1. A system (100) for classification of Parkinson’s disease, the system (100) comprising: a data acquisition unit (102) adapted to receive magnetic resonance imaging data, speech recordings, and gait sensor data associated with a subject; a preprocessing unit (104) adapted to preprocess the magnetic resonance imaging data through image resizing, normalization, and augmentation, extract acoustic features from the speech recordings, and perform temporal gait analysis on the gait sensor data; a classification unit (106) adapted to classify processed magnetic resonance imaging data using an EfficientNet-B4 convolutional neural network, classify processed speech features using ensemble classifiers, and classify processed gait features using a machine learning classifier to generate modality-specific prediction scores; and a processing unit (112), characterized in that the processing unit (112) is configured to: receive magnetic resonance imaging data, speech recordings, and gait sensor data from the data acquisition unit (102); process the magnetic resonance imaging data, the speech recordings, and the gait sensor data through the preprocessing unit (104); generate modality-specific prediction scores through the classification unit (106); optimize hyperparameters and generate a final diagnostic prediction through a fusion unit (108), when the modality-specific prediction scores exceed from a predefined threshold; and generate clinical reports and secure deployment outputs through a reporting unit (110).

2. The system (100) as claimed in claim 1, wherein the fusion unit (108) is adapted to optimize hyperparameters associated with the classification unit (106) using a Horned Lizard Optimization algorithm and perform weighted late fusion on the modality-specific prediction scores to generate the final diagnostic prediction associated with Parkinson’s disease.

3. The system (100) as claimed in claim 1, wherein the reporting unit (110) is adapted to generate a clinical output report comprising confidence scores and interpretability information, encrypt subject data, anonymize personally identifiable information, and provide secured deployment through a Streamlit-based web application.

4. The system (100) as claimed in claim 1, wherein the multimodal deep learning architecture is configured to support modular addition of further clinical modalities for Parkinson’s disease diagnosis.

5. The system (100) as claimed in claim 1, wherein the ensemble classifiers comprise a Random Forest classifier and a CatBoost classifier.

6. The system (100) as claimed in claim 1, wherein the processing unit (112) is configured to store the final diagnostic prediction and associated interpretability information in a secured clinical database.

7. The system (100) as claimed in claim 1, wherein the processing unit (112) is configured to generate the final diagnostic prediction in real time through simultaneous processing of magnetic resonance imaging data, speech recordings, and gait sensor data.

8. The system (100) as claimed in claim 1, wherein the reporting unit (110) is adapted to encrypt subject data using an AES-256 encryption technique before storage and processing.

9. The system (100) as claimed in claim 1, wherein the reporting unit (110) is adapted to generate graphical representations of prediction confidence levels and modality contribution values.

10. A computer-implemented method (400) for classification of Parkinson’s disease using a multimodal deep learning architecture, the method (400) is characterized by steps of: receiving magnetic resonance imaging data, speech recordings, and gait sensor data from a data acquisition unit (102); processing the magnetic resonance imaging data, the speech recordings, and the gait sensor data through the preprocessing unit (104); generating modality-specific prediction scores through the classification unit (106); optimizing hyperparameters and generate a final diagnostic prediction through a fusion unit (108), when the modality-specific prediction scores exceed from a predefined threshold; and generating clinical reports and secure deployment outputs through a reporting unit (110). Date: May 12, 2026 Place: Noida Nainsi Rastogi Patent Agent (IN/PA-2372) Agent for the Applicant

Description:BACKGROUND
Field of Invention
[001] Embodiments of the present invention generally relate to a medical diagnostic system and particularly to a system and method for classification of Parkinson’s disease.
Description of Related Art
[002] Parkinson’s disease (PD) is a progressive neurodegenerative disorder that primarily affects motor functions due to the degeneration of dopamine-producing neurons in the brain. The disease is characterized by symptoms such as tremors, rigidity, bradykinesia, postural instability, speech impairments, and gait abnormalities. Early and accurate diagnosis of Parkinson’s disease is critical for effective disease management and timely therapeutic intervention. However, diagnosing Parkinson’s disease, particularly in its early stages, remains a significant clinical challenge.
[003] Conventional diagnostic approaches for Parkinson’s disease largely rely on clinical observation, neurological examinations, and subjective assessment scales such as the Unified Parkinson’s disease Rating Scale (UPDRS). These methods are highly dependent on clinician expertise and vary across practitioners, leading to inconsistent diagnostic outcomes. Additionally, early-stage symptoms of Parkinson’s disease are often subtle and heterogeneous, further complicating reliable diagnosis using traditional techniques.
[004] Magnetic Resonance Imaging (MRI) based deep learning models have been explored to identify structural and functional brain changes associated with Parkinson’s disease. Similarly, speech-based classifiers and gait-sensor-based systems have been proposed to capture motor and vocal impairments related to the disease. While these approaches have demonstrated promising results, most existing solutions are limited to a single modality or rely on simple fusion techniques. Single-modality systems often fail to capture the full spectrum of Parkinson’s disease biomarkers, as the disease manifests differently across patients and affects multiple physiological systems.
[005] Furthermore, many existing multi-modal systems employ early-stage feature fusion or naïve averaging techniques, that leads to feature misalignment, information loss, and reduced classification performance. In addition, existing approaches frequently depend on manual or grid-based hyperparameter tuning, that results in suboptimal model performance and poor generalization to real-world clinical data. Another significant limitation of current Parkinson’s disease classification systems is their lack of clinical deploy ability. Many proposed solutions remain academic prototypes and do not address essential considerations such as data privacy, security, regulatory compliance, real-time processing, and interpretability of results.
[006] There is thus a need for an improved and advanced system and method for classification of Parkinson’s disease that can administer the aforementioned limitations in a more efficient manner.
SUMMARY
[007] Embodiments in accordance with the present invention provide a system for classification of Parkinson’s disease. The system comprising a data acquisition unit adapted to receive magnetic resonance imaging data, speech recordings, and gait sensor data associated with a subject. The system further comprising a preprocessing unit adapted to preprocess the magnetic resonance imaging data through image resizing, normalization, and augmentation, extract acoustic features from the speech recordings, and perform temporal gait analysis on the gait sensor data. The system further comprising a classification unit adapted to classify processed magnetic resonance imaging data using an EfficientNet-B4 convolutional neural network, classify processed speech features using ensemble classifiers comprising a Random Forest classifier and a CatBoost classifier, and classify processed gait features using a machine learning classifier to generate modality-specific prediction scores. The system further comprising a processing unit. The processing unit is configured to receive the magnetic resonance imaging data, the speech recordings, and the gait sensor data from the data acquisition unit, process the magnetic resonance imaging data, the speech recordings, and the gait sensor data through the preprocessing unit, generate modality-specific prediction scores through the classification unit, optimize hyperparameters and generate a final diagnostic prediction through a fusion unit, when the modality-specific prediction scores exceed from a predefined threshold, and generate clinical reports and secure deployment outputs through a reporting unit.
[008] Embodiments in accordance with the present invention further provide a method for classification of Parkinson’s disease. The method comprising steps of receiving magnetic resonance imaging data, speech recordings, and gait sensor data from a data acquisition unit; processing the magnetic resonance imaging data, the speech recordings, and the gait sensor data through the preprocessing unit; generating modality-specific prediction scores through the classification unit; optimizing hyperparameters and generate a final diagnostic prediction through a fusion unit, when the modality-specific prediction scores exceed from a predefined threshold; and generating clinical reports and secure deployment outputs through a reporting unit.
[009] Embodiments of the present invention may provide a number of advantages depending on their particular configuration. First, embodiments of the present application may provide a system for classification of Parkinson’s disease.
[0010] Next, embodiments of the present application may provide a system for classification of Parkinson’s disease that may support interpretability through the generation of confidence scores and explainable outputs, thereby increasing clinician trust in automated diagnostic results.
[0011] Next, embodiments of the present application may provide a system for classification of Parkinson’s disease that may be capable of reducing subjectivity in Parkinson’s disease diagnosis by leveraging automated machine learning and deep learning-based classification models.
[0012] Next, embodiments of the present application may provide a system for classification of Parkinson’s disease that may provide improved decision-making by combining modality-specific predictions using weighted ensemble voting, thereby reducing the impact of noise or uncertainty in any single modality.
[0013] Next, embodiments of the present application may provide a system for classification of Parkinson’s disease that may be scalable and extensible to support additional data modalities or neurological conditions without significant architectural modifications.
[0014] Next, embodiments of the present application may provide a system for classification of Parkinson’s disease that may be deployable in clinical environments through a secure and user-friendly interface, facilitating real-time decision support for healthcare professionals.
[0015] These and other advantages will be apparent from the present application of the embodiments described herein.
[0016] The preceding is a simplified summary to provide an understanding of some embodiments of the present invention. This summary is neither an extensive nor exhaustive overview of the present invention and its various embodiments. The summary presents selected concepts of the embodiments of the present invention in a simplified form as an introduction to the more detailed description presented below. As will be appreciated, other embodiments of the present invention are possible utilizing, alone or in combination, one or more of the features set forth above or described in detail below.
BRIEF DESCRIPTION OF THE DRAWINGS
[0017] The above and still further features and advantages of embodiments of the present invention will become apparent upon consideration of the following detailed description of embodiments thereof, especially when taken in conjunction with the accompanying drawings, and wherein:
[0018] FIG. 1A illustrates a block diagram of a system for classification of Parkinson’s disease, according to an embodiment of the present invention;
[0019] FIG. 1B illustrates an architecture diagram of the system for classification of Parkinson’s disease, according to an embodiment of the present invention;
[0020] FIG. 2 illustrates components of a processing unit of the system for classification of Parkinson’s disease, according to an embodiment of the present invention;
[0021] FIG. 3A illustrates a home dashboard interface of a Streamlit-based web application of the system for classification of Parkinson’s disease, according to an embodiment of the present invention;
[0022] FIG. 3B illustrates an educational interface of the Streamlit-based web application of the system for classification of Parkinson’s disease, according to an embodiment of the present invention;
[0023] FIG. 3C illustrates a conversational chatbot interface of the Streamlit-based web application of the system for classification of Parkinson’s disease, according to an embodiment of the present invention;
[0024] FIG. 3D illustrates a news interface of the Streamlit-based web application of the system for classification of Parkinson’s disease, according to an embodiment of the present invention;
[0025] FIG. 3E illustrates a motor assessment of the Streamlit-based web application of the system for classification of Parkinson’s disease, according to an embodiment of the present invention;
[0026] FIG. 3F illustrates a Magnetic Resonance Imaging (MRI) analysis interface of the Streamlit-based web application of the system for classification of Parkinson’s disease, according to an embodiment of the present invention;
[0027] FIG. 3G illustrates an enhanced motor assessment interface of the web-based of the Streamlit-based web application of the system for classification of Parkinson’s disease, according to an embodiment of the present invention;
[0028] FIG. 3H illustrates a first clinical gait analysis interface with normal result of the Streamlit-based web application of the system for classification of Parkinson’s disease according to an embodiment of the present invention;
[0029] FIG. 3I illustrates a second clinical gait analysis interface with moderate gait changes detection interface of the Streamlit-based web application of the system for classification of Parkinson’s disease, according to an embodiment of the present invention;
[0030] FIG. 3J illustrates a first speech analysis interface of the Streamlit-based web application of the system for classification of Parkinson’s disease, according to an embodiment of the present invention;
[0031] FIG. 3K illustrates a second speech analysis interface of the Streamlit-based web application of the system for classification of Parkinson’s disease according to an embodiment of the present invention; and
[0032] FIG. 4 illustrates a flowchart of a method for classification of Parkinson’s disease, according to an embodiment of the present invention.
DETAILED DESCRIPTION
[0033] FIG. 1A illustrates a block diagram of a system 100 for classification of Parkinson’s disease using a multimodal deep learning architecture, according to an embodiment of the present invention. In an embodiment of the present invention, the system 100 may be adapted to support early-stage Parkinson’s disease diagnosis through multimodal analysis of the magnetic resonance imaging data, the speech recordings, and the gait sensor data associated with heterogeneous clinical data sources. The system 100 may be configured to process non-homogeneous patient datasets received from medical imaging systems, wearable sensing devices, remote healthcare platforms, and clinical repositories for improving diagnostic robustness under real-world clinical conditions. Further, the system 100 may be adapted to support transparent and scalable diagnostic operations through generation of explainable prediction outputs, visual explanations, and modality-specific confidence scores associated with Parkinson’s disease classification.
[0034] In an embodiment of the present invention, the system 100 may provide a multimodal and intelligent diagnostic mechanism for identification of Parkinson’s disease using magnetic resonance imaging data, speech recordings, and gait sensor data. The system 100 may be configured to preprocess multimodal clinical data, perform modality-specific classification, optimize prediction performance, and generate an interpretable diagnostic output suitable for clinical decision support. The system 100 may operate in a continuous or on-demand manner to classify Parkinson’s disease based on newly received patient data. The system 100 may provide scalable deployment across diverse clinical environments while ensuring robustness, adaptability, and reliable decision support for healthcare professionals.
[0035] In an embodiment of the present invention, the system 100 may be configured to improve multimodal Parkinson’s disease classification accuracy relative to conventional single-modality diagnostic systems through weighted late fusion of the magnetic resonance imaging data, the speech recordings, and the gait sensor data using the Horned Lizard Optimization algorithm and multimodal ensemble classification operations.
[0036] According to the embodiments of the present invention, the system 100 for classification of Parkinson’s disease may incorporate non-limiting hardware components to enhance the processing speed and efficiency such as the system 100 for classification of Parkinson’s disease may comprise a data acquisition unit 102, a preprocessing unit 104, a classification unit 106, a fusion unit 108, a reporting unit 110, and a processing unit 112. In an embodiment of the present invention, the hardware components of the system 100 may be integrated with computer-executable instructions for overcoming the challenges and the limitations of the existing systems.
[0037] In an embodiment of the present invention, the data acquisition unit 102 may be adapted to receive multimodal patient data associated with a subject. The multimodal patient data may comprise magnetic resonance imaging (MRI) data, speech recordings, and gait sensor data. In an embodiment of the present invention, the magnetic resonance imaging (MRI) data may comprise structural brain scan images associated with neurological evaluation of the subject. The speech recordings may comprise vocal utterances, sustained phonation samples, conversational speech samples, and articulation recordings associated with speech impairment analysis. Further, the gait sensor data may comprise motion-related signals generated from wearable sensors, inertial measurement units, accelerometers, gyroscopes, pressure sensing devices, mobile sensing devices, and so forth, adapted to monitor gait characteristics of the subject. The system 100 may be configured for deployment in hospitals, diagnostic laboratories, neurological assessment centres, telemedicine environments, and remote healthcare monitoring platforms. In an embodiment of the present invention, the system 100 may be configured to support lightweight clinical deployment operations for enabling multimodal Parkinson’s disease diagnosis through secured healthcare networks, remote diagnostic platforms, and multi-user neurological assessment environments.
[0038] The multimodal patient data may comprise heterogeneous clinical information corresponding to neurological, vocal, and motion characteristics of the subject. In an embodiment of the present invention, the multimodal patient data may be indicative of early-stage and advanced-stage Parkinsonian symptoms associated with the subject. The data acquisition unit 102 may be adapted to receive data from local clinical repositories, medical imaging systems, wearable sensing systems, cloud-based healthcare platforms, and real-time monitoring devices. The system 100 may further be configured to support modular integration of additional clinical modalities including handwriting analysis data, electroencephalogram data, and facial expression data without structural modification of the system 100.
[0039] In an embodiment of the present invention, the data acquisition unit 102 may be configured to receive heterogeneous multimodal patient data from non-homogeneous healthcare environments comprising distributed clinical repositories, remote sensing platforms, mobile healthcare devices, and hospital-based neurological assessment systems for enabling real-world Parkinson’s disease diagnosis operations.
[0040] The data acquisition unit 102 may be, but not limited to, a medical data collection device, a clinical data interface, a wearable sensor interface, a medical imaging acquisition interface, a remote healthcare monitoring interface, and so forth. Embodiments of the present invention are intended to include or otherwise cover any type of the data acquisition unit 102, including known, related art, and/or later developed technologies.
[0041] In an embodiment of the present invention, the preprocessing unit 104 may be adapted to receive the multimodal patient data from the data acquisition unit 102 and preprocess the received multimodal patient data for downstream analysis. The preprocessing unit 104 may be adapted to preprocess the magnetic resonance imaging (MRI) data through resizing, normalization, augmentation, skull stripping, intensity balancing, and contrast enhancement operations. In an embodiment of the present invention, the preprocessing unit 104 may be adapted to remove imaging artifacts and standardize image dimensions associated with the magnetic resonance imaging (MRI) data. Further, the preprocessing unit 104 may be adapted to preprocess the speech recordings through filtering, silence removal, segmentation, frequency normalization, noise suppression operations, and so forth.
[0042] In an embodiment of the present invention, the preprocessing unit 104 may be configured to process noisy speech recordings and heterogeneous acoustic datasets associated with real-world healthcare environments for improving robustness of speech impairment detection associated with Parkinson’s disease diagnosis.
[0043] The preprocessing unit 104 may be adapted to extract acoustic features including Mel Frequency Cepstral Coefficients (MFCC), pitch variation, shimmer, jitter, harmonic-to-noise ratio, spectral centroid, and statistical speech descriptors from the speech recordings.
[0044] In another embodiment of the present invention, the preprocessing unit 104 may be adapted to preprocess the gait sensor data through temporal synchronization, normalization, smoothing, signal correction, and motion stabilization operations. The preprocessing unit 104 may be adapted to perform temporal gait analysis for extraction of gait parameters including stride length, cadence, gait symmetry, acceleration variance, step periodicity, and motion irregularity indicators. The preprocessing unit 104 may further be configured to generate standardized multimodal feature representations suitable for downstream classification operations.
[0045] In an embodiment of the present invention, the preprocessing unit 104 may be configured to process heterogeneous gait sensor datasets received from wearable sensing devices, accelerometers, gyroscopes, and mobile healthcare monitoring devices for improving gait abnormality detection associated with Parkinson’s disease classification operations.
[0046] In an embodiment of the present invention, the preprocessing unit 104 may be adapted to perform data-independent preprocessing operations on the magnetic resonance imaging data, the speech recordings, and the gait sensor data for enabling standardized feature extraction across heterogeneous patient datasets. The preprocessing unit 104 may be configured to extract modality-specific feature representations comprising Mel Frequency Cepstral Coefficients, spectral speech descriptors, temporal gait statistics, neurological texture features, and motion irregularity indicators associated with early-stage Parkinsonian symptoms.
[0047] The preprocessing unit 104 may be, but not limited to, an image preprocessing engine, a speech signal preprocessing engine, a gait signal conditioning engine, a feature normalization engine, and so forth. Embodiments of the present invention are intended to include or otherwise cover any type of the preprocessing unit 104, including known, related art, and later developed technologies.
[0048] In an embodiment of the present invention, the classification unit 106 may be adapted to receive processed multimodal feature representations from the preprocessing unit 104 and classify the processed multimodal feature representations to generate modality-specific prediction scores associated with Parkinson’s disease. The classification unit 106 may be adapted to classify processed magnetic resonance imaging (MRI) data using the EfficientNet-B4 convolutional neural network. The EfficientNet-B4 convolutional neural network may be adapted to extract deep spatial and texture-based neurological features associated with Parkinsonian degeneration. In an embodiment of the present invention, the classification unit 106 may be configured to classify processed speech features using ensemble classifiers comprising a Random Forest classifier and a CatBoost classifier. The ensemble classifiers may be adapted to identify speech impairments associated with vocal instability, articulation variation, and frequency irregularities.
[0049] In another embodiment of the present invention, the classification unit 106 may be adapted to classify processed gait features using machine learning classifiers including support vector machine classifiers, decision tree classifiers, gradient boosting classifiers, k-nearest neighbour classifiers, and so forth. The classification unit 106 may be adapted to generate modality-specific prediction scores corresponding to each clinical modality. The modality-specific prediction scores may comprise prediction probabilities, classification confidence values, and disease severity indicators associated with Parkinson’s disease diagnosis. The classification unit 106 may further be configured to support parallel processing of multimodal data streams for enabling real-time diagnostic prediction generation.
[0050] In an embodiment of the present invention, the EfficientNet-B4 convolutional neural network associated with classification of the processed magnetic resonance imaging data may generate classification accuracy up to 97.6%. Further, the Random Forest classifier, the CatBoost classifier, and the Extreme Gradient Boosting classifier associated with classification of the processed speech features may generate classification accuracy up to 94.87%, and the machine learning classifier associated with classification of the processed gait features may generate classification accuracy up to 91.5%. In another embodiment of the present invention, the fusion unit 108 may generate multimodal classification accuracy up to 96.2% through weighted late fusion and weighted ensemble voting operations associated with the modality-specific prediction scores.
[0051] In an embodiment of the present invention, the modality-specific prediction scores may be adapted to identify early-stage Parkinsonian abnormalities associated with neurological impairment, speech instability, and gait irregularities. The classification unit 106 may be configured to generate disease severity indicators associated with progression analysis of Parkinson’s disease using the processed magnetic resonance imaging data, the processed speech features, and the processed gait features.
[0052] In an embodiment of the present invention, the classification unit 106 may be configured to generate multimodal prediction outputs through ensemble machine learning classification of the processed speech features and the processed gait features for improving detection of Parkinsonian speech impairments and motion abnormalities associated with the subject.
[0053] In an embodiment of the present invention, the classification unit 106 may be adapted to perform modular model training operations using the EfficientNet-B4 convolutional neural network for classification of the processed magnetic resonance imaging data, the Random Forest classifier, the CatBoost classifier, and the Extreme Gradient Boosting classifier for classification of the processed speech features, and the machine learning classifier for classification of the processed gait features. The classification unit 106 may be configured to generate modality-specific prediction outputs associated with early-stage Parkinson’s disease detection and multimodal clinical decision support.
[0054] The classification unit 106 may be, but not limited to, a deep learning classification engine, an ensemble learning engine, a multimodal inference engine, a predictive analytics engine, and so forth. Embodiments of the present invention are intended to include or otherwise cover any type of the classification unit 106, including known, related art, and later developed technologies.
[0055] In an embodiment of the present invention, the fusion unit 108 may be adapted to receive modality-specific prediction scores from the classification unit 106 and generate a final diagnostic prediction associated with Parkinson’s disease. The fusion unit 108 may be adapted to optimize hyperparameters associated with the classification unit 106 using a Horned Lizard Optimization algorithm.
[0056] In an embodiment of the present invention, the Horned Lizard Optimization algorithm may be adapted to optimize multimodal classification parameters associated with the EfficientNet-B4 convolutional neural network, the Random Forest classifier, the CatBoost classifier, the Extreme Gradient Boosting classifier, and the machine learning classifier for improving multimodal prediction consistency and diagnostic accuracy associated with Parkinson’s disease classification operations. The Horned Lizard Optimization algorithm may be adapted to optimize learning rate, training epochs, batch size, classifier depth, regularization coefficients, feature selection parameters, ensemble voting weights, and other operational parameters associated with multimodal classification. In an embodiment of the present invention, the fusion unit 108 may be adapted to perform weighted late fusion on the modality-specific prediction scores to generate the final diagnostic prediction.
[0057] The fusion unit 108 may be adapted to dynamically assign optimized weights to modality-specific prediction outputs based on predictive reliability, historical performance metrics, confidence scores, and modality relevance. In an embodiment of the present invention, the fusion unit 108 may be adapted to generate a probability score associated with occurrence of Parkinson’s disease. The probability score may correspond to a likelihood value associated with disease presence and severity. In an embodiment of the present invention, the probability score associated with occurrence of Parkinson’s disease may be adapted to represent neurological abnormality likelihood derived from weighted late fusion of the processed magnetic resonance imaging data, the processed speech features, and the processed gait features. The fusion unit 108 may further be configured to support modular integration of additional clinical modalities without modification of the weighted late fusion architecture.
[0058] In an embodiment of the present invention, the fusion unit 108 may be configured to generate weighted optimization outputs associated with the modality-specific prediction scores for reducing prediction uncertainty associated with individual clinical modalities during Parkinson’s disease classification operations.
[0059] In an embodiment of the present invention, the fusion unit 108 may be adapted to support scalable multimodal expansion operations through integration of handwriting analysis data, electroencephalogram data, facial expression data, and additional neurological assessment modalities associated with Parkinson’s disease diagnosis without structural modification of the system 100.
[0060] In an embodiment of the present invention, the fusion unit 108 may be configured to perform weighted ensemble voting on the modality-specific prediction scores using optimized modality weights generated through the Horned Lizard Optimization algorithm. The fusion unit 108 may be adapted to improve multimodal classification accuracy relative to single-modality diagnostic systems and bimodal diagnostic systems through weighted late fusion of the processed magnetic resonance imaging data, the processed speech features, and the processed gait features.
[0061] The fusion unit 108 may be, but not limited to, a multimodal fusion engine, a hyperparameter optimization engine, a weighted ensemble engine, a probabilistic inference engine, and so forth. Embodiments of the present invention are intended to include or otherwise cover any type of the fusion unit 108, including known, related art, and later developed technologies.
[0062] In an embodiment of the present invention, the reporting unit 110 may be adapted to receive the final diagnostic prediction from the fusion unit 108 and generate clinical output reports associated with Parkinson’s disease diagnosis. The reporting unit 110 may be adapted to generate graphical representations of prediction confidence levels, modality contribution values, disease severity indicators, and interpretability information. In an embodiment of the present invention, the graphical representations may comprise heat maps, confidence distribution plots, modality contribution charts, prediction probability indicators, and SHapley Additive exPlanations (SHAP) based visualization outputs associated with Parkinson’s disease classification operations.
[0063] In an embodiment of the present invention, the reporting unit 110 may be adapted to generate clinician-readable reports suitable for neurological assessment and diagnostic review. In an embodiment of the present invention, the clinical output report may comprise the confidence scores, the modality contribution values, the probability score, the SHapley Additive exPlanations (SHAP) based interpretability information, and the disease severity indicators associated with the final diagnostic prediction for assisting clinician-based Parkinson’s disease assessment operations.
[0064] The reporting unit 110 may further be adapted to encrypt subject data using an AES-256 encryption technique before storage and processing. In an embodiment of the present invention, the AES-256 encryption technique may be adapted to secure the magnetic resonance imaging data, the speech recordings, the gait sensor data, the modality-specific prediction scores, and the final diagnostic prediction during storage, transmission, and remote clinical deployment operations.
[0065] The reporting unit 110 may be adapted to anonymize personally identifiable information associated with the subject for ensuring compliance with healthcare privacy standards and secure clinical deployment requirements. In an embodiment of the present invention, the reporting unit 110 may be configured to generate visual explanations associated with the final diagnostic prediction through feature contribution mapping, modality contribution analysis, confidence visualization outputs, and SHapley Additive exPlanations (SHAP) based interpretability operations for assisting clinician-readable neurological assessment review.
[0066] In an embodiment of the present invention, the reporting unit 110 may be adapted to generate SHAP values, feature contribution visualizations, and graphical interpretability outputs associated with the final diagnostic prediction for improving clinical transparency and diagnostic review operations. Further, the reporting unit 110 may be configured to support role-based access operations, Health Insurance Portability and Accountability Act compliance operations, General Data Protection Regulation compliance operations, and Food and Drug Administration compliant deployment operations associated with secured clinical usage of the system 100.
[0067] In another embodiment of the present invention, the reporting unit 110 may be adapted to provide secured deployment through interfaces 300-320 (as shown in FIG. 3A-3K) of a Streamlit-based web application. The Streamlit-based web application may enable real-time interaction, remote diagnostic access, visualization of prediction outputs, and monitoring of multimodal patient data. The reporting unit 110 may further be adapted to transmit the generated clinical reports and deployment outputs to healthcare professionals, remote diagnostic systems, and secured healthcare databases.
[0068] In an embodiment of the present invention, the reporting unit 110 may be configured to support clinical decision support operations through secured transmission of the clinical reports, the probability scores, the modality contribution values, and the interpretability information to authorized healthcare professionals for neurological assessment review and treatment planning operations.
[0069] In an embodiment of the present invention, the Streamlit-based web application may be adapted to support lightweight multi-user clinical deployment operations for enabling simultaneous diagnostic access by healthcare professionals across remote healthcare environments, neurological assessment centres, and hospital-based diagnostic systems. The reporting unit 110 may be, but not limited to, a clinical reporting engine, a visualization engine, a secure deployment engine, an encryption engine, and so forth. Embodiments of the present invention are intended to include or otherwise cover any type of the reporting unit 110, including known, related art, and later developed technologies.
[0070] In an embodiment of the present invention, the processing unit 112 may be operatively coupled to the data acquisition unit 102, the preprocessing unit 104, the classification unit 106, the fusion unit 108, and the reporting unit 110. The processing unit 112 may be configured to coordinate acquisition, preprocessing, classification, fusion, optimization, reporting, and storage operations associated with the multimodal deep learning architecture. In an embodiment of the present invention, the processing unit 112 may be configured to receive magnetic resonance imaging (MRI) data, speech recordings, and gait sensor data from the data acquisition unit 102 and process the received multimodal patient data through the preprocessing unit 104. Further, the processing unit 112 may be configured to enable generation of modality-specific prediction scores through the classification unit 106.
[0071] In an embodiment of the present invention, the processing unit 112 may be adapted to coordinate modular model training operations, multimodal prediction generation operations, weighted optimization operations, and real-time diagnostic deployment operations associated with the multimodal deep learning architecture of the system 100.
[0072] In an exemplary scenario, when the modality-specific prediction scores exceed a predefined threshold, the processing unit 112 may be configured to enable the fusion unit 108 to optimize hyperparameters and generate the final diagnostic prediction associated with Parkinson’s disease. In another exemplary scenario, the processing unit 112 may be configured to generate the final diagnostic prediction in real time through simultaneous processing of magnetic resonance imaging (MRI) data, speech recordings, and gait sensor data. The processing unit 112 may further be configured to store the final diagnostic prediction, interpretability information, probability scores, and associated clinical outputs in a secured clinical database. In an embodiment of the present invention, the secured clinical database may be adapted to store the final diagnostic prediction, the modality-specific prediction scores, the confidence scores, the interpretability information, and encrypted patient records associated with longitudinal Parkinson’s disease monitoring and diagnostic review operations.
[0073] In an embodiment of the present invention, the processing unit 112 may be configured to process noisy clinical datasets and heterogeneous multimodal patient data associated with real-world healthcare environments for improving classification reliability and diagnostic generalization performance associated with Parkinson’s disease detection operations. Further, the processing unit 112 may be adapted to coordinate multimodal prediction generation and weighted optimization operations for improving clinical decision support accuracy associated with the final diagnostic prediction.
[0074] The processing unit 112 may be, but not limited to, a microprocessor, a graphics processing unit, an embedded artificial intelligence processor, a cloud processing engine, a multimodal inference processor, and so forth. Embodiments of the present invention are intended to include or otherwise cover any type of the processing unit 112, including known, related art, and later developed technologies. The processing unit 112 may further be explained in detail in conjunction with FIG. 2.
[0075] FIG. 1B illustrates an architecture diagram of the system 100 for classification of Parkinson’s disease, according to an embodiment of the present invention. In an exemplary embodiment, a user X may provide the magnetic resonance imaging data, the speech recordings, and the gait sensor data to the data acquisition unit 102 for Parkinson’s disease assessment. The data acquisition unit 102 may be configured to receive the multimodal patient data and transmit the received multimodal patient data to the preprocessing unit 104.
[0076] In an embodiment of the present invention, the preprocessing unit 104 may preprocess the magnetic resonance imaging data through resizing and normalization operations, extract the acoustic features from the speech recordings, and perform temporal gait analysis on the gait sensor data to generate the processed multimodal feature representations. Further, the preprocessing unit 104 may transmit the processed multimodal feature representations to the classification unit 106.
[0077] In an embodiment of the present invention, the classification unit 106 may classify the processed magnetic resonance imaging data using the EfficientNet-B4 convolutional neural network, classify the processed speech features using the ensemble classifiers comprising the Random Forest classifier and the CatBoost classifier, and classify the processed gait features using the machine learning classifier to generate the modality-specific prediction scores associated with Parkinson’s disease. Further, the classification unit 106 may transmit the modality-specific prediction scores to the fusion unit 108.
[0078] In an embodiment of the present invention, the fusion unit 108 may optimize the hyperparameters associated with the classification unit 106 using the Horned Lizard Optimization algorithm and perform weighted late fusion on the modality-specific prediction scores to generate the final diagnostic prediction associated with Parkinson’s disease for the user X.
[0079] In an embodiment of the present invention, the reporting unit 110 may generate the clinical reports comprising the confidence scores and the interpretability information associated with the final diagnostic prediction and provide secured deployment through the Streamlit-based web application for enabling clinical review of the diagnostic outputs associated with the user X.
[0080] FIG. 2 illustrates components of the processing unit 112 of the system 100 for classification of Parkinson’s disease, according to an embodiment of the present invention. The processing unit 112 may comprise a data processing module 200, a feature extraction module 202, a hyperparameter optimization module 204, and a decision fusion module 206. The modules may collectively enable preprocessing, feature extraction, optimization, and fusion of multi-modal patient data for accurate Parkinson’s disease classification.
[0081] According to embodiments of the present invention, the data processing module 200 may be configured to receive magnetic resonance imaging (MRI) data, speech recordings, and gait sensor data from the data acquisition unit 102. The data processing module 200 may further be configured to preprocess the magnetic resonance imaging (MRI) data through resizing, normalization, augmentation, skull stripping, and contrast enhancement operations to generate standardized MRI image data. According to embodiments of the present invention, the data processing module 200 may be configured to preprocess the speech recordings through filtering, silence removal, segmentation, and noise reduction operations to generate enhanced speech signal data. Further, the data processing module 200 may be configured to preprocess the gait sensor data through temporal synchronization, smoothing, normalization, and motion correction operations to generate standardized gait signal data. Furthermore, the data processing module 200 may be configured to transmit the standardized MRI image data, enhanced speech signal data, and standardized gait signal data to the feature extraction module 202.
[0082] The feature extraction module 202 may be configured to receive the standardized MRI image data, enhanced speech signal data, and standardized gait signal data from the data processing module 200. In an embodiment of the present invention, the feature extraction module 202 may be configured to extract deep spatial and texture-based features from the magnetic resonance imaging (MRI) data using convolutional feature mapping associated with the EfficientNet-B4 architecture. According to embodiments of the present invention, the feature extraction module 202 may further be configured to extract acoustic features including Mel Frequency Cepstral Coefficients (MFCC), pitch, shimmer, jitter, harmonic-to-noise ratio, and statistical descriptors from the speech recordings. In another embodiment of the present invention, the feature extraction module 202 may be configured to extract gait-related parameters including stride length, cadence, gait symmetry, acceleration variance, and motion periodicity from the gait sensor data. Further, the feature extraction module 202 may be configured to generate modality-specific feature vectors based on the extracted features. Furthermore, the feature extraction module 202 may be configured to transmit the modality-specific feature vectors to the hyperparameter optimization module 204.
[0083] The hyperparameter optimization module 204 may be configured to receive the modality-specific feature vectors from the feature extraction module 202 and optimize operational parameters associated with multimodal classification of Parkinson’s disease, according to an embodiment of the present invention. In a preferred embodiment of the present invention, the hyperparameter optimization module 204 may be configured to employ the Horned Lizard Optimization algorithm to determine optimized hyperparameters associated with the classification unit 106. The optimized hyperparameters may comprise learning rate, batch size, training epochs, classifier depth, regularization coefficients, feature selection parameters, and ensemble voting weights. In an exemplary scenario, the hyperparameter optimization module 204 may be configured to evaluate candidate hyperparameter combinations based on classification accuracy, sensitivity, specificity, precision, recall, and loss minimization criteria. Further, the hyperparameter optimization module 204 may be configured to generate optimized classification outputs based on the optimized hyperparameters. Furthermore, the hyperparameter optimization module 204 may be configured to transmit the optimized classification outputs to the decision fusion module 206.
[0084] The decision fusion module 206 may be configured to receive the optimized classification outputs from the hyperparameter optimization module 204. According to embodiments of the present invention, the decision fusion module 206 may be configured to perform weighted late fusion on magnetic resonance imaging (MRI) prediction scores, speech prediction scores, and gait prediction scores to generate a final diagnostic prediction associated with Parkinson’s disease. In an embodiment of the present invention, the decision fusion module 206 may be configured to dynamically assign optimized weights to modality-specific prediction outputs based on predictive reliability, confidence scores, and historical performance metrics. Further, the decision fusion module 206 may be configured to generate the probability score associated with occurrence of Parkinson’s disease and interpretability information corresponding to modality contribution values. In another embodiment of the present invention, if the decision fusion module 206 determines that the generated probability score exceeds the predefined diagnostic threshold, then the decision fusion module 206 may be configured to generate the Parkinson’s disease detection output. Further, the decision fusion module 206 may be configured to transmit the Parkinson’s disease detection output and associated interpretability information to the reporting unit 110 for generation of clinical output reports.
[0085] FIG. 3A illustrates a home dashboard interface 300 of the Streamlit-based web application of the system 100 for classification of Parkinson’s disease, according to an embodiment of the present invention. The home dashboard interface 300 may be rendered via a web application framework and may provide an overview of system capabilities, operational statistics, and navigation controls.
[0086] The home dashboard interface 300 may be rendered via a web application framework and may provide an overview of system capabilities, operational statistics, and navigation controls. The home dashboard interface 300 may include a navigation panel positioned on a lateral portion of the display, providing selectable options including, but not limited to, Home, Learn, Detect Models, Chatbot, Research, and About. The navigation panel may enable a user to switch between functional modules of the system 100. The home dashboard interface 300 may further display summary statistics, including detection accuracy, processing speed, number of supported modalities, and number of users. The home dashboard interface 300 may present selectable detection options corresponding to Magnetic Resonance Imaging (MRI) brain scan analysis, drawing-based motor assessment, speech analysis, and gait analysis. Embodiments of the present invention are intended to include or otherwise cover any type of dashboard visualization or statistical indicators.
[0087] FIG. 3B illustrates an educational interface 302 of the Streamlit-based web application of the system 100 for classification of Parkinson’s disease, according to an embodiment of the present invention. The educational interface 302 may be configured to provide informational content related to Parkinson’s disease. The educational interface 302 may present structured educational material, including disease overview, symptoms, diagnosis stages, medications, and frequently asked questions. The educational interface 302 may include tab-based navigation controls allowing a user to access various educational sections such as Overview, Motor Symptoms, Non-Motor Symptoms, Medications, Diagnosis & Stages, FAQ, and Resources. The educational content may be presented in a scrollable format to enhance readability. The educational interface 302 may serve as an awareness and support module, enabling users to better understand disease characteristics alongside diagnostic outputs. Embodiments of the present invention are intended to include or otherwise cover any type of informational or educational interface.
[0088] FIG. 3C illustrates a conversational chatbot interface 304 of the Streamlit-based web application of the system 100 for classification of Parkinson’s disease, according to an embodiment of the present invention. The conversational chatbot interface 304 may be configured to respond to user queries related to Parkinson’s disease symptoms, treatments, detection methods, and general information. The conversational interface may include a conversation display region showing user-submitted queries and corresponding automated responses generated by an artificial intelligence chatbot. The conversational chatbot interface 304 may provide text-based responses derived from a curated knowledge base or trained language model. An input field may be provided to allow users to submit queries, along with a submission control to initiate response generation. Embodiments of the present invention are intended to include or otherwise cover any type of conversational or interactive support interfaces.
[0089] FIG. 3D illustrates a news interface 306 of the Streamlit-based web application of the system 100 for classification of Parkinson’s disease, according to an embodiment of the present invention. The news interface 306 may be configured to present updated scientific information related to Parkinson’s disease and the system 100 for classification of Parkinson’s disease. The news interface 306 may display sections such as recent research breakthroughs, clinical trials, detection methodologies, key publications, and resources. The content may include structured summaries of discoveries, impact statements, treatment developments, and trial statuses. The research interface may enable users to remain informed about ongoing advancements in Parkinson’s disease research. Embodiments of the present invention are intended to include or otherwise cover any type of research dissemination or content aggregation mechanisms.
[0090] FIG. 3E illustrates a motor assessment interface 308 of the Streamlit-based web application of the system 100 for classification of Parkinson’s disease, according to an embodiment of the present invention. The motor assessment interface 308 may be configured to evaluate motor control impairments. The motor assessment interface 308 may allow users to upload or draw spiral or wave patterns for analysis. The motor assessment interface 308 may include input options enabling file upload or real-time drawing input. The uploaded drawing data may be processed by the feature extraction module 202 configured to compute motor-related features such as tremor intensity, entropy, or line irregularity. The motor assessment interface 308 may further display analysis results, including abnormality detection, confidence score, and extracted feature values. A visual alert may be presented when abnormal motor patterns are detected. Embodiments of the present invention are intended to include or otherwise cover any type of drawing-based motor assessment tools.
[0091] FIG. 3F illustrates a Magnetic Resonance Imaging (MRI) analysis interface 310 of the Streamlit-based web application of the system 100 for classification of Parkinson’s disease, according to an embodiment of the present invention. The Magnetic Resonance Imaging (MRI) brain scan analysis may be configured to perform artificial intelligence-based analysis of uploaded Magnetic Resonance Imaging (MRI) images. The Magnetic Resonance Imaging (MRI) brain scan analysis interface may include an upload module allowing a user to submit a Magnetic Resonance Imaging (MRI) brain scan image for analysis. The uploaded image may be processed by a convolutional neural network configured to identify neurological patterns indicative of Parkinson’s disease. The Magnetic Resonance Imaging (MRI) brain scan analysis interface may display analysis results, including classification status (normal or abnormal), confidence score, and probability distribution visualization. The Magnetic Resonance Imaging (MRI) image may be displayed alongside analytical outputs for reference. Embodiments of the present invention are intended to include or otherwise cover any type of medical image analysis or visualization interfaces.
[0092] FIG. 3G illustrates an enhanced motor assessment interface 312 of the web-based of the Streamlit-based web application of the system 100 for classification of Parkinson’s disease, according to an embodiment of the present invention. The enhanced motor assessment interface 312 may be configured to evaluate fine motor control through spiral or wave drawing analysis. The enhanced motor assessment interface 312 may include a navigation panel positioned along a lateral portion of the display, providing selectable options including Home, Learn, Detect Models, Chatbot, Research, and About. The navigation panel may further include selectable detection models such as MRI brain scan analysis, drawing-based motor assessment, gait analysis, and speech analysis. The enhanced motor assessment interface 312 may include an input options section allowing a user to either upload a drawing or perform a real-time drawing using an interactive canvas. In the illustrated embodiment, a spiral drawing pattern is displayed within the drawing area for analysis. In an embodiment of the present invention, the enhanced motor assessment interface 312 may further include an analysis results section configured to present the outcome of the drawing analysis. The results may indicate a normal drawing pattern, accompanied by a confidence score and extracted motor-related features such as tremor score and entropy. The results may be visually highlighted using color-coded indicators to convey motor function status. Embodiments of the present invention are intended to include or otherwise cover any type of drawing-based motor control assessment, visualization, or feedback mechanisms.
[0093] FIG. 3H illustrates a first clinical gait analysis interface 314 with normal result of the Streamlit-based web application of the system 100 for classification of Parkinson’s disease according to an embodiment of the present invention. The first clinical gait analysis interface 314 may be configured to evaluate gait parameters using wearable sensor data or structured datasets. The first clinical gait analysis interface 314 may include a gait parameters section providing multiple input methods, including manual parameter entry, CSV file upload, or a quick test option. In the illustrated embodiment, a CSV file containing gait features is uploaded for analysis. The first clinical gait analysis interface 314 may further include an analysis results section displaying a normal gait classification, indicating that gait parameters fall within healthy or expected ranges. The results may include an overall risk score, a risk level indicator, and a count of identified findings. A component analysis section may be provided to present a breakdown of individual gait-related metrics contributing to the overall assessment. The results may be visually emphasized using green or other affirmative indicators to convey low clinical risk. Embodiments of the present invention are intended to include or otherwise cover any type of gait data ingestion, processing, or normal-range assessment interfaces.
[0094] FIG. 3I illustrates a second clinical gait analysis interface 316 with moderate gait changes detection interface of the Streamlit-based web application of the system 100 for classification of Parkinson’s disease, according to an embodiment of the present invention. The second clinical gait analysis interface 316 with moderate gait changes detection may be configured to assess manually entered gait parameters and detect deviations from normative patterns. The second clinical gait analysis interface 316 with moderate gait changes detection may allow a user to manually input gait parameters such as cadence, stride length, and stride velocity using adjustable controls or sliders. Parameters exceeding predefined thresholds may be visually highlighted to indicate potential abnormalities. The analysis results section may display a warning indication identifying moderate gait changes detected. The second clinical gait analysis interface 316 with moderate gait changes detection may further present an overall risk score, a moderate risk level classification, and the number of detected findings requiring attention or monitoring. The second clinical gait analysis interface 316 with moderate gait changes detection may additionally include a component analysis section presenting disease-specific gait markers, such as a Parkinson’s gait markers score, to assist in identifying clinically relevant movement patterns. Embodiments of the present invention are intended to include or otherwise cover any type of gait abnormality detection, risk stratification, or clinical decision-support visualization mechanisms.
[0095] FIG. 3J illustrates a first speech analysis interface 318 of the Streamlit-based web application of the system 100 for classification of Parkinson’s disease, according to an embodiment of the present invention. The first speech analysis interface 318 may be rendered via a web application framework and may be configured to enable acquisition of voice data for speech-based neurological assessment. The first speech analysis interface 318 may include a navigation panel positioned on a lateral portion of the display, providing selectable options including, but not limited to, Home, Learn, Detect Models, Chatbot, Research, and About. The navigation panel may enable a user to navigate between different functional modules of the system 100. The first speech analysis interface 318 may further include a detection model selection section allowing a user to choose among multiple diagnostic modalities, such as MRI Brain Scan, Drawing Test, Gait Analysis, and Speech Analysis.
[0096] In an embodiment of the present invention, the central portion of the first speech analysis interface 318 may present a speech analysis workspace comprising selectable input options for recording audio or uploading an audio file. The first speech analysis interface 318 may include a voice recording control enabling the user to initiate and terminate real-time audio capture through a microphone-enabled device. Upon successful recording of speech input, the first speech analysis interface 318 may display a confirmation indicator notifying the user that the audio data has been received by the system 100. The first speech analysis interface 318 may further include a model status panel indicating the availability or operational readiness of various detection models associated with the platform.
[0097] FIG. 3K illustrates a second speech analysis interface 320 of the Streamlit-based web application of the system 100 for classification of Parkinson’s disease according to an embodiment of the present invention. The second speech analysis interface 320 may be configured to present recorded speech information along with diagnostic analysis results generated by the system 100. The second speech analysis interface 320 may include a playback section enabling a user to review the recorded speech sample. The playback section may comprise media controls including play, pause, duration display, and volume adjustment. In certain embodiments, the second speech analysis interface 320 may display system-generated notifications or warnings related to audio preprocessing or data handling. Such notifications may be informational in nature and may not affect the diagnostic outcome presented to the user. The speech analysis interface 320 may further include an analysis results section configured to display the output of the speech-based Parkinson’s disease detection model. The results may be presented as a classification outcome indicating a neurological condition status, such as Healthy or Indicative of Parkinson’s disease. The second speech analysis interface 320 may display a confidence level associated with the generated classification result. The confidence level may be expressed as a numerical or percentage value representing the estimated reliability of the system 100. Embodiments of the present invention are not limited to the specific layout or visual representation and may include variations in arrangement, labelling, or presentation while maintaining substantially similar functionality.
[0098] FIG. 4 illustrates a flow chart of a method 400 for classification of Parkinson’s disease, according to an embodiment of the present invention.
[0099] At step 402, the system 100 may receive the magnetic resonance imaging data, the speech recordings, and the gait sensor data from the data acquisition unit 102.
[00100] At step 404, the system 100 may process the magnetic resonance imaging data, the speech recordings, and the gait sensor data through the preprocessing unit 104.
[00101] At step 406, the system 100 may generate the modality-specific prediction scores through the classification unit 106.
[00102] At step 408, the system 100 may compare the modality-specific prediction scores with the predefined threshold. Upon comparison, if the modality-specific prediction scores exceed the predefined threshold, then the method 400 may proceed to a step 410. Else, the method 400 may revert to the step 402.
[00103] At step 410, the system 100 may optimize the hyperparameters and generate the final diagnostic prediction through the fusion unit 108.
[00104] At step 412 the system 100 may generate the clinical reports and secure deployment outputs through the reporting unit 110. , Claims:CLAIMS
I/We Claim:
1. A system (100) for classification of Parkinson’s disease, the system (100) comprising:
a data acquisition unit (102) adapted to receive magnetic resonance imaging data, speech recordings, and gait sensor data associated with a subject;
a preprocessing unit (104) adapted to preprocess the magnetic resonance imaging data through image resizing, normalization, and augmentation, extract acoustic features from the speech recordings, and perform temporal gait analysis on the gait sensor data;
a classification unit (106) adapted to classify processed magnetic resonance imaging data using an EfficientNet-B4 convolutional neural network, classify processed speech features using ensemble classifiers, and classify processed gait features using a machine learning classifier to generate modality-specific prediction scores; and
a processing unit (112), characterized in that the processing unit (112) is configured to:
receive magnetic resonance imaging data, speech recordings, and gait sensor data from the data acquisition unit (102);
process the magnetic resonance imaging data, the speech recordings, and the gait sensor data through the preprocessing unit (104);
generate modality-specific prediction scores through the classification unit (106);
optimize hyperparameters and generate a final diagnostic prediction through a fusion unit (108), when the modality-specific prediction scores exceed from a predefined threshold; and
generate clinical reports and secure deployment outputs through a reporting unit (110).
2. The system (100) as claimed in claim 1, wherein the fusion unit (108) is adapted to optimize hyperparameters associated with the classification unit (106) using a Horned Lizard Optimization algorithm and perform weighted late fusion on the modality-specific prediction scores to generate the final diagnostic prediction associated with Parkinson’s disease.
3. The system (100) as claimed in claim 1, wherein the reporting unit (110) is adapted to generate a clinical output report comprising confidence scores and interpretability information, encrypt subject data, anonymize personally identifiable information, and provide secured deployment through a Streamlit-based web application.
4. The system (100) as claimed in claim 1, wherein the multimodal deep learning architecture is configured to support modular addition of further clinical modalities for Parkinson’s disease diagnosis.
5. The system (100) as claimed in claim 1, wherein the ensemble classifiers comprise a Random Forest classifier and a CatBoost classifier.
6. The system (100) as claimed in claim 1, wherein the processing unit (112) is configured to store the final diagnostic prediction and associated interpretability information in a secured clinical database.
7. The system (100) as claimed in claim 1, wherein the processing unit (112) is configured to generate the final diagnostic prediction in real time through simultaneous processing of magnetic resonance imaging data, speech recordings, and gait sensor data.
8. The system (100) as claimed in claim 1, wherein the reporting unit (110) is adapted to encrypt subject data using an AES-256 encryption technique before storage and processing.
9. The system (100) as claimed in claim 1, wherein the reporting unit (110) is adapted to generate graphical representations of prediction confidence levels and modality contribution values.
10. A computer-implemented method (400) for classification of Parkinson’s disease using a multimodal deep learning architecture, the method (400) is characterized by steps of:
receiving magnetic resonance imaging data, speech recordings, and gait sensor data from a data acquisition unit (102);
processing the magnetic resonance imaging data, the speech recordings, and the gait sensor data through the preprocessing unit (104);
generating modality-specific prediction scores through the classification unit (106);
optimizing hyperparameters and generate a final diagnostic prediction through a fusion unit (108), when the modality-specific prediction scores exceed from a predefined threshold; and
generating clinical reports and secure deployment outputs through a reporting unit (110).

Date: May 12, 2026
Place: Noida

Nainsi Rastogi
Patent Agent (IN/PA-2372)
Agent for the Applicant