SYSTEM & METHOD FOR RADIOMICS-BASED PNEUMONIA DIAGNOSIS USING MACHINE LEARNING

Abstract

ABSTRACT SYSTEM & METHOD FOR RADIOMICS-BASED PNEUMONIA DIAGNOSIS USING MACHINE LEARNING The present disclosure relates, in general, to the field of feature extraction and advanced machine learning techniques. More specifically, embodiments of the present invention relate to a system that processes chest X-ray images to detect pneumonia. It includes an input module (104) that receives X-ray images, followed by a pre-processing module (106) applying rules to enhance image quality. A radiomics feature extraction module (108) then analyzes the images, utilizing techniques like GLSZM, GLCM, GLDM, and GLRLM to extract texture and radiomics features. These features are further processed using a Power Spectral Density (PSD) analysis module (110), which applies methods such as Burg, Yule-Walker, and Welch PSD estimates to capture frequency-related characteristics. The extracted and analyzed features are then classified by a machine learning module using classifiers like Bernoulli Naïve Bayes, Quadratic Discriminant, Random Subspace Boost, and Gradient Boosting. Finally, the output module (114) displays the classification results, providing real-time diagnostic feedback on whether the X-ray images indicate pneumonia or a healthy condition.

Specification

Description:FIELD
[0001] The present disclosure relates, in general, to the field of advanced healthcare.
[0002] More specifically, embodiments of the present invention relate to the field of feature extraction and advanced machine learning techniques.
BACKGROUND
[0003] The background information herein below relates to the present disclosure but is not necessarily prior art.
[0004] Pneumonia, a leading cause of morbidity and mortality worldwide, is traditionally diagnosed using chest X-ray imaging, which heavily relies on the expertise of radiologists. However, manual interpretation of these images often introduces variability and subjectivity, depending on the radiologist's experience and skill level. Inconsistent readings can lead to misdiagnosis or delayed treatment, especially in regions with limited access to highly trained professionals.
[0005] Current Computer-Aided Diagnosis (CAD) systems have sought to alleviate this issue by assisting radiologists in identifying pneumonia from chest X-rays. However, these systems typically employ basic image processing techniques that fail to capture the intricate textural and structural features of lung tissue. As a result, their diagnostic capabilities remain limited, particularly when dealing with complex cases of pneumonia.
[0006] There is, therefore, felt a need for a system for radiomics-based pneumonia diagnosis that can leverage the power of advanced machine learning techniques for the early diagnosis of pneumonia using chest X-ray images.
OBJECTS
[0007] Some of the objects of the present disclosure, which at least one embodiment herein satisfies, are as follows.
[0008] It is an object of the present disclosure to ameliorate one or more problems of the prior art or to at least provide a useful alternative.
[0009] The main object of the present disclosure is to provide a system for radiomics-based pneumonia diagnosis using machine learning. More particularly, the object of the present disclosure is to provide a radiomics-based pneumonia diagnosis system integrating feature extraction and machine learning classifiers.
[0010] An object of the present disclosure is to provide a system that addresses key challenges in traditional pneumonia diagnosis, including the variability in radiologist expertise, limited sensitivity of early detection methods, and the manual, time-consuming nature of image interpretation.
[0011] Another object of the present disclosure is to provide a system that solves the problem of imbalanced datasets and overfitting in machine learning models by employing advanced classifiers that can handle high-dimensional data effectively.
[0012] Yet another object of the present disclosure is to mitigate issues related to inconsistent diagnostic outcomes by automating the feature extraction and classification processes, thereby reducing human error and improving diagnostic precision.
[0013] Other objects and advantages of the present disclosure will be more apparent from the following description, which is not intended to limit the scope of the present disclosure.
SUMMARY
[0014] This summary is provided to introduce concepts related to the field of radiomics feature extraction and advanced machine learning techniques. More specifically, embodiments of the present invention relate to a radiomics-based pneumonia diagnosis system integrating feature extraction and machine learning classifiers. The concepts are further described hereinbelow in the detailed description. This summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used to limit the scope of the claimed subject matter.
[0015] The present disclosure envisages a system for radiomics-based pneumonia diagnosis using machine learning is provided. The system broadly comprises an input module, a pre-processing module, a radiomics feature extraction module, a power spectral density (PSD) analysis module, a classification module, and an output module.
[0016] The input module is configured to receive a plurality of X-ray images of human chest as input data from an image capturing device. The pre-processing module is configured to receive the captured X-ray images and further configured to implement a set of pre-processing rules to generate pre-processed X-ray images. The radiomics feature extraction module is configured to extract texture and radiomics features from the pre-processed X-ray images by implementing radiomics techniques including Gray Level Size Zone Matrix (GLSZM), Gray Level Co-Occurrence Matrix (GLCM), Gray Level Dependence Matrix (GLDM), and Gray Level Run Length Matrix (GLRLM). The power spectral density (PSD) analysis module is configured to analyse the extracted radiomics features by implementing PSD analysis, including Burg Power Spectral Density Estimate, Yule-Walker Power Spectral Density Estimate, and Welch Power Spectral Density Estimate, to capture frequency characteristics that enrich feature data. The classification module is configured to process the extracted and PSD analysed features by implementing a plurality of machine learning based classifiers, including Bernoulli Naïve Bayes, Quadratic Discriminant, Random Subspace Boost, and Gradient Boosting, for classifying the X-ray images as either pneumonia-affected or healthy based on the analysed radiomics features and an output module is configured to display classification results on a user interface, providing real-time diagnostic feedback regarding the presence or absence of pneumonia in the analyzed chest X-ray images.
[0017] In an embodiment, the classification module is configured to output classification accuracy metrics, including Cohen's Kappa, Matthews Correlation Co-efficient, Youden's Index, Log Loss, and Brier Score, to evaluate classifier performance.
[0018] In an embodiment, the classification module is further configured to handle high-dimensional radiomics data using a plurality of machine learning classifiers, including Bernoulli Naïve Bayes, Quadratic Discriminant, Random Sub-space Boost, and Gradient Boosting, to accurately distinguish between pneumonia and non-pneumonia cases.
[0019] In an embodiment, the set of pre-processing rules comprises noise reduction filters and normalization techniques to standardize chest X-ray images for optimal feature extraction.
[0020] In an embodiment, the system further includes a repository configured to store the set of predefined instructions and the set of pre-processing rules, and a microprocessor coupled to the repository to execute the set of predefined instructions for executing one or more processing units.
[0021] In an embodiment, the radiomics feature extraction module further comprises a set of extraction techniques for computing radiomic parameters including cluster shade, auto-correlation, contrast, and zone entropy for improved texture analysis.
[0022] In an embodiment, the PSD analysis module is configured to apply a Welch PSD method to provide additional insight into the spatial frequencies of lung textures indicative of pneumonia.
[0023] In an embodiment, the GLCM features extracted from the X-ray images include metrics containing auto-correlation, cluster prominence, cluster shade, correlation, energy, and entropy, which provide insights into the texture and spatial patterns of the lung tissues.
[0024] In an embodiment, the GLSZM features extracted from the X-ray images include small area emphasis, large area emphasis, gray level non-uniformity, and zone entropy, enabling detailed analysis of size zones in gray-level regions within the lung tissues.
[0025] In an embodiment, the GLRLM features extracted include parameters such as short run emphasis, long run emphasis, run length non-uniformity, and gray level variance, providing insights into the length and uniformity of runs of consecutive pixels at the same gray level.
[0026] In an embodiment, the GLDM features extracted include small dependence emphasis, large dependence emphasis, gray level variance, and dependence entropy, which measure the dependence of pixels at varying gray levels within the lung tissue regions.
[0027] In an embodiment, the output module presents a graphical representation of the classifier's performance, including Precision-Recall Curves (AUC-PR), for further diagnostic evaluation by healthcare professionals.
[0028] The present disclosure further envisages a method for radiomics-based pneumonia diagnosis using machine learning, comprising the steps of :
• receiving, by an input module, a plurality of X-ray images of human chest as input data from an image capturing device;
• pre-processing, by a pre-processing module, the X-ray images by implementing a set of pre-processing rules to generate pre-processed X-ray images;
• extracting, by a radiomics feature extraction module, texture and radiomics features from the pre-processed X-ray images by implementing radiomics techniques including Gray Level Size Zone Matrix (GLSZM), Gray Level Co-Occurrence Matrix (GLCM), Gray Level Dependence Matrix (GLDM), and Gray Level Run Length Matrix (GLRLM);
• analysing, by a power spectral density analysis module, the extracted radiomics features by implementing PSD analysis, including Burg Power Spectral Density Estimate, Yule-Walker Power Spectral Density Estimate, and Welch Power Spectral Density Estimate, to capture frequency characteristics that enrich feature data;
• processing, by a classification module, the extracted and PSD analysed features by implementing a plurality of machine learning based classifiers, including Bernoulli Naïve Bayes, Quadratic Discriminant, Random Subspace Boost, and Gradient Boosting, for classifying the X-ray images as either pneumonia-affected or healthy based on the analysed radiomics features; and
• displaying, by an output module, classification results on a user interface, providing real-time diagnostic feedback regarding the presence or absence of pneumonia in the analyzed chest X-ray images.
[0029] In an embodiment, the steps of implementing the set of pre-processing rules comprises the steps of denoising the X-ray images by removal of unwanted noise from the X-ray images while preserving critical structural detail, re-sizing the X-ray images to a dimension of 224 multiplied by 224 pixels, batch normalizing the re-scaled the X-ray images by dividing the pixel values by 225; and augmenting the batches of the normalized X-ray images.
BRIEF DESCRIPTION OF THE ACCOMPANYING DRAWING
[0030] A system for radiomics-based pneumonia diagnosis using machine learning will now be described with the help of an accompanying drawings, in which:
[0031] Figure 1 depicts a high-level block diagram of a system 100 for radiomics-based pneumonia diagnosis using machine learning, in accordance with an embodiment of the present disclosure;
[0032] Figure 2 illustrates exemplary radiomics-based statistical feature matrices, in accordance with an embodiment of the present disclosure;
[0033] Figure 3 illustrates a graphical depiction of power spectral density (PSD) analysis of GLCM feature matrix, in accordance with an embodiment of the present disclosure;
[0034] Figure 4 illustrates a graphical depiction of PSD analysis of GLSZM feature matrix, in accordance with an embodiment of the present disclosure;
[0035] Figure 5 illustrates a graphical depiction of PSD analysis of GLRLM feature matrix, in accordance with an embodiment of the present disclosure;
[0036] Figure 6 illustrates a graphical depiction of PSD analysis of GLDM feature matrix, in accordance with an embodiment of the present disclosure;
[0037] Figure 7 illustrates a workflow of pneumonia diagnosis using radiomics feature extraction from dataset to image classification, in accordance with an embodiment of the present disclosure;
[0038] Figures 8A-8D illustrate radiomic feature extraction results for a test x-ray image for the four feature extraction techniques, viz., Gray Level Co-Occurrence Matrix (GLCM), Gray Level Size Zone Matrix (GLSZM), Gray Level Run Length Matrix (GLRLM), and Gray Level Dependence Matrix (GLDM), in accordance with an embodiment of the present disclosure;
[0039] Figures 9A-9D illustrate the area as depicted by the precision-recall curves (AUC-PR) for different machine learning based classifiers, viz., Gradient Boosting, Quadratic Discriminant, Random Sub-space Boost, and Bernoulli Naïve Bayes, in accordance with an embodiment of the present disclosure, and
[0040] Figure 10 illustrates a method for radiomics-based pneumonia diagnosis using machine learning in accordance with an embodiment of the present disclosure.
LIST OF REFERENCE NUMERALS USED IN THE DESCRIPTION AND DRAWINGS :
100 SYSTEM
102 INPUT CAPTURING DEVICE
104 INPUT MODULE
106 PRE-PROCESSING UNIT
108 RADIOMICS FEATURE EXTRACTION MODULE
110 POWER SPECTRAL DENSITY (PSD) ANALYSIS MODULE
112 CLASSIFICATION MODULE
114 OUTPUT MODULE
116 USER INTERFACE
118 REPOSITORY
120 MICROPROCESSOR

DETAILED DESCRIPTION
[0001] Embodiments, of the present disclosure, will now be described with reference to the accompanying drawing.
[0002] Embodiments are provided so as to thoroughly and fully convey the scope of the present disclosure to the person skilled in the art. Numerous details are set forth, relating to specific components and methods to provide a complete understanding of the embodiments of the present disclosure. It will be apparent to the person skilled in the art that the details provided in the embodiments should not be construed to limit the scope of the present disclosure. In some embodiments, well-known apparatus structures, and well-known techniques are not described in detail.
[0003] The terminology used, in the present disclosure, is only for the purpose of explaining a particular embodiment and such terminology shall not be considered to limit the scope of the present disclosure. As used in the present disclosure, the forms "a", "an", and "the" may be intended to include the plural forms as well, unless the context clearly suggests otherwise. The terms, "comprises", "comprising", "including" and "having" are open-ended transitional phrases and therefore, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not forbid the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.
[0004] When an element is referred to as being "embodied thereon", "engaged to", "coupled to" or "communicatively coupled to" another element, it may be directly on, engaged, connected, or coupled to the other element. As used herein, the term "and/or" includes any and all combinations of one or more of the associated listed elements.
[0005] Pneumonia, a leading cause of morbidity and mortality worldwide, is traditionally diagnosed using chest X-ray imaging, which heavily relies on the expertise of radiologists. However, manual interpretation of these images often introduces variability and subjectivity, depending upon the radiologist's experience and skill level. Inconsistent readings can lead to misdiagnosis or delayed treatment, especially in regions with limited access to highly trained professionals.
[0006] Current Computer-Aided Diagnosis (CAD) systems have sought to alleviate this issue by assisting radiologists in identifying pneumonia from chest X-rays. However, these systems typically employ basic image processing techniques that fail to capture the intricate textural and structural features of lung tissue. As a result, their diagnostic capabilities remain limited, particularly when dealing with complex cases of pneumonia.
[0007] The present disclosure addresses these shortcomings by incorporating advanced radiomics-based feature extraction techniques and machine learning classifiers into the pneumonia diagnosis workflow. Radiomics, which allows for the extraction of high-dimensional, quantitative data from medical images, significantly improves the depth of analysis. Key techniques such as Gray Level Size Zone Matrix (GLSZM), Gray Level Co-Occurrence Matrix (GLCM), Gray Level Dependence Matrix (GLDM), and Gray Level Run Length Matrix (GLRLM) are employed to capture detailed textural information from chest X-rays. These features enable a more nuanced analysis of lung abnormalities, beyond what basic image processing can offer.
[0008] Additionally, the integration of Power Spectral Density (PSD) analysis and machine learning classifiers further enhances the extraction of relevant features from the medical images. Coupled with advanced machine learning classifiers, this invention provides a robust, automated system that improves the precision, consistency, and efficiency of pneumonia diagnosis. The combination of radiomics and machine learning represents a significant advancement over existing diagnostic technologies, offering a comprehensive tool for accurate and timely detection of pneumonia.
[0041] The present disclosure relates, in general, to the field of advanced healthcare. More specifically, embodiments of the present invention relate to the field of feature extraction and advanced machine learning techniques.
[0042] This disclosure is a combination of a new diagnostic process, software algorithms, and a system for medical image analysis. It enhances the traditional diagnostic workflow by incorporating radiomics feature extraction from chest X-ray images, combined with machine learning classifiers.
[0043] This radiomics-based approach allows for more precise identification of pneumonia-affected regions in the lungs, improving the accuracy and consistency of diagnoses compared to manual interpretations.
[0044] The use of radiomics, along with advanced classifiers like Bernoulli Naïve Bayes, Random Subspace Boost, Quadratic Discriminant, and Gradient Boosting provide earlier detection and more reliable outcomes in pneumonia diagnosis.
[0045] The disclosure also stands out for its use of a diverse set of machine-learning classifiers, including Bernoulli Naïve Bayes, Random Subspace Boost, Quadratic Discriminant, and Gradient Boosting. These classifiers are specifically tailored to process the complex and high-dimensional radiomics data, resulting in more accurate differentiation between pneumonia and non-pneumonia cases.
[0046] The classifiers excel in capturing non-linear relationships in the data, consistently achieving superior performance across various evaluation metrics such as Cohen's Kappa, Matthews Correlation Coefficient, Youden's Index, Log Loss, and Brier Score. The integration of these advanced classifiers with radiomics-based feature extraction marks a significant improvement in pneumonia diagnosis, offering more reliable and consistent outcomes compared to traditional methods.
[0047] A preferred embodiment of a system 100 for radiomics-based pneumonia diagnosis using machine learning, will now be described in detail with reference to Figures 1 to 9. The preferred embodiment does not limit the scope and ambit of the present disclosure.
[0048] Figure 1 depicts a high-level block diagram of a system 100 for radiomics-based pneumonia diagnosis using machine learning, in accordance with an embodiment of the present disclosure.
[0009] The system 100 of the present disclosure broadly comprises an input module 104, a pre-processing module 106, a radiomics feature extraction module 108, a power spectral density (PSD) analysis module 110, a classification module 112, and an output module 114.
[0010] The input module 104 is configured to receive a plurality of X-ray images of a human chest as input data from an image capturing device. The pre-processing module 106 is configured to receive the captured X-ray images and further configured to implement a set of pre-processing rules to generate pre-processed X-ray images. The radiomics feature extraction module 108 is configured to extract texture and radiomics features from the pre-processed X-ray images by implementing radiomics techniques including Gray Level Size Zone Matrix (GLSZM), Gray Level Co-Occurrence Matrix (GLCM), Gray Level Dependence Matrix (GLDM), and Gray Level Run Length Matrix (GLRLM). The power spectral density (PSD) analysis module 110 is configured to analyse the extracted radiomics features by implementing PSD analysis, including Burg Power Spectral Density Estimate, Yule-Walker Power Spectral Density Estimate, and Welch Power Spectral Density Estimate, to capture frequency characteristics that enrich feature data. The classification module 112 is configured to process the extracted and PSD analysed features by implementing a plurality of machine learning based classifiers, including Bernoulli Naïve Bayes, Quadratic Discriminant, Random Subspace Boost, and Gradient Boosting, for classifying the X-ray images as either pneumonia-affected or healthy based on the analysed radiomics features and an output module 114 is configured to display classification results on a user interface, providing real-time diagnostic feedback regarding the presence or absence of pneumonia in the analyzed chest X-ray images.
[0011] In an exemplary embodiment, the classification module 112 is configured to output classification accuracy metrics, including Cohen's Kappa, Matthews Correlation Co-efficient, Youden's Index, Log Loss, and Brier Score, to evaluate classifier performance.
[0012] In an exemplary embodiment, the classification module 112 is further configured to handle high-dimensional radiomics data using a plurality of machine learning classifiers, including Bernoulli Naïve Bayes, Quadratic Discriminant, Random Sub-space Boost, and Gradient Boosting, to accurately distinguish between pneumonia and non-pneumonia cases.
[0013] In an exemplary embodiment, the set of pre-processing rules comprises noise reduction filters and normalization techniques to standardize chest X-ray images for optimal feature extraction.
[0014] In an exemplary embodiment, the system 100 further includes a repository 118 configured to store the set of predefined instructions and the set of pre-processing rules, and a microprocessor 120 coupled to the repository 118 to execute the set of predefined instructions for executing one or more processing units.
[0015] In an exemplary embodiment, the radiomics feature extraction module 108 further comprises a set of extraction techniques for computing radiomic parameters including cluster shade, auto-correlation, contrast, and zone entropy for improved texture analysis.
[0016] In an exemplary embodiment, the PSD analysis module 110 is configured to apply a Welch PSD method to provide additional insight into the spatial frequencies of lung textures indicative of pneumonia.
[0017] In an exemplary embodiment, the GLCM features extracted from the X-ray images include metrics containing auto-correlation, cluster prominence, cluster shade, correlation, energy, and entropy, which provide insights into the texture and spatial patterns of the lung tissues.
[0018] In an exemplary embodiment, the GLSZM features extracted from the X-ray images include small area emphasis, large area emphasis, gray level non-uniformity, and zone entropy, enabling detailed analysis of size zones in gray-level regions within the lung tissues.
[0019] In an exemplary embodiment, the GLRLM features extracted include parameters such as short run emphasis, long run emphasis, run length non-uniformity, and gray level variance, providing insights into the length and uniformity of runs of consecutive pixels at the same gray level.
[0020] In an exemplary embodiment, the GLDM features extracted include small dependence emphasis, large dependence emphasis, gray level variance, and dependence entropy, which measure the dependence of pixels at varying gray levels within the lung tissue regions.
[0021] In an exemplary embodiment, the output module 114 presents a graphical representation of the classifier's performance, including Precision-Recall Curves (AUC-PR), for further diagnostic evaluation by healthcare professionals.
[0022] In an exemplary embodiment, the system 100 can include one or more processors and may be implemented as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, logic circuitries, and/or any devices that manipulate data based on operational instructions. Among other capabilities, the one or more processors are configured to fetch and execute computer-readable instructions stored in a memory of the system 100. The memory may store one or more computer-readable instructions or routines, which may be fetched and executed for executing the instructions. The memory may include any non-transitory storage device including, for example, volatile memory such as RAM, or non-volatile memory such as EPROM, flash memory, and the like. The functions of one or more processor(s) may be provided through the use of dedicated hardware as well as hardware capable of executing machine-readable instructions. In other examples, one or more processors may be implemented by electronic circuitry or printed circuit board. One or more processors may be configured to execute functions of various modules of the system 100.
[0023] In an alternative aspect, the memory may be an external data storage device coupled to the system 100 directly or through one or more offline/online data servers.
[0024] In an exemplary embodiment, the system 100 further comprises a network interface to receive real-time data inputs from external sources such as databases, APIs, and sensors, which are used by the system 100. The network interface may include a variety of interfaces, for example, interfaces for data input and output devices, referred to as I/O devices, transceivers, storage devices, and the like. The network interface may facilitate communication of the system 100 with various devices coupled to the system 100. The network interface may also provide a communication pathway for one or more components of the system 100. Examples of such components include, but are not limited to, processing module(s) and data storage.
[0025] The module(s) of the system 100 may be implemented as a combination of hardware and programming (for example, programmable instructions) to implement one or more functionalities of the processing module(s). In the examples described herein, such combinations of hardware and programming may be implemented in several different ways. For example, the programming for the module(s) may be processor-executable instructions stored on a non-transitory machine-readable storage medium and the hardware for the module(s) may include a processing resource (for example, one or more processors), to execute such instructions. In the present examples, the machine-readable storage medium may store instructions that, when executed by the processing resource, implement the module(s). In such examples, the system 100 may include the machine-readable storage medium storing the instructions and the processing resource to execute the instructions, or the machine-readable storage medium may be separate but accessible to the system 100 and the processing resource. In other examples, the module(s) may be implemented by electronic circuitry.
[0026] Figure 2 illustrates radiomics-based statistical feature matrices, in accordance with an embodiment of the present disclosure. A detailed breakdown of radiomics features used in medical image analysis, specifically illustrating the extraction of texture features from images using four key methods: GLCM (Gray Level Co-Occurrence Matrix), GLSZM (Gray Level Size Zone Matrix), GLRLM (Gray Level Run Length Matrix), and GLDM (Gray Level Dependence Matrix) is provided.
[0027] Here's a step-by-step breakdown of the content :
[0028] 1. GLCM Features : The Gray Level Co-Occurrence Matrix (GLCM) is a method used to extract texture features from images by calculating the frequency of pairs of pixel values occurring together at a specific spatial relationship. A total of 23 GLCM features are listed, such as:
• Auto-correlation
• Cluster Prominence
• Contrast
• Dissimilarity
• Entropy
• Information Measures of Correlation (IMC1 and IMC2)
• Sum Variance
[0029] These features are calculated for multiple images labeled as Img1, Img2, all the way to ImgN, creating a GLCM feature matrix with dimensions, N x 23. Each row corresponds to an image, and each column to a specific GLCM feature.
[0030] 2. GLSZM Features: The Gray Level Size Zone Matrix (GLSZM) captures texture by measuring the size of connected regions with the same gray level in an image. The matrix measures features across regions that are not necessarily adjacent. This method extracts 16 GLSZM features, including:
• Gray Level Non-Uniformity
• Zone Entropy
• Small and Large Area Emphasis
• Zone Variance
[0031] Like the GLCM, this extraction method generates a feature matrix of size N x 16 where N is the number of images, and the columns correspond to different GLSZM features.
[0032] 3. GLRLM Features: The Gray Level Run Length Matrix (GLRLM) quantifies the number of consecutive pixels (or runs) having the same gray level. The 16 features extracted using this matrix include:
• Gray Level Non-Uniformity
• Run Entropy
• Short and Long Run Emphasis
• Run Variance
[0033] The feature matrix for GLRLM follows a similar structure, with dimensions N x 16, indicating the number of images (N) and corresponding feature values for each.
[0034] 4. GLDM Features: The Gray Level Dependence Matrix (GLDM) calculates the gray-level dependencies in an image, helping to describe its homogeneity. The 14 GLDM features include:
• Dependence Entropy
• Low Gray Level Emphasis
• Dependence Variance
• Gray Level Variance
[0035] The GLDM feature matrix is organized as N x 14, where N refers to the number of images, and each column represents one of the extracted features.
[0036] Feature Matrices Overview: Each radiomics method (GLCM, GLSZM, GLRLM, and GLDM) produces a feature matrix for a collection of images. The number of images is represented by N, and the columns correspond to the specific features calculated from the images. These matrices are used to compile detailed quantitative data from medical images, which can then be input into machine learning models to support more precise diagnostic decisions, particularly in conditions like pneumonia.
[0037] Final Summary:
• GLCM produces a matrix of N x 23.
• GLSZM produces a matrix of N x 16.
• GLRLM produces a matrix of N x 16.
• GLDM produces a matrix of N x 14.
[0038] Each of these methods provides a unique approach to quantifying textural and structural information within the image, offering comprehensive feature extraction for use in advanced image analysis workflows.
[0039] Figure 3 illustrates a graphical depiction of three distinct Power Spectral Density (PSD) estimation methods-Burg, Yule-Walker, and Welch-of GLCM feature matrix for analyzing the frequency distribution of different features extracted from medical images, in accordance with an embodiment of the present disclosure. Below is a detailed breakdown of each PSD estimate plot and its significance.
1. Burg Power Spectral Density Estimate (Top Plot):
• Burg's method is an autoregressive (AR) model-based approach that provides high resolution in spectral estimation, especially for short data sequences. The plot displays various curves representing the PSD across a normalized frequency range from 0 to 1 (in radians per sample).
• On the Y-axis, the PSD values are displayed in decibels (dB per rad/sample).
• Each curve corresponds to a different feature, as indicated by the color legend at the bottom (e.g., Auto-correlation, Cor-relation, Cluster Prominence, Energy, Entropy).
• The plot reveals how the power of each feature varies across different frequencies. For instance, some features (like Auto-correlation in blue) have a higher power in the lower frequency range, while others show more stable or less prominent variations.
2. Yule-Walker Power Spectral Density Estimate (Middle Plot):
• The Yule-Walker method is another AR model-based technique that estimates PSD by solving the Yule-Walker equations. This method is often used when the autoregressive parameters of a system need to be estimated.
• The layout is similar to the Burg estimate, with normalized frequency on the X-Axis and PSD in dB on the Y-Axis.
• The curves follow a similar general trend as the Burg estimate but with slight differences in magnitude and distribution. This plot offers insight into how the chosen model and estimation technique influence the spectral characteristics of the features.
• The features represented by each color maintain consistency across all three plots, allowing for comparative analysis between methods.
3. Welch Power Spectral Density Estimate (Bottom Plot):
• Welch's method is a modified version of the periodogram method. It averages the PSD estimates from overlapping segments of the data, which reduces noise and provides a smoother spectral estimate.
• In this plot, the PSD is more stabilized across the frequency spectrum compared to the other two methods. The curves are flatter, especially in the higher frequency ranges, indicating that Welch's method is more effective at reducing noise and variance in PSD estimation.
• As with the other two plots, each curve represents a different feature. Features like Cluster Tendency, Contrast, and Entropy show relatively low variation in their power across the frequency spectrum, while others (like Auto-correlation and Correlation) exhibit more pronounced changes.
[0040] Color Legend and Feature Descriptions:
At the bottom of the image is a comprehensive legend that maps each color to a specific image feature. These features represent different statistical metrics calculated from medical images, such as:
• Auto-correlation: Measures the correlation between an image and a shifted version of itself.
• Cluster Prominence and Cluster Shade: These describe the skewness and asymmetry of clusters of gray-level values.
• Energy: Indicates the sum of squared pixel values, reflecting texture uniformity.
• Entropy: Measures the randomness or complexity in an image's pixel intensity distribution.
• Homogeneity, Contrast, and Correlation: These describe how similar or dissimilar pixel intensities are within an image, helping to identify patterns and structures.
[0041] This legend allows for easy identification of the corresponding curves in the PSD estimates, providing a full view of how each feature behaves across frequencies using different spectral estimation techniques. This image compares three PSD estimation techniques (Burg, Yule-Walker, Welch) for various image features. Each method shows how the power of extracted features from medical images varies across normalized frequencies. The Burg and Yule-Walker methods display more variation in high-frequency ranges, while Welch provides a smoother and more stable PSD estimate. This analysis is critical in understanding the frequency-domain characteristics of radiomics features, which can be used in medical image classification and diagnosis tasks, such as pneumonia detection.
[0042] Figure 4 illustrates a graphical depiction of three distinct Power Spectral Density (PSD) estimation methods-Burg, Yule-Walker, and Welch-of GLSZM feature matrix for analyzing the frequency distribution of different features extracted from medical images, in accordance with an embodiment of the present disclosure. The breakdown of each PSD estimate plot and its significance is similar to that as in Figure 3.
[0043] Color Legend and Feature Descriptions:
[0044] The legend at the bottom of the image represents different categories or variables corresponding to the colored lines on the plots. These labels might refer to signal features, such as:
o Small Area Emphasis, Large Area Emphasis, Zone Variance, Zone Entropy, Gray Level Non-Uniformity, etc.
o These could be linked to texture analysis, likely based on the Gray-Level Co-occurrence Matrix (GLCM), often used in image processing and feature extraction.
[0045] The visual comparison of these three methods helps highlight the differences in power spectral density estimation, where:
• The Burg and Yule-Walker methods provide relatively smooth and well-separated lines, indicating a clear frequency response across the range.
• The Welch method introduces more noise or variations, potentially due to its averaging approach.
[0046] Figure 5 illustrates a graphical depiction of three distinct Power Spectral Density (PSD) estimation methods-Burg, Yule-Walker, and Welch-of GLRLM feature matrix for analyzing the frequency distribution of different features extracted from medical images, in accordance with an embodiment of the present disclosure. The breakdown of each PSD estimate plot and its significance is similar to that of the previous two figures.
[0047] Color Legend and Feature Descriptions:
[0048] The legend identifies the different colored lines corresponding to specific features, mostly related to run-length features, which are typically used in texture analysis. The terms include:
o Short Run Emphasis
o Long Run Emphasis
o Gray Level Non-Uniformity
o Run Length Non-Uniformity
o Run Percentage
o Run Variance
o Run Entropy
o High Gray Level Run Emphasis
o Low Gray Level Run Emphasis
[0049] These terms are often associated with gray-level run-length matrices (GLRLM), used in image processing to quantify texture by analyzing consecutive runs of pixels with similar intensity. The PSD of these features helps understand the frequency distribution of these texture-related parameters.
Comparative Insights:
• Burg and Yule-Walker estimates both show a smoother, monotonic decrease in power as frequency increases, with less noise and fluctuations.
• The Welch method captures more fluctuations and noise at the lower frequency ranges, which may indicate finer granularity in the power distribution of the signal.

[0050] These plots likely compare how different features (related to run-length texture analysis) behave across different frequency domains using various PSD estimation methods. This can be useful for signal or image analysis in fields like medical imaging, remote sensing, or pattern recognition.
[0051] Figure 6 illustrates a graphical depiction of three distinct Power Spectral Density (PSD) estimation methods-Burg, Yule-Walker, and Welch-of GLDM feature matrix for analyzing the frequency distribution of different features extracted from medical images, in accordance with an embodiment of the present disclosure. The breakdown of each PSD estimate plot and its significance is similar to those provided for as in the previous figures.
[0052] Color Legend and Feature Descriptions :
[0053] The legend contains various texture-related features represented by the colored lines, primarily derived from gray-level dependence matrices (GLDM). These include :
o Small Dependence Emphasis: Measures small dependencies between pixels.
o Large Dependence Emphasis: Quantifies larger-scale dependencies in the texture.
o Gray-Level Non-Uniformity: Assesses how uniform or non-uniform the gray-level distributions are.
o Dependence Non-Uniformity: Captures the variation in the pixel dependencies.
o Dependence Entropy: Represents the randomness or complexity of dependencies in the image.
o Gray-Level Variance: Evaluate the variance in gray levels across the image.
[0054] These texture features play a vital role in image analysis and classification applications, such as distinguishing between different texture types, analyzing patterns, or detecting anomalies in images. The three methods of spectral analysis (Burg, Yule-Walker, Welch) allow for a detailed comparison of how these features are distributed across the frequency spectrum.
[0055] Figure 7 illustrates a workflow of pneumonia diagnosis using radiomics feature extraction from dataset to image classification, in accordance with an embodiment of the present disclosure. The diagram represents a detailed pipeline for a radiomics-based approach to classifying pneumonia from chest X-ray images using various texture features and machine learning classifiers. Below is a step-by-step breakdown of the process:
1. Dataset: The dataset consists of chest X-ray images categorized into two classes:
o Pneumonia: Images depicting signs of pneumonia in the lungs.
o Normal: Images showing no signs of pneumonia, representing healthy lungs.
This is the raw input data that the system processes to extract features for classification.

2. Pre-processing: This is a crucial step to improve the quality of the input images before extracting any features. It may include steps such as:
o Re-scaling: Adjusting the image resolution or intensity values.
o Normalization: Ensuring that the pixel values fall within a specific range.
3. De-noising: This step reduces noise in the images, which may arise due to artifacts, sensor errors, or external interference. By de-noising, the images become clearer and more suitable for feature extraction.

4. Synthetic Data Augmentation:
o Synthetic data generation techniques are used to create additional training samples by transforming existing images (e.g., rotation, flipping, cropping, etc.).
o Augmentation helps to increase the size of the dataset, improve model generalization, and prevent overfitting.

5. Radiomics Feature Extraction:
Radiomics refers to the extraction of a large number of quantitative features from medical images. In this process, different texture features are computed using various matrix-based techniques. Discussed below are the ones displayed in the figures:

5.1. GLCM (Gray Level Co-occurrence Matrix) Features:
GLCM calculates how often pairs of pixel intensities with a specified spatial relationship occur in an image.
Features derived from GLCM include:
o Auto-correlation: Measures the correlation between pixel pairs.
o Cluster Shade/Prominence: Quantifies the skewness or asymmetry of clusters of pixels.
o Contrast, Correlation, Homogeneity: Different metrics for texture smoothness and uniformity.
o Entropy: Measures the randomness or complexity of the texture.
o IDN (Inverse Difference Normalized): Measures texture smoothness based on pixel intensity differences.

5.2. GLSZM (Gray Level Size Zone Matrix) Features :
GLSZM quantifies zones of connected pixels that share the same gray level.
Features include:
o Small Area Emphasis/Large Area Emphasis: Captures the importance of small or large areas with constant gray levels.
o Gray Level Non-Uniformity (GLNU): Measures variability in gray levels.
o Zone Entropy/Variance: Represents complexity and dispersion of pixel intensities across zones.

5.3. GLRLM (Gray Level Run Length Matrix) Parameters:
GLRLM measures the number of consecutive pixels (runs) with the same gray level along specific directions.
Features include:
o Short/Long Run Emphasis: Highlights textures with short or long continuous regions of similar pixel intensity.
o Run Length Non-Uniformity: Measures variability in run lengths.
o Run Percentage: Fraction of runs relative to total pixels.

5.4. GLDM (Gray Level Dependence Matrix) Parameters:
GLDM computes the dependency of pixels based on their spatial distance from each other.
Features include:
o Small Dependence Emphasis/Large Dependence Emphasis: Emphasizes areas with either small or large pixel dependencies.
o Dependence Variance/Entropy: Represents the variability and randomness in dependencies between pixels.

6. Feature Evaluation Metrics:
To analyze the frequency-domain characteristics of the radiomics features, power spectral density (PSD) estimation techniques are applied:

• Burg Power Spectral Density Estimate : A method for estimating the PSD based on an autoregressive model.
• Yule-Walker Power Spectral Density Estimate : Uses the Yule-Walker equations to compute the PSD.
• Welch Power Spectral Density Estimate : A non-parametric method that averages periodograms from overlapping data segments.

These PSD estimation techniques provide insights into how texture features behave across different frequencies, which is useful for classifying different textures and anomalies.

7. Classifiers:
The extracted radiomics features are fed into various machine learning classifiers for pneumonia detection:
• Bernoulli Naïve Bayes: A probabilistic classifier based on Bayes' theorem, suitable for binary feature vectors.
• Quadratic Discriminant : A classifier that models the decision boundary as a quadratic function of the features.
• Random Subspace Boost : A boosting algorithm that operates in a random subspace of the features.
• Gradient Boosting : An ensemble method that builds weak learners sequentially to improve accuracy.

These classifiers help in distinguishing between pneumonia and normal cases based on the extracted texture features.

8. Performance Metrics :
To evaluate the performance of the classifiers, several metrics are used:
• Cohen's Kappa: Measures agreement between predicted and actual classifications.
• Matthews Correlation Coefficient (MCC): A balanced measure for binary classification.
• Youden's Index: Combines sensitivity and specificity to provide a single metric for classifier performance.
• Log Loss: A loss function that penalizes incorrect classifications with a higher cost.
• Brier Score: Measures the accuracy of probabilistic predictions.

These metrics offer a comprehensive evaluation of the model's ability to correctly classify pneumonia and normal cases.

[0056] Figures 8A-8D illustrate radiomic feature extraction results for a test x-ray image for the four feature extraction techniques, viz., Gray Level Co-Occurrence Matrix (GLCM), Gray Level Size Zone Matrix (GLSZM), Gray Level Run Length Matrix (GLRLM), and Gray Level Dependence Matrix (GLDM), in accordance with an embodiment of the present disclosure;
[0057] Figures 9A-9D illustrate areas as depicted by the Precision-Recall Curves (AUC-PR) for different machine learning based classifiers, viz., Gradient Boosting, Quadratic Discriminant, Random Sub-space Boost, and Bernoulli Naïve Bayes, in accordance with an embodiment of the present disclosure.
Precision-Recall Curve Analysis:
• The Precision-Recall (PR) curve is a graphical representation used to evaluate the performance of the machine learning classifiers based on their precision (positive predictive value) and recall (true positive rate).
• This graph provides insights into the classifiers' ability to correctly identify pneumonia cases from chest X-ray images by focusing on the balance between precision and recall.
Curve Interpretation:
• Precision (Y-axis): The fraction of relevant instances among the retrieved instances (i.e., how many of the positive predictions made are actually correct).
• Recall (X-axis): The fraction of relevant instances that have been retrieved over the total amount of relevant instances (i.e., how many actual positives were captured by the model).
• Each point on the curve corresponds to a precision-recall pair at a particular decision threshold.
AUC-PR (Area Under the Precision-Recall Curve):
• AUC-PR is used as a scalar value to summarize the PR curve, with higher values indicating better performance.
[0058] Figure 9A illustrates a precision-recall curve (PR curve) for different machine learning classifiers, using GLCM (Gray Level Co-occurrence Matrix) features for pneumonia classification from chest X-ray images,
[0059] Figure 9B illustrates a precision-recall curve (PR curve) for different machine learning classifiers, using GLSZM (Gray Level Co-occurrence Matrix) features for pneumonia classification from chest X-ray images,
[0060] Figure 9C illustrates a precision-recall curve (PR curve) for different machine learning classifiers, using GLRLM (Gray Level Co-occurrence Matrix) features for pneumonia classification from chest X-ray images, and
[0061] Figure 9D illustrates a precision-recall curve (PR curve) for different machine learning classifiers, using GLSZM (Gray Level Co-occurrence Matrix) features for pneumonia classification from chest X-ray images,
Application in Pneumonia Classification :
• The high precision-recall performance of these classifiers, especially Gradient Boosting and Quadratic Discriminant, makes them particularly suited for medical applications like pneumonia detection.
• In clinical settings, these classifiers can assist radiologists by reducing false negatives (missed pneumonia cases) and maintaining high diagnostic precision, improving overall patient outcomes.
[0062] Figure 10 illustrates a method 1000 for radiomics-based pneumonia diagnosis using machine learning. The method 1000 includes the following steps of:
[0063] In method step 1002, the method 1000 includes receiving, by an input module 104, a plurality of X-ray images of a human chest as input data from an image capturing device 102;
[0064] In method step 1004, the method 1000 includes pre-processing 106, by a pre-processing module (106), the X-ray images by implementing a set of pre-processing rules to generate pre-processed X-ray images;
[0065] In method step 1006, the method 1000 includes extracting, by a radiomics feature extraction module 108, texture and radiomics features from the pre-processed X-ray images by implementing radiomics techniques including Gray Level Size Zone Matrix (GLSZM), Gray Level Co-Occurrence Matrix (GLCM), Gray Level Dependence Matrix (GLDM), and Gray Level Run Length Matrix (GLRLM);
[0066] In method step 1008, the method 1000 includes analysing, by a power spectral density analysis module 110, the extracted radiomics features by implementing PSD analysis, including Burg Power Spectral Density Estimate, Yule-Walker Power Spectral Density Estimate, and Welch Power Spectral Density Estimate, to capture frequency characteristics that enrich feature data;
[0067] In method step 1010, the method 1000 includes processing, by a classification module 112, the extracted and PSD analysed features by implementing a plurality of machine learning based classifiers, including Bernoulli Naïve Bayes, Quadratic Discriminant, Random Subspace Boost, and Gradient Boosting, for classifying the X-ray images as either pneumonia-affected or healthy based on the analysed radiomics features; and
[0068] In method step 1012, the method 1000 includes displaying, by an output module 114, classification results on a user interface 116, providing real-time diagnostic feedback regarding the presence or absence of pneumonia in the analyzed chest X-ray images.
[0069] In an exemplary embodiment, the method 1000 for radiomics-based pneumonia diagnosis using machine learning further comprises the steps of implementing the set of pre-processing rules which includes the steps of:
• denoising of the X-ray images by removal of unwanted noise from the X-ray images while preserving critical structural detail,
• re-sizing the X-ray images to a dimension of 224 multiplied by 224 pixels,
• batch normalizing the re-scaled the X-ray images by dividing the pixel values by 225; and
• augmenting the batches of the normalized X-ray images.

[0070] The following table reports the performance evaluation audit of multiple classifiers for different Radiomic Feature Extraction algorithms
Feature Extraction Classifiers Cohen's Kappa MCC Youden's Index Log Loss Brier Score


GLCM Bernoulli Naïve Bayes 0.72 0.64 0.56 0.53 0.18
Random Subspace Boost 0.79 0.72 0.65 0.48 0.14
Quadratic Discriminant 0.85 0.79 0.73 0.38 0.09
Gradient Boosting 0.91 0.86 0.80 0.29 0.07


GLRLM Bernoulli Naïve Bayes 0.75 0.68 0.60 0.51 0.16
Random Subspace Boost 0.80 0.73 0.66 0.47 0.13
Quadratic Discriminant 0.87 0.82 0.76 0.34 0.08
Gradient Boosting 0.92 0.87 0.81 0.28 0.06


GLDM Bernoulli Naïve Bayes 0.74 0.67 0.59 0.52 0.17
Random Subspace Boost 0.78 0.71 0.64 0.49 0.15
Quadratic Discriminant 0.84 0.77 0.71 0.39 0.11
Gradient Boosting 0.90 0.85 0.79 0.30 0.08


GLSZM Bernoulli Naïve Bayes 0.78 0.71 0.64 0.49 0.15
Random Subspace Boost 0.82 0.75 0.68 0.46 0.12
Quadratic Discriminant 0.86 0.80 0.74 0.36 0.10
Gradient Boosting 0.93 0.88 0.82 0.27 0.05

[0071] The following table presents the results of the radiomic feature extraction for a test X-ray image, outlining the calculated parameters for each texture analysis method.

Radiomic Features Radiomic Parameters Feature Results











GLCM

Fig. Pneumonia Test X-ray Cluster Shade 0.0207897352
IDM 0.425738728
IDMN 0.999576066
Difference Average 3.212071558
Cluster Prominence 0.0207897352
Joint Average 245.95892348
Sum Entropy 8.751276632
Max Probability 0.028439316
Inverse Variance 2.57499780
Autocorrelation 0.999999962
Difference Entropy 3.28822679
Joint Energy 0.001251162
Max Correlation Coefficient 0.996731385
IMC2 53.2493140
Cluster Tendency 583427562.914
ID 0.42573872
Correlation 0.9967313852
Contrast 28.38712935
IMC1 0.996731385
IDN 0.999950994
Joint Entropy 11.58475786
Sum of Squares 77836.9216
Cluster Shade 0.0207897352







GLSZM

Fig. Pneumonia Test X-ray Zone Variance -598261007754
Small Area High Gray Level Emphasis 157144870
Size-Zone Non-Uniformity 1693423973818492416
Gray Level Non-Uniformity Normalized 0.11
Zone Entropy 3.50
Small Area Emphasis 16728728.76
Large Area Emphasis 75452003316
Gray Level Non-Uniformity 3412660668
Zone Percentage 0.000052218493819
Small Area Low Gray Level Emphasis 17274.19
Low Gray Level Zone Emphasis 28987.23
High Gray Level Zone Emphasis 4114822
Large Area Low Gray Level Emphasis 20294.92
Large Area High Gray Level Emphasis 24516968.00
Size-Zone Non-Uniformity Normalized 53324993.07
Gray Level Variance -41156602284720






GLRLM

Fig. Pneumonia Test X-ray Run Entropy -613628.52
Long Run Emphasis 121229
Run Length Non-Uniformity Normalized 424.19
Short Run Emphasis 70826.89
Run Length Non-Uniformity 27799399.00
Gray Level Non-Uniformity 27799399
Short Run High Gray Level Emphasis 1764111893
Run Variance 621140153040944
High Gray Level Run Emphasis 1459482205
Gray Level Variance 6857633138811517952
Low Gray Level Run Emphasis 715.66
Short Run Low Gray Level Emphasis 710.65
Long Run Low Gray Level Emphasis 2197.34
Run Percentage 1.19
Gray Level Non-Uniformity Normalized 424.19
Long Run High Gray Level Emphasis 15720983





GLDM

Fig. Pneumonia Test X-ray Dependence Entropy -6368297.12
Small Dependence Low Gray Level Emphasis 98454861.00
Large Dependence High Gray Level Emphasis 21704118103404179456
Small Dependence Emphasis 0.000496346876455413
High Gray Level Emphasis 98454861.00
Dependence Non-Uniformity Normalized 12312.63
Dependence non-uniformity 806920561.00
Large Dependence Low Gray Level Emphasis 15572578951.00
Large Dependence Emphasis 19440.82
Gray Level Non-Uniformity 801025.00
Dependence Variance 298931115780553.62
Small Dependence High Gray Level Emphasis 2772355743489.00
Low Gray Level Emphasis 98454861.00
Gray Level Variance 3471384038.82


[0072] The functions described herein may be implemented in hardware, software executed by a processor, firmware, or any combination thereof. If implemented in software executed by a processor, the functions may be stored on or transmitted over as one or more instructions or code on a computer-readable medium. Other examples and implementations are within the scope and spirit of the disclosure and appended claims. For example, due to the nature of software, functions described above can be implemented using software executed by a processor, hardware, firmware, hardwiring, or combinations of any of these. Features implementing functions may also be physically located at various positions, including being distributed such that portions of functions are implemented at different physical locations.
[0073] The foregoing description of the embodiments has been provided for purposes of illustration and is not intended to limit the scope of the present disclosure. Individual components of a particular embodiment are generally not limited to that particular embodiment, but, are interchangeable. Such variations are not to be regarded as a departure from the present disclosure, and all such modifications are considered to be within the scope of the present disclosure.
TECHNICAL ADVANCEMENTS AND ECONOMIC SIGNIFICANCE
[0074] The present disclosure described herein above has several technical advantages including, but not limited to, the realization of a system (and a method) for radiomics-based pneumonia diagnosis using machine learning, which:
• provides more accurate and earlier detection of pneumonia by analysing subtle variations in texture and structure within chest X-rays using radiomics techniques,
• automates the diagnostic process, improving both, speed and consistency while reducing the reliance on manual interpretation,
• uses machine learning classifiers like Bernoulli Naïve Bayes, Random Sub-space Boost, Quadratic Discriminant, and Gradient Boosting which enhances predictive performance, offering higher diagnostic accuracy as demonstrated by superior metrics such as Cohen's Kappa, Matthews Correlation Coefficient, Youden's Index, Log Loss, and Brier Score, and
• provides deeper insights into the image data, making it a more reliable tool for medical professionals and resulting in better patient outcomes.
[0075] The foregoing disclosure has been described with reference to the accompanying embodiments which do not limit the scope and ambit of the disclosure. The description provided herein is purely by way of example and illustration.
[0076] The embodiments herein and the various features and advantageous details thereof are explained with reference to the non-limiting embodiments in the following description. Descriptions of well-known components and processing techniques are omitted so as to not unnecessarily obscure the embodiments herein. The examples used herein are intended merely to facilitate an understanding of ways in which the embodiments herein may be practiced and to further enable those of skill in the art to practice the embodiments herein. Accordingly, the examples should not be construed as limiting the scope of the embodiments herein.
[0077] The foregoing description of the specific embodiments so fully reveals the general nature of the embodiments herein that others can, by applying current knowledge, readily modify and/or adapt for various applications such specific embodiments without departing from the generic concept, and, therefore, such adaptations and modifications should and are intended to be comprehended within the meaning and range of equivalents of the disclosed embodiments. It is to be understood that the phraseology or terminology employed herein is for the purpose of description and not of limitation. Therefore, while the embodiments herein have been described in terms of preferred embodiments, those skilled in the art will recognize that the embodiments herein can be practiced with modification within the spirit and scope of the embodiments as described herein.
[0078] Throughout this specification, the word "comprise", or variations such as "comprises" or "comprising", will be understood to imply the inclusion of a stated element, or group of elements, but not the exclusion of any other element, or group of elements.
[0079] Any discussion of documents, acts, materials, devices, articles, or the like that has been included in this specification is solely for the purpose of providing a context for the disclosure. It is not to be taken as an admission that any or all of these matters form a part of the prior art base or were common general knowledge in the field relevant to the disclosure as it existed anywhere before the priority date of this application.
[0080] The numerical values mentioned for the various physical parameters, dimensions, or quantities are only approximations and it is envisaged that the values higher/lower than the numerical values assigned to the parameters, dimensions or quantities fall within the scope of the disclosure, unless there is a statement in the specification specific to the contrary.
[0081] While considerable emphasis has been placed herein on the components and component parts of the preferred embodiments, it will be appreciated that many embodiments can be made and that many changes can be made in the preferred embodiments without departing from the principles of the disclosure. These and other changes in the preferred embodiment as well as other embodiments of the disclosure will be apparent to those skilled in the art from the disclosure herein, whereby it is to be distinctly understood that the foregoing descriptive matter is to be interpreted merely as illustrative of the disclosure and not as a limitation. , Claims:WE CLAIM:
1. A system (100) for radiomics-based pneumonia diagnosis using machine learning, the system (100) comprising:
an input module (104) configured to receive a plurality of X-ray images of human chest as input data from an image capturing device (102);
a pre-processing module (106) configured to receive the captured X-ray images, and further configured to implement a set of pre-processing rules to generate pre-processed X-ray images;
a radiomics feature extraction module (108) configured to extract texture and radiomics features from the pre-processed X-ray images by implementing radiomics techniques including Gray Level Size Zone Matrix (GLSZM), Gray Level Co-Occurrence Matrix (GLCM), Gray Level Dependence Matrix (GLDM), and Gray Level Run Length Matrix (GLRLM);
a power spectral density (PSD) analysis module (110) configured to analyse the extracted radiomics features by implementing PSD analysis, including Burg Power Spectral Density Estimate, Yule-Walker Power Spectral Density Estimate, and Welch Power Spectral Density Estimate, to capture frequency characteristics that enrich feature data;
a classification module (112) configured to process the extracted and PSD analysed features by implementing a plurality of machine learning based classifiers, including Bernoulli Naïve Bayes, Quadratic Discriminant, Random Subspace Boost, and Gradient Boosting, for classifying the X-ray images as either pneumonia-affected or healthy based on the analysed radiomics features; and
an output module (114) configured to display classification results on a user interface, providing real-time diagnostic feedback regarding the presence or absence of pneumonia in the analyzed chest X-ray images.
2. The system (100) as claimed in claim 1, wherein the classification module (112) is further configured to output classification accuracy metrics, including Cohen's Kappa, Matthews Correlation Coefficient, Youden's Index, Log Loss, and Brier Score, to evaluate classifier performance.
3. The system (100) as claimed in claim 1, wherein the classification module (112) is configured to handle high-dimensional radiomics data using a plurality of machine learning classifiers, including Bernoulli Naïve Bayes, Quadratic Discriminant, Random Sub-space Boost, and Gradient Boosting, to accurately distinguish between pneumonia and non-pneumonia cases.
4. The system (100) as claimed in claim 1, wherein the set of pre-processing rules comprises noise reduction filters and normalization techniques to standardize chest X-ray images for optimal feature extraction.
5. The system (100) as claimed in claim 1, wherein the system (100) further includes:
a repository (118) configured to store the set of predefined instructions and the set of pre-processing rules, and
a microprocessor (120) coupled to the repository (118) to execute the set of predefined instructions for executing one or more processing units.
6. The system (100) as claimed in claim 1, wherein the radiomics feature extraction module (108) further comprises a set of extraction techniques for computing radiomic parameters including Cluster Shade, Autocorrelation, Contrast, and Zone Entropy for improved texture analysis.
7. The system (100) as claimed in claim 1, wherein the PSD analysis module (110) is configured to apply the Welch PSD method to provide additional insight into the spatial frequencies of lung textures indicative of pneumonia.
8. The system (100) as claimed in claim 1, wherein the GLCM features extracted from the X-ray images include metrics containing auto-correlation, cluster prominence, cluster shade, correlation, energy, and entropy, which provide insights into the texture and spatial patterns of the lung tissues.
9. The system (100) as claimed in claim 1, wherein the GLSZM features extracted from the X-ray images include small area emphasis, large area emphasis, gray level non-uniformity, and zone entropy, enabling detailed analysis of size zones in gray-level regions within the lung tissues.
10. The system (100) as claimed in claim 1, wherein the GLRLM features extracted include parameters such as short run emphasis, long run emphasis, run length non-uniformity, and gray level variance, providing insights into the length and uniformity of runs of consecutive pixels at the same gray level.
11. The system (100) as claimed in claim 1, wherein the GLDM features extracted include small dependence emphasis, large dependence emphasis, gray level variance, and dependence entropy, which measure the dependence of pixels at varying gray levels within the lung tissue regions.
12. The system (100) as claimed in claim 1, wherein the output module (114) presents a graphical representation of the classifier's performance, including Precision-Recall Curves (AUC-PR), for further diagnostic evaluation by healthcare professionals.
13. A method (1000) for radiomics-based pneumonia diagnosis using machine learning, comprising:
receiving (1002), by an input module (104), a plurality of X-ray images of human chest as input data from an image capturing device (102);
pre-processing (1004), by a pre-processing module (106), the X-ray images by implementing a set of pre-processing rules to generate pre-processed X-ray images;
extracting (1006), by a radiomics feature extraction module (108), texture and radiomics features from the pre-processed X-ray images by implementing radiomics techniques including Gray Level Size Zone Matrix (GLSZM), Gray Level Co-Occurrence Matrix (GLCM), Gray Level Dependence Matrix (GLDM), and Gray Level Run Length Matrix (GLRLM);
analysing (1008), by a power spectral density analysis module (110), the extracted radiomics features by implementing PSD analysis, including Burg Power Spectral Density Estimate, Yule-Walker Power Spectral Density Estimate, and Welch Power Spectral Density Estimate, to capture frequency characteristics that enrich feature data;
processing (1010), by a classification module (112), the extracted and PSD analysed features by implementing a plurality of machine learning based classifiers, including Bernoulli Naïve Bayes, Quadratic Discriminant, Random Subspace Boost, and Gradient Boosting, for classifying the X-ray images as either pneumonia-affected or healthy based on the analysed radiomics features; and
displaying (1012), by an output module (114), classification results on a user interface (116), providing real-time diagnostic feedback regarding the presence or absence of pneumonia in the analyzed chest X-ray images.
14. The method (1000) as claimed in claim 9, wherein the steps of implementing the set of pre-processing rules comprises the steps of:
denoising of the X-ray images by removal of unwanted noise from the X-ray images while preserving critical structural detail,
re-sizing the X-ray images to a dimension of 224 multiplied by 224 pixels,

batch normalizing the re-scaled the X-ray images by dividing the pixel values by 225; and
augmenting the batches of the normalized X-ray images.

Dated this 06th Day of November, 2024

_______________________________
MOHAN RAJKUMAR DEWAN, IN/PA - 25
OF R. K. DEWAN & CO.
AUTHORIZED AGENT OF APPLICANT

TO,
THE CONTROLLER OF PATENTS
THE PATENT OFFICE, AT CHENNAI

Documents