Description:FIELD OF THE INVENTION
The invention relates to the field of medical imaging and artificial intelligence. More particularly, it concerns an automated brain tumor classification system and method that integrates Swin Transformer architecture with generative adversarial networks, metaheuristic optimization, and interpretability modules for accurate, explainable, and clinically deployable tumor diagnosis from MRI scans.
BACKGROUND OF THE INVENTION
Classifying brain tumors from MRI scans presents significant challenges due to the complex spatial structures of tumors and the technical limitations of existing computational models. Conventional convolutional methods primarily focus on local feature extraction and are limited in their ability to capture global contextual and hierarchical patterns necessary for differentiating tumor subtypes such as glioma, meningioma, and pituitary adenoma.
Additionally, available MRI datasets are often imbalanced, with certain tumor classes being underrepresented. This imbalance can lead to biased learning outcomes and reduced model generalization. Traditional data augmentation techniques may not provide sufficient anatomical diversity or realism to address these issues effectively.
Model performance is also highly sensitive to hyperparameter settings. Manual tuning procedures are computationally demanding and may result in suboptimal configurations. Moreover, existing models often lack transparency, which poses challenges for clinical application due to limited interpretability.The key limitations in current approaches include:
Inability to capture global and hierarchical features from MRI scans using standard convolutional architectures.
Severe class imbalance in tumor datasets, especially underrepresentation of pituitary tumors.
Limited effectiveness of conventional augmentation methods, which fail to generate clinically realistic variations.
Manual hyperparameter tuning is inefficient and often leads to poor model convergence.
Lack of interpretability, making the models unsuitable for clinical use without explanation of predictions.
US2004077967A1: A method and system for automated real time interpretation of brain waves in an acute brain injury of a patient using correlations between brain wave frequency power ratio and wave morphology, determined by a measure of the rhythmicity and variability of the brain wave as a function of the slope of the brain wave upstroke, the arc of the brain wave, and the synchronicity of the brain wave. A system is provided with a central processing unit (16), an electroencephalogram acquisition unit (18), a reference database (24), a quantitative electroencephalogram analysis program (14), and a display device (26) for communicating the classification of the acquired electroencephalographic signals. Artifact rejection is also provided.
US11227387B2: Methods, systems, and computer readable media to detect and model a brain tumor in an electronic image and to predict features of the brain tumor based on the model. The method can include classifying one or more magnetic resonance imaging (MRI) images of a brain into one or more of one or more tumorous images containing an image of a tumor or one or more non-tumorous images, wherein the classification is performed using a deep learning CNN system. The method can also include segmenting a tumor region from one of the one or more tumorous images. The segmenting can include a neighboring Fuzzy C-Means (FCM) process. The method can further include classifying the segmented tumor region into one of four classes of brain tumor types. The segmented tumor region is classified as a particular brain tumor type using the deep learning CNN system. The method can also include reconstructing a 3D model of the tumor region and measuring one or more of a location of the tumor, a shape of the tumor, or a volume of the tumor.
Conventional convolutional neural network approaches for brain tumor classification are constrained in their ability to capture global and hierarchical patterns. This results in limited performance in distinguishing between subtypes such as glioma, meningioma, and pituitary adenoma. Furthermore, datasets are often imbalanced, with pituitary tumors underrepresented, leading to biased training. Augmentation techniques commonly used fail to provide clinically realistic diversity. Model performance also depends on sensitive hyperparameters, which when tuned manually, are computationally inefficient and prone to suboptimal results. Moreover, most existing methods operate as black boxes without providing interpretability, limiting their clinical trust and adoption. The present invention addresses all these challenges by combining Swin Transformer-based hierarchical feature extraction, DCGAN-based synthetic augmentation, Grey Wolf Optimizer-driven hyperparameter tuning, and SHAP-based interpretability into a unified framework.
SUMMARY OF THE INVENTION
This summary is provided to introduce a selection of concepts, in a simplified format, that are further described in the detailed description of the invention.
This summary is neither intended to identify key or essential inventive concepts of the invention and nor is it intended for determining the scope of the invention.
The invention presents an automated system for multiclass brain tumor classification using MRI scans. The framework introduces four integrated modules: a synthetic data augmentation unit based on deep convolutional generative adversarial networks to balance datasets and improve tumor subtype representation; a Swin Transformer-based feature extractor and classifier that captures both local and global contextual relationships; a Grey Wolf Optimizer algorithm for automatic hyperparameter tuning to improve convergence and efficiency; and an interpretability module based on SHAP that provides heatmaps to highlight decision-relevant regions in MRI scans.
The system receives MRI images, generates additional anatomically realistic tumor samples when needed, processes the data through hierarchical patch-based Swin Transformer layers, and classifies tumor types into glioma, meningioma, or pituitary adenoma. Optimized hyperparameters ensure reliable performance across datasets, while SHAP-based visual overlays provide transparency for clinical validation.
The invention thus overcomes limitations of CNN-based classifiers, improves robustness to class imbalance, and provides interpretability necessary for medical applications. Its modular and scalable design allows deployment in hospital diagnostic platforms, cloud-based radiology services, and clinical decision support systems.
To further clarify advantages and features of the present invention, a more particular description of the invention will be rendered by reference to specific embodiments thereof, which is illustrated in the appended drawings. It is appreciated that these drawings depict only typical embodiments of the invention and are therefore not to be considered limiting of its scope. The invention will be described and explained with additional specificity and detail with the accompanying drawings.
The invention presents an integrated and intelligent deep learning framework for multiclass brain tumor classification utilizing MRI scans. It is specifically developed to address current challenges, including limited global feature extraction capabilities, dataset imbalance, manual hyperparameter tuning inefficiencies, and limited model interpretability, all of which have historically restricted the performance and practical deployment of traditional methods.
BRIEF DESCRIPTION OF THE DRAWINGS
The illustrated embodiments of the subject matter will be understood by reference to the drawings, wherein like parts are designated by like numerals throughout. The following description is intended only by way of example, and simply illustrates certain selected embodiments of devices, systems, and methods that are consistent with the subject matter as claimed herein, wherein:
FIGURE 1: SYSTEM ARCHITECTURE
The figures depict embodiments of the present subject matter for the purposes of illustration only. A person skilled in the art will easily recognize from the following description that alternative embodiments of the structures and methods illustrated herein may be employed without departing from the principles of the disclosure described herein.
DETAILED DESCRIPTION OF THE INVENTION
The detailed description of various exemplary embodiments of the disclosure is described herein with reference to the accompanying drawings. It should be noted that the embodiments are described herein in such details as to clearly communicate the disclosure. However, the amount of details provided herein 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 scope of the present disclosure as defined by the appended claims.
It is also to be understood that various arrangements may be devised that, although not explicitly described or shown herein, embody the principles of the present disclosure. Moreover, all statements herein reciting principles, aspects, and embodiments of the present disclosure, as well as specific examples, are intended to encompass equivalents thereof.
The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments. As used herein, the singular forms “a",” “an” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms “comprises,” “comprising,” “includes” and/or “including,” when used herein, specify the presence of stated features, integers, steps, operations, elements and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and/or groups thereof.
It should also be noted that in some alternative implementations, the functions/acts noted may occur out of the order noted in the figures. For example, two figures shown in succession may, in fact, be executed concurrently or may sometimes be executed in the reverse order, depending upon the functionality/acts involved.
In addition, the descriptions of "first", "second", “third”, and the like in the present invention are used for the purpose of description only, and are not to be construed as indicating or implying their relative importance or implicitly indicating the number of technical features indicated. Thus, features defining "first" and "second" may include at least one of the features, either explicitly or implicitly.
Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which example embodiments belong. It will be further understood that terms, e.g., those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.
The invention presents an integrated and intelligent deep learning framework for multiclass brain tumor classification utilizing MRI scans. It is specifically developed to address current challenges, including limited global feature extraction capabilities, dataset imbalance, manual hyperparameter tuning inefficiencies, and limited model interpretability, all of which have historically restricted the performance and practical deployment of traditional methods.
The invention combines four innovative modules:
1. Data Augmentation Using Deep Convolutional GAN (DCGAN):
To address the common issue of class imbalance, particularly the underrepresentation of tumor classes like pituitary adenoma, a DCGAN (Deep Convolutional Generative Adversarial Network) is trained on existing MRI images to generate synthetic, anatomically realistic tumor samples. DCGAN consists of two networks:
• A Generator, which learns to produce fake MRI images from noise vectors.
• A Discriminator, which learns to distinguish between real and fake images.
These networks are trained in opposition (adversarial training), allowing the generator to progressively improve the realism of synthesized images. The newly generated images are combined with real ones to form a balanced and diversified training set, leading to more robust and unbiased classification.
2. Feature Extraction and Classification Using Swin Transformer:
The classification model is based on the Swin Transformer, a novel hierarchical vision transformer architecture that operates on non-overlapping image patches. Unlike CNNs, which are limited to local receptive fields, Swin Transformer applies shifted window-based multi-head self-attention (SW-MSA) across different layers, enabling it to:
• Capture both local fine-grained features and global contextual relationships across the entire image.
• Process medical images at multiple scales, which is critical for recognizing varying tumor shapes and sizes.
The transformer processes MRI images in a hierarchical manner, starting from smaller patch embeddings and progressively fusing them at higher levels to extract representations that are more abstract. These are passed to fully connected layers to predict the tumor class (e.g., glioma, meningioma, or pituitary).
3. Hyperparameter Tuning Using Grey Wolf Optimizer (GWO):
The performance of deep models such as Swin Transformer is highly sensitive to hyperparameters like:
• Learning rate
• Patch size
• Embedding dimension
• Dropout rate
• Number of transformer layers
To automatically optimize these parameters, the invention integrates a metaheuristic optimization algorithm the Grey Wolf Optimizer (GWO). Inspired by the leadership and hunting behavior of grey wolves, GWO dynamically explores the solution space and updates the parameters to minimize classification loss. It maintains a balance between exploration and exploitation to avoid local minima and ensures faster convergence toward optimal settings. This step replaces manual and grid-based tuning approaches, significantly improving accuracy and training efficiency.
4. Explainability Using SHAP (SHapley Additive exPlanations):
To ensure that the classification results are clinically interpretable, the system includes a SHAP-based post-hoc explainability module. SHAP assigns each pixel (or region) in the MRI image an importance score indicating its contribution to the final classification. These scores are visualized as heatmaps superimposed on the original MRI scan, enabling radiologists to:
• Understand which regions influenced the decision (e.g., tumor boundaries, edema).
• Validate the prediction rationale.
• Improve trust and acceptance of the automated system in clinical workflows.
This level of explainability is critical for use in medical diagnosis, ensuring transparency and accountability.
This invention is ideally suited for integration into hospital diagnostic platforms, clinical decision support systems, or as an AI assistant for radiologists. It is capable of handling real-world variability in brain tumor imaging while maintaining performance, adaptability, and transparency.
The novelty of the invention lies in the integration of cutting-edge components into a unified and integrated system optimized for multiclass brain tumor classification. The Swin Transformer-based architecture constitutes the first instance of such an architecture for extracting contextual and hierarchical features from MRI scans to enable precise discrimination between tumor subclasses. The method enhances dataset balance using synthetic images produced by DCGAN, in this case, to compensate for underrepresented classes. Unlike conventional approaches, it leverages the Grey Wolf Optimizer (GWO) for dynamic hyperparameter adjustment to improve model performance without manual hyperparameter tuning. The use of SHAP also allows for region-level interpretability, the transparency necessary for clinical adoption. The modular and scalable design is a new state of the art for explainable, high-accuracy solutions to medical imaging.
The invention provides a system and method for brain tumor classification using MRI scans that integrates advanced deep learning and optimization techniques. The system first employs a preprocessing stage that normalizes MRI images and prepares them for training and inference. To address dataset imbalance, particularly the underrepresentation of certain tumor classes, a deep convolutional generative adversarial network is used. This module generates synthetic MRI scans that are anatomically realistic and class-specific, thereby ensuring sufficient variability and diversity in the training dataset.
The generative adversarial network consists of two components: a generator that produces synthetic MRI images from noise and a discriminator that evaluates the authenticity of generated versus real images. Through adversarial training, the generator progressively improves, resulting in a balanced dataset that enhances classification performance across all tumor classes.
Following data augmentation, MRI images are processed using a Swin Transformer-based architecture. This architecture divides images into non-overlapping patches and applies shifted window-based multi-head self-attention. Unlike traditional CNNs, which are limited to local receptive fields, the Swin Transformer captures both local and global relationships, learning hierarchical contextual representations critical for identifying tumor morphology.
The extracted features pass through hierarchical layers, where progressively larger receptive fields enable the model to understand fine as well as global patterns. These features are then aggregated and passed through classification layers that predict the tumor subtype, such as glioma, meningioma, or pituitary adenoma.
The classification performance of the system is highly dependent on the choice of hyperparameters, including learning rate, embedding dimension, patch size, and number of transformer layers. Instead of relying on manual tuning, the invention integrates a Grey Wolf Optimizer. Inspired by the social hierarchy and hunting strategy of grey wolves, this algorithm explores the parameter search space dynamically, balancing exploitation and exploration to converge toward optimal configurations. The optimization process enhances accuracy, reduces computational cost, and ensures faster training convergence.
To further ensure transparency in predictions, the invention incorporates an explainability module based on SHAP. SHAP assigns contribution scores to individual image regions, identifying the pixels or areas most responsible for classification outcomes. These scores are presented as visual heatmaps superimposed on the original MRI scans, enabling radiologists to interpret and validate the automated classification. This explainability is critical for clinical adoption and trust, as it allows users to understand how and why a particular classification decision was made.
The modular design of the system supports scalability and adaptability. New datasets can be incorporated, and the model retrained or fine-tuned without re-engineering the architecture. It supports integration with hospital PACS systems, enabling seamless deployment in diagnostic workflows. Cloud-based deployment is also possible, supporting telemedicine and remote diagnostic applications.
The invention differs from conventional CNN-based systems by addressing limitations in local feature extraction, scalability, dataset imbalance, and interpretability. By combining Swin Transformer-based contextual modeling, GAN-based augmentation, Grey Wolf Optimizer-driven tuning, and SHAP explainability, it provides a comprehensive solution for automated tumor classification.
BEST METHOD OF WORKING
The best method of working involves using the invention within a clinical radiology environment. MRI images are first normalized and preprocessed, after which DCGAN generates synthetic samples for underrepresented tumor types. The Swin Transformer processes these images through hierarchical layers, capturing local and global tumor features. The Grey Wolf Optimizer automatically selects optimal hyperparameters, ensuring model efficiency and accuracy. The final classification output is produced alongside SHAP-based heatmaps that highlight decision-relevant tumor regions. Deployment is best achieved on GPU-enabled workstations or cloud platforms integrated into PACS systems, enabling real-time classification, interpretability, and clinical reliability.
 
, Claims:1. A system for automated brain tumor classification from MRI scans comprising:
a data augmentation module using a generative adversarial network configured to generate synthetic tumor images for dataset balancing; a feature extraction and classification module comprising a Swin Transformer configured to capture local and global features through hierarchical shifted window attention;
an optimization module using a Grey Wolf Optimizer configured to tune hyperparameters of the classification module; an interpretability module based on SHAP configured to generate visual heatmaps indicating decision-relevant regions; and
an output unit configured to classify tumor types and provide interpretable visualizations for clinical support.
2. The system as claimed in claim 1, wherein the generative adversarial network comprises a generator for producing synthetic MRI images and a discriminator for distinguishing between real and synthetic images.
3. The system as claimed in claim 1, wherein the Swin Transformer processes non-overlapping image patches through shifted window-based multi-head self-attention.
4. The system as claimed in claim 1, wherein the optimization module dynamically adjusts learning rate, patch size, embedding dimension, dropout rate, and number of layers.
5. The system as claimed in claim 1, wherein the interpretability module assigns pixel-level or region-level importance scores for clinical validation.
6. The system as claimed in claim 1, wherein the output unit classifies tumor subtypes selected from glioma, meningioma, and pituitary adenoma.
7. The system as claimed in claim 1, wherein the system is integrated with clinical PACS infrastructure for deployment in diagnostic workflows.
8. The system as claimed in claim 1, wherein the system is deployable on GPU-enabled workstations or cloud platforms for real-time inference.
9. A method for automated brain tumor classification comprising: preprocessing MRI images and generating synthetic tumor images using a generative adversarial network; extracting local and global features using a Swin Transformer;
optimizing model hyperparameters using a Grey Wolf Optimizer; classifying tumor type using the extracted features; and generating interpretability heatmaps using SHAP to highlight decision-relevant regions.
10. The method as claimed in claim 9, wherein the tumor classification identifies glioma, meningioma, and pituitary adenoma subtypes with interpretability overlays for clinical decision support.