Data Processing System for Improving Tomographic Reconstruction Using Machine Learning and Method Thereof

Abstract

ABSTRACT: Title: Data Processing System for Improving Tomographic Reconstruction Using Machine Learning and Method Thereof The present disclosure proposes a data processing system (100) that employs machine learning to enhance the quality of projection data for improved tomographic imaging. The data processing system (100) for improving tomographic reconstruction comprises a computing device (102) having a processor (104) and a memory (106) for storing one or more instructions executable by the processor (104). In addition, the computing device (102) is in communication with an application server (122) via a network (120). The processor (104) is configured to execute plurality of modules (108) for performing multiple functions. The plurality of modules (108) comprises a data acquisition module (110), an extraction module (112), a processing module (114), a generation module (116) and a reconstruction module (118). The proposed data processing system (100) produces high-quality projection data characterized by reduced noise, enhanced spatial resolution, and improved contrast.

Specification

Description:DESCRIPTION:
Field of the invention:
[0001] The present disclosure generally relates to the technical field of tomography, and in specific, relates to a data processing system employing machine learning to enhance the quality of projection data for improved tomographic imaging.
Background of the invention:
[0002] Tomographic reconstruction is a pivotal technique used in various imaging modalities, including computed tomography (CT), magnetic resonance imaging (MRI), and positron emission tomography (PET). These methods rely on projection data to generate cross-sectional or three-dimensional images for applications in medical diagnostics, industrial inspections, and research studies. High-quality projection data is essential for producing accurate and reliable tomographic images. Conventional tomographic systems face significant challenges in acquiring high-quality projection data. For instance, in medical imaging, achieving high data quality often necessitates the use of elevated radiation doses.

[0003] Prolonged exposure to such doses poses substantial health risks to patients. When radiation doses are reduced to prioritize safety, the resulting projection data frequently exhibit lower quality, characterized by increased noise, diminished spatial resolution, and poor contrast. Another limitation lies in the prevalence of artifacts, such as streaks and blurring, in lower-quality projection data. These artifacts compromise the accuracy of reconstructed images, affecting diagnostic reliability. Although traditional noise reduction and artifact correction techniques have been developed, they are often inadequate in addressing these issues comprehensively.

[0004] Computational complexity further compounds these challenges. Established reconstruction methods, such as filtered back-projection and iterative reconstruction, require substantial computational resources and are time-intensive, particularly when processing noisy or artifact-prone data. Additionally, conventional systems lack adaptability to dynamically enhance projection data quality during real-time imaging operations, further restricting their utility in advanced applications. Recent advancements in machine learning have introduced transformative possibilities in the field of tomographic imaging.

[0005] Machine learning models, such as convolutional neural networks (CNNs), demonstrate exceptional capabilities in mapping lower-quality input data to higher-quality outputs. These models, when trained on paired datasets, can effectively reduce noise, enhance spatial resolution, and correct artifacts autonomously. The integration of machine learning in imaging systems represents a promising approach to overcoming limitations in traditional techniques. Despite its potential, the application of machine learning in tomographic reconstruction has been constrained by several factors.

[0006] One significant challenge is the limited availability of comprehensive paired datasets representing diverse projection data qualities. This insufficiency hinders the training and generalization of machine learning models. Furthermore, real-time processing demands considerable computational power, which poses challenges for hardware implementation. Current systems also lack seamless integration between machine learning-based data enhancement and advanced reconstruction techniques, such as compressed sensing and iterative methods, limiting the scope of their applicability.

[0007] There is a pressing need for a comprehensive system that leverages the strengths of machine learning and tomographic reconstruction techniques. Such a system should enable the enhancement of lower-quality projection data, particularly those acquired at reduced radiation doses, while minimizing noise and artifacts. Additionally, it should preserve or improve spatial resolution and contrast, seamlessly integrate with advanced reconstruction methods, and provide real-time processing capabilities for both medical and industrial imaging applications.
Objectives of the invention:
[0008] The primary objective of the invention is to provide a data processing system employing machine learning to enhance the quality of projection data for improved tomographic imaging.

[0009] Another objective of the invention is to provide a data processing system to achieve improved tomographic reconstruction even when input projection data is acquired at a significantly reduced radiation dose, ranging from 1% to 90% of the standard dose.

[0010] The other objective of the invention is to provide a data processing system that employ machine learning models, including neural networks such as convolutional neural networks, deep neural networks, or recurrent neural networks, to process and enhance projection data quality.

[0011] The other objective of the invention is to provide a data processing system that utilizes a machine learning model trained on paired datasets of lower-quality and higher-quality projection data to process extracted data and generate improved projection data.

[0012] Yet another objective of the invention is to provide a data processing system that enables pre-processing of input projection data through noise reduction, normalization, artifact correction, or interpolation techniques to improve the data's quality before machine learning processing.

[0013] Further objective of the invention is to provide a data processing system that produces high-quality projection data characterized by reduced noise, enhanced spatial resolution, and improved contrast.
Summary of the invention:
[0014] The present disclosure proposes a data processing system for improving tomographic reconstruction using machine learning and method thereof. The following presents a simplified summary in order to provide a basic understanding of some aspects of the claimed subject matter. This summary is not an extensive overview. It is not intended to identify key/critical elements or to delineate the scope of the claimed subject matter. Its sole purpose is to present some concepts in a simplified form as a prelude to the more detailed description that is presented later.

[0015] In order to overcome the above deficiencies of the prior art, the present disclosure is to solve the technical problem to provide a data processing system employing machine learning to enhance the quality of projection data for improved tomographic imaging.

[0016] According to an aspect, the invention provides a data processing system for improving tomographic reconstruction using machine learning. In one embodiment herein, the data processing system comprises a computing device having a processor and a memory for storing one or more instructions executable by the processor. In addition, the computing device is in communication with an application server via a network. The processor is configured to execute a plurality of modules for performing multiple functions. The plurality of modules comprises a data acquisition module, an extraction module, a processing module, a generation module, and a reconstruction module.

[0017] In one embodiment herein, the data acquisition module is configured to acquire input projection data of lower quality from an imaging system. The input projection data comprises one or more sinograms, 2D images, or 3D images. The input projection data is acquired from one or more detectors, a storage system, a cloud-based server, or a remote source. The input projection data is pre-processed by applying noise reduction, normalization, or artifact correction algorithms.

[0018] In one embodiment herein, the extraction module is configured to extract multiple parameters of the input projection data. The multiple parameters include features, regions, and pixels. The processing module is configured to process the extracted data using a machine learning model, which is trained to map lower-quality projection data to higher-quality projection data. The lower-quality projection data is acquired at a radiation dose ranging from 1% to 90% of the standard dose used in imaging systems.

[0019] In one embodiment herein, the generation module is configured to generate an output projection data of higher quality based on the processing by the machine learning model. The reconstruction module is configured to utilize the output projection data and reconstruct tomographic data using one or more reconstruction techniques. In particular, the reconstruction techniques include one or more of back-projection, filtered back-projection, iterative reconstruction, or compressed sensing methods. The output projection data of higher quality is characterized by reduced noise, enhanced spatial resolution, or improved contrast.

[0020] In one embodiment herein, the input projection data is acquired from one or more detectors, a storage system, a cloud-based server, or a remote source. The machine learning model is pre-trained using paired datasets comprising lower-quality projection data and corresponding higher-quality projection data. The machine learning model is a neural network selected from one or more convolutional neural networks, deep neural networks, or recurrent neural networks.

[0021] According to another aspect, the invention provides a method for operating the data processing system for improving tomographic reconstruction. At one step, the data acquisition module acquires the input projection data of lower quality from the imaging system. At one step, the extraction module extracts the multiple parameters of the input projection data.

[0022] At one step, the processing module processes the extracted data using a machine learning model, which is trained to map lower-quality projection data to higher-quality projection data. At one step, the generation module generates the output projection data of higher quality based on the processing by the machine learning model. At one step, the reconstruction module utilizes the output projection data and reconstructs tomographic data using one or more reconstruction techniques.

[0023] Further, objects and advantages of the present invention will be apparent from a study of the following portion of the specification, the claims, and the attached drawings.
Detailed description of drawings:
[0024] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate an embodiment of the invention, and, together with the description, explain the principles of the invention.

[0025] FIG. 1 illustrates a block diagram of a data processing system for improving tomographic reconstruction, in accordance to an exemplary embodiment of the present invention.

[0026] FIG. 2 illustrates a flowchart of a method for operating a data processing system for improving tomographic reconstruction, in accordance to an exemplary embodiment of the present invention.
Detailed invention disclosure:
[0027] Various embodiments of the present invention will be described in reference to the accompanying drawings. Wherever possible, same or similar reference numerals are used in the drawings and the description to refer to the same or like parts or steps.

[0028] The present disclosure has been made with a view towards solving the problem with the prior art described above, and it is an object of the present invention to provide a data processing system employing machine learning to enhance the quality of projection data for improved tomographic imaging.

[0029] According to an exemplary embodiment of the invention, FIG. 1 refers to a data processing system 100 for improving tomographic reconstruction using machine learning. In one embodiment herein, the data processing system comprises a computing device 102 having a processor 104 and a memory 106 for storing one or more instructions executable by the processor 104. In addition, the computing device 102 is in communication with an application server 122 via a network 120.

[0030] The processor 104 is configured to execute a plurality of modules 108 for performing multiple functions. The plurality of modules 108 comprises a data acquisition module 110, an extraction module 112, a processing module 114, a generation module 116, and a reconstruction module 118.

[0031] In one embodiment herein, the data acquisition module 110 is configured to acquire input projection data of lower quality from an imaging system. The input projection data comprises one or more sinograms, 2D images, or 3D images. The input projection data is acquired from one or more detectors, a storage system, a cloud-based server, or a remote source. The input projection data is pre-processed by applying noise reduction, normalization, or artifact correction algorithms.

[0032] In one embodiment herein, the extraction module 112 is configured to extract multiple parameters of the input projection data. The multiple parameters include features, regions, and pixels. The processing module 114 is configured to process the extracted data using a machine learning model, which is trained to map lower-quality projection data to higher-quality projection data. The lower-quality projection data is acquired at a radiation dose ranging from 1% to 90% of the standard dose used in imaging systems.

[0033] In one embodiment herein, the generation module 116 is configured to generate an output projection data of higher quality based on the processing by the machine learning model. The reconstruction module 118 is configured to utilize the output projection data and reconstruct tomographic data using one or more reconstruction techniques. In particular, the reconstruction techniques include one or more of back-projection, filtered back-projection, iterative reconstruction, or compressed sensing methods. The output projection data of higher quality is characterized by reduced noise, enhanced spatial resolution, or improved contrast.

[0034] In one embodiment herein, the input projection data is acquired from one or more detectors, a storage system, a cloud-based server, or a remote source. The machine learning model is pre-trained using paired datasets comprising lower-quality projection data and corresponding higher-quality projection data. The machine learning model is a neural network selected from one or more convolutional neural networks, deep neural networks, or recurrent neural networks.

[0035] In one embodiment herein, the computing device 102 includes, but not limited to, a computer, a smartphone, a laptop and a personal digital assistant (PDA). The network 120 includes, but not limited to, a local area network (LAN), a wide area network (WAN) and cloud-based network. The application server 122 includes, but not limited to, a centralized processing unit, machine learning frameworks and communication interfaces.

[0036] In one embodiment, a tomographic imaging system, such as a computed tomography (CT) system, acquires raw projection data in the form of signals, images, or volumes. These data are obtained by transmitting electromagnetic waves (e.g., X-rays, visible light, ultraviolet light, and infrared light) or sound waves (e.g., ultrasound) through an object. The waves carry information specific to the object's internal properties, such as the attenuation coefficients of its materials. The system employs reconstruction algorithms, such as filtered back-projection (FBP), inverse Radon transform, iterative reconstruction (IR), and algebraic reconstruction technique (ART), to transform the acquired raw projection data into tomographic images.

[0037] In one embodiment, the present invention introduces a reconstruction-based method for improving the quality of tomographic images. Unlike acquisition-based methods that optimize parameters during data collection or image-domain-based methods that process fully reconstructed images, this method focuses on enhancing raw projection data before reconstruction. The technique applies machine learning to transform low-quality projection data, such as noisy, low-dose sinograms, into high-quality data with reduced noise and artifacts. The enhanced projection data are then processed using standard reconstruction algorithms to generate high-quality tomographic images.

[0038] In one embodiment, a machine learning model is used to transform raw projection data. The model utilizes an input local window to extract regions or patches from the raw data and an output local window to generate transformed patches. The input local window shifts across the data with overlapping regions, ensuring comprehensive coverage. The size of the input window is generally larger than or equal to the output window. The transformed patches are seamlessly combined to form enhanced projection data. Preferred models include artificial neural network regression, though other models such as support vector regression, convolutional neural networks, and Gaussian process regression may also be employed.

[0039] In one embodiment, the machine learning model undergoes a supervised training process. The training involves paired datasets of low-quality projection images (e.g., low-dose, noisy sinograms) and high-quality projection images (e.g., high-dose, artifact-free sinograms). The model learns to transform low-quality images into high-quality counterparts by adjusting its parameters. Once trained, the model processes new low-dose projection images, producing high-quality output images that are free from noise and artifacts.

[0040] In one embodiment, the invention applies to X-ray imaging, where low-dose projection images are transformed into high-dose-like projection images. Low-dose images typically have high noise levels and numerous artifacts due to reduced radiation exposure. The machine learning model transforms these images into high-dose-like images, providing improved signal-to-noise ratios and reduced artifacts. This transformation allows the use of safer, low-dose imaging protocols without compromising diagnostic image quality.

[0041] In one embodiment, the transformed high-quality projection data are processed using reconstruction algorithms such as FBP, inverse Radon transform, IR, or ART. These algorithms reconstruct tomographic images from the enhanced projection data. For example, 2D tomographic images can be reconstructed from 1D projection data, or 3D volumetric images can be reconstructed from 2D projections. By preprocessing the projection data using machine learning, the invention mitigates noise and artifacts, enabling these algorithms to produce superior image quality even with low-dose data.

[0042] In one embodiment, the invention comprises two distinct steps. The first step involves training the machine learning model to transform lower-quality projection images into higher-quality counterparts. This is achieved using supervised learning with paired datasets of low-dose and high-dose sinograms. The second step involves applying the trained model to new low-dose projection data, producing high-quality data suitable for reconstruction. The reconstructed images closely resemble those obtained from high-dose imaging, with noise and artifacts significantly reduced.

[0043] In one embodiment, the invention is used for diagnostic applications in radiology. The machine learning model processes low-dose projection images to produce high-quality output, which is then reconstructed into tomographic images using algorithms such as FBP or IR. The reconstructed images exhibit improved diagnostic quality, enhancing radiologists' sensitivity and specificity in detecting lesions. This advancement enables safer imaging protocols by reducing radiation exposure without compromising image quality, potentially improving patient outcomes.

[0044] According to another embodiment of the invention, FIG. 2 refers to a flowchart 200 of a method for operating the data processing system 100 for improving tomographic reconstruction. At step 202, the data acquisition module 110 acquires the input projection data of lower quality from the imaging system. At step 204, the extraction module 112 extracts the multiple parameters of the input projection data.

[0045] At step 206, the processing module 114 processes the extracted data using a machine learning model, which is trained to map lower-quality projection data to higher-quality projection data. At step 208, the generation module 116 generates the output projection data of higher quality based on the processing by the machine learning model. At step 210, the reconstruction module 118 utilizes the output projection data and reconstructs tomographic data using one or more reconstruction techniques.

[0046] Numerous advantages of the present disclosure may be apparent from the discussion above. In accordance with the present disclosure, a data processing system 100 for improving tomographic reconstruction using machine learning is disclosed. The proposed data processing system 100 employs machine learning to enhance the quality of projection data for improved tomographic imaging. The proposed data processing system 100 achieves improved tomographic reconstruction even when input projection data is acquired at a significantly reduced radiation dose, ranging from 1% to 90% of the standard dose.

[0047] The proposed data processing system 100 employs machine learning models, including neural networks such as convolutional neural networks, deep neural networks, or recurrent neural networks, to process and enhance projection data quality. The proposed data processing system 100 utilizes a machine learning model trained on paired datasets of lower-quality and higher-quality projection data to process extracted data and generate improved projection data.

[0048] The proposed data processing system 100 enables the pre-processing of input projection data through noise reduction, normalization, artifact correction, or interpolation techniques to improve the data's quality before machine learning processing. The proposed data processing system 100 produces high-quality projection data characterized by reduced noise, enhanced spatial resolution, and improved contrast.

[0049] It will readily be apparent that numerous modifications and alterations can be made to the processes described in the foregoing examples without departing from the principles underlying the invention, and all such modifications and alterations are intended to be embraced by this application.
, Claims:CLAIMS:
I / We Claim:
1. A data processing system (100) for improving tomographic reconstruction, comprising:
a computing device (102) having a processor (104) and a memory (106) for storing one or more instructions executable by the processor (104), wherein the computing device (102) is in communication with an application server (122) via a network (120), wherein the processor (104) is configured to execute plurality of modules (108) for performing multiple functions, wherein the plurality of modules (108) comprises:
a data acquisition module (110) configured to acquire input projection data of lower quality from an imaging system;
an extraction module (112) configured to extract multiple parameters of the input projection data;
a processing module (114) configured to process the extracted data using a machine learning model, which is trained to map lower-quality projection data to higher-quality projection data;
a generation module (116) configured to generate an output projection data of higher quality based on the processing by the machine learning model; and
a reconstruction module (118) configured to utilize the output projection data and reconstruct tomographic data using one or more reconstruction techniques.
2. The data processing system (100) as claimed in claim 1, wherein the input projection data comprises one or more of sinograms, 2D images, or 3D images, wherein the input projection data is acquired from one or more of a detector, a storage system, a cloud-based server, or a remote source.
3. The data processing system (100) as claimed in claim 1, wherein the multiple parameters include features, regions and pixels.
4. The data processing system (100) as claimed in claim 1, wherein the lower-quality projection data is acquired at a radiation dose ranging from 1% to 90% of the standard dose used in imaging systems.
5. The data processing system (100) as claimed in claim 1, wherein the machine learning model is pre-trained using paired datasets comprising lower-quality projection data and corresponding higher-quality projection data.
6. The data processing system (100) as claimed in claim 1, wherein the machine learning model is a neural network selected from one or more of convolutional neural networks, deep neural networks, or recurrent neural networks.
7. The data processing system (100) as claimed in claim 1, wherein the reconstruction techniques include one or more of back-projection, filtered back-projection, iterative reconstruction, or compressed sensing methods.
8. The data processing system (100) as claimed in claim 1, wherein the input projection data is pre-processed by applying noise reduction, normalization, or artifact correction algorithms.
9. The data processing system (100) as claimed in claim 1, wherein the output projection data of higher quality is characterized by reduced noise, enhanced spatial resolution, or improved contrast.
10. A method for operating a data processing system (100) for improving tomographic reconstruction, comprising:
acquiring, a data acquisition module (110), input projection data of lower quality from an imaging system;
extracting, an extraction module (112), multiple parameters of the input projection data;
processing, a processing module (114), the extracted data using a machine learning model, which is trained to map lower-quality projection data to higher-quality projection data;
generating, a generation module (116), an output projection data of higher quality based on the processing by the machine learning model; and
utilizing, a reconstruction module (118), the output projection data and reconstruct tomographic data using one or more reconstruction techniques.

Documents