SYSTEM AND METHOD FOR DETECTING BRAIN ABNORMALITY FROM MAGNETIC RESONANCE (MR) IMAGES

Abstract

The present disclosure pertains to a system (100) to detect brain abnormality from magnetic resonance (MR) images. The system (100) includes an input unit (102) configured to acquire MR images of the brain. The system (100) is configured to receive the MR images from the input unit (102), pre-process the received MR images, identify regions of interest (ROIs) corresponding to brain abnormalities, apply a classification model to the identified ROIs, use deep learning techniques to analyze and classify the brain abnormalities, evaluate performance of the classification model taking into consideration one or more metrics and generate output of the classification and display to the input unit (102). The system (100) enables early detection and classification of brain abnormalities by an efficient MR image analysis.

Specification

Description:TECHNICAL FIELD
[0001] The present invention relates to the field of image processing. In particular, it relates to a system and method for detecting brain abnormality from magnetic resonance (MR) images.

BACKGROUND
[0002] Background description includes information that may be useful in understanding the present disclosure. It is not an admission that any of the information provided herein is prior art or relevant to the presently claimed disclosure, or that any publication specifically or implicitly referenced is prior art.
[0003] Digital imaging and magnetic resonance (MR) imaging have been utilized for identifying brain disorders. The presence of noise and artefacts in the MR images complicates the identification process. Manual analysis of the MR images is laborious and prone to variability. Traditional methods of deep learning used for identification of brain abnormalities lack efficiency during computation and do not have precision of localization. Existing methods of segmentation and classification lack robustness and accuracy. These limitations have highlighted the need for a robust, accurate and efficient brain MR image analysis.
[0004] There is, therefore a need to provide an efficient and improved solution for detecting brain abnormality from magnetic resonance (MR) images.

OBJECTS OF THE INVENTION
[0005] Some of the objects of the present disclosure, which at least one embodiment herein satisfies are as listed herein below.
[0006] A general object of the present disclosure is to provide a system and a method for detecting brain abnormality from magnetic resonance (MR) images.
[0007] Another object of the present disclosure is to provide accuracy, enabling refined extraction of features and patterns from MR images.
[0008] Another object of the present disclosure is to provide robustness, ensuring accurate segmentation of the images in various sizes and complexities.
[0009] Another object of the present disclosure is to provide efficiency, ensuring an accelerated segmentation process for speedy diagnosis of abnormalities.
[0010] Yet another object of the present disclosure is to provide reliability by a single stage detector model, enhancing analysis of brain MR images.

SUMMARY
[0011] Aspects of the present disclosure relate to a system to detect brain abnormality from magnetic resonance (MR) images. An object detection model integrated with deep learning models is utilized for precise segmentation and classification of MR images for detecting brain abnormalities. This improves accuracy, robustness and generalization of datasets, enabling an adaptable and practical analysis of brain MR images.
[0012] In an aspect, the system includes an input unit configured to acquire magnetic resonance (MR) images of an entity, and a processing unit to analyse the received MR images. The processing unit may be configured to receive the MR images from the input unit, pre-process the received MR images, and identify regions of interest (ROIs) corresponding to brain abnormalities from the pre-processed MR images, In addition, the processing unit applies a classification model to the identified ROIs, where deep learning techniques are utilized to analyze and classify the brain abnormalities, evaluate performance of the classification model taking into consideration one or more metrics and generate output of the classification and display to the input unit, where the output includes the segmented and classified MR images displaying the brain abnormalities.
[0013] In an aspect, the processing unit pre-process the MR images is configured to remove non-brain tissues from the MR images, apply an anisotropic diffusion filter to reduce noise in the MR images, utilize Contrast Limited Adaptive Histogram Equalization (CLAHE) to adjust contrast of the MR images and apply Generative Adversarial Networks (GANs) for data segmentation in the MR images.
[0014] In an aspect, the identification of regions of interest (ROIs) is performed using You Look Only Once version 9 (YOLOv9).
[0015] In an aspect, the classification model is trained using a training dataset of the preprocessed MR images, and the performance of the classification model is validated using a testing dataset. In addition, the classification model includes a hybrid classification ensemble that includes one or more convolutional neural networks (CNNs).
[0016] In an aspect, transfer learning is applied to fine-tune a pre-trained model for segmentation and classification of the identified ROIs.
[0017] Another aspect of the present disclosure pertains to a method for detecting brain abnormality from magnetic resonance (MR) images. The method may include the steps of receiving the MR images from an input unit, pre-processing the received MR images, identifying regions of interest (ROIs) corresponding to brain abnormalities from the pre-processed MR images, applying a classification model to the identified ROIs, where deep learning techniques are utilized to analyze and classify the brain abnormalities, evaluating performance of the classification model taking into consideration one or more metrics and generating output of the classification and display to the input unit, where the output includes the segmented and classified MR images displaying the brain abnormalities.
[0018] In an aspect, the pre-processing of MR images includes the steps of removing non-brain tissues from the MR images, applying an anisotropic diffusion filter to reduce noise in the MR images, utilizing Contrast Limited Adaptive Histogram Equalization (CLAHE) to adjust contrast of the MR images and applying Generative Adversarial Networks (GANs) for data segmentation in the MR images.
[0019] Various objects, features, aspects, and advantages of the inventive subject matter will become more apparent from the following detailed description of preferred embodiments, along with the accompanying drawing figures in which like numerals represent like components.

BRIEF DESCRIPTION OF DRAWINGS
[0020] The accompanying drawings are included to provide a further understanding of the present disclosure, and are incorporated in and constitute a part of this specification. The drawings illustrate exemplary embodiments of the present disclosure and, together with the description, serve to explain the principles of the present disclosure.
[0021] FIG. 1 illustrates an exemplary network architecture of a proposed system to detect brain abnormality from magnetic resonance (MR) images, in accordance with an embodiment of the present disclosure.
[0022] FIG. 2 illustrates an exemplary block diagram of the proposed system, in accordance with an embodiment of the present disclosure.
[0023] FIG. 3 illustrates an exemplary flow chart for pre-processing MR images, in accordance with an embodiment of the present disclosure.
[0024] FIG. 4 illustrates an exemplary architecture of YOLOv9, in accordance with an embodiment of the present disclosure.
[0025] FIG. 5 illustrates an exemplary architecture of Programmable Gradient Information, in accordance with an embodiment of the present disclosure.
[0026] FIG. 6 illustrates an exemplary architecture of a generalized efficient layer aggregation network, in accordance with an embodiment of the present disclosure.
[0027] FIG. 7 illustrates an exemplary flow chart for detecting brain abnormalities from MR images, in accordance with an embodiment of the present disclosure.
[0028] FIG. 8A illustrates exemplary views of pre-processing steps with MR images, in accordance with an embodiment of the present disclosure.
[0029] FIG. 8B illustrates exemplary views of output images, in accordance with an embodiment of the present disclosure.
[0030] FIG. 9 illustrates an exemplary flow diagram representing a method for detecting brain abnormality from magnetic resonance (MR) images, in accordance with an embodiment of the present disclosure.
DETAILED DESCRIPTION
[0031] The following is a detailed description of embodiments of the disclosure depicted in the accompanying drawings. The embodiments are in such detail as to clearly communicate the disclosure. However, the amount of detail offered is not intended to limit the anticipated variations of embodiments; on the contrary, the intention is to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the present disclosure as defined by the appended claims.
[0032] Embodiments of the present disclosure pertain to a system to detect brain abnormality from magnetic resonance (MR) images. An object detection model integrated with deep learning models is utilized for precise segmentation and classification of MR images for detecting brain abnormalities. This improves accuracy, robustness and generalization of datasets, enabling an adaptable and practical analysis of brain MR images.
[0033] In an embodiment, the system includes an input unit configured to acquire magnetic resonance (MR) images of an entity, and a processing unit for analyzing the MR images. The processing unit may be configured to receive the MR images from the input unit, pre-process the received MR images, identify regions of interest (ROIs) corresponding to brain abnormalities from the pre-processed MR images, apply a classification model to the identified ROIs, where deep learning techniques are utilized to analyze and classify the brain abnormalities, evaluate performance of the classification model taking into consideration one or more metrics and generate output of the classification and display to the input unit, where the output includes the segmented and classified MR images displaying the brain abnormalities.
[0034] In an embodiment, the processing unit pre-process the MR images is configured to remove non-brain tissues from the MR images, apply an anisotropic diffusion filter to reduce noise in the MR images, utilize Contrast Limited Adaptive Histogram Equalization (CLAHE) to adjust contrast of the MR images and apply Generative Adversarial Networks (GANs) for data segmentation in the MR images.
[0035] In an embodiment, the identification of regions of interest (ROIs) is performed using You Look Only Once version 9 (YOLOv9).
[0036] In an embodiment, the classification model is trained using a training dataset of the preprocessed MR images, and the performance of the classification model is validated using a testing dataset. In addition, the classification model includes a hybrid classification ensemble that includes one or more convolutional neural networks (CNNs).
[0037] In an embodiment, transfer learning is applied to fine-tune a pre-trained model for segmentation and classification of the identified ROIs.
[0038] Another embodiment of the present disclosure pertains to a method for detecting brain abnormality from magnetic resonance (MR) images. The method may include the steps of receiving the MR images from an input unit, pre-processing the received MR images, identifying regions of interest (ROIs) corresponding to brain abnormalities from the pre-processed MR images, applying a classification model to the identified ROIs, where deep learning techniques are utilized to analyze and classify the brain abnormalities, evaluating performance of the classification model taking into consideration one or more metrics and generating output of the classification and display to the input unit, where the output includes the segmented and classified MR images displaying the brain abnormalities.
[0039] In an embodiment, the pre-processing the MR images includes the steps of removing non-brain tissues from the MR images, applying an anisotropic diffusion filter to reduce noise in the MR images, utilizing Contrast Limited Adaptive Histogram Equalization (CLAHE) to adjust contrast of the MR images and applying Generative Adversarial Networks (GANs) for data segmentation in the MR images.
[0040] Referring to FIGs. 1 and 2, a system 100 to detect brain abnormality from magnetic resonance (MR) images is disclosed. The system 100 includes an input unit 102 configured to acquire magnetic resonance (MR) images of an entity. The acquired MR images include images of brain, describing internal structure of the brain. The input unit 102 can include a user interface that allows for interaction with the system 100, and it can be any device such as, but not limited to a smartphone, laptop, tablet, or desktop computer. The input unit 102 facilitates the input of MR images, for further processing of this information efficiently. The versatility of the input unit 102 ensures that users can access and utilize the system 100 from a wide range of devices, accommodating various user preferences and technological environments.
[0041] In an embodiment, the system 100 includes a processing unit 104. The processing unit 104 may include one or more processor(s) 202 that may be implemented as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, logic circuitries, and/or any devices that manipulate data based on operational instructions. Among other capabilities, the one or more processor(s) 202 may be configured to fetch and execute computer-readable instructions stored in a memory 204 of the processing unit 104. The memory 204 may store one or more computer-readable instructions or routines, which may be fetched and executed to create or share the data units over a network service. The memory 204 may include any non-transitory storage device including, for example, volatile memory such as Random Access Memory (RAM), or non-volatile memory such as an Erasable Programmable Read-Only Memory (EPROM), flash memory, and the like.
[0042] In an embodiment, the system 100 is communicatively coupled to the learning engine 106 and a server 110 through a communication unit 108. The communication unit 108 may be wired communication means, or wireless communication means, or a combination thereof. In some embodiments, the wired communication means may include, but not limited to, wires, cables, data buses, optical fibre cables, and the like. In some embodiments, the wireless communication means may include, but not be limited to, telecommunication ks, Near Field Communication (NFC), Bluetooth, Internet, Local Area Networks (LAN), Wide Area Networks (WAN), Light Fidelity (Li-FI) networks, a carrier network, and the like. In some embodiments, the form factor of the data transmitted through the communication means may be any one or combination of including, but not limited to, analogue signals, electrical signals, digital signals, radio signals, infrared signals, data packets, and the like.
[0043] Further, the server 110 serves as a central hub for processing, storing, and managing the data collected by the system 100. Through the communication unit 108, data in real-time are transmitted to the server 110. Once the data arrives, the server 110 undertakes various essential tasks. The server 110 undertakes activities such as pre-processing, identification, evaluation, and result generation to improve the quality and usability of the data. The server 110 applies advanced learning models to analyze the data further. These models help in identifying and classifying brain abnormalities with high accuracy by leveraging the computational power of the server 110. The use of the server 110 enables large-scale data storage and future analysis. It ensures that the system 100 can handle vast amounts of data while providing a robust platform for continuous learning and improvement of the algorithms over time. This centralization of processing and data management enhances the scalability of the system 100 and allows for future refinements based on newly acquired data or updated learning models.
[0044] In addition, the processing unit 104 may also include an interface(s) 206. The interface(s) 206 may include a variety of interfaces, for example, interfaces for data input and output devices, referred to as I/O devices, storage devices, and the like. The interface(s) 206 may provide a communication pathway for one or more components of the processing unit 104. Examples of such components include but are not limited to, processing engine(s) 208 and a database 224.
[0045] In an embodiment, the processing engine 208 is implemented as a combination of hardware and programming to implement one or more functionalities of the processing engine 208. Such combinations of hardware and programming may be implemented in several different ways. For example, the programming for the processing engine 208 is processor-executable instructions stored on a non-transitory machine-readable storage medium, and the hardware for the processing engine 208 includes a processing resource (for example, one or more processors 202), to execute such instructions.
[0046] In an embodiment, the processing engine 208 can include a receiving module 210, a pre-processing module 212, an identification module 214, a classification module 216, an output generation module 218, a training module 220 and other module(s) 222. The other module(s) 222 implements functionalities that supplement applications or functions performed by the system 100 or the processing engine 208. The database 224 serves, among other things, as a repository for storing data processed, received, and generated by one or more of the modules.
[0047] In an embodiment, the receiving module 210 is configured to receive the MR images from the input unit 102. The MR images of the brain are acquired by the input unit 102 for detecting abnormalities. In an embodiment, the pre-processing module 212 is configured to pre-process the received MR images. This pre-processing prepares the MR images to ensure they are suitable for the subsequent stages of analysis.
[0048] Further, for pre-processing of MR images, the processing unit 104 is configured to remove non-brain tissues from the MR images, apply an anisotropic diffusion filter to reduce noise in the MR images, utilize Contrast Limited Adaptive Histogram Equalization (CLAHE) to adjust contrast of the MR images and apply Generative Adversarial Networks (GANs) for data segmentation in the MR images. To focus on the brain, non-brain tissues such as scalp, skull and the like are removed from the MR images and is known as skull stripping. The anisotropic diffusion filter reduces noise while preserving edge details. CLAHE enhances the contrast of the MR images to improve the visibility of brain structures. GANs are applied for data augmentation, to separate different regions of the brain within the MR images.
[0049] In an exemplary implementation, for pre-processing the received MR images, following steps are follows as shown in a flow chart 300 for FIG. 3. At step 302, skull stripping is caried out for removing non-brain tissues. At step 304, denoising of the MR images is done using the anisotropic diffusion filter while preserving edge details. At step 306, contrast enhancement of the denoised MR images is done using CLAHE and at step 308, data augmentation for further refining is carried out using GANs.
[0050] In an embodiment, the identification module 214 is configured to identify regions of interest (ROIs) corresponding to brain abnormalities from the pre-processed MR images. ROIs represent areas in the brain where abnormalities, such as tumors, lesions, or the like, are present. Further, the identification of regions of interest (ROIs) is performed using You Look Only Once version 9 (YOLOv9). YOLO is an object detection algorithm that identifies objects in images or video frames. The processing unit 104 applies the detection method to the MR images to identify specific areas or ROIs in the brain that indicate abnormalities. It helps to identify and localize regions of interest in abnormalities from MR images.
[0051] In an exemplary embodiment, a YOLO architecture 400 (as shown in FIG. 4) includes a backbone, a neck, and a head. The backbone serves as a feature extractor which is essential for identification for different levels of information, such as edges, textures, and complicated pattern from the MR images. After the feature extraction, neck of the YOLO architecture processes and refines these extracted features, using multi-scale feature integration to ensure the model can accurately segment structures of varying sizes and complexities. The neck architecture which processes feature maps from the backbone, benefits from the groundbreaking techniques such as PGI (Programmable Gradient Information) and GELAN (Generalized Efficient Layer Aggregation Network). The head of the YOLO architecture generates the final output, which in the case of MR image is a pixel-wise segmentation mask that shows the edges of certain anatomical structures or abnormalities. Further, referring to FIGs. 5 and 6, an architecture of Programmable Gradient Information (PGI) 600 and an architecture of Generalized Efficient Layer Aggregation Network 700 is shown as a combination of Cross Stage Partial (CSP) and Efficient Layer Aggregation Network (ELAN).
[0052] In an embodiment, the classification module 216 is configured to apply a classification model to the identified ROIs and deep learning techniques are utilized to analyze and classify the brain abnormalities. The classification model differentiates between types of brain abnormalities and classifies them into appropriate conditions such as tumors, lesions or the like. The classification aids in identifying the appropriate brain disorder.
[0053] Further, the classification model includes a hybrid classification ensemble that includes one or more convolutional neural networks (CNNs). The CNN can include models such as Residual Network (ResNet), Densely Connected Convolutional Network (DenseNet), MobileNet, and EfficientNetV2 or the like.
[0054] Further, the processing unit 104 is configured to evaluate performance of the classification model taking into consideration one or more metrics. The one or more metrics can include parameters such as accuracy, confusion matrix, log-loss, AUC-ROC or the like to evaluate the classification models based on the datasets.
[0055] Further, the classification model is trained using a training dataset of the preprocessed MR images, and the performance of the classification model is validated using a testing dataset. The training dataset includes a set of pre-processed MR images used to train the classification model. The testing dataset can include another set of pre-processed MR images (not used during training) to validate and test the model's accuracy and reliability, ensuring that it generalizes to the new data.
[0056] Further, transfer learning is applied to fine-tune a pre-trained model for segmentation and classification of the identified ROIs. Transfer learning is employed to fine-tune the pre-trained model. Transfer learning takes an existing model that has already learned features from a large dataset and adapts it for a specific task, such as MR image segmentation and classification. This reduces training time and improves the model's accuracy, when working with smaller, specialized datasets.
[0057] In an embodiment, the training module 220 is configured to train the learning engine 106 by a deep learning technique. The training module 220 is configured to train the deep learning model on a dataset of MR images of the brain to identify ROIs corresponding to brain abnormalities from pre-processed MR images. Further, implement the trained deep learning model to apply the classification model to the identified ROIs to classify the brain abnormalities. Further, evaluating the performance of the classification model.
[0058] In an exemplary embodiment, the deep learning techniques enable the system 100 to automatically learn and improve from experience without being explicitly programmed. These techniques use statistical techniques to identify patterns and relationships within data and make predictions or decisions based on that information. These techniques are used to analyze the data received during the detection of brain abnormalities. These algorithms are specifically trained to recognize patterns and characteristics in the data, such as parameters associated with brain abnormality and the like. Once trained, these techniques or algorithms can process the received data, identify relevant patterns, and make adjustments to the output generation process accordingly.
[0059] In an embodiment, the output generation module 218 is configured to generate output of the classification and display to the input unit 102, and the output includes the segmented and classified MR images displaying the brain abnormalities. The output includes the MR images with the brain abnormalities clearly marked and classified. The output also includes the classified abnormality. The processed and classified images are sent back to the input This ensures that the output images and classification results are displayed in an easily interpretable manner, enabling professionals to make informed decisions based on the analysis.
[0060] Referring to FIG. 7, an exemplary flow chart 700 for detecting brain abnormalities from MR images is disclosed. At step 402, the MR images are acquired by the input unit 102. At step 704, the MR images are preprocessed. At step 706, the dataset is divided into a training dataset (step 708) and a testing dataset (step 710) for segmentation and classification of the MR images. At step 712, the ROIs are identified and the identified ROIs are extracted (step 714). At step 716, transfer learning is applied to fine-tune the pre-trained model for segmentation and classification of the identified ROIs. At step 718, the hybrid classification ensemble including one or more CNNs is applied to the classification model. At step 420, output of the classification is generated corresponding to the brain abnormalities. At step 422, performance of the classification model is validated using the testing dataset.
[0061] In an exemplary embodiment, referring to FIG. 8A, the input MR images are pre-processed by skull stripping, denoising and enhancing contrast of the MR images to improve the accuracy of brain abnormality detection. Further, the output of the classification is shown in FIG. 8B. the output includes various brain segments such as Contrast enhanced Coronal image, Contrast enhanced Sagittal image, Contrast enhanced Meningioma image, Contrast enhanced Glioma image, Segmented Coronal image, Segmented Sagittal image, Segmented Meningioma image and Segmented Glioma image.
[0062] In an exemplary implementation, the system 100 acquires MR images of the brain using an input unit 102. Once the images are acquired, the system 100 pre-processes the received MR images. The system 100 removes non-brain tissues from the MR images to focus only on the brain regions. The anisotropic diffusion filter is used to reduce noise while preserving the important edges in the image, improving the clarity of the brain structures. CLAHE is employed to enhance the contrast in the MR images, making the brain abnormalities more distinguishable. GANs are applied for fine-grained segmentation of the brain, separating different regions for better analysis of potential abnormalities.
[0063] After pre-processing, the system 100 uses the You Only Look Once version 9 (YOLOv9) model to identify specific Regions of Interest (ROIs) in the MR images. Once the ROIs are identified, the system 100 applies a deep learning-based hybrid classification ensemble, which includes one or more convolutional neural networks (CNNs) to analyze the ROIs and classify the type of abnormality based on patterns learned from the training dataset. To improve accuracy and reduce training time, the system 100 uses transfer learning for fine-tuning the pre-trained model to generalize better to new cases of brain abnormalities. The system 100 evaluates the performance of the classification model by calculating various metrics. The system 100 generates the output in the form of segmented and classified MR images. These images highlight the brain abnormalities, along with the classification. The results are displayed back to the input unit 102 for diagnosis.
[0064] Referring to FIG. 9, a method 900 for detecting fraudulent activity during an assessment is disclosed. At block 902, the method 900 includes a step of receiving the MR images from an input unit. This step ensures that the MR images acquired are ready for further processing.
[0065] At block 904, the method 900 includes the step of pre-processing the received MR images. It ensures the images are optimally prepared for subsequent analysis.
[0066] In an embodiment, the method 900 to pre-process the images includes step of removing non-brain tissues from the MR images to focus on relevant brain areas, and applying an anisotropic diffusion filter to reduce noise in the MR images for improving the clarity. Further, utilizing Contrast Limited Adaptive Histogram Equalization (CLAHE) to adjust contrast of the MR images to improve the visibility of different brain structures and applying Generative Adversarial Networks (GANs) for data segmentation in the MR images. This ensures that the MR images are clean and enhanced for accurate analysis.
[0067] At block 906, the method 900 includes the step of identifying regions of interest (ROIs) corresponding to brain abnormalities from the pre-processed MR images. This step ensures that specific areas within the brain or ROIs, are identified. These ROIs correspond to abnormalities. By segmenting the images, the system isolates these critical areas for more focused analysis. The method uses specialized algorithms (like YOLOv9) to accurately identify the regions that are most relevant to the condition being studied.
[0068] At block 908, the method 900 includes the step of applying a classification model to the identified ROIs, and deep learning techniques are utilized to analyze and classify the brain abnormalities. This ensures that the identified ROIs are classified using deep learning techniques for determining type of abnormality.
[0069] At block 910, the method 900 includes the step of evaluating performance of the classification model taking into consideration one or more metrics. This validates the reliability of the model and helps fine-tune it, ensuring the classification results are accurate and reliable.
[0070] At block 912, the method 900 includes the step of generating output of the classification and display to the input unit. The output includes the segmented and classified MR images displaying the brain abnormalities. This ensures that the results of the classification, including visualizations, are displayed to the input unit for review, diagnosis, or further action.
[0071] In an embodiment, the classification model is trained using a training dataset of the preprocessed MR images, and the performance of the classification model is validated using a testing dataset. This ensures that the model learns how to classify brain abnormalities accurately and ensures that the model can be generalized to new data.
[0072] In an embodiment, the classification model includes a hybrid classification ensemble that includes one or more convolutional neural networks (CNNs). This ensures that the system captures a wide range of details, enhancing the accuracy and robustness of the classification process.
[0073] Thus, the present disclosure discloses the system and method for detecting brain abnormality from magnetic resonance (MR) images. By incorporating object detection model with deep learning model, this solutions provide an efficient, robust and accurate system to classify and analyze MR images of brain for detecting abnormalities in the brain.
[0074] While the foregoing describes various embodiments of the disclosure, other and further embodiments of the disclosure may be devised without departing from the basic scope thereof. The scope of the disclosure is determined by the claims that follow. The disclosure is not limited to the described embodiments, versions or examples, which are included to enable a person having ordinary skill in the art to make and use the disclosure when combined with information and knowledge available to the person having ordinary skill in the art.



ADVANTAGES OF THE INVENTION
[0075] The present disclosure provides a system and a method for detecting brain abnormality from magnetic resonance (MR) images.
[0076] The present disclosure provides accuracy, enabling refined extraction of features and patterns from MR images.
[0077] The present disclosure provides robustness, ensuring accurate segmentation of the images in various sizes and complexities.
[0078] The present disclosure provides efficiency, ensuring an accelerated segmentation process for speedy diagnosis of abnormalities.
[0079] The present disclosure provides reliability by a single stage detector model, enhancing analysis of brain MR images.
 
, Claims:1. A system (100) to detect brain abnormality from magnetic resonance (MR) images, the system (100) comprising:
an input unit (102) configured to acquire magnetic resonance (MR) images of an entity; and
one or more processors (202) in communication with the input unit (102), and the one or more processor (202) coupled with a memory (204), wherein the memory (204) stores instructions executable by the one or more processors (202) to:
receive the MR images from the input unit (102);
pre-process the received MR images;
identify regions of interest (ROIs) corresponding to brain abnormalities from the pre-processed MR images;
apply a classification model to the identified ROIs, wherein deep learning techniques are utilized to analyze and classify the brain abnormalities;
evaluate performance of the classification model taking into consideration one or more metrics; and
generate output of the classification and display to the input unit (102), wherein the output comprises the segmented and classified MR images displaying the brain abnormalities.

2. The system (100) as claimed in claim 1, wherein to pre-process the MR images, the one or more processors (202) are configured to:
remove non-brain tissues from the MR images;
apply an anisotropic diffusion filter to reduce noise in the MR images;
utilize Contrast Limited Adaptive Histogram Equalization (CLAHE) to adjust contrast of the MR images; and
apply Generative Adversarial Networks (GANs) for data segmentation in the MR images.

3. The system (100) as claimed in claim 1, wherein the identification of regions of interest (ROIs) is performed using You Look Only Once version 9 (YOLOv9).
4. The system (100) as claimed in claim 1, wherein the classification model is trained using a training dataset of the preprocessed MR images, and the performance of the classification model is validated using a testing dataset.

5. The system (100) as claimed in claim 1, wherein the classification model comprises a hybrid classification ensemble that includes one or more convolutional neural networks (CNNs).

6. The system (100) as claimed in claim 1, wherein transfer learning is applied to fine-tune a pre-trained model for segmentation and classification of the identified ROIs.

7. A method (900) for detecting brain abnormality from magnetic resonance (MR) images, the method (900) comprises the steps of:
receiving (902), by one or more processors, the MR images from an input unit;
pre-processing (904), by the one or more processors, the received MR images;
identifying (906), by the one or more processors, regions of interest (ROIs) corresponding to brain abnormalities from the pre-processed MR images;
applying (908), by the one or more processors, a classification model to the identified ROIs, wherein deep learning techniques are utilized to analyze and classify the brain abnormalities;
evaluating (910), by the one or more processors, performance of the classification model taking into consideration one or more metrics; and
generating (912), by the one or more processors, output of the classification and display to the input unit, wherein the output comprises the segmented and classified MR images displaying the brain abnormalities.
8. The method (900) as claimed in claim 7, wherein pre-processing the MR images comprises the steps of:
removing, by the one or more processors, non-brain tissues from the MR images;
applying, by the one or more processors, an anisotropic diffusion filter to reduce noise in the MR images;
utilizing, by the one or more processors, Contrast Limited Adaptive Histogram Equalization (CLAHE) to adjust contrast of the MR images; and
applying, by the one or more processors, Generative Adversarial Networks (GANs) for data segmentation in the MR images.

9. The method (900) as claimed in claim 7, wherein the classification model is trained using a training dataset of the preprocessed MR images, and the performance of the classification model is validated using a testing dataset.

10. The method (900) as claimed in claim 7, wherein the classification model comprises a hybrid classification ensemble that includes one or more convolutional neural networks (CNNs).

Documents