Claims:1. A method for improving image reconstruction of a magnetic resonance imaging (MRI) (104), the method comprising:
receiving a user instruction data through an electronic user interface (204);
sending, by a processor (202), an optimized sequence setting for accelerating an image scan acquisition;
receiving, by the processor (202), an anonymized image scan pertaining to a patient and a scan format of the anonymized image scan, wherein the anonymized image scan is acquired by under sampling in K space of the MRI (104) using the optimized sequence setting and wherein the optimized sequence setting is recommended based on a load factor;
predicting, by the processor (202), a high frequency image scan from the anonymized image scan in image domain, wherein a deep learning algorithm is used for predicting the high frequency image scan;
reconstructing, by the processor (202), the high frequency image scan wherein the reconstruction adds at least one quality parameter to the high frequency image scan using the deep learning algorithm;
outputting, by the processor (202), the reconstructed high frequency image scan of the patient for analysis.
2. The method of claim 1, wherein the anonymized image scan is acquired by the deep learning algorithm using an AI engine (222) and the AI engine (222) is used while acquiring scan of an organ system and a part of the organ system of the patient.

3. The method of claim 1, wherein the optimized sequence setting is recommended using an AI engine (222) having the deep learning algorithm.

4. The method of claim 1, wherein the anonymized image scan is acquired by under sampling in frequency domain of the MRI (104).

5. The method of claim 1, wherein the anonymized image scan is as acquired image scan.
6. The method of claim 1, wherein the load factor is a patient volume load and the optimised sequence setting is changed through a remote plugin (112) for the MRI (104).

7. The method of claim 1, wherein the anonymized image scan is acquired by an anonymization software deployed remotely at a diagnostic center (110) having the MRI (104).

8. The method of claim 1, wherein receiving the anonymized image scan from a diagnostic center (110) having a group of devices including an MRI (104) machine, a scanning software machine (102), and a picture archiving and communication system (PACS) (108) through a remote plugin (112).

9. The method of claim 1, wherein the high frequency image scan has a missing frequency image scan.

10. The method of claim 1, wherein the scan format is a communication standard and the scan format is sent over a network (120) from a diagnostic center (110).

11. The method of claim 1, wherein the quality parameter includes parameters to sharpen, denoise, and dering using an AI engine (222) having the deep learning algorithm.

12. The method of claim 1, further comprising: preparing, by the processor (202), a report (402) from the reconstructed high frequency image scan of the patient for diagnosis and quantification.

13. The method of claim 1, further comprising: analysing, by the processor (202), the reconstructed high frequency image scan for anatomical structures, anatomical anomalies and anatomical features of the patient.

14. The method of claim 1, further comprising: evaluating, by the processor (202), the reconstructed high frequency image scan of the patient while receiving feedback on quality.

15. A system for improving image reconstruction of a magnetic resonance imaging (MRI) (104), the system comprising:
a processor (202); and
a memory (206) coupled to the processor (202), wherein the processor (202) executes a plurality of modules (208) stored in the memory (206), and wherein the plurality of modules (208) comprising:
an acquisition module (210), to send an optimized sequence setting for accelerating an image scan acquisition from the MRI (104) and receive an anonymized image scan pertaining to a patient and a scan format of the anonymized image scan, wherein the anonymized image scan is acquired by under sampling in K space of the MRI (104) using the optimized sequence setting and wherein the optimized sequence setting is recommended based on a load factor;
an AI engine module (212), to predict a high frequency image scan from the anonymized image scan in image domain, wherein a deep learning algorithm is used for predicting the high frequency image scan;
a reconstruction module (214), to reconstruct the high frequency image scan wherein the reconstruction adds at least one quality parameter to the high frequency image scan using the deep learning algorithm;
a report module (216), to output the reconstructed high frequency image scan of the patient for analysis.
16. The system of claim 15, wherein the acquisition module (210) acquires the anonymized image scan by the deep learning algorithm using an AI engine (222) and the AI engine (222) is used while acquiring scan of an organ systems and a part of the organ systems of the patient.

17. The system of claim 15, wherein the optimized sequence setting is recommended using an AI engine (222) having the deep learning algorithm.

18. The system of claim 15, wherein the acquisition module (210) acquires the anonymized image scan by under sampling in the frequency domain of the MRI (104).

19. The system of claim 15, wherein the anonymized image scan is as acquired image scan.

20. The system of claim 15, wherein the load factor is a patient volume load and the optimised sequence setting is changed through a remote plugin (112) for the MRI (104).

21. The system of claim 15, wherein the anonymized image scan is acquired by an anonymization software deployed remotely, in communication with the acquisition module (210), at a diagnostic center (110) having the MRI (104).

22. The system of claim 15, wherein the acquisition module (210) receives the anonymized image scan from a diagnostic center (110) having a group of devices including an MRI (104) machine, a scanning software machine (102), and a picture archiving and communication system (PACS) (108) through a remote plugin (112).

23. The system of claim 15, wherein the high frequency image scan has a missing frequency image scan.

24. The system of claim 15, wherein the scan format is a communication standard and the scan format is sent over a network (120) from a diagnostic center (110).

25. The system of claim 15, wherein the quality parameter includes parameters to sharpen, denoise, and dering using an AI engine (222) having the deep learning algorithm.

26. The system of claim 15, wherein the report module (216) prepares a report (402) from the reconstructed high frequency image scan of the patient for diagnosis and quantification.

27. The system of claim 15, further comprising: an analysis module (218), to analyse the reconstructed high frequency image scan for anatomical structures, anatomical anomalies and anatomical features of the patient.

28. The system of claim 15, further comprising: an evaluation module (220), to evaluate the reconstructed high frequency image scan of the patient while receiving feedback on quality.

29. The system of claim 15, wherein the plurality of modules (208) are hosted on a data center that is in communication with a remote plugin (112).

30. The system of claim 15, wherein the plurality of modules (208) are hosted on a cloud (130) that is in communication with a remote plugin (112) through a network (120).
, Description:SYSTEM AND METHOD FOR IMPROVING IMAGE RECONSTRUCTION

FIELD OF THE INVENTION
The present disclosure generally relates to the field of magnetic resonance imaging (MRI), more specifically the present disclosure relates to a system and method for improving image reconstruction using artificial intelligence delivering a faster environment, reducing hassle being taken by technicians, while maintaining quality standards.
BACKGROUND OF THE INVENTION
Magnetic resonance imaging (MRI) is a medical imaging technique commonly used to visualize organs or various parts of the organ systems of a patient. The MRI technique provides clinical image with good resolution, contrast between the different soft tissues of the body, acquires images of the bone, disk, joint, nerves, ligaments, and organs etc. and is therefore referred as an ideal imaging technique. The MRI technique has the advantage of using various control ratios by controlling various parameters to obtain images of multiple control ratios in the same region in clinical diagnosis. However, since a series of images require a long scanning time, at times they cause inconvenience to a patient, elongates the patient queue and overall reduces the usability of the costly MRI machines. Such long scan time for MRI may result in high imaging cost and limit the patient volume and accessibility.
In various interviews held with large MRI diagnostic centers, radiologists having years of significant experience in acquiring, sharing feedback and analyzing thousands of reports every year have expressed views on the available solutions and challenges. The diagnostic centers are in high demand, have higher patient waiting times and are looking for technical solutions to reduce patient turnaround time. Most of the available MRI machines do not use any or much of intelligence tools but the diagnostic centers/users are willing to try new developments in technology. For patients with conditions that require low level of specialty scans usually meets with standard solutions and is easily detectable but for patients with complex conditions require higher speciality requirements and may not be easily detected without technological assistance. The diagnostic centers have MRI machines from different manufacturers and there is a need to clear regulatory hurdles for accepting any changes as well. To reduce patient load, accommodate all patients or in emergency situations, the technicians find it difficult to obtain high image scans while maintaining the quality and also face situations of rescans or patient dissatisfaction leading to overall reduced quality of service being offered by small or large hospitals/ diagnostic centers alike.
Some MRI manufacturers have addressed issues based on changing software inside MRI machine but generally it is very specific to that machine and sometimes to a model and is less of a technological solution but mostly a reduction in quality. Current solutions require costly MRI hardware upgrade or software inside MRI upgrade and are offered as premium solutions by manufacturers. The current solutions do not use advanced technological solutions or artificial intelligence. Therefore, in acquiring magnetic resonance images, it is vital to provide a manufacturer agnostic, technical intelligence based solution that reduces acquisition time of the patient image scans and improves the quality of the reconstructed images.
Due to the complex parameters and effort required in image reconstruction of a magnetic resonance imaging (MRI), along with a faster acquisition time, there exists a need for developing a system and method for improving image reconstruction of a magnetic resonance imaging (MRI) which generates a high quality reconstructed image scan without or minimally requiring changes in existing machinery/ equipment.

SUMMARY OF THE INVENTION
This summary is provided to introduce concepts related to systems and methods for improving image reconstruction of a magnetic resonance imaging (MRI) and the concepts are further described below in the detailed description. This summary is neither intended to identify essential features of the claimed subject matter nor is it intended for use in determining or limiting the scope of the claimed subject matter.
In one implementation, a method for improving image reconstruction of a magnetic resonance imaging (MRI) is disclosed. The method comprises, receiving a user instruction data through an electronic user interface and sending, by a processor, an optimized sequence setting for accelerating an image scan acquisition. The method further comprising receiving, by the processor, an anonymized image scan pertaining to a patient and a scan format of the anonymized image scan, wherein the anonymized image scan is acquired by under sampling in K space of the MRI using the optimized sequence setting and wherein the optimized sequence setting is recommended based on a load factor. The method further comprising predicting, by the processor, a high frequency image scan from the anonymized image scan in image domain, wherein a deep learning algorithm is used for predicting the high frequency image scan and reconstructing, by the processor, the high frequency image scan wherein the reconstruction adds at least one quality parameter to the high frequency image scan using the deep learning algorithm and further outputting, by the processor, the reconstructed high frequency image scan of the patient for analysis.
In yet another implementation, the method having the anonymized image scan that is acquired by the deep learning algorithm using an AI engine and the AI engine is used while acquiring scan of an organ system and a part of the organ system of the patient.
In yet another implementation, the method having the optimized sequence setting that is recommended using an AI engine having the deep learning algorithm.
In yet another implementation, the method having the anonymized image scan that is acquired by under sampling in the frequency domain of the MRI.
In yet another implementation, the method having the anonymized image scan that is as acquired image scan.
In yet another implementation, the method having the load factor that is a patient volume load and the optimised sequence setting is changed through a remote plugin for the MRI.
In yet another implementation, the method having the anonymized image scan that is acquired by an anonymization software deployed remotely at a diagnostic center having the MRI.
In yet another implementation, the method having receiving the anonymized image scan from a diagnostic center having a group of devices including an MRI machine, a scanning software machine, and a picture archiving and communication system (PACS) through a remote plugin.
In yet another implementation, the method having the high frequency image scan that has a missing frequency image scan.
In yet another implementation, the method having the scan format that is a communication standard and the scan format is sent over a network from a diagnostic center.
In yet another implementation, the method having the quality parameter that includes parameters to sharpen, denoise, and dering using an AI engine having the deep learning algorithm.
In another implementation, the method comprises, preparing, by the processor, a report from the reconstructed high frequency image scan of the patient for diagnosis and quantification.
In another implementation, the method comprises, analysing, by the processor, the reconstructed high frequency image scan for anatomical structures, anatomical anomalies and anatomical features of the patient.
In another implementation, the method comprises, evaluating, by the processor, the reconstructed high frequency image scan of the patient while receiving feedback on usage quality.
In one implementation, a system for improving image reconstruction of a magnetic resonance imaging (MRI) is disclosed. The system comprising a processor and a memory coupled to the processor, wherein the processor executes a plurality of modules stored in the memory, and wherein the plurality of modules comprising an acquisition module, to send an optimized sequence setting for accelerating an image scan acquisition from the MRI and receive an anonymized image scan pertaining to a patient and a scan format of the anonymized image scan, wherein the anonymized image scan is acquired by under sampling in K space of the MRI using the optimized sequence setting and wherein the optimized sequence setting is recommended based on a load factor. The plurality of modules further comprises an AI engine module, to predict a high frequency image scan from the anonymized image scan in image domain, wherein a deep learning algorithm is used for predicting the high frequency image scan. The plurality of modules further comprises a reconstruction module, to reconstruct the high frequency image scan wherein the reconstruction adds at least one quality parameter to the high frequency image scan using the deep learning algorithm and further a report module, to output the reconstructed high frequency image scan of the patient for analysis.
In yet another implementation, the system has the acquisition module that acquires the anonymized image scan by the deep learning algorithm using an AI engine and the AI engine is used while acquiring scan of an organ systems and a part of the organ systems of the patient.
In yet another implementation, the system has the optimized sequence setting that is recommended using an AI engine having the deep learning algorithm.
In yet another implementation, the system has the acquisition module that acquires the anonymized image scan by under sampling in the frequency domain of the MRI.
In yet another implementation, the system has the anonymized image scan that is as acquired image scan.
In yet another implementation, the system has the load factor that is a patient volume load and the optimised sequence setting is changed through a remote plugin for the MRI.
In yet another implementation, the system has the anonymized image scan that is acquired by an anonymization software deployed remotely, in communication with the acquisition module, at a diagnostic center having the MRI.
In yet another implementation, the system has the acquisition module that receives the anonymized image scan from a diagnostic center having a group of devices including an MRI machine, a scanning software machine, and a picture archiving and communication system (PACS) through a remote plugin.
In yet another implementation, the system has the high frequency image scan that has a missing frequency image scan.
In yet another implementation, the system has the scan format that is a communication standard and the scan format is sent over a network from a diagnostic center.
In yet another implementation, the system has the quality parameter that includes parameters to sharpen, denoise, and dering using an AI engine having the deep learning algorithm.
In yet another implementation, the system has the report module that prepares a report from the reconstructed high frequency image scan of the patient for diagnosis and quantification.
In another implementation, the system comprises an analysis module, to analyse the reconstructed high frequency image scan for anatomical structures, anatomical anomalies and anatomical features of the patient.
In another implementation, the system comprises an evaluation module, to evaluate the reconstructed high frequency image scan of the patient while receiving feedback on usage quality.
In yet another implementation, the system has the plurality of modules that are hosted on a data center that is in communication with a remote plugin.
In yet another implementation, the system has the plurality of modules are hosted on a cloud that is in communication with a remote plugin through a network.
It is primary object of the subject matter to provide a system and method for improving image reconstruction of a magnetic resonance imaging (MRI) from an anonymized image scan pertaining to a patient that is acquired by under sampling in K space of the MRI using an optimized sequence setting based on a patient load factor and more specifically predict a high frequency image scan from the anonymized image scan in image domain using a deep learning algorithm and thereby reconstructing the high frequency image scan while adding quality parameters.
It is another object of the subject matter to provide a system and method for improving image reconstruction of a magnetic resonance imaging (MRI) that reduces time spent by a patient with an MRI machine, reduces probability of motion artefacts that would generally lead to rescans, decreases the MRI scan time by 1/2 to 1/4 while using cloud-enabled AI reconstruction deep learning algorithms on fast acquired MRI images.
It is another object of the subject matter to provide a system and method for improving image reconstruction of a magnetic resonance imaging (MRI) that increases revenue per machine and delivers better ROI, increases the lifetime value of the MRI machine, helps in adoption of whole body MRI which can screen multiple diseases/ cancer at a time, increases use of the MRI as a modality since the MRI is golden standard for imaging and is safer than other modalities.
It is another object of the subject matter to provide a system and method for improving image reconstruction of a magnetic resonance imaging (MRI) that operates in the image space of the image scan for scan enhancement.
It is another object of the subject matter to provide a system and method for improving image reconstruction of a magnetic resonance imaging (MRI) that is based on an innovative AI algorithm using a deep learning algorithm depending upon a deep neural network and can work on different sources of MRI scans.
It is another object of the subject matter to provide a system and method for improving image reconstruction of a magnetic resonance imaging (MRI) that is based on selection of clinical operator settings, getting an accelerated scan and use that scan for further analysis, can provide live feedback option to technicians based on AI analysis whether to use the scan or not.
It is another object of the subject matter to provide a system and method for improving image reconstruction of a magnetic resonance imaging (MRI) that takes care of a patient load in the diagnostic center and use that as an alert to technicians on quantum of acceleration using a recommended optimised sequence setting. The MRI scans are generated with optimized sequences with the change in settings available to technician/ clinician in a scanner machine software which results in the acquisition of a subset of samples in the K space or frequency domain of the MRI.
It is another object of the subject matter to provide a system and method for improving image reconstruction of a magnetic resonance imaging (MRI) that provides MRI enhancement solution in the image space and accelerates MRI scanning by up to 4 times. The settings are done in the MRI scanning software to acquire scans in an accelerated way and the enhancement is done by the software in the image space to improve the scan.
It is another object of the subject matter to provide a system and method for improving image reconstruction of a magnetic resonance imaging (MRI) that requires no change in a radiologist workflow and is compatible with 1.5t/3t and further supports scans for all body areas.
It is another object of the subject matter to provide a system and method for improving image reconstruction of a magnetic resonance imaging (MRI) that supports MRI machines from all major manufacturers, is MRI scan machine agnostic and does not require change in original MRI manufacturer software.
It is another object of the subject matter to provide a system and method for improving image reconstruction of a magnetic resonance imaging (MRI) that enables faster scan with good scan quality of MRI and provides 2 to 4 times reduction in time compared to normal scan. Saves time, reduces cost of operation and provides convenience to patients. Generate reduced motion artefact in MRI output which results in better diagnosis and lower repeatability of bad scans. Provide excellent return on investment by aiding radiologists for faster analysis and oncologists for surgical planning.
It is another object of the subject matter to provide a system and method for improving image reconstruction of a magnetic resonance imaging (MRI) that helps in faster diagnosis using computer automated segmentation, annotation and classification of tumours, tears in MRI scans. Helps to prioritize referral (triage) for doctors and provide input for examination.
It is another object of the subject matter to provide a number of advantages depending on the particular aspect, embodiment, implementation and/or configuration.
It is another object of the subject matter to provide a solution that provides reliable execution, scalability, and value-added services, while controlling operating effort and costs.
It is another object of the subject matter to efficiently manage numerous instances simultaneously, work in different regulatory requirements, enable resources to collaborate and work together closely, efficiently and collectively with user friendly interfaces.
These and other implementations, embodiments, processes and features of the subject matter will become more fully apparent when the following detailed description is read with the accompanying experimental details. However, both the foregoing summary of the subject matter and the following detailed description of it represent one potential implementation or embodiment and are not restrictive of the present disclosure or other alternate implementations or embodiments of the subject matter.

BRIEF DESCRIPTION OF THE DRAWINGS
A clear understanding of the key features of the subject matter summarized above may be had by reference to the appended drawings, which illustrate the method and system of the subject matter, although it will be understood that such drawings depict preferred embodiments of the subject matter and, therefore, are not to be considered as limiting its scope with regard to other embodiments which the subject matter is capable of contemplating. Accordingly:
FIGURE.1 illustrates a system diagram describing the working of an exemplary image reconstruction of a magnetic resonance imaging (MRI) system, in accordance with an embodiment of the present subject matter.
FIGURE.2 illustrates a schematic module diagram depicting a method of improving image reconstruction of a magnetic resonance imaging (MRI), in accordance with an embodiment of the present subject matter.
FIGURE.3 illustrates an exemplary flowchart of a method of improving image reconstruction of a magnetic resonance imaging (MRI), in accordance with an embodiment of the present subject matter.
FIGURE.4 illustrates an exemplary report interface automatically generated by an exemplary image reconstruction of a magnetic resonance imaging (MRI) system, in accordance with an embodiment of the present subject matter.
FIGURE.4a illustrates another part of an exemplary report automatically generated by an exemplary image reconstruction of a magnetic resonance imaging (MRI) system, in accordance with an embodiment of the present subject matter.
DETAILED DESCRIPTION OF THE INVENTION
The following is a detailed description of implementations of the present disclosure depicted in the accompanying drawings. The implementations are in such detail as to clearly communicate the disclosure. However, the amount of detail offered is not intended to limit the implementations but it is to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the present disclosure. While aspects of described systems and methods for improving image reconstruction of a magnetic resonance imaging (MRI) can be implemented in any number of different computing systems, environments, and/or configurations, the embodiments are described in the context of the following exemplary system(s) and exemplary method(s).
The present disclosure provides a method for improving image reconstruction of a magnetic resonance imaging (MRI). It receives a user instruction data through an electronic user interface and sends an optimized sequence setting for accelerating an image scan acquisition. Further, it receives an anonymized image scan pertaining to a patient and a scan format of the anonymized image scan, the anonymized image scan is acquired by under sampling in K space of the MRI using the optimized sequence setting. The optimized sequence setting is recommended based on a load factor. It further predicts a high frequency image scan from the anonymized image scan in image domain using a deep learning algorithm and reconstructs the high frequency image scan while adding at least one quality parameter to the high frequency image scan using the deep learning algorithm and outputs the reconstructed high frequency image scan of the patient for analysis.
The present disclosure provides a mechanism for accelerating a patient image scan for a MRI machine/ scanner. The acceleration can be set dynamically based on the time of the day when there is high demand such as high patient volume/ inflow. The acceleration factor can be anything like 3x times or 4x times while keeping the same reconstruction algorithm. The accelerated scan results in the acquisition of a subset of samples in the K space or image domain of the MRI. The accelerated scan format exists in typical image encoding standard and communication standard such as digital imaging and communications in medicine (DICOM) and the DICOMs are sent over the network from the scanner machine while anonymising the low frequency image scans. A cloud based method/ application using artificial intelligence (AI) algorithms, along with enhancement of standard of care scans, recommends acceleration setting where an optimized sequence is constructed using a change in the setting available for clinicians in a scanner workstation.
The present disclosure provides a mechanism for reconstruction of an image scan that is done in the image domain instead of frequency domain using a deep learning algorithm which predicts a high frequency from a low frequency in the image domain. The reconstruction is done in the image domain instead of frequency domain by a reconstruction software using the deep learning algorithm which also estimates missing frequency samples using AI algorithms. The reconstruction is done on the anonymised image scan by predicting a high frequency image scan and adding quality parameters on the cloud. In this mechanism a deep learning algorithm is used to reconstruct the scans and the AI model for reconstruction algorithm remains the same for different organs or parts of organs or organ systems.
The present solution reduces the workload of a radiologist/ a doctor, diagnosis time is reduced as a difference obtained after comparing image report and model prediction result is available in advance. The MRI scan machine can cater to a higher count of patients at the same time, making the scan more accessible, affordable. AI based solution brings intelligence, links existing medical imaging diagnosis workflow without adding additional hardware, no change in existing software solutions. Brings more standardization, quality control, big data and deep learning algorithms to further AI enabled research and development. By providing a solution in the image domain of the scans, data transmission quantity is multiplied, overheads of equipment, network and PACS server updates are annulled. The solution aims to bring medical professionals closer to intelligent solutions and human expertise + AI intelligence to medical diagnostics. A component wise structure and a step wise method is listed below.
FIGURE.1 illustrates a system diagram 100 describing the working of an exemplary image reconstruction of a magnetic resonance imaging (MRI) system, in accordance with an embodiment of the present subject matter.
The present disclosure depicts two important activities which are sourcing an anonymised image scan for assessment and reconstruction of a high frequency image scan to the desired requirements. These two activities are explained below in the system diagram 100.
In one embodiment, an application for improving image reconstruction of a magnetic resonance imaging (MRI) resides on a server. The application is linked to various supporting components, including but not limited to, servers, databases, network, diagnostic centers through a remote plugin, and the like. Functioning of the application linked to the image reconstruction of a magnetic resonance imaging (MRI) system is further detailed out.
The system diagram 100 illustrates a system that has a diagnostic center 110 having apparatuses including but not limited to a magnetic resonance imaging (MRI) 104, a scanning software machine 102, a picture archiving and communication system (PACS) 108, a remote plugin 112, and a copier printer 106 among others. The diagnostic center 110 is a remotely placed center where a patient would visit generally situated in a hospital or a standalone lab and is connected through a network 120. A user 140 can access the various parts of the system through the network 120. The cloud 130 is generally connected to one or a plurality of the diagnostic centers 110 spread across locations globally.
In one embodiment, a diagnostic center 110 is connected through a network 120 further connected to a data center that includes but not limited to servers such as a server 132a, a server 132b, and alike, databases such as a database 134a, a database 134b, and alike, and other computing, network resources. The data center may be locally or remotely placed and is accessible directly by the diagnostic center 110 resources.
In one embodiment, a diagnostic center 110 is connected through a network 120 to a cloud 130, the cloud 130 including but not limited to servers such as a server 132a, a server 132b, and alike, databases such as a database 134a, a database 134b, and alike, and other computing, network resources.
The MRI 104 of the diagnostic center 110 is a magnetic resonance imaging (MRI) system comprising an imaging accelerator, a magnet system, a patient transport table connected to the magnet system, and a controller connected to the other parts of the MRI 104. In one example of the present solution, a patient may visit the diagnostic center 110 and gets scanned by the MRI 104 through the patient transport table and the magnet system. In one example, the MRI 104 comprises a computing system having at least a processor and a database and implements the present solution.
The scanning software machine 102 of the diagnostic center 110 may have a display that can be a CRT monitor, an LCD monitor, an LED monitor, a console monitor, a conventional workstation commonly used in medical industry, diagnostic centers, hospitals to view medical images provided from the DICOM/ picture archiving and communication system (PACS) or directly from a medical imaging device or from some other source, or other display device. An image interface of the scanning software machine 102 provides view of the tomographic images. The scanning software machine 102 has a computing device comprising at least a processor and a memory that can host a scan software for the MRI 104 hosted locally i.e. a MRI manufacturer software and/ or a remotely located plugin 112 of a reconstruction software of the present solution through the network 120.
The picture archiving and communication system (PACS) 108 of the diagnostic center 110 is based on a popular imaging technology used mostly in medical diagnosis to securely store and digitally transmit electronic images and clinical reports. The use of PACS eliminates the need to manually file and store, retrieve and send sensitive information, films and reports. Instead, medical documentation and images can be securely housed in off-site servers and safely accessed essentially from anywhere in the world using PACS software, workstations and mobile devices. The PACS can handle volume of digital medical images as the need grows throughout the medical industry and data analytics of those images becomes more prevalent. In one example of the present solution, anonymised low frequency acquired image scans of a patient are prepared in the DICOM format, and stored in the picture archiving and communication system.
The remote plugin 112 of the diagnostic center 110 allows you to use a container, remote machine, as a full-featured environment. No additional source code needs to be on a local machine (such as 102/ 104 etc.) to get these benefits. Each such extension in the remote mode can run commands and other extensions directly inside a container, in WSL, or on a remote machine so that everything feels like it does when you run locally. Any popular code spaces, remote, virtual machine, docker containers can be used to implement the remote plugin 112. The plugin is used or a locally installed small software pack is referred to as the same under a plugin as the same belongs to the present solution being managed and run through a cloud implemented method and system for improving image reconstruction. The remote plugin software has an anonymization software that converts the low frequency image scan into an anonymised image. The anonymised image is an encrypted image using a block-chain or a similar cryptographic hash technique to detect unauthorized modification or corruption of records. The patient data can be anonymized so identity is protected unless the individual gives permission or authorization. Anonymizing the data includes removing any identifying information about the individual patients in the data set, hence making the re-identification of those individuals very difficult. Anonymization of image data may be a compliance requirement for transfer of the data out of a hospital by software systems. In an example, medical data is anonymized inside the hospital network, and then transferred out to external systems, for example, a cloud network, to provide a more extensible information access point to the user, or to provide some software services that leverages cloud computing. One method for anonymizing medical imaging data includes removing all patient identifying data and assigning the medical imaging data an ID code. The ID codes may include a string of random or assorted alphanumeric characters or uses any such popular technique. A scan format is used to scan and the scan format is shared with the anonymised image prepared from the low frequency image scan of the patient, most popularly, the digital imaging and communications in medicine (DICOM) standard aids the distribution and viewing of medical images, such as CT scans, MRIs, and ultrasound. A DICOM file has image information pointers, image data. The DICOM image data can be compressed or encapsulated to reduce the image size. Files can be compressed using lossy or lossless variants of the JPEG format, as well as a lossless Run-Length Encoding format as the case may be. DICOM is the most common standard for receiving scans from a hospital.
The copier printer 106 of the diagnostic center 110 produces high-quality colour and mono prints. The copier printer 106 combines copying, printing with embedded DICOM software, printing directly from medical equipment/ apparatus without need of a conversion software or external hardware. It can copy, print from nuclear medicine to scans (e.g. MRI) and ultrasounds, is DICOM-embedded and produces high-quality medical images for clinical assessment and record keeping.
In one implementation, sourcing of an anonymised image scan is from a diagnostic center 110 where it is acquired from an MRI 104. A patient visits the diagnostic center 110 and gets scanned for the organ or part of the organ system as required for a medical diagnosis. An image scan is acquired by a scanner software machine 102 by accelerating a data acquisition from the MRI 104 where the settings are made in such a way that only low frequency samples are selected to get the scan in an accelerated way. The low frequency image scans are generated with optimized sequences with the change in settings available to clinicians or done automatically in the scanner machine software which results in the acquisition of a subset of samples in the K space of image domain or frequency domain of MRI. The decision to accelerate or not is decided by the AI reconstruction software dynamically using a load factor, such as a patient queue information and communicated to clinicians. The load factor is a patient volume load and the setting is changed in a scanner workstation of the MRI 104. The decision to accelerate is conveyed to the MRI 104 in form of the optimised sequence setting, the decision is either received from the server 132a as decided by AI model having a deep learning algorithm or is taken locally by a remote plugin 112. In both cases the optimisation helps in faster acquisition of the image scans at low frequency of the patient.
The anonymized low frequency image scan of the patient and a scan format of the anonymized low frequency image scan is sent through the remote plugin 112 to AI model hosted on a cloud 130 via a network 120. The anonymized low frequency image scan is acquired by the deep learning algorithm using an AI module and the AI module is used while acquiring scan of an organ system and a part of the organ system of the patient. The anonymized low frequency image scan is acquired by under sampling in the K space of image domain of the MRI. The anonymized low frequency image scan is acquired by an anonymization software deployed remotely at a diagnostic center having the MRI. The scan format is a communication standard and the scan format is sent over a network from a diagnostic center.
The network 120 can be a local area network (LAN), a wide area network (WAN), the internet, a server client architecture, a cloud network or another specialised network. The network 120 connects the diagnostic center 110 and the associated apparatuses such as the magnetic resonance imaging (MRI) 104, the scanning software machine 102, the picture archiving and communication system (PACS) 108, the remote plugin 112, and the copier printer 106, and alike to the cloud 130 and its components. The cloud 130 includes one or more computer servers, such as the server 132a, and the server 132b, and alike which can enable distributed computing, such as cloud computing. The cloud 130 in some cases with the aid of the computer system, can implement a peer-to-peer network, which may enable devices coupled to the computer system to behave as a client or a server. The cloud 130 includes one or more databases such as the database 134a, the database 134b, and alike. The database is a storage device including but not limited to a hard disk, a read only memory (ROM), a random access memory (RAM), a distributed storage over a network, and a cloud memory. The cloud 130 is a computing platform may include a storage-based cloud platform based on storage data, a computing-based cloud platform that processes data, and an integrated cloud computing platform that takes into account data storage and processing. The cloud 130 platform used in the system and method for improving the image reconstruction of the MRI may be a public cloud, a private cloud, a community cloud, or a hybrid cloud. For example, according to actual needs, a part of the information received by system and method for improving the image reconstruction of the MRI may be calculated and/or stored by the cloud platform; and another part of the information may be calculated and/or stored by a local processing device and/or a storage device.
In one implementation, reconstruction of a high frequency image scan is performed at the server 132a of the cloud 130 from the received anonymized low frequency image scan pertaining to the patient and the scan format of the anonymized low frequency image scan. The server 132a has an AI model that predicts the high frequency image scan from the anonymized low frequency image scan in image domain using the deep learning algorithm. The high frequency image scan can also have or be a missing frequency image scan. It further reconstructs the high frequency image scan by adding quality parameters to the high frequency image scan using the deep learning algorithm. The quality parameters include parameters for sharpening, denoising, and deringing using an AI module of the deep learning algorithm. It further outputs the reconstructed high frequency image scan of the patient for analysis and prepares a report from the reconstructed high frequency image scan of the patient for diagnosis and quantification. The reconstructed high frequency image scan is analysed for anatomical structures, anatomical anomalies and anatomical features of the patient and can be evaluated for receiving feedback on usage quality from the MR technicians.
The user 140 is connected to the network 120 including but not limited to a lab MRI professional, a machine maintenance or test professional, an MR technician, a clinician, a radiologist, a console operator, a MRI manufacturer trained professional, a doctor, a patient, a patent attendant, a paramedic professional, a hospital management information system (HMIS) interface and associated professionals/ users, an application developer, a quality analyst may use a user access device to access a system 230 on the cloud 130 hosted via at least a server and a database and other required computing resources.
A method for improving image reconstruction of a magnetic resonance imaging (MRI) is employed on a server of the cloud 130 such as on the server 132a and employs a database, such as the database 134a. The method has an AI engine employing a deep learning algorithm. An optimized acquisition process in MRI with inputs from patient load/ volume data. Based on the input either manually or automatically the settings are changed to acquire low frequency K space data which accelerates the MRI acquisition process. The K space is converted to DICOM Image data and stored by the MRI machine. The DICOM image data is passed through a deep learning reconstruction algorithm. The deep learning algorithm provides an optimal estimate of high frequency coefficients. The estimate prediction is done in the image domain.
If, LF(i) is the K space coefficient with optimal acquisition process.
Im1(i) is the image reconstructed from the LF(i).
Im2(i) is the image reconstructed with normal acquisition process.
Im1_e(i) is the reconstructed output from the deep learning algorithm.
The deep learning algorithm uses deep neural networks and minimizes reconstruction error between Im1_e(i) and Im2(i). The reconstruction error can be MSE (Im1_e(i) - Im2(i))^2. But any other error metric can also be used. Any data normalisation or pre-processing is done by the system 230 at the server 232a.
The deep learning algorithm applies neural networks to image scans. A convolutional neural network (CNN) is a particular kind of artificial neural network that preserves spatial relationships in the data, with very few connections between the layers. An input to a CNN is arranged in a grid structure and then fed through layers that preserve these relationships, each layer operating on a small region of the previous layer. The CNNs are able to form a highly efficient representation of the input data and are best designed to suit image-oriented tasks. The CNNs typically have fully connected layers at the end, which compute the final outputs. Such as it has convolutional layers, the activations from the previous layers are convolved with a set of small parameterized filters, frequently of size 3 × 3, collected in a tensor W(j,i), where j is the filter number and i is the layer number. Activation layer, the feature maps from a convolutional layer are fed through nonlinear activation functions. This makes it possible for the entire neural network to approximate almost any nonlinear function. Pooling, each feature map produced by feeding the data through one or more convolutional layer is then typically pooled in a pooling layer. Dropout regularization, an averaging technique based on stochastic sampling of neural networks. By randomly removing neurons during training one ends up using slightly different networks for each batch of training data, and the weights of the trained network are tuned based on optimization of multiple variations of the network. Batch normalization, these layers are typically placed after activation layers, producing normalized activation maps by subtracting the mean and dividing by the standard deviation for each training batch.
In one example of the present solution, a reconstruction software on the server can take scans directly from a scanning software machine or PACS machine with a scanning instruction and the reconstruction software both running in the cloud.
In one example of the present solution, a reconstruction software on the server can take scans from a scanning software machine or PACS machine remotely at the diagnostic center, where a scanning is remote and the reconstruction software is running in the cloud.
This results in significant improvement in accelerating the low frequency image scan of the MRI by using a recommended optimised sequence setting and reconstructing a high quality image scan and adding quality parameters using an AI model employing a deep learning algorithm, which in turn reduces overall cost and ensure high quality image availability for radiologist analysis and evaluation.
FIGURE.2 illustrates a schematic module diagram 200 depicting a method of improving image reconstruction of a magnetic resonance imaging (MRI), in accordance with an embodiment of the present subject matter.
A system 230 for improving image reconstruction of a magnetic resonance imaging (MRI) implements a method of improving image reconstruction of a magnetic resonance imaging (MRI) on a server 232a, the system 230 includes a processor(s) 202, an interface(s) 204, an AI Engine 222, and a memory 206 coupled to the processor(s) 202. The processor(s) 202 may be implemented as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, state machines, logic circuitries, distributing processing, cloud computing, and/or any devices that manipulate signals based on the system 230 instructions. Among other capabilities, the processor(s) 202 is configured to fetch and execute computer-readable instructions stored in the memory 206.
Although the present disclosure is explained by considering a scenario that the system is implemented as an application on a server, the systems and methods can be implemented in a variety of computing systems. The computing systems that can implement the described method(s) include, but are not restricted to, cloud network, mainframe computers, workstations, personal computers, desktop computers, minicomputers, servers, multiprocessor systems, laptops, tablets, SCADA systems, smartphones, mobile computing devices and the like.
The interface(s) 204 may include a variety of software and hardware interfaces, for example, a web interface, a graphical user interface, etc., allowing the system 230 to interact with a user. Further, the interface(s) 204 may enable the system 230 to communicate with other computing devices, such as web servers and external data servers (not shown in figure). The interface(s) 204 can facilitate multiple communications within a wide variety of networks and protocol types, including wired networks, for example LAN, cable, etc., and wireless networks such as WLAN, cellular, or satellite. The interface(s) 204 may include one or more ports for connecting a number of devices to each other or to another server over a network 120, to other network, within a cloud 130 or to other clouds.
In one implementation, the interface(s) 204 software also provides a web based viewer with simple to use intuitive graphical user interface (GUI). The features including but not limited to are administrator login, input, output, matrix views, 3D view – axial, coronal, sagittal, operations - pan, zoom, mark, measure, invert colour, scale colour, search, etc. among others. In one implementation, the interface 204 is accessed by the user 140.
A network such as the network 120 used for communicating between all elements in an application server and cloud environment may be a wireless network, a wired network or a combination thereof. The network can be implemented as one of the different types of networks, such as intranet, local area network LAN, wide area network WAN, the internet, and the like. The network may either be a dedicated network or a shared network. The shared network represents an association of the different types of networks that use a variety of protocols, for example, Hypertext Transfer Protocol HTTP, Transmission Control Protocol/Internet Protocol TCP/IP, Wireless Application Protocol WAP, and the like, to communicate with one another. Further the network may include a variety of network devices, including routers, bridges, servers, computing devices. The network 120 further has access to storage devices residing at a client site computer, a host site server or computer, over the cloud 130, or a combination thereof and the like. The storage has one or many local and remote computer storage media, including one or many memory storage devices, databases, and the like.
The memory 206 can include any computer-readable medium known in the art including, for example, volatile memory (e.g., RAM), and/or non-volatile memory (e.g., EPROM, flash memory, etc.). In one embodiment, the memory 206 includes module(s) 208 and databases 224 (such as 224a, 224b, 224c, 224d, …, 224n). All such databases (224a, 224b, 224c, 224d, …, 224n) may be saved at a common location or spread over the network 120 or the cloud environment 130.
The modules 208 further includes an acquisition module 210, an AI engine module 212, a reconstruction module 214, a report module 216, an analysis module 218, and other modules and sub-modules 220 including but not limited to an evaluation module and the like. It will be appreciated that such modules may be represented as a single module or a combination of different modules. Additionally, the memory 206 further includes databases 224 that serves, amongst other things, as a repository for storing data fetched, processed, received and generated by one or more of the modules 208. The databases 224 includes, for example, operational data, workflow data, and other data at the databases 224. The databases 224 has the databases represented by 224a, 224b, …, 224n, as the case may be. In one embodiment, the databases 224 has access to the other databases over a web or cloud network. The databases 224 includes multiple databases including but not limited to an acquisition module data, an AI engine module data, a reconstruction module data, a report module data, an analysis module data, patient data, diagnostic center data, and other modules and sub-modules data, and other components including but not limited to libraries, link identifiers, a database adapter, a database parser, a technical dictionary, a pattern recognizer, and the like. It will be appreciated that such databases may be represented as a single database or a combination of different databases. In one embodiment data may be stored in the memory 206 in the form of data structures. Additionally, the aforementioned data can be organized using data models, such as relational or hierarchical data models.
In one of embodiments, the server 232a is a remote or local data centre and is further connected to the diagnostic center 110 by a network 120 or is locally hosted.
In one of other embodiments, the server 232a is further connected to the cloud environment 130 or can be equated to a server 132a in the cloud environment 130.
It’s imperative that the server 232a has systems and resources. The computing systems and databases communicate with each other under cloud computing rules and also with the server 232a under the web or other available communication mediums/ protocols. The computing systems generally are distributed processing systems including multiple computing devices connected by and communicating over the network 120. A software application may be executed in the cloud by configuring them to execute across one or more of the computing devices in a particular cloud computing environment 130. The computing devices of the cloud computing system may each execute separate copies of the software application, or, in some cases, the operations of the software application may be split among different computing devices and executed in parallel. The cloud computing system may include a plurality of cloud computing instances representing resources available for executing applications. Each instance may be a physical computing device having particular capabilities (storage size, processing speed, network bandwidth, etc.), or may be a virtual computing device having particular capabilities. A particular cloud computing system may offer different instance types having different sets of capabilities for executing software applications.
In one implementation, at first, a user 140 including but not limited to an MR technician, a clinician, a radiologist, and others listed in description above may use a user access device to access the system 230 via the interface 204 using the server 232a for various different purposes. The working of the system 230 and related method and associated modules, sub modules, methods may be explained in detail also using FIG.1, FIG.3, FIG.4, and FIG.4a, explained in this description.
In one embodiment, the system 230 receives a user instruction data by the interface 204 on the server 232a. The modules 208 of the system 230 processes the instructions using the processor 202 while using the databases 224, the AI engine 222 and supporting components. An acquisition module 210 for improving image reconstruction of the magnetic resonance imaging (MRI) sends an optimized sequence setting for accelerating an image scan acquisition from the MRI 104 and the acquisition module 210 receives an anonymized image scan pertaining to a patient and also a scan format of the anonymized image scan. The acquisition module 210 communicates with a remote plugin 112 so the anonymized image scan is acquired by under sampling in K space of the MRI 104 using the optimized sequence setting. The acquisition module 210 communicates with the AI engine 222 to recommend the optimized sequence setting based on a load factor. The load factor is a patient volume load and the setting is changed in a scanner workstation 102 of the MRI 104. The acquisition module 210 acquires the anonymized image scan by the deep learning algorithm using an AI engine module 212 and the AI engine module 212 is also used while acquiring scan of an organ systems and a part of the organ systems of the patient. In one example, the anonymized image scan is acquired by an anonymization software deployed remotely, in communication with the acquisition module 210, at a diagnostic center 110 having the MRI 104. The acquisition module 210 receives the anonymized image scan from a diagnostic center 110 having a group of devices including an MRI 104 machine, a scanning software machine 102, and a picture archiving and communication system (PACS) 108 through a remote plugin 112. The scan format used is a communication standard and the scan format is sent over a network 120 from a diagnostic center 110.
The AI engine module 212 for improving image reconstruction of the magnetic resonance imaging (MRI) predicts a high frequency image scan from the anonymized image scan in the image domain. The AI engine module 212 uses a deep learning algorithm for predicting the high frequency image scan using an AI engine 222 having the deep learning algorithm. In one implementation, the high frequency image scan can have or is a missing frequency image scan. In one embodiment, an RPA bot is configured to generate high frequency image scan using a deep learning and neural network based predicting algorithm through an AI engine 222. In one embodiment, the RPA bot is a software bot or a combination of a software and hardware bot. In an embodiment, the software bot is a computer program enabling a processor to perform robotic process automation by utilizing AI. In another embodiment, the bot as a combination of hardware and software, where the hardware includes memory, processor, controller and other associated chipsets especially dedicated to perform functions that enable robotic process automation for improving image reconstruction of a magnetic resonance imaging (MRI).
The reconstruction module 214 for improving image reconstruction of the magnetic resonance imaging (MRI) reconstructs the high frequency image scan by adding quality parameters to the high frequency image scan using the AI engine 222 having the deep learning algorithm. The reconstruction module 214 adds quality parameters that includes sharpening, denoising, and deringing and similar procedures using an AI engine 222 having the deep learning algorithm. The other quality parameters include histogram normalization, contract adjustment, segmenting, enhancing regions or sub-regions, error metric resolutions or improvements, quality metrics like structural similarity (SSIM), peak signal to noise ratio (PSNR), normalize mean square error (NMSE), and alike. In one example, at least one quality parameter is added from multiple such parameters to sharpen, denoise, dering and other such functions to improve image scan quality.
The report module 216 for improving image reconstruction of the magnetic resonance imaging (MRI) prepares a report using the reconstructed high frequency image scan of the patient for diagnosis and quantification. The report module 216 outputs the reconstructed high frequency image scan of the patient for analysis. The analysis module 218 for improving image reconstruction of the magnetic resonance imaging (MRI) analyses the reconstructed high frequency image scan for anatomical structures, anatomical anomalies and anatomical features of the patient and/or provide the same to the user 140. The evaluation module 220 of other modules and sub-modules 220 for improving image reconstruction of the magnetic resonance imaging (MRI) evaluates the reconstructed high frequency image scan of the patient and receives feedback on usage quality from lab technicians or radiologists alike.
In one implementation, the system 230 has a plurality of modules 208 that are hosted on a data center that is in communication with the remote plugin 112.
In yet another implementation, the system 230 has a plurality of modules 208 that are hosted on a cloud 130 that is in communication with the remote plugin 112 through a network 120.
In all examples, embodiments, implementations, an anonymised image scan can be an as-is acquired image scan i.e. a low frequency/ fast scan acquired as per the optimised sequence setting enabled by AI algorithms,
In one example, a reference architecture of an AKS software having a system 230 is described. It has a configuration where an AKS plugin is installed at a customer site/ diagnostic center and an AKS server software on a cloud. The local workstation OS Ubuntu 18.04/20.04 for anonymization/ de-anonymization runs on Docker and non-root user. An I/O configuration includes an AKS software that can be connected to different PACS to take input in DICOM format. Input - fast acquired DICOM image sequence (.dcm sequence) from MRI. The image acquisition can be 2 times to 4 times faster. Settings for acceleration will be provided separately using an optimised sequence setting recommendation. Scanners used for MRI: 1.5T/ 3T from different manufacturers like GE/ Philips/ Siemens. Output - AKS output which can be used for diagnosis. PACS support - multiple PACS/ MRI scanner support which provide access to DICOM data. Anonymization support provided to protect patient data. AKS Server - Server deployment is Kubernetes based on cloud for scalability. The Kubernetes Cluster includes pipeline architecture, elastically scalable, secured data storage, elastically scalable, remote diagnostics, easy integration, cyber-security, enterprise domain, SaaS Model. The hosting includes hybrid: local dc/cloud secure: HIPAA compliant and anonymised; pluggable: remote integration; GPU/CPU: choice of hardware; insightful: deep data analysis; interoperable: connected to multiple scanners and PACS; separate hardware is required for on premise system; local data center hosting for large installation only.
FIGURE.3 illustrates an exemplary flowchart 300 of a method of improving image reconstruction of a magnetic resonance imaging (MRI), in accordance with an embodiment of the present subject matter.
A method 300 for improving image reconstruction of a magnetic resonance imaging (MRI) is shown. The method may be described in the general context of computer executable instructions. Generally, computer executable instructions can include routines, programs, objects, components, data structures, procedures, modules, functions, etc., that perform particular functions or implement particular abstract data types. The method may also be practiced in a distributed computing environment where functions are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, computer executable instructions may be located in both local and remote computer storage media, including memory storage devices. The order in which the method is described and is not intended to be construed as a limitation, and any number of the described method blocks can be combined in any order to implement the method or alternate methods. Additionally, individual blocks may be deleted from the method without departing from the spirit and scope of the disclosure described herein. Furthermore, the method may be implemented in any suitable hardware, software, firmware, or combination thereof. However, for ease of explanation, in the embodiments described below, the method may be implemented in any of the described system(s).
At step/ block 302, receive a user instruction data through an electronic user interface (204).
At step/ block 304, send an optimized sequence setting for accelerating an image scan acquisition. In one implementation, the optimized sequence setting may be send by an acquisition module 210 using a processor 202.
At step/ block 306, receive an anonymized image scan pertaining to a patient and a scan format of the anonymized image scan, the anonymized image scan is acquired by under sampling in K space of the MRI 104 using the optimized sequence setting recommended based on a load factor. In one implementation, the anonymized image scan may be received by an acquisition module 210 using the processor 202.
At step/ block 308, predict a high frequency image scan from the anonymized image scan in image domain using a deep learning algorithm. In one implementation, the high frequency image scan may be predicted by an AI engine module 212 using the processor 202.
At step/ block 310, reconstruct the high frequency image scan by adding at least one quality parameter to the high frequency image scan using the deep learning algorithm. In one implementation, the high frequency image scan may be reconstructed by a reconstruction module 214 using the processor 202.
At step/ block 312, output the reconstructed high frequency image scan of the patient for analysis. In one implementation, the reconstructed high frequency image scan may be outputted by a report module 216 using the processor 202.
Thus, the method 300 helps in improving image reconstruction of a magnetic resonance imaging (MRI) by recommending an optimized sequence setting, predicting a high frequency image scan from the anonymized image scan in image domain using a deep learning algorithm and reconstructing the high frequency image scan by adding at least one quality parameter to the high frequency image scan using the deep learning algorithm.
FIGURE.4 and illustrates an exemplary report interface 400 and FIGURE. 4a illustrates another part of an exemplary report 402 automatically generated by an exemplary image reconstruction of a magnetic resonance imaging (MRI) system, in accordance with an embodiment of the present subject matter.
A report module 216 generates a report 402 of a patient assessment to improve image reconstruction of a magnetic resonance imaging (MRI) by recommending an optimized sequence setting, predicting a high frequency image scan from an anonymized image scan in image domain using a deep learning algorithm and reconstructing the high frequency image scan by adding quality parameters to the high frequency image scan using the deep learning algorithm. The report 402 provides automated insights into the MRI by using a strong magnetic field and radio waves to create detailed images of the parts of organ systems, organs and tissues of a patient.
The patient can be examined using the non-invasive tool for various anomalies of the brain and spinal cord, tumours, cysts, and other anomalies in various parts of the body, breast cancer screening for women who face a high risk of it, injuries or abnormalities of the joints, such as the back and knee, certain types of heart problems, diseases of the liver and other abdominal organs, and alike. The list is only indicative and not exhaustive. The reports for all such scans will be different. However, the present subject matter has the ability to work in all such situations and generate the report 402 covering various aspects of the patient assessment. Generally, the report 402 will have sections like 404, 406, 408 covering the status or reading of parts of to be scanned, graphical findings in 410 and an image of original scans 412 (by MRI under optimised sequence setting) and reconstructed scans (by using an AI engine 222 employing a deep learning algorithm). The section of reports changes with the type of scan and requirement of the radiologist settings or manufacturer but mostly the sections remains spread over the similar plan.
In one implementation, the report 402 has a clinical presentation 404 covering the reasons or history of the patient reported or observed by a doctor or specialist having points listed in 404a, 404b, 404c, …, 404n as the case may be. Further, the report 402 has findings section 406 reported after the scan is analysed by a radiologist, a rules engine, an AI engine or as the case may be having pointers listed in 406a, 406b, 406c, …, 406n. The report 402 will have an impression section 408 having the pointers of the recommendations or observation results by a radiologist or alike over pointers covered in 408a, 408b, 408c, …, 408n. A section will be covering the graphical parts of the scan to report findings over a period of time in charts such as 410a, 410b and may also cover the past scans using a medical history details of the patient or any part of the scan where a chart may be required or helpful. In one example, effects of repetition time (TR) and echo time (TE) on a medical resonance (MR) signal is shown via 410a. An original scan may be reconstructed using the present method and system may be shown in 412a, 412b for various organs or organ systems or parts of organ systems as the case may be. The FIG. 4a shows the original acquired scans 412a and 412b and the reconstructed scans 412aX, 412bX for respective image scans of a patient. The 412aX, 412bX and alike are reconstructed scans by the present system and method to improve image reconstruction of the magnetic resonance imaging (MRI) by predicting a high frequency image scan from the anonymized image scan in image domain using a deep learning algorithm and reconstructing the high frequency image scan by adding quality parameters to the high frequency image scan using the deep learning algorithm. In one example, 412 will represent image scans 412a as original scan acquired with low frequency and 412aX as reconstructed high frequency image scan, such as left to right - original, reconstructed 3x, reconstructed 4x image scans respectively of a patient with stroke, a patient with epilepsy, a normal patient yet to be diagnosed with a condition, a patient with tumour etc.
The report 402 may have other parts such as patient details (age, gender, reference number etc.), scan details (zoom level, resolution, dimensions, weight, LOC, series etc.), date and time, place, MRI machine manufacturer settings, sequences, gradients, planes, artefacts, imaging techniques, and B values etc. Similarly, the sections may be added or deleted based on the report style, patient condition, type of scan, diagnostic center, regulatory requirements, etc.
In one example, qualitative and quantitative comparisons of the images generated by a deep learning algorithm of the present subject matter from simulated under sampled MR brain and standard of care images is performed with following details. A retrospective study was conducted on 50 MRI pathological scans acquired on Siemens Magnetom1.5T. Information about the dataset - the model is tested on 50 T2 axial series (t2_tse_tra_512) obtained from different patients across 7 pathologies. It is done for the acceleration factor of 3 and 4.
Table A: Number of cases in each pathology
Pathology Number of Cases
Alzheimer’s 1
Bleed, Hemorrhage, Hematoma 9
Epilepsy 9
Normal Case 7
Other Cases 11
Stroke 9
Tumor 4

Two types of under sampling were done, under sampling by 3 and under sampling by 4 in the K space which simulated fast MR acquisition. These images were then converted to resemble under sampled images. Under sampling by 3 and 4 will be equivalent to acquiring an MRI signal by 1/3rd and 1/4th of the time respectively. The up-resolution is done on this simulated under sampled data by a customized CNN network based on encoder-decoder CNN with 14.8 million parameters by an AKS server (such as 232a). The three sets of images were randomized, and presented to two neuro radiologists for evaluation of their overall quality on a five-point Likert scale (1- non diagnostic, 2 -poor quality, 3- diagnostic, 4-good, 5-excellent). The parameters for evaluation were grey-white matter differentiation (for CNR), delineation of pathologies, delineation of normal deep grey matter anatomy, and artefacts. The overall image quality score was then estimated for each set of images and compared. Structural similarity (SSIM) and peak signal to noise ratio (PSNR) are used as the metrics to evaluate the performance. The Peak SNR and SSIM values were also calculated for each set of images with respect to the standard of care (SOC) images. The method used is average SSIM for both the 1/3rd under sampled and 1/4th under sampled images was 0.95 & 0.97. The PSNR for the 1/3rd under sampled and 1/4th under sampled images were 36.03 dB and 33.90 dB. The average qualitative scores for the SOC, 1/3rd under sampled, and 1/4th under sampled images were 11.16, 11.11 & 10.98 out of 20.
Table B: Average subjective scoring based on the radiologists scores.
Same Original Higher 3X Higher/ Reconstructed 4X Higher/ Reconstructed
artefact - artefact 20 9 12 9
dilineation_normal - delineation of normal deep grey matter anatomy 10 11 13 16
dilineation_pathology - delineation of pathologies 17 7 16 10
gw_diff - Grey-white matter differentiation (for CNR) 16 8 13 13

It was evaluated and reported that the super-resolved under sampled images (by AI Engine 222 employing the deep learning algorithm) are comparable to the routine SOC images both in quantitative and qualitative comparisons.
Although implementations of system and method for improving image reconstruction of a magnetic resonance imaging (MRI) have been described in language specific to structural features and/or methods, it is to be understood that the appended claims are not necessarily limited to the specific features or methods described. Rather, the specific features and methods are disclosed as examples of implementations for improving image reconstruction of the magnetic resonance imaging (MRI).