Integrated clinical embedded ai system for real-time label-free disease detection: a platform-independent approach with multispectral analysis

Abstract

A system and method for real-time, label-free, in vitro, and cost-effective early disease detection and classification are disclosed. The system integrates multiple spectroscopic techniques, coupled with a firmware¬ controlled terminal gateway and an advanced clinical Al application. It features a sensor unit for sample stability and an actuator to ensure uniform metabolite distribution in liquid samples. The embedded controller dynamically optimizes spectroscopy parameters and processes data in conjunction with the clinical Al application. This enables the precise detection of diseases, including asymptomatic stages. The system's platform-independent architecture ensures adaptability to diverse healthcare settings, such as point-of-care diagnostics and remote environments, while reducing operational complexity and costs. Its innovative design revolutionizes disease monitoring and diagnostics, enabling early intervention.

Description

INTEGRATED CLINICAL EMBEDDED Al SYSTEM FOR REAL-TIME LABEL-FREE DISEASE DETECTION: A PLATFORM-INDEPENDENT APPROACH WITH MULTISPECTRAL ANALYSISTECHNICAL FIELD

[0001] The present invention relates to medical diagnostics and healthcare technology, providing an advanced integrated system and method that combines multispectral analytical techniques, including Raman spectroscopy, infrared (IR) spectroscopy, and mass spectroscopy, with clinical artificial intelligence (Al) application modules. This system enables real-time, label-free, and non-invasive detection and classification of diseases by analyzing biological changes at the micro level within cells and tissues. It offers detailed insights into molecular structures, vibrational modes, and biochemical compositions, supporting comprehensive multi-omics analysis such as genomics, proteomics, metabolomics, and lipidomics. The system identifies subtle genetic, epigenetic, and metabolic alterations, enhancing diagnostic accuracy through ensemble deep learning and reinforcement learning for interpreting complex multispectral datasets. Firmware-controlled gateways ensure seamless hardware integration, dynamic data processing, and improved real-time performance, making it a highly efficient solution for early disease detection and analysis.BACKGROUND ART

[0002] In the clinical practice or the medical domain, the conventional methods for analyzing molecular-level changes in fields such as lipidomics, metabolomics, proteomics, glycomics, and fluxomics often rely on complex, time-consuming, and costly operations, including extensive sample preparation and labeling. Several existing patents have contributed to the development of diagnostic systems using spectroscopic techniques. For instance, W02005052558A1 discloses a method for classifying tissue based on features derived from Raman spectra and background fluorescence spectra. This method is particularly applicable for screening skin cancer and involves the use of principal component analysis (PCA) and linear discriminant analysis (LDA) to classify tissues as abnormal or normal. However, this patent focuses only on the use of Raman and background fluorescence spectra, lacking integration with other spectroscopic techniques and Al-driven data processing for broad disease detection.

[0003] Similarly, WO2014027967A1 describes a diagnostic instrument using Raman spectroscopy for tissue measurement, where light is transmitted and collected from the tissue using optical fibers. The collected light is analyzed to identify abnormal tissue. However, this patent is limited to Raman spectroscopy and does not integrate additional spectroscopic methods or Al models for real-time analysis and classification of a wide range of diseases.

[0004] Additionally, US20030081206A1 introduces a multi-pass sampling system for Raman spectroscopy to enhance the collected signal and improve the system's sensitivity. The system utilizes optical elements such as mirrors and filters for enhancing the Raman signal but focuses primarily on signal enhancement, rather than offering a multi-spectral or Al-based comprehensive diagnostic solution.

[0005] Prior systems, though advanced, lack a unified multi-spectral approach and real-time Al-driven analysis necessary for accurate and early-stage disease detection. The present invention integrates Raman spectroscopy, infrared spectroscopy, and mass spectrometry to enable detailed analysis of biological samples at the micro level, providing insights into molecular structures, functional groups, and metabolites leading to multi-omics screening. The system is adaptable to diverse healthcare settings, including point-of-care diagnostics and remote environments, and offers a cost-effective and efficient solution for detecting diseases, even in asymptomatic stages. By combining advanced analytical techniques with real-time Al-based analysis, the invention delivers a comprehensive approach to disease detection and addresses the limitations of prior technologies.SUMMARY OF THE INVENTION

[0006] The present invention discloses a novel system and method for realtime, label-free, non-invasive, and cost-effective disease detection and classification, utilizing advanced clinical Al application modules and multispectral analytical techniques, including Raman spectroscopy (surface- enhanced Raman spectroscopy), IR spectroscopy, and mass spectrometry. The system enables rapid and precise diagnostics at the molecular level without the need for extensive sample preparation or biomarker labelling, addressing challenges in early-stage disease detection.

[0007] It is therefore the primary objective of the invention is to provide a platform-independent system and method for real-time, label-free, and in vitro detection and classification of diseases, offering high accuracy and efficiency while reducing diagnostic costs and complexity. The system is capable of analyzing multi-omics data, including lipidomics, metabolomics, proteomics, glycomics, and fluxomics, enabling comprehensive profiling of disease-related molecular alterations. This multi-omics approach allows for deeper insights into the biochemical and molecular constituents of diseases, enhancing the system’s diagnostic capabilities across a broad spectrum of conditions.

[0008] It is yet another object of the present invention to provide a method that incorporates a clinical Al application module, employing multitier ensemble deep learning and reinforcement learning models to dynamically adapt to novel metabolomic profiles, ensuring reliable and scalable diagnostics over time.

[0009] Another object of the invention is to optimize spectroscopic parameters such as laser power, ionization settings, and interferometer resolution using a firmware-controlled terminal gateway, thereby enhancing data acquisition accuracy across diverse biological samples.

[0010] The invention further includes specialized sensors, such as temperature, humidity, and light spectrum analyzers, along with actuators like sonic vibrators, to ensure effective environmental monitoring and uniform metabolite distribution, thereby improving sample reliability and reducing processing time. Additionally, an embedded controller serves as the analytical core, capable of automatically optimizing parameters for spectroscopic modalities. This enables real-time performance, making the system suitable for use in clinical, laboratory, and remote healthcare settings.DESCRIPTION OF ACCOMPANYING FIGURES

[0011] FIG.1 Block Diagram of the System and Method Architecture for Real- Time, Label-Free, and Non-lnvasive Disease Detection

[0012] FIG.2 Detailed Components of the Sensor Unit, Including Temperature, Humidity, and Light Spectrum Analyzers

[0013] FIG.3 Components of the Spectroscopy Unit, Featuring Raman (SERS), Infrared (IR), and Mass Spectrometry Modules

[0014] FIG.4 Components of the Embedded Controller, Comprising Memory, Control, and Processing Units

[0015] FIG. 5 Components of the clinical Al Application Module, with multitier Deep Learning and Reinforcement Learning models included in the processing unitDETAILED DESCRIPTION OF EMBODIMENTS

[0016] An automated system and method have been developed, the invention addresses the critical need for early-stage, real-time, label-free, and cost-effective disease detection by providing a platform-independent diagnostic system capable of identifying and classifying diseases, including those in asymptomatic stages. This advanced system integrates Clinical Al Application (7) modules with multispectral analytical techniques, such as Raman spectroscopy, IR spectroscopy, and mass spectrometry, to offer a comprehensive, non-invasive diagnostic approach.

[0017] Leveraging the strengths of these multi-spectroscopic methodologies driven by the Clinical Al Application (7) module and Firmware (5) controlled Terminal gateway (6), the system provides detailed insights into molecular structures, vibrational modes, functional groups, and biochemical compositions. Raman spectroscopy detects molecular vibrations and subtle biochemical changes, IR spectroscopy analyzes functional groups, and mass spectrometry profiles metabolites and biomolecules at the micro-level with precision. Together, these techniques enable a holistic, multi-omics approach, encompassing genomics, proteomics, metabolomics, and lipidomics, to understand disease mechanisms and progression.

[0018] Raman spectroscopy, I spectroscopy, and mass spectrometry each play a crucial role in analyzing lipidomics, metabolomics, proteomics, glycomics, and fluxomics. Raman spectroscopy identifies molecular vibrations, aiding lipidomic and glycomic profiling by detecting lipid species and glycan structures. IR spectroscopy reveals functional groups in lipids, proteins, and metabolites, enhancing the study of lipidomic, proteomic, and metabolomic profiles. Mass spectrometry provides high-resolution identification and quantification of metabolites, lipids, proteins, and glycans, offering detailed insights into metabolic pathways, protein structures, and fluxomics by tracking metabolic transformations. Together, these techniques along with the clinical Al application module enable comprehensive multi-omics analysis, and resulting to the output stage (8) for clinical purposes.

[0019] The system’s ability to detect genetic, epigenetic, and metabolic alterations makes it effective for diagnosing a wide range of medical conditions, including cancers, neurodegenerative disorders, cardiovascular diseases, metabolic disorders, and infectious diseases. By eliminating the need for extensive sample preparation or biomarker labelling, the invention enhances accessibility, reduces diagnostic time, and improves cost efficiency.

[0020] This innovative platform combines advanced clinical Al module with cutting-edge analytical techniques, providing a transformative tool for early- stage disease detection and monitoring across clinical, laboratory, and remote healthcare settings. This system’s versatility, accuracy, and adaptability make it a transformative solution for modem healthcare, offering precise diagnostics in clinical, laboratory, and remote settings without the need for extensive sample preparation or biomarker labelling.

[0021] The system's diagnostic capabilities are further enhanced by clinical Al application driven ensemble deep learning (23) module in FIG.1 and FIG.5, trained on extensive datasets to identify intricate patterns and correlations in molecular changes. These modules adapt dynamically to new metabolomic profiles using the reinforcement learning (24) module in FIG.1 and FIG.5, ensuring robust and accurate diagnostics over time. The firmware (6) controlled terminal gateway (5) seamlessly integrates sensor unit (3), actuator (11), and spectroscopies unit (1 ), enabling real-time data processing and optimal management of multispectral inputs. This coordination ensures the system is adaptable to a variety of diagnostic workflows and healthcare settings.

[0022] Any part of the spectroscopy unit can operate independently in remote settings, providing unmatched flexibility. For instance, Raman or IR spectroscopy can be deployed in field conditions to collect spectral data. The Clinical Al Application (7) module correlates these spectral features with mass spectrometry data using unique molecular-level correlation factors. This process ensures precise molecular profiling and anomaly detection, even when different spectroscopy units are geographically dispersed.

[0023] The Clinical Al Application (7) module identifies metabolites indicative of anomalies and correlates their spectral features across modalities, delivering a comprehensive diagnostic output. This capability enables the system to function as a platform-independent diagnostic solution, suitable for diverse environments. For example, Raman and IR spectroscopy performed in remote locations can provide initial insights, which are subsequently refined and validated against mass spectrometry data through Al-driven molecular analysis.

[0024] This decentralized platform-independent architecture ensures accurate disease detection and classification while maintainingscalability and accessibility, making the system invaluable for point-of- care diagnostics, remote healthcare settings, and large-scale epidemiological studies.

[0025] The platform-independent embedded clinical Al Application (7) module and a firmware (6) module-controlled terminal gateway (5) that connects to diverse peripherals, sensors unit (3), and spectroscopic unit (1) as in FIG.1. The system is designed to detect diseases in various samples, including asymptomatic conditions, by analyzing chemical profiling in the early stages.

[0026] The system and method involve multiple peripheral units, including sensor unit (3) referring to FIG.1 and FIG.2, actuators (11), embedded controllers (9), and spectroscopies (1). The sensor unit comprises a temperature sensor (13), humidity sensor (14), and light spectrum analyzer sensor (15). The temperature sensor (13) measures the sample analysis unit called the input (4) module’s temperature attached to the spectroscopy's transmitter-receiver end, application through the connected terminal gateway (5). The input (4) module is associated with the spectroscopy unit (1), sensor unit (3), user input (12), and actuator(11), thus enabling efficient spectral read-out of input sample (2). The humidity sensor (14) analyzes the humidity of the environment in the input (2) sample analysis unit and sends the values to the processing unit (18) referring to FIG.4 of the embedded controller (9), informing the user for accurate analysis. The light spectrum analyzer sensor (15) measures electromagnetic wave interference in the sample analysis unit and conveys the interference values to the processing unit (18) for precise analysis, these sensor analytical values are transmitted to the processing unit (18) in the embedded controller (9) referring to FIG.1 and FIG.4 associated with the clinical Al application (7) module where processing unit (18) warns the system if any variations in the sensor measurement exist.

[0027] Incorporated within the system is an actuator (11) attached to the body of the input (4), specifically a sonic vibrator, which works to distribute metabolites uniformly within liquid samples, reducing the sample incubation time. The spectroscopy unit (1) includes various spectroscopies like Raman spectroscopy (19) specifically surface- enhanced Raman spectroscopy (SERS), IR spectroscopy (20), andMass spectroscopy (21), operating collaboratively or independently based on the sample nature and investigation parameters.

[0028] The embedded controller (9), serving as the system's core, comprises a memory unit (16), control unit (17), and processing unit (18) referring to (10) of FIG.1 and particularly in FIG.4. It can function as a computer, executing tasks related to the analysis. When an input sample (2) is introduced into the analysis unit within the input unit (4), the combined action of the spectroscopies sends analytical values to the system. The processing unit (18) analyzes the spectral values, determining the input sample's nature (solid or liquid). Based on user input (12) and the sample's nature, the control unit (17) in the embedded controller (9) directs through the terminal gateway (5) the combination of spectroscopies or single spectroscopy to perform the required analytical tasks.

[0029] The processing unit (18) plays a crucial role in configuring control parameters based on the nature of the sample under investigation. These control parameters are subsequently transmitted through the terminal gateway (5) by the control unit (17), a process executedautomatically by the embedded controller (9). These control parameters are sent to the spectroscopy unit (1) as in FIG.1 and FIG.3 for effective analysis by the corresponding spectroscopy / spectroscopies. Specifically, for Raman spectroscopy (19), control parameters encompass laser- related factors such as power, integration time, and wattage. In the realm of mass spectrometry (21), the system dynamically adjusts various parameters, including ionization source elements like spray voltage and solvent flow rate, mass analyzer parameters such as ion acceleration voltage and RF / DC voltages, as well as detector settings like channeltron voltage or conversion dynode voltage.

[0030] For IR spectroscopy (20), the processing unit takes charge of managing an array of parameters. These include source-related aspects such as type, sample handling characteristics like thickness and state, interferometer settings encompassing mirror speed and resolution, as well as detector specifications like type and temperature. Additionally, the processing unit oversees supplementary accessories such as ATR pressure or gas cell conditions in this context. The automated adjustment of these parameters is instrumental in ensuring optimal conditions fordata acquisition, a process that significantly contributes to enhancing the accuracy and sensitivity of the analytical techniques employed.

[0031] Raman spectroscopy (19), specifically SERS, which relies on the inelastic scattering of monochromatic light, stands out for its ability to provide detailed information about molecular vibrations. This technique excels in examining biochemical changes at the molecular level, offering valuable insights into the structures and compositions of cellular components. In contrast, IR spectroscopy (20) focuses on molecular vibrations and absorptions, delivering comprehensive details about functional groups within molecules. Particularly useful for identifying biomolecular structures and alterations, IR spectroscopy serves as a valuable tool in disease classification efforts. Meanwhile, Mass spectroscopy (21) measures the mass-to-charge ratio of ions, enabling precise identification of molecules based on their mass and fragmentation patterns. This technique excels in the detection and quantification of metabolites, proteins, and peptides in complex biological samples.

[0032] The firmware (6) is a specialized set of hardware instructions embedded within the embedded controller (9) and connected to the terminal gateway (5). Its primary function is to seamlessly integrate and control the sensors, spectroscopies, and user input (12) through the terminal gateway, enhancing overall system performance. The firmware facilitates communication among these units by providing instructions for data collection to sensors, coordinating actuator actions based on system requirements, and managing the embedded controller for overall system control. It also interfaces with spectroscopic devices for accurate data analysis. This integration is crucial in point-of-care healthcare systems, where rapid and reliable diagnostics are essential, ensuring efficient collaboration among system components, including the Clinical Al Application (7) module, for effective operation in real-world healthcare scenarios.

[0033] Moreover, the firmware's integration with the gateway terminal is crucial for effective functioning, especially in settings like point-of-care healthcare systems. The clinical Al application module's processing unit (22) employs multitier ensemble deep learning modules (23) for disease classification and metabolomic profiling. Reinforcement learningmodules (24) further enhance diagnostic accuracy by adapting to new metabolomic characteristics with additional user input labeling.

[0034] The deep learning modules (23) are trained individually for each category class, and these models are stored in the memory unit (16) of the embedded controller. Each classification model is specifically designed with trained spectral features corresponding to its respective disease classes. During the inference process, when a new sample is under investigation, the control unit (9) directs the spectroscopies with the results obtained from the sensors and retrieves the readout of the spectral features of the subjected sample.

[0035] The processing unit (18) within the embedded controller then processes these spectrogram features channeled through the terminal gateway and quantitatively compares them with the spectral features within the pre-trained classification models in the memory unit (16). Given that the spectrogram encompasses a wide range of molecular fingerprints, the system can detect all or part of the abnormalities present in the sample.Y1

[0036] In cases where the spectrogram of the sample is novel and unfamiliar to the system, the support of user input (12) (Label) is employed. The new spectral features of the sample are then trained using a reinforcement learning module (24). This training process is effective and is appropriately combined with the relevant results obtained from the deep learning classifiers. This integration occurs during the processing phase by the processing unit of the embedded controller.

[0037] The system’s real-time capabilities result from the integration of sensor units, spectroscopy units, and the clinical Al application, enabling label-free disease detection without extensive sample preparation. This approach is cost- effective and efficient, making it suitable for diverse settings such as point-of- care systems and laboratories. The firmware coordinates these components, ensuring adaptability and reliability across different environments, including emergency diagnostics. In critical care, the system provides real-time insights for swift decision-making, improving patient outcomes. Additionally, its elimination of sample preparation and biomarker labeling reduces diagnostic time and costs, addressing challenges in resource-limited settings and minimizing reliance on highly trained personnel.

Claims

CLAIMS1. A real-time, label-free, and early diagnostic system for health monitoring, comprising:• a spectroscopy unit (1) integrating Raman spectroscopy, infrared spectroscopy, and mass spectrometry;• a sensor unit (3) comprising temperature, humidity, and light spectrum analysers to monitor environmental conditions of the analytical area called input during analysis;• an actuator (11) comprising a sonic vibrator to ensure uniform distribution of metabolites in liquid samples for investigation;• an embedded controller (9) comprising a memory unit, a control unit, and a processing unit configured to optimize spectroscopy parameters based on input samples;• a firmware-controlled terminal gateway to facilitate seamless integration and data exchange between the sensor unit, actuator, spectroscopy unit, and embedded controller; and• a clinical Al application module (7) employing multitier ensemble deep learning and reinforcement learning models for disease classification and metabolomic profiling,• wherein the system enables real-time, label-free, non- invasive disease detection across diverse clinical and remote healthcare settings.

2. The system as claimed in claim 1 , wherein the spectroscopy unit dynamically adjusts parameters, including laser power for Raman spectroscopy, ionization settings for mass spectrometry, and interferometer resolution for IR spectroscopy by the embedded controller, to ensure optimal data acquisition and diagnostic accuracy based on the nature of input sample.

3. The system as claimed in claim 1 , wherein the clinical Al application module correlates spectral features across Raman, IR, and mass spectrometry modalities, providing a holistic molecular analysis for disease detection and classification.

4. The system as claimed in claim 1 , wherein the deep learning module is pre-trained with spectral features for multiple disease classes, stored in the memory unit of the embedded controller, and dynamically updated using reinforcement learning for novel spectrograms.

5. The system as claimed in claim 1 , wherein the firmware facilitates real-time processing and control by:• configuring spectroscopy parameters based on sample properties by analysing spectrogram features;• integrating sensor measurements to adapt to environmental variations; and• synchronizing data flow between the spectroscopy unit and clinical Al application module.

6. The system as claimed in claim 1 , wherein the sensor unit, is configured to monitor environmental conditions where the sample is kept during analysis and transmit analytical values to the processing unit (18). The sensor unit alerts the system to variationsin these conditions, enabling automatic adjustments to maintain diagnostic precision and reduce analysis errors.

7. A method for real-time, label-free disease detection and classification, comprising:• introducing a biological sample into the input (4) of the system associated with the spectroscopy unit for spectral analysis;• dynamically optimizing spectroscopy parameters through an embedded controller based on sensor feedback and sample properties;• analyzing spectral data using a clinical Al application module pre-trained with disease-specific spectral features; and• classifying the sample and identifying potential anomalies through ensemble deep learning and reinforcement learning and the result is passed to the output stage for clinical interpretation.

8. The method as claimed in claim 7, wherein the spectroscopy unit operates independently or collaboratively in remote or clinical settings, enabling scalable and flexible diagnostic workflows.

9. The method as claimed in claim 7, further comprises the step of deploying sonic vibrations to ensure homogenous distribution of metabolites within liquid samples for accurate analysis.

10. The system or method as claimed in any preceding claims, wherein the diagnostic capability extends to a wide range of diseases, including but not limited to cancers, neurodegenerative disorders, metabolic conditions, inflammatory and infectious diseases, and even in asymptomatic stages.

Images