System for AI-supported segmentation and quantitative analysis of multi-contrast magnetic resonance imaging data

Abstract

A system for AI-supported segmentation and quantitative analysis of multi-contrast magnetic resonance imaging data, consisting of: a housing configured to enclose and support a variety of electronic components; a housing-mounted data acquisition unit configured to receive multi-contrast magnetic resonance imaging datasets with a variety of imaging contrasts via one or more communication interfaces; a preprocessing processor that is operationally coupled with the data acquisition unit and is configured to perform the spatial alignment, intensity normalization, and cross-contrast registration of the received data sets to produce a uniform volumetric representation; a storage unit that is electrically connected to the preprocessing processor and is configured to store the received data sets, the data processed in the meantime, and the trained model parameters; a segmentation processor arranged in the housing, which is operationally connected to the preprocessing processor and the storage unit, wherein the segmentation processor includes a parallel computing circuit configured to execute trained artificial intelligence models to generate voxel-level segmentation outputs corresponding to anatomical or pathological regions based on the uniform volumetric representation; a feature quantification unit that is operationally coupled with the segmentation processor and configured to calculate quantitative parameters such as volume measurements, boundary features, and intensity distribution metrics from the segmented regions; a visualization unit that is operationally coupled with the segmentation processor and the feature quantification unit and is configured to generate visual overlays, multidimensional representations, and structured output data according to the segmented regions and calculated quantitative parameters; and a communication interface configured to transmit the structured output data to external systems, where the preprocessing processor and the segmentation processor are configured to work together in a coordinated manner to perform cross-contrast data fusion to improve segmentation accuracy and quantitative analysis.

Description

Application area of ​​the invention:

[0001] The present disclosure relates to the field of medical imaging systems and computer-aided diagnostic devices, in particular a structurally integrated machine for AI-assisted segmentation and quantitative analysis of multi-contrast magnetic resonance imaging (MRI) data. The disclosure specifically relates to a hardware-based image analysis system with dedicated processing circuitry, memory architecture, and sensor-based acquisition modules for extracting clinically relevant features from multiparametric MRI datasets and for generating quantifiable biomarkers for diagnostic and monitoring applications. Background of the invention

[0002] Magnetic resonance imaging (MRI) is frequently used for the non-invasive visualization of internal anatomical structures and offers various contrast techniques such as T1-weighted, T2-weighted, FLAIR, and diffusion-weighted imaging. Although these techniques provide complementary diagnostic information, the interpretation of such multi-contrast datasets remains heavily reliant on manual segmentation and subjective assessment by clinicians. This process is time-consuming, prone to inter-examiner discrepancies, and often insufficient to extract subtle quantitative biomarkers needed for early disease detection and monitoring.

[0003] Existing computer-aided tools for MRI analysis are predominantly software-based and exhibit insufficient integration with dedicated hardware architectures. This limits processing speed, scalability, and reliability in real-time clinical operation. Furthermore, conventional segmentation methods often fail to effectively utilize correlations between contrasts, reducing the accuracy in delineating pathological areas such as tumors, lesions, or degenerative tissue boundaries.

[0004] There is a need for a robust, machine-based system that integrates the acquisition of multi-contrast MRI data with powerful, hardware-accelerated artificial intelligence, enabling precise segmentation and quantitative analysis. Such a system should minimize manual intervention, improve reproducibility, and deliver clinically relevant metrics through a structurally defined and functionally coherent device.

[0005] Magnetic resonance imaging (MRI) has become one of the most powerful and widely used modalities for non-invasive soft tissue imaging due to its superior contrast resolution and the ability to generate various image contrasts, such as T1-weighted, T2-weighted, FLAIR, DWI, and perfusion images. These multi-contrast datasets provide complementary information about tissue composition, microstructural properties, and pathological changes, making MRI particularly valuable in neurological diseases, oncology, musculoskeletal diagnostics, and cardiovascular diagnostics. However, the increasing complexity and growing volume of multiparametric MRI data present significant challenges for interpretation, especially in clinical workflows that rely heavily on manual or semi-automated analysis techniques.

[0006] Conventional MRI analysis methods typically involve the manual segmentation of anatomical or pathological regions by trained radiologists or clinicians. This process requires the inspection and contouring of each individual slice, which is not only time-consuming but also leads to intra- and inter-individual variability. Interpretational inconsistencies arise from differing expertise, fatigue, and subjective judgment, resulting in inconsistent segmentation findings and reduced reproducibility in clinical practice. Furthermore, manual segmentation is not scalable for large datasets or longitudinal studies requiring repeated measurements to monitor disease progression or treatment response.

[0007] To overcome these limitations, various semi- and fully automated segmentation methods have been developed. Traditional image processing techniques such as thresholding, region growth, edge detection, and deformable models have been applied to MRI data to facilitate segmentation. While these methods offer basic segmentation capabilities, they often rely on predefined parameters and assumptions about image intensity distributions. This makes them sensitive to noise, intensity inhomogeneities, and the variability of different MRI scanners and acquisition protocols. Consequently, these methods often require manual initialization or correction, limiting their effectiveness in fully automated workflows.

[0008] In recent years, machine learning and deep learning have gained importance in medical image segmentation. Convolutional neural networks (CNNs), including architectures such as U-Net and its variants, have significantly improved segmentation accuracy by directly learning hierarchical feature representations from image data. These models are capable of capturing complex spatial patterns and contextual information across different imaging modalities, thus enabling more precise delineation of tumors, lesions, and anatomical structures. Despite these advances, most existing implementations are predominantly software-based and utilize general-purpose computing platforms that may not be optimized for processing large datasets or for real-time clinical use.

[0009] One of the biggest limitations of existing AI-based solutions lies in their reliance on significant computing resources, including graphics processing units (GPUs) or cloud infrastructure. This reliance leads to latency, increases operating costs, and raises concerns about data privacy and security, especially when patient data needs to be transferred to external servers for processing. Furthermore, the variability of hardware configurations and software environments can result in inconsistent performance, making it difficult to standardize these solutions across different healthcare settings.

[0010] Another significant drawback of current systems is the lack of seamless integration between the acquisition and analysis of multi-contrast MRI data. Many existing solutions process each contrast modality independently or combine them using simple chaining techniques, without fully exploiting the underlying relationships between the different contrasts. This can lead to suboptimal segmentation performance, especially when pathological features are only detectable through the combined interpretation of multiple contrasts. Furthermore, inconsistencies in spatial orientation and intensity scaling between the modalities can further impair the accuracy of automated analysis.

[0011] Preprocessing steps such as image registration, normalization, and artifact correction are crucial for reliable segmentation results. However, in many conventional systems, these steps are implemented as separate software modules, often requiring manual configuration and intervention. Errors in preprocessing can impact subsequent analysis steps, leading to inaccurate segmentations and unreliable quantitative measurements. The lack of a unified, hardware-integrated preprocessing pipeline exacerbates these problems, as data transfer between modules introduces additional latency and potential sources of error.

[0012] The quantitative analysis of MRI data, including the extraction of volumetric, morphological, and intensity-based biomarkers, represents another area where existing solutions reach their limits. While some software tools offer basic measurement capabilities, they often lack the ability to consistently and reproducibly calculate advanced metrics such as texture features, diffusion parameters, and multiparametric indices. Furthermore, these tools are typically not optimized for processing large datasets or performing real-time analyses, limiting their usefulness in time-critical clinical scenarios.

[0013] Another challenge with current systems is the limited interpretability and validatability of AI-generated results. Many deep learning models function as black boxes, delivering segmentation results without comprehensible explanations of the underlying decision-making process. This lack of transparency can hinder clinical application, as users may be hesitant to rely on automated results that are not easily verified or understood. Furthermore, variations in training data and model architectures can lead to performance differences between different patient groups, raising concerns about generalizability and robustness.

[0014] Hardware aspects also play a crucial role in the implementation of MRI analysis systems. Existing solutions often rely on external computer systems that are not physically integrated into the imaging devices. This leads to fragmented workflows and increased setup complexity. The lack of dedicated hardware architectures specifically for multi-contrast MRI analysis limits the possibilities for low-latency processing and real-time feedback. Furthermore, standard hardware may not offer the necessary optimization for the efficient execution of complex AI models, resulting in increased power consumption and thermal management issues.

[0015] Data management and storage present additional challenges in current approaches. Multicontrast MRI datasets are inherently large and require efficient processing to ensure smooth data handling and rapid retrieval. Conventional systems frequently encounter data bottlenecks, particularly when transferring high-resolution volume data between storage units and processing modules. This can lead to delays and reduced system throughput, especially in high-volume clinical environments.

[0016] In summary, while significant progress has been made in the development of automated and AI-based MRI analysis methods, existing solutions have several limitations. These include reliance on manual intervention, a lack of hardware integration, inefficiencies in multi-contrast data fusion, high computational costs, and limited capabilities for robust quantitative analyses. These drawbacks underscore the need for a structurally integrated system that combines dedicated hardware components with advanced AI processing capabilities to enable efficient, accurate, and reproducible segmentation and analysis of multi-contrast MRI data in routine clinical practice. Summary of the invention

[0017] This disclosure describes a system for AI-assisted segmentation and quantitative analysis of multi-contrast magnetic resonance imaging (MRI) data. The system comprises a robust housing with a multi-channel data acquisition interface, a preprocessing unit, a hardware-accelerated inference processor, a feature quantification module, and a visualization interface. It receives multi-contrast MRI datasets from an external imaging device or storage medium via standardized communication interfaces and processes them using a sequence of hardware-executed operations.

[0018] The preprocessing engine is configured to spatially normalize the received MRI datasets, harmonize the intensity, and register across modalities, thereby generating aligned, multidimensional image volumes. The hardware-accelerated inference processor includes parallel processing cores that execute trained artificial intelligence models, including convolutional neural networks and attention-based architectures, for automated segmentation of anatomical and pathological regions in the multi-contrast input data.

[0019] The feature quantification module calculates volumetric, morphological, and intensity-based parameters of the segmented regions, including lesion volume, interface irregularities, diffusion metrics, and tissue heterogeneity indices. The visualization interface is operationally linked to the processing unit and presents segmented overlays, three-dimensional reconstructions, and quantitative reports for user interpretation.

[0020] The system thus offers a coherent machine-level implementation that increases segmentation accuracy, reduces processing latency, and enables standardized quantitative analysis of multi-contrast MRI data.

[0021] The main objective of the present invention is a system in the form of an integrated device for AI-assisted segmentation and quantitative analysis of multi-contrast magnetic resonance imaging (MRI) data. The system overcomes the limitations of manual and semi-automated analyses by enabling precise, reproducible, and rapid delineation of anatomical and pathological regions. A further objective of the invention is a hardware-integrated architecture comprising a data acquisition interface, preprocessing circuits, high-performance processors, and visualization modules. This ensures seamless and deterministic processing of multi-contrast MRI datasets within a single, structurally unified device.

[0022] A further objective of the invention is to provide a system for cross-modality alignment and normalization of multiple MRI contrasts using dedicated preprocessing hardware. This ensures consistency in the spatial and intensity domains prior to segmentation. Another objective of the invention is the efficient execution of artificial intelligence models through hardware-accelerated processing units configured for parallel computations. This reduces latency and enables near real-time analysis suitable for clinical environments. A further objective of the invention is to facilitate precise segmentation by utilizing hardware-level multi-contrast data fusion mechanisms. This improves the detection and delineation of subtle pathological features that may not be visible in individual imaging modalities.

[0023] A further objective of the invention is to provide a quantitative analysis module for calculating volumetric, morphological, and intensity-based parameters from segmented regions. This allows for the extraction of clinically relevant biomarkers for diagnosis, prognosis, and therapy monitoring. Another objective of the invention is to provide an integrated visualization interface for displaying segmented overlays, three-dimensional reconstructions, and structured analysis results. This improves interpretability and usability for clinicians and medical professionals.

[0024] A further objective of the invention is to ensure reliability and operational stability through the integration of energy management circuits, thermal regulation mechanisms, and fault-tolerant hardware components into the device structure. Furthermore, the invention aims to minimize dependence on external computing infrastructure by providing a self-contained system that enables the continuous processing of MRI data. This increases data security, reduces transmission latency, and allows for use in various healthcare settings.

[0025] A further objective of the invention is to provide a scalable and modular system architecture that can be adapted to different imaging protocols, clinical requirements, and computational loads without any loss of performance. Finally, the invention aims to enable standardized and reproducible analysis of multi-contrast MRI datasets through tightly integrated hardware and processing workflows, thereby eliminating the fluctuations and inefficiencies present in existing solutions. BRIEF DESCRIPTION OF THE IMAGE

[0026] These and other features, aspects and advantages of the present invention will be better understood if the following detailed description is read with reference to the accompanying drawing, in which the same symbols represent the same parts: Fig. Figure 1 shows a block diagram of a system for AI-assisted segmentation and quantitative analysis of multi-contrast magnetic resonance imaging data.

[0027] Furthermore, those skilled in the art will recognize that the elements in the drawing are simplified and not necessarily drawn to scale. For example, the flowcharts illustrate the process by highlighting the main steps to facilitate understanding of the present disclosure. With regard to the construction of the device, one or more components may be represented in the drawing by conventional symbols. The drawing may show only those specific details relevant to understanding the embodiments of the present disclosure, so as not to clutter the drawing with details that are already apparent to those skilled in the art from the description contained herein. Detailed description of the invention

[0028] To facilitate understanding of the principles of the invention, reference is made below to the embodiment shown in the drawing, which is described using specific terms. It is understood, however, that this does not limit the scope of protection of the invention. Rather, modifications and further developments of the depicted system, as well as further applications of the inventive principles shown therein, are conceivable, insofar as they would normally occur to a person skilled in the art in the field of the invention.

[0029] It will be clear to those skilled in the art that the foregoing general description and the following detailed description are exemplary and explanatory of the invention and are not to be understood as a limitation thereof.

[0030] References to “an aspect”, “another aspect”, or similar phrases in this description mean that a particular feature, structure, or property described in connection with the embodiment is included in at least one embodiment of the present disclosure. Therefore, phrases such as “in one embodiment”, “in another embodiment”, and similar expressions in this description may, but do not necessarily, all refer to the same embodiment.

[0031] The terms "includes," "comprehensive," or similar expressions denote non-exclusive inclusion. Thus, a procedure or method containing a list of steps does not only include those steps but may also include further steps not explicitly listed or inherent in the procedure or method. Likewise, the statement "includes..." for one or more devices, subsystems, elements, structures, or components, without further limitations, does not preclude the existence of other devices, subsystems, elements, structures, or components.

[0032] Unless otherwise defined, all technical and scientific terms used herein have the same meanings generally known to those skilled in the art in the field to which this invention belongs. The systems, methods, and examples described herein serve only for illustration and are not to be understood as limiting.

[0033] Embodiments of the present disclosure are described in detail below with reference to the attached drawing.

[0034] Fig.Figure 1 shows a block diagram of a system for AI-assisted segmentation and quantitative analysis of multi-contrast magnetic resonance imaging (MRI) data. The system 100 comprises: a housing (102) for accommodating and mounting several electronic components; a data acquisition unit (104) located within the housing for receiving multi-contrast MRI datasets via one or more communication interfaces; a preprocessing processor (106) connected to the data acquisition unit for spatial alignment, intensity normalization, and cross-contrast registration of the received datasets to generate a uniform volumetric representation; and a storage unit (108) connected to the preprocessing processor for storing the received datasets, the data processed in the meantime, and the trained model parameters.A segmentation processor (110) housed within the enclosure, connected to the preprocessing processor and the storage unit, comprises a parallel computing circuit for executing trained AI models to generate voxel-based segmentation outputs for anatomical or pathological regions based on the uniform volumetric representation. A feature quantification unit (112) connected to the segmentation processor computes quantitative parameters such as volume measurements, boundary features, and intensity distribution metrics of the segmented regions. A visualization unit (114) connected to the segmentation processor and the feature quantification unit generates visual overlays, multidimensional representations, and structured output data for the segmented regions and the computed quantitative parameters. A communication interface (116) transmits the structured output data to external systems.The preprocessing processor and segmentation processor work together in a coordinated manner to perform cross-contrast data fusion to improve segmentation accuracy and quantitative analysis.

[0035] In one embodiment, the data acquisition unit (104) comprises a plurality of high-speed input interfaces, including optical communication ports and serial data ports, wherein the data acquisition unit further comprises a buffer circuit configured to temporarily store incoming multi-contrast data sets to enable continuous data streaming and synchronized processing across multiple image contrasts.

[0036] In one embodiment, the preprocessing processor (106) comprises programmable logic circuits configured to perform rigid and non-rigid registration across the plurality of image contrasts, wherein the preprocessing processor is further configured to perform voxel-wise interpolation and resampling operations to ensure uniform spatial resolution across the entire volumetric representation.

[0037] In one embodiment, the preprocessing processor (106) further comprises an intensity calibration circuit configured to compensate for scanner-induced fluctuations and inhomogeneities of signal intensity at different contrasts in magnetic resonance imaging, thereby enabling consistent input conditions for the segmentation processor.

[0038] In one embodiment, the segmentation processor (110) comprises a plurality of tensor computation units and matrix processing units arranged to perform parallel convolution operations, feature extraction, and hierarchical representation learning based on the trained artificial intelligence models stored in the memory unit.

[0039] In one embodiment, the segmentation processor is configured to perform feature aggregation on multiple scales by integrating information from different spatial resolutions and multiple image contrasts, thereby improving the delineation of regions with low-contrast boundaries or heterogeneous intensity profiles.

[0040] In one embodiment, the segmentation processor (110) further comprises a weighting circuit configured to assign adaptive significance to each image contrast based on local signal characteristics, thereby enabling dynamic cross-contrast fusion during segmentation.

[0041] In one embodiment, the feature quantification unit (112) comprises an arithmetic circuit configured to calculate volumetric parameters based on voxel counting, surface parameters based on boundary extraction, and statistical parameters including mean intensity, variance, and distribution profiles over segmented regions.

[0042] In one embodiment, the feature quantification unit (112) is further configured to calculate diffusion-related parameters and texture-based heterogeneity indices using spatial correlation and intensity variation analysis within the segmented regions.

[0043] In one embodiment, the visualization unit (114) comprises a graphic processing circuit configured to generate layer-wise overlays of segmented regions on original magnetic resonance imaging data and to render three-dimensional volumetric representations using reconstructed voxel data.

[0044] The system components are arranged within the housing as tangible, physically realized electronic and electromechanical structures and interconnected via conductive traces to enable deterministic signal transmission and processing. The data acquisition unit comprises physical interface circuits with transceivers, optical receivers, and serial input ports on printed circuit boards, as well as buffer elements such as capacitive and semiconductor memories for temporary data storage. The preprocessing processor and the segmentation processor are implemented as integrated circuits with logic gates, arithmetic circuits, matrix arrays, and parallel processing cores on semiconductor substrates. Each operation, including spatial alignment, normalization, convolution, and feature extraction, is performed through electrical signal propagation and state transitions within these circuits.

[0045] The storage unit consists of physical memory elements, including volatile and non-volatile semiconductor cells in addressable arrays for storing image data and parameter values. The feature quantification unit comprises dedicated arithmetic circuits with adders, multipliers, and accumulators for performing volumetric and statistical calculations using hardware operations. The visualization unit includes a graphics processing circuit, display drivers, and signal converters that generate electrical signals for image overlays and volumetric representations for output to a display device. The communication interface is implemented through network interfaces, including physical layer transceivers and protocol processing circuits for data transmission to external systems.

[0046] The system for AI-assisted segmentation and quantitative analysis of multi-contrast magnetic resonance imaging (MRI) data operates with a coordinated sequence of hardware-executed computational processes. These implement a structured technique for receiving, preprocessing, segmenting, and quantitatively analyzing volumetric image datasets. The data acquisition unit first receives multi-contrast MRI data comprising multiple image sequences corresponding to different tissue contrast mechanisms. This data is received in a digitized volume format via high-speed communication interfaces and temporarily buffered within the acquisition circuitry to ensure the synchronized availability of all contrast channels for subsequent processing.The cached data is then transferred to the preprocessing processor, which deterministically performs a series of spatial and intensity harmonization operations.

[0047] The preprocessing processor first performs spatial alignment. Each image contrast undergoes rigid transformation operations, including translation and rotation. This is followed by non-rigid deformation adjustments to compensate for anatomical misalignments and patient motion artifacts. These transformations are implemented using programmable logic circuits configured to iteratively minimize spatial discrepancy metrics between corresponding voxel intensities across different contrasts. After alignment, the preprocessing processor performs voxel-wise resampling to ensure uniform spatial resolution and raster consistency across all image datasets. Simultaneously, intensity normalization is performed.The intensity values ​​are scaled and standardized using reference distributions derived from the input data, thereby compensating for scanner-related variations and field inhomogeneities. The result of these preprocessing operations is a uniform volumetric representation in which all image contrasts are spatially coregistered and intensity-normalized.

[0048] The unified volumetric representation is then passed to the segmentation processor, which executes an AI-based segmentation procedure implemented by hardware-accelerated parallel computers. The segmentation processor extracts hierarchical features by applying a sequence of convolution operations to the volumetric data, extracting local spatial features at various receptive field scales. These features are propagated through successive transformation layers, with intermediate representations capturing both fine-grained structural details and global contextual information. The segmentation processor also performs cross-contrast feature integration, combining feature maps derived from individual image contrasts using adaptive weighting mechanisms.These weighting mechanisms dynamically assign a relative importance to each contrast channel based on local signal characteristics and context relevance, thus enabling improved differentiation between anatomical and pathological regions.

[0049] The segmentation process also includes a multiscale aggregation procedure that combines features of different spatial resolutions to refine the boundaries and improve robustness against noise and intensity ambiguities. The processor uses the learned model parameters stored in memory for voxel-level classification, assigning each voxel in the volumetric representation to a corresponding tissue class or pathological category. This classification involves calculating probability values ​​for each voxel and subsequently selecting the most probable class based on predefined decision criteria. Additionally, spatial consistency conditions are enforced through hardware-based smoothing operations at the neighborhood level, reducing segmentation artifacts and ensuring the continuity of anatomical structures.

[0050] Following segmentation, the resulting labeled volume data are transferred to the feature quantification unit. This unit performs a series of deterministic computational procedures to derive quantitative parameters from the segmented regions. The quantification process begins with volume analysis, in which the number of voxels in each segmented region is counted and converted into physical volume measurements based on voxel dimensions. Next, contour extraction techniques are applied to determine surface properties such as surface texture and contour irregularities by analyzing transitions between adjacent voxel classes. The feature quantification unit also calculates intensity-based statistical measures such as mean intensity, variance, and higher-order distribution properties within each segmented region, providing insights into tissue composition and heterogeneity.

[0051] In addition to basic geometric and statistical parameters, the quantification method includes the calculation of advanced biomarkers derived from spatial and intensity relationships within the segmented regions. These include diffusion-related parameters obtained by analyzing signal intensity variations in diffusion-weighted image data, as well as texture-based metrics derived from spatial correlation patterns of neighboring voxels. The calculation of these parameters is performed using dedicated arithmetic circuits that execute matrix operations and neighborhood-based calculations in parallel, thus ensuring efficient processing of large volumetric datasets.

[0052] The processed segmentation results and calculated quantitative parameters are then transferred to the visualization unit, which presents the results in a clinically interpretable format. The visualization unit overlays segmented regions with the original MRI slices, enabling a direct comparison between raw data and processed results. Additionally, the visualization unit generates three-dimensional volume reconstructions by aggregating voxel data into surface or volume representations, thereby providing a better spatial understanding of anatomical and pathological structures. Structured output data, including quantitative measurements and classification results, are formatted into standardized data structures suitable for transmission via the communication interface.

[0053] The communication interface enables the transfer of processed data to external systems such as clinical workstations, storage servers, and diagnostic reports using standardized communication protocols. The system also ensures continuous and stable operation through its energy management unit, which regulates voltage levels, monitors power consumption, and dynamically adjusts processing parameters to thermal conditions. Temperature sensors provide real-time feedback to the control electronics, enabling adaptive modulation of the processing frequency and power distribution to prevent overheating and maintain operational efficiency.

[0054] The system's methodology is characterized by a tightly integrated sequence of hardware-implemented operations. This begins with the synchronized acquisition of multi-contrast image data, followed by precise preprocessing for spatial and intensity alignment, advanced segmentation using AI models with cross-contrast feature fusion, and comprehensive quantitative analysis of the segmented regions. The integration of these processes into a unified device architecture enables rapid, accurate, and reproducible analysis of multi-contrast magnetic resonance imaging (MRI) data. This overcomes the limitations of conventional software-based approaches and improves clinical decision-making.

[0055] The system for AI-assisted segmentation and quantitative analysis of multi-contrast magnetic resonance imaging (MRI) data is implemented as a structurally integrated device. It consists of a protective housing made of thermally conductive material that contains the internal electronic assemblies and facilitates heat dissipation during the processing of large data volumes. The housing encloses a multi-interface data acquisition module. This module features high-speed inputs, including serial communication interfaces, optical data links, and connectors for removable storage media. This allows for the reception of multi-contrast MRI datasets from external imaging devices or archiving systems.

[0056] A preprocessing unit is housed within the casing and electrically connected to the data acquisition module. This preprocessing unit comprises dedicated signal conditioning circuits and programmable logic units for performing spatial resampling, voxel-wise intensity normalization, and rigid and non-rigid registration across multiple MRI contrasts. The preprocessing unit also includes buffer memory elements for the temporary storage of volumetric intermediate data sets during alignment and transformation processes.

[0057] A central processing unit is mounted within the package and comprises several parallel processing units, including tensor cores and matrix multiplication accelerators, arranged on a circuit substrate. This processing unit is configured to perform AI inference operations using pretrained segmentation models stored in non-volatile memory. The processing unit also includes a memory controller connected to high-bandwidth memory modules to enable fast data access and model execution.

[0058] The segmentation function is achieved by performing feature extraction operations at different scales of the aligned MRI datasets. The processing unit generates voxel-level classification results corresponding to different anatomical or pathological regions. The system is configured to fuse information from multiple contrast channels using weighted feature aggregation mechanisms implemented in hardware, thereby improving segmentation accuracy in regions with low contrast or indistinct borders.

[0059] A quantitative analysis module is electrically coupled to the processing unit and includes arithmetic circuits for calculating geometric and statistical features of the segmented regions. These features include volume measurements from voxel counts, surface area estimates using contour extraction methods, and intensity distribution metrics calculated over segmented tissue classes. The module also includes circuits for calculating advanced biomarkers such as apparent diffusion coefficients and texture-based heterogeneity indices.

[0060] The device features an integrated visualization and output interface, consisting of a display controller, a graphics unit, and external outputs. This interface generates real-time visual overlays of segmented areas onto original MRI slices, as well as three-dimensional volume representations. Additionally, it produces structured output data in the form of quantitative reports, which can be transferred to external clinical systems via network interfaces.

[0061] The system also includes an energy management system with voltage regulators, current control circuits, and temperature monitoring sensors, ensuring stable operation under varying workloads. Optionally, the device can be equipped with a control panel featuring input controls and status indicators for easier operation and monitoring.

[0062] During operation, the system receives multi-contrast MRI datasets, performs preprocessing to align and normalize the data, segments it using AI and hardware-accelerated processors, and calculates quantitative features of the segmented regions. The results are then visualized and output via integrated display and communication interfaces. The structural integration of acquisition, processing, and output modules in a single device enables efficient, reliable, and reproducible analysis of multi-contrast MRI data, thus overcoming the limitations of conventional software-centric approaches.

[0063] The present disclosure relates to medical imaging devices and computer-aided diagnostic systems, in particular a hardware-integrated system for AI-assisted segmentation and quantitative analysis of multi-contrast magnetic resonance imaging (MRI) data. The disclosure specifically relates to a structurally unified device comprising data acquisition circuits, preprocessing processors, parallel computers, storage units, and visualization components. The system is configured to process multiparametric MRI datasets to generate segmented representations and quantitative biomarkers for clinical evaluation, diagnosis, and monitoring.

[0064] The drawing and the preceding description illustrate embodiments. Those skilled in the art will recognize that one or more of the described elements can be combined to form a single functional element. Alternatively, certain elements can be divided into several functional elements. Elements of one embodiment can be added to another. For example, the process flows described here can be modified and are not limited to the manner described herein. Furthermore, the actions of a flowchart need not be performed in the sequence shown; nor do all actions necessarily need to be carried out. Actions that do not depend on other actions can be performed in parallel with the other actions. The scope of protection of the embodiments is in no way limited by these specific examples. Numerous variations, whether explicitly stated in the description or not, such as...Differences in structure, dimensions, and materials are possible. The scope of protection of the embodiments is at least as comprehensive as described by the following claims.

[0065] The advantages, other benefits, and problem solutions have been described above with reference to specific embodiments. However, the advantages, benefits, problem solutions, and any components that can effect or enhance an advantage, benefit, or solution are not to be construed as critical, necessary, or essential features or components of the claims. REFERENCES 100 A system for AI-supported segmentation and quantitative analysis of multi-contrast magnetic resonance imaging data. 102 Housing structure 104 Data acquisition unit 106 Preprocessing processor 108 storage units 110 Segmentation processor 112 Feature quantification unit 114 Visualization Unit 116 Communication interface

Claims

[1] A system for AI-assisted segmentation and quantitative analysis of multi-contrast magnetic resonance imaging data, consisting of: a housing configured to enclose and support a variety of electronic components; a housing-mounted data acquisition unit configured to receive multi-contrast magnetic resonance imaging datasets with a variety of imaging contrasts via one or more communication interfaces; a preprocessing processor that is operationally coupled with the data acquisition unit and is configured to perform the spatial alignment, intensity normalization, and cross-contrast registration of the received data sets to produce a uniform volumetric representation; a storage unit that is electrically connected to the preprocessing processor and is configured to store the received data sets, the data processed in the meantime, and the trained model parameters; a segmentation processor arranged in the housing, which is operationally connected to the preprocessing processor and the storage unit, wherein the segmentation processor includes a parallel computing circuit configured to execute trained artificial intelligence models to generate voxel-level segmentation outputs corresponding to anatomical or pathological regions based on the uniform volumetric representation; a feature quantification unit that is operationally coupled with the segmentation processor and configured to calculate quantitative parameters such as volume measurements, boundary features, and intensity distribution metrics from the segmented regions; a visualization unit that is operationally coupled with the segmentation processor and the feature quantification unit and is configured to generate visual overlays, multidimensional representations, and structured output data according to the segmented regions and calculated quantitative parameters; and a communication interface configured to transmit the structured output data to external systems, where the preprocessing processor and the segmentation processor are configured to work together in a coordinated manner to perform cross-contrast data fusion to improve segmentation accuracy and quantitative analysis. [2] System according to claim 1, wherein the data acquisition unit comprises a plurality of high-speed input interfaces, including optical communication ports and serial data ports, and wherein the data acquisition unit further comprises a buffer circuit configured to temporarily store incoming multi-contrast data sets to enable continuous data streaming and synchronized processing across multiple image contrasts. [3] System according to claim 1, wherein the preprocessing processor comprises a programmable logic circuit configured to perform rigid and non-rigid registration across the plurality of image contrasts, and wherein the preprocessing processor is further configured to perform voxel-wise interpolation and resampling operations to ensure uniform spatial resolution across the entire volumetric representation. [4] System according to claim 1, wherein the preprocessing processor further comprises an intensity calibration circuit configured to compensate for scanner-related variations and inhomogeneities of signal intensity across different contrasts of magnetic resonance imaging, thereby enabling consistent input conditions for the segmentation processor. [5] System according to claim 1, wherein the segmentation processor comprises a plurality of tensor computation units and matrix processing units arranged to perform parallel convolution operations, feature extraction and hierarchical representation learning based on the trained artificial intelligence models stored in the memory unit. [6] System according to claim 5, wherein the segmentation processor is configured to perform feature aggregation on multiple scales by integrating information from different spatial resolutions and multiple image contrasts, thereby improving the delineation of regions with low-contrast boundaries or heterogeneous intensity profiles. [7] System according to claim 1, wherein the segmentation processor further comprises a weighting circuit configured to assign adaptive significance to each image contrast based on local signal characteristics, thereby enabling dynamic cross-contrast fusion during segmentation. [8] System according to claim 1, wherein the feature quantification unit comprises an arithmetic circuit configured to calculate volumetric parameters based on voxel counting, surface parameters based on boundary extraction, and statistical parameters including mean intensity, variance, and distribution profiles over segmented regions. [9] System according to claim 8, wherein the feature quantification unit is further configured to calculate diffusion-related parameters and texture-based heterogeneity indices using spatial correlation and intensity variation analysis within the segmented regions. [10] System according to claim 1, wherein the visualization unit comprises a graphic processing circuit configured to generate layer-wise overlays of segmented regions on original magnetic resonance image data and render three-dimensional volumetric representations using reconstructed voxel data.

Images