR (false discovery rate) 0.166 (0.119, 0.225) ACC (accuracy) 0.840 (0.794, 0.879)

[1673] Table 27 below shows the overall performance of the test set. The AUROC (Area Under the Receiver Operating Characteristic Curve) value was measured at 0.778, demonstrating that the model effectively distinguishes between classes (Figure 8b).

[1674] Indicator 95% confidence interval AUROC 0.778 - TPR (sensitivity) 0.333 (0.165, 0.537), TNR (specificity) 0.905 (0.840, 0.947), PPV (precision) 0.350 (0.164, 0.590), NPV (negative predictive value) 0.899 (0.837, 0.940), FPR (false positive rate) 0.095 (0.053, 0.160), FNR (false negative rate) 0.667 (0.463, 0.835), FDR (false discovery rate) 0.650 (0.410, 0.836), ACC (accuracy) 0.829 (0.760, 0.883)

[1675] Table 28 below shows the results of calculating the confusion matrix of the training data set, and Table 29 below shows the results of calculating the confusion matrix of the validation data set. TP below is the number of cases in which the model correctly predicted a subject with an actual BRCA mutation as a subject with a BRCA mutation; TN below is the number of cases in which the model correctly predicted a subject without an actual BRCA mutation as a subject without a BRCA mutation; FP below is the number of cases in which the model incorrectly predicted a subject without an actual BRCA mutation as a subject with a BRCA mutation; and FN below is the number of cases in which the model incorrectly predicted a subject with an actual BRCA mutation as a subject without a BRCA mutation.

[1676]

[1677] Predicted: BRCA mutation present Predicted: BRCA mutation present Total Actual: BRCA mutation absent 138 (TN) 28 (FP) 166 Actual: BRCA mutation present 25 (FN) 141 (TP) 166 Total 163 169 332

[1678] Predicted: BRCA mutation status Predicted: BRCA mutation status Total Actual: BRCA mutation status None 124 (TN) 13 (FP) 137 Actual: BRCA mutation status 14 (FN) 7 (TP) 21 Total 138 20 158

[1679] (4) Feature Importance Analysis

[1680] Feature Importance analysis based on permutation importance was performed on the trained model using the Shapley Additive Explanations (SHAP) library. The analysis was performed based on the training dataset, and the results showed that the top features contributing to the prediction decision were wavelet-HHH_glszm_ZoneVariance, exponential_firstorder_RobustMeanAbsoluteDeviation, and diagnostics_Image-original_Maximum, in that order. The average importance of each feature was calculated as 0.194, 0.099, and 0.052, respectively. In addition, feature importance analysis based on permutation importance was also performed on the test dataset (Fig. 8c). According to the results derived from the above analysis, the main features with high contribution to the prediction decision were wavelet-HHH_glszm_ZoneVariance, wavelet-HLL_glszm_ZoneVariance, wavelet-LLL_glrlm_RunLengthNonUniformity, and logarithm_glrlm_RunLengthNonUniformity, in that order. The average importance of each was calculated to be 0.125, 0.039, 0.014, and 0.010, respectively, and it can be confirmed that the results are different from the importance ranking compared to the training data (Fig. 8d).

[1681]

[1682] 8. Combination of FFS feature selection method and RF classifier

[1683] (1) Information on the prediction model for BRCA gene mutation

[1684] Forward Feature Selection (FFS) was adopted as a feature selection method, and the presence of BRCA gene mutations was predicted using a Random Forest (RF) classifier. To ensure high reliability, only image information with an ICC value of 0.95 or higher was extracted. As a result, 32 image information with an ICC value of 0.95 were extracted from a total of 1,422 image information.

[1685] (2) Data set composition

[1686] The training dataset consisted of 199 MRI images of prostate cancer patients obtained from Severance Hospital in Sinchon, Korea. 33 samples harbored BRCA mutations and 166 wild-type samples were non-BRCA mutations. The validation dataset consisted of 158 MRI images of prostate cancer patients obtained from Seoul National University Bundang Hospital, 21 samples harbored BRCA mutations and 137 wild-type samples were non-BRCA mutations.

[1687] (3) Summary of overall performance indicators

[1688] Tables 30 to 33 below show the overall performance of the combined model of the FFS feature selection method and the RF classifier. Table 30 shows the overall performance of the training data set. The AUROC (Area Under the Receiver Operating Characteristic Curve) value was measured to be 0.914. This demonstrates excellent classification performance based on training data, suggesting that the model effectively learned the characteristics of BRCA gene mutations compared to other combinations with AUROC measurements less than 0.9. The model with an AUROC value of 0.914 based on the training data is evaluated as a combination with excellent learning suitability for BRCA mutation classification, which is interpreted as an indicator showing the possibility that the model can also exhibit high performance in actual prediction (Fig. 9a).

[1689] Index measurement value 95% confidence interval AUROC 0.914 - TPR (sensitivity) 0.892 (0.835, 0.930) TNR (specificity) 0.807 (0.741, 0.860) PPV (precision) 0.822 (0.760, 0.871) NPV (negative predictive value) 0.882 (0.821, 0.924) FPR (false positive rate) 0.193 (0.140, 0.259) FNR (false negative rate) 0.108 (0.070, 0.165) FDR (false discovery rate) 0.178 (0.129, 0.240) ACC (accuracy) 0.849 (0.807, 0.884)

[1690] Table 31 below shows the overall performance of the test dataset. The AUROC (Area Under the Receiver Operating Characteristic Curve) value was measured to be 0.810, demonstrating that the model effectively performs class discrimination. This indicator quantitatively demonstrates the discriminatory power between classes and suggests that the model simultaneously secures high classification accuracy and stability. Therefore, it is clear that the BRCA mutation model of this combination achieved outstanding performance improvement with statistical significance in predicting BRCA gene mutations compared to other combinations whose AUROC values ​​on the test dataset were measured to be less than 0.8 (Figure 9b).

[1691] Indicator 95% confidence interval AUROC 0.810 - TPR (sensitivity) 0.333 (0.172, 0.546) TNR (specificity) 0.898 (0.836, 0.938) PPV (precision) 0.333 (0.172, 0.546) NPV (negative predictive value) 0.898 (0.836, 0.938) FPR (false positive rate) 0.102 (0.062, 0.164) FNR (false negative rate) 0.667 (0.454, 0.828) FDR (false discovery rate) 0.667 (0.454, 0.828) ACC (accuracy) 0.823 (0.756, 0.874)

[1692] Table 32 below shows the results of calculating the confusion matrix of the training data set, and Table 33 shows the results of calculating the confusion matrix of the validation data set. TP below is the number of cases in which the model correctly predicted a subject with an actual BRCA mutation as a subject with a BRCA mutation; TN below is the number of cases in which the model correctly predicted a subject without an actual BRCA mutation as a subject without a BRCA mutation; FP below is the number of cases in which the model incorrectly predicted a subject without an actual BRCA mutation as a subject with a BRCA mutation; and FN below is the number of cases in which the model incorrectly predicted a subject with an actual BRCA mutation as a subject without a BRCA mutation.

[1693]

[1694] Predicted: BRCA mutation present Predicted: BRCA mutation present Total Actual: BRCA mutation absent 134 (TN) 32 (FP) 166 Actual: BRCA mutation present 18 (FN) 148 (TP) 166 Total 152 180 332

[1695] Predicted: BRCA mutation present Predicted: BRCA mutation present Total Actual: BRCA mutation absent 123 (TN) 14 (FP) 137 Actual: BRCA mutation present 14 (FN) 7 (TP) 21 Total 137 21 158

[1696] (4) Feature Importance Analysis

[1697] For the trained model, feature importance analysis based on permutation importance was performed using the Shapley Additive Explanations (SHAP) library. The analysis was performed based on the training dataset (Train Set), and the results showed that the top features contributing to the prediction decision were wavelet-HHH_glszm_ZoneVariance, wavelet-LLL_glszm_RunLengthNonUniformity, and wavelet-HLL_ngtdm_Busyness, in that order. The average importance of each feature was calculated as 0.042, 0.032, and 0.024, respectively (Fig. 9c). In addition, feature importance analysis based on permutation importance was also performed on the validation dataset (Test Set). According to the results derived from the above analysis, the main features with high contribution to the prediction decision were wavelet-HHH_glszm_ZoneVariance, wavelet-LLL_glrlm_RunLengthNonUniformity, log-sigma-3-0-mm-3D_firstorder_Energy, and original_glrlm_RunLengthNonUniformity, in that order. The average importance of each was calculated to be 0.074, 0.069, 0.020, and 0.016, respectively, and it can be confirmed that the results are different from the importance ranking compared to the training data (Fig. 9d).

[1698]

[1699]

[1700] While specific aspects of the present invention have been described in detail above, it should be apparent to those skilled in the art that these specific descriptions are merely preferred embodiments and do not limit the scope of the present invention. Therefore, the substantial scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for providing information necessary for predicting BRCA gene mutations in cancer patients, comprising the following steps: (a) A step of extracting digitized image information from a magnetic resonance image (MRI) of a cancer patient with segmented lesions; (b) a step of performing consistency analysis on the extracted image information; (c) a step of performing a feature selection process on the image information for which the above consistency has been analyzed; (d) A step of applying a machine learning model that predicts the presence or absence of a BRCA gene mutation using the above-mentioned selected features as input values.

2. A method according to claim 1, characterized in that the cancer is prostate cancer.

3. A method characterized in that, in the first paragraph, the consistency analysis of step (b) is performed by extracting image information having an ICC (Intraclass Correlation Coefficient) value of 0.8 or higher.

4. A method according to claim 3, characterized in that the consistency analysis of step (b) additionally performs positive sample amplification.

5. A method characterized in that, in the first paragraph, the feature selection process of step (c) is performed by a method selected from the group consisting of No FS (NO Feature Selection), Corr (Correlation-based Selection), LASSO (Least Absolute Shrinkage and Selection Operator), and FFS (Forward Feature Selection).

6. A method characterized in that, in the fifth paragraph, the feature selection process of step (c) is performed by the FFS (Forward Feature Selection) method.

7. A method characterized in that, in the first paragraph, the step of applying a machine learning model for predicting the presence or absence of a BRCA gene mutation in step (d) is performed by a binary classifier selected from the group consisting of Adab (Adaptive Boosting), RF (Random Forest), SGD (Stochastic Gradient Descent), and LDA (Linear Discriminant Analysis).

8. A method characterized in that, in the 7th paragraph, the step of applying a machine learning model for predicting the presence or absence of a BRCA gene mutation in step (e) is performed by RF (Random Forest).

9. A device for predicting BRCA gene mutation in a cancer patient, comprising the following composition. (a) A receiving unit for receiving a magnetic resonance image (MRI) of a cancer patient with segmented lesions; (b) an extraction unit that extracts digitized image information from the segmented lesion; (c) an analysis unit that performs a consistency analysis process on the extracted image information; (d) a feature selection unit that performs a feature selection process on the image information for which the above consistency has been analyzed; (e) A prediction unit that applies a machine learning model to predict the presence or absence of a BRCA gene mutation using the above-selected features as input values.

10. A prediction device, characterized in that in paragraph 9, the consistency analysis process is performed by extracting image information having an ICC (Intraclass Correlation Coefficient) value of 0.8 or higher.

11. A prediction device according to claim 10, characterized in that the consistency analysis additionally performs positive sample amplification.

12. A prediction device according to claim 9, characterized in that the feature selection process is performed by the FFS (Forward Feature Selection) method.

13. A prediction device according to claim 9, characterized in that the machine learning model for predicting the presence or absence of a BRCA gene mutation is performed by RF (Random Forest).

14. A method for learning an artificial intelligence model for predicting BRCA gene mutations in cancer patients, comprising the following steps. (a) receiving a magnetic resonance image (MRI) of a cancer patient with segmented lesions; (b) a step of extracting digitized image information from the segmented lesion; (c) a step of performing a consistency analysis process on the extracted image information; (d) a step of performing a feature selection process on the image information for which the above consistency has been analyzed; (e) A step of training a machine learning model that predicts the presence or absence of a BRCA gene mutation using the above-mentioned selected features as input values.

15. A method according to claim 14, characterized in that the consistency analysis process is performed by extracting image information having an ICC (Intraclass Correlation Coefficient) value of 0.8 or higher.

16. A method according to claim 15, characterized in that the consistency analysis process additionally performs positive sample amplification.

17. A method according to claim 14, characterized in that the feature selection process is performed by the FFS (Forward Feature Selection) method.

18. A method according to claim 14, characterized in that the machine learning model for predicting the presence or absence of a BRCA gene mutation is performed by RF (Random Forest).

Images