A portable peripheral blood multi-modal intelligent auxiliary diagnosis system and method

Abstract

The application discloses a portable peripheral blood multi-modal intelligent auxiliary diagnosis system and method, and belongs to the cross technical field of medical examination, artificial intelligence and multi-modal data fusion. The system comprises a microfluidic chip, which is used for physically separating white blood cells from a whole blood sample and arranging the white blood cells in an imaging cavity; a multi-modal information acquisition unit, which comprises a computational optical microscopic module for performing multi-mode imaging on the white blood cells in the imaging cavity, a physiological parameter sensor for acquiring physiological parameters of a user, and a clinical information acquisition module for acquiring text information of a medical record of the user; and a multi-modal data fusion processing unit, which obtains an auxiliary diagnosis report through cross-modal joint reasoning according to received image, physiological parameter and text data. The application effectively improves the accuracy of auxiliary diagnosis by fusing cell morphological characteristics and clinical symptoms and sign information, and is suitable for portable scenes such as home self-examination, primary medical treatment and field rescue.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of medical testing, artificial intelligence, and multimodal data fusion, specifically to a portable peripheral blood multimodal intelligent auxiliary diagnostic system and method. It is used to achieve physical separation of leukocytes, in-situ high-resolution imaging, and automated five-part differential and abnormal cell identification from a small amount of whole blood sample. It also integrates multi-dimensional information such as body temperature and medical record text to achieve auxiliary diagnosis of infectious and hematological diseases. It is especially suitable for portable application scenarios such as home self-testing, primary healthcare, and field rescue. Background Technology

[0002] Early diagnosis of infectious diseases and hematological disorders relies heavily on white blood cell count and differential diagnosis. White blood cell differential count is a core diagnostic tool for infectious diseases, hematological disorders, and immune-related diseases. However, existing technologies have significant limitations in their application scenarios: Traditional manual microscopic examination relies on Wright staining and manual morphological identification by pathologists, which is time-consuming, highly subjective, and difficult to standardize. Although fully automated blood analyzers have achieved automated five-part differential analysis, the equipment is bulky, expensive (usually hundreds of thousands to millions of yuan), requires specialized reagents and regular maintenance, and is completely unsuitable for the on-site, real-time testing needs in resource-constrained environments such as homes, primary clinics, or field rescues.

[0003] While existing portable hematology analyzers have achieved miniaturization, they still rely on liquid reagent packs such as sheath fluid, hemolysin, and staining solutions, as well as precision microfluidic control modules. The instruments typically weigh several kilograms, resulting in high costs per test. Furthermore, the core detection unit still requires precision optical components, making it difficult to further miniaturize them to handheld or home-use levels.

[0004] Existing microfluidic separation technologies or portable microscope solutions mostly focus on a single step—either only completing cell separation or only achieving portable imaging—lacking a system-level solution that deeply integrates and synergistically optimizes the three stages of "sample preparation → microscopic imaging → AI diagnosis." Especially in the imaging stage, traditional high-magnification microscopy relies on precision objective groups and multi-axis focusing mechanisms, which cannot simultaneously achieve miniaturization and high resolution.

[0005] Currently, there are some AI classification models based on blood cell images, such as patent CN120746986A. However, these models only use single image data, and the input information dimensions are limited. They cannot make comprehensive judgments like clinicians by integrating multiple dimensions of information such as "visual (cell morphology under a microscope), tactile (body temperature), and questioning (symptom complaints)." For example, even if the same symptom is "elevated neutrophils," if there is also text information about high fever and cough, it suggests a significantly increased possibility of bacterial respiratory infection; if there is no fever and the user complains of fatigue, it may point to chronic inflammation or other conditions. A single image modality cannot make such precise distinctions.

[0006] Therefore, there is an urgent need for an end-to-end system that integrates physical separation, miniaturized high-resolution imaging, and multimodal AI analysis, starting from a tiny fingertip blood sample, to truly realize "sample in - result out" applications in scenarios such as home self-testing, primary healthcare, and wilderness rescue. Summary of the Invention

[0007] To overcome the shortcomings of existing technologies, this invention aims to provide an intelligent auxiliary diagnostic system and method that integrates peripheral blood images with multi-dimensional clinical information, in order to achieve the following objectives: 1. Physically separate leukocytes with high purity from a small amount of fingertip blood and perform in-situ high-resolution imaging; 2. Supports lensless or miniaturized microscopic imaging in multiple imaging modes (bright field, fluorescence, phase, etc.), and the imaging module size can be reduced to the level of a palm or mobile phone accessory. 3. Collect and integrate multi-dimensional, multi-modal data such as peripheral blood cell images, body temperature, and medical record texts (symptoms, present illness, past medical history, etc.); 4. Through multimodal AI large model, perform joint reasoning on fused data to output test results including white blood cell five-part differential and abnormal cell screening, and provide auxiliary diagnostic opinions (such as suspected disease direction, severity assessment, medical treatment suggestions, etc.). 5. The system is highly integrated, easy to operate, portable and low-power, making it suitable for non-laboratory environments such as home self-testing, primary clinics and field rescue.

[0008] To achieve the above objectives, the present invention provides the following solution: An intelligent auxiliary diagnostic system integrating peripheral blood images and multi-dimensional clinical information includes: Microfluidic chips are used to physically separate white blood cells from whole blood samples and arrange the separated white blood cells within their imaging chambers; The multimodal information acquisition unit includes: a computational optical microscopy imaging module, used for multimodal imaging of white blood cells in the imaging cavity; Physiological parameter sensors are used to acquire the user's physiological parameters; The clinical information acquisition module is used to obtain users' medical record text information; The multimodal data fusion processing unit is used to receive and fuse image data, physiological parameter data and clinical information text data output by the multimodal information acquisition unit, and perform joint reasoning based on the fused data to output a diagnostic report containing white blood cell classification results and / or auxiliary diagnostic opinions.

[0009] Furthermore, the microfluidic chip includes: A lower substrate, the surface of which is provided with a flow channel system, the flow channel system including a sample injection area, a filtration area and a waste liquid collection area; A porous membrane is placed over the filtration zone. The average pore size of the porous membrane is smaller than the average diameter of the target white blood cells to be retained but larger than the average diameter of red blood cells and platelets. The porous membrane allows visible light to pass through. The upper substrate, located above the porous membrane, is made of an optically transparent material and together with the lower substrate and the porous membrane, forms an imaging cavity located upstream of the porous membrane. The imaging cavity provides an optical path to an external optical imaging module through the upper substrate, and the outer surface of the upper substrate is provided with a standardized optical interface for docking with the computational optical microscopy imaging module.

[0010] Furthermore, the multimodal data fusion processing unit is a multimodal fusion diagnostic engine based on a large language model architecture, comprising: The image coding branch is used to extract image features from cell image data; The numerical coding branch is used to map physiological parameter data into numerical feature vectors; The text encoding branch is used for semantic understanding and medical terminology standardization of medical record text information. The cross-modal fusion layer is used to perform cross-modal attention fusion on the outputs of the image encoding branch, numerical encoding branch and text encoding branch through an attention mechanism to generate diagnostic reasoning text.

[0011] Furthermore, the physiological parameter sensor includes an infrared body temperature sensor for acquiring the user's body temperature value; the clinical information acquisition module includes a voice interaction or structured questionnaire interface for acquiring the user's chief complaint, present medical history, and past medical history text information.

[0012] Furthermore, the multi-mode imaging includes at least two of bright field, dark field, fluorescence, phase difference / phase, and polarization images, and the computational optical microscopy imaging module uses a lensless holographic scheme or a miniature objective scheme for imaging.

[0013] Furthermore, the microfluidic chip has a capillary-driven structure, which enables the blood sample to complete the entire process of injection, filtration and washing entirely by relying on capillary forces.

[0014] Furthermore, the diagnostic report includes a list of differential diagnoses, an assessment of disease severity, and recommendations for medical treatment.

[0015] The intelligent assisted diagnostic system also includes a remote review module. When the output confidence level of the multimodal data fusion processing unit is lower than the threshold or abnormal cells are detected, the remote review module automatically sends the anonymized diagnostic data to the cloud for review by professional physicians.

[0016] The present invention also provides a method for auxiliary diagnosis using the above-mentioned intelligent auxiliary diagnostic system, comprising: Step 1: Physically separate white blood cells from a whole blood sample using a microfluidic chip and arrange them in the imaging chamber; Step 2: Obtain multi-mode image data of white blood cells within the imaging cavity using a computational optical microscopy imaging module; Step 3: Acquire the user's physiological parameter data through the physiological parameter sensor, and acquire the user's medical record text information through the clinical information acquisition module; Step 4: Input the multimodal image data, physiological parameter data and medical record text information into the multimodal data fusion processing unit, and output a diagnostic report containing white blood cell classification results and / or auxiliary diagnostic suggestions through cross-modal joint reasoning.

[0017] Furthermore, the multimodal data fusion processing unit continuously optimizes the model using federated learning technology while protecting user data privacy.

[0018] Compared with the prior art, the present invention discloses the following technical effects: This invention elevates simple cell classification output to a comprehensive auxiliary diagnosis that is closer to clinical practice, enabling non-professional users to understand the meaning of test results and obtain clear action suggestions, thus effectively improving diagnostic value.

[0019] This invention collects and integrates multimodal image data, physiological parameter data, and medical record text information. Cell images can provide objective cytological evidence, body temperature data can provide real-time physiological status, and textual complaints can reflect the course of the disease and subjective feelings. Cross-validation of the three can significantly improve the specificity and sensitivity of diagnosis and reduce single-modal misjudgment.

[0020] This invention automates the entire process from sampling to outputting results, eliminating the need for manual interpretation and overcoming limitations imposed by medical resources. Furthermore, the capillary drive structure of the microfluidic chip and the computational optical microscopy imaging module employ a lensless holographic scheme or a miniature objective lens scheme for imaging, enabling the miniaturization and portability of the diagnostic system and ultimately achieving true home-based blood health monitoring.

[0021] The present invention also includes a federated learning or local fine-tuning mechanism, which can achieve continuous optimization of the model while protecting users' sensitive health data from being leaked. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a structural framework diagram of the intelligent assisted diagnostic system provided in an embodiment of the present invention; Figure 2 This is a flowchart illustrating the intelligent assisted diagnosis method provided in an embodiment of the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] This invention provides an intelligent auxiliary diagnostic system that integrates peripheral blood images with multi-dimensional clinical information, such as... Figure 1 As shown, it includes: a microfluidic chip, a multimodal information acquisition unit, and a multimodal data fusion processing unit.

[0026] (a) The microfluidic chip includes: The lower substrate has a flow channel system on its surface, which includes at least a sample injection area, a filtration area, and a waste liquid collection area. A porous membrane covers at least a portion of the flow channel in the filtration zone. The average pore size of the porous membrane is smaller than the average diameter of the target white blood cells to be retained but larger than the average diameter of red blood cells and platelets. The porous membrane allows visible light to pass through. The upper substrate, located above the porous membrane, is made of an optically transparent material and together with the lower substrate and the porous membrane, forms an imaging cavity located upstream of the porous membrane. The imaging cavity provides an optical path to an external optical imaging module through the upper substrate, and the outer surface of the upper substrate is provided with a standardized optical interface (such as a positioning groove, magnetic ring or snap-fit ​​structure) for docking with the computational optical microscopy imaging module.

[0027] The microfluidic chip also includes: Reagent pre-storage and release structure: A reagent chamber containing a dry fluorescent dye (e.g., Hoechst 33342 or acridine orange) is set upstream of the sample injection area or filtration area. Its inner wall is coated with a protective agent. When a blood sample flows in, the dye dissolves and labels the leukocyte nucleic acid. Cell monolayer guidance structure: The imaging cavity height (the distance between the porous membrane and the upper substrate) is optimized to 10-30μm based on the average diameter of white blood cells, which naturally restricts the cells to be arranged in a monolayer, avoids overlap, and ensures that each cell can be independently and clearly imaged. Capillary-driven structure: The inner wall of the flow channel is hydrophilic, an open liquid storage tank is set at the inlet, and a highly absorbent waste liquid absorption pad or capillary pump array is set at the outlet, so that the blood sample can complete the entire process of injection, filtration and cleaning entirely by capillary force. Quality control reference microparticle region: Reference microspheres with known optical characteristics (size, fluorescence intensity, phase value) are fixed at one or more points in the imaging cavity for automatic focusing, optical calibration and self-verification of AI model confidence in the imaging system.

[0028] (ii) The multimodal information acquisition unit includes: a computational optical microscopy imaging module, a physiological parameter sensor, and a clinical information acquisition module; The computational optical microscopy imaging module includes: Image sensor: High-sensitivity CMOS or CCD area array sensor, pixel size 1.1-2.2μm, resolution 5 million-20 million pixels; The lighting unit includes a multi-wavelength LED light source (e.g., white LEDs for bright field, 385 nm ultraviolet LEDs for fluorescence excitation, and 450 nm blue LEDs for phase imaging), and the lighting mode can be switched electronically or automatically. Imaging optical path: Miniaturized high-resolution imaging is achieved by employing one or a combination of the following schemes: Scheme A (lens-free holographic scheme): The sample is directly attached to the surface of the image sensor, and the LED coherent light source illuminates it obliquely from above. The phase and amplitude images are reconstructed from the hologram through the angular spectrum propagation algorithm, achieving submicron resolution without any lens; Option B (Miniature Objective Lens): A miniature objective lens group consisting of 1-3 aspherical plastic or glass lenses, with a total length of <20 mm and a numerical aperture of 0.3-0.5, is used in conjunction with a voice coil motor to achieve autofocus; Option C (Programmable Illumination Scheme): Fourier layered imaging is achieved using a programmable LED array or spatial light modulator, and high-resolution images exceeding the numerical aperture limit of the objective lens are synthesized using multi-angle illumination. Computational reconstruction processor: an embedded GPU or FPGA accelerator that runs computational imaging algorithms such as holographic reconstruction, layer synthesis, denoising, and deconvolution in real time to output high-quality cell images; Chip interface: A slot or magnetic holder that matches the standardized optical interface on the chip substrate to ensure that the imaging cavity automatically aligns with the center field of view of the sensor after the chip is inserted, with a tolerance of ±100μm; Housing and power management: The overall size can be controlled to be the size of a POS or smaller (e.g., 200×80×30 mm). 3 It weighs less than 300 g, has a built-in rechargeable lithium battery, and supports USB-C charging and data transfer.

[0029] The physiological parameter sensor includes an infrared body temperature sensor (e.g., a patch-type or non-contact infrared temperature measurement array) to acquire the user's real-time body temperature (accurate to 0.1℃) during the test; or the user can input body temperature data via voice or text.

[0030] The clinical information collection module provides a voice input or structured questionnaire interface for collecting users' chief complaints (e.g., "fever and cough for 3 days"), present medical history, past medical history, medication information, and other natural language text. This module can be deployed on a local embedded screen or on a mobile app / mini-program connected via Bluetooth / Wi-Fi.

[0031] (III) The multimodal data fusion processing unit includes the following functional modules: Multimodal feature fusion encoder: It is used to receive any one or more combinations of modalities such as bright field, dark field, fluorescence, phase difference / phase, polarization image from the imaging module as input, and to map the image features of different modalities to a unified embedding space through a multi-branch feature extraction network (such as a shared backbone network + modality-specific adaptation layer); White blood cell detection and segmentation module: Based on the target detection architecture (such as YOLO series or DETR series), white blood cells are located in the panoramic scan image of the imaging cavity, and the precise bounding box coordinates of each cell are output. The instance segmentation network (such as MaskR-CNN or SAM adapted version) is used to segment the cell nucleus and cytoplasm. Five-class classification module: Based on Vision Transformer or hybrid architecture, each segmented white blood cell is classified into five categories (neutrophils, lymphocytes, monocytes, eosinophils, and basophils), and the class probability vector and confidence score are output. Abnormal cell screening module: Based on the five-part differential diagnosis, it further identifies abnormal cell types such as primitive cells, immature granulocytes, atypical lymphocytes, reactive lymphocytes, and nucleated red blood cells. When the proportion of abnormal cells exceeds the clinical threshold, an alarm is generated. The auxiliary diagnostic report generation module integrates the five-part differential count results, abnormal cell detection, and blood cell morphology characteristics to automatically generate a structured blood routine interpretation report, including the total white blood cell count, various percentages and absolute values, a list of abnormal findings, and preliminary diagnostic suggestions (such as "neutrophilia, consistent with possible bacterial infection" or "primitive cells found, further consultation with the hematology department is recommended"). Privacy protection and continuous learning framework: The model supports federated learning or local adaptive fine-tuning. User image data can be inferred locally and only desensitized feature vectors are uploaded, protecting privacy while supporting continuous model optimization.

[0032] Furthermore, the detailed architecture of the multimodal data fusion processing unit is as follows: Input layer: Receives at least three modalities of data—multi-channel cell images (I), physiological values ​​such as body temperature (V), and chief complaint / medical history text (T).

[0033] Image encoding branch: Use hierarchical Vision Transformer to extract fine-grained morphological features (such as nucleocytoplasmic ratio, granularity, degree of lobulation, etc.) of each cell from the panoramic cell image and output a global image representation vector.

[0034] Numerical coding branch: Body temperature and other optional physiological indicators (such as heart rate, blood oxygen, etc.) are mapped to fixed-dimensional vectors through a learnable numerical embedding layer.

[0035] Text encoding branch: Based on large language models such as BERT or LLaMA that are adaptively pre-trained in the medical field, semantic understanding of colloquial complaints and standardization of medical terminology are performed.

[0036] Cross-modal fusion layer: Employing a multi-head cross-attention mechanism and gated fusion units, the text "fever" automatically enhances the attention weight of the "granulocytosis" feature in the image, allowing body temperature values ​​to corroborate the textual description of the disease course, achieving fine-grained semantic alignment and collaborative reasoning between different modalities.

[0037] Multi-task output layer: Simultaneously outputs cell classification results and diagnostic inference text. The diagnostic inference task employs an autoregressive decoding method, concatenating image and numerical information as special tokens into the text sequence for joint generation.

[0038] The present invention also provides a method for auxiliary diagnosis using the above-mentioned intelligent auxiliary diagnostic system, such as... Figure 2 As shown, it includes: 1. Sample loading and separation: The user collects blood from the fingertip and dilutes it before dripping it into the microfluidic chip. The chip relies on capillary action or a manual micropump to filter, wash and selectively stain white blood cells, forming a monolayer of white blood cells.

[0039] 2. Multimodal data synchronous acquisition: Insert the chip into the system host (or connect it to a mobile phone accessory), and the system will automatically trigger the following parallel acquisition: Image acquisition: The computational optics module scans the imaging cavity to acquire a panoramic view of cells in multiple channels (bright field / fluorescence / phase, etc.); Body temperature measurement: Infrared sensors synchronously record the user's real-time body temperature; Text entry: The interface prompts the user to briefly describe the main symptoms, or to input the chief complaint of discomfort by voice, and the system automatically transcribes it into medical record text.

[0040] 3. Multimodal Fusion Inference: The fusion diagnostic engine performs cell detection and classification on the image, and simultaneously inputs cytological findings (such as a significant increase in neutrophils accompanied by toxic granules) and clinical information (such as a body temperature of 39.2℃ and complaints of chills and sore throat) as prompts. The large model then infers and generates an auxiliary diagnostic report containing the following: Total white blood cell count, white blood cell differential count and percentage; Abnormal cell detection status (such as primitive cells, atypical lymphocytes, etc.); A comprehensive diagnostic impression based on multimodal information (e.g., blood counts and clinical features consistent with acute bacterial upper respiratory tract infection or unexplained cytosis with low-grade fever, suggesting further investigation by a hematologist). Differential diagnosis list, probability ranking, severity assessment, and medical advice and precautions.

[0041] 4. Results Display and Feedback: The report is presented on a display screen or through a mobile app / mini-program, and users can choose to share it with an online doctor for review with one click.

[0042] This invention also provides a microfluidic chip optimization method based on differences in cell size and deformation, comprising: 1. Multi-level gradient aperture enables subgroup hierarchical separation.

[0043] To further physically separate leukocyte subsets, this invention connects two or more membranes with different pore sizes in series along the flow channel on the same chip: First-level membrane: pore size 8-10 μm, retaining neutrophils (average diameter >10 μm) and eosinophils; Secondary membrane: pore size 4.5-5 μm, trapping lymphocytes and monocytes.

[0044] The spacing between the two membrane stages is 100-500 μm to ensure that upstream cell retention does not interfere with downstream filtration. This multi-stage gradient pore size design can better physically separate leukocyte subpopulations.

[0045] 2. Optimization of the pre-separation filtration structure for red blood cells and platelets.

[0046] To reduce the risk of clogging of the main filtration membrane, a pre-separation structure for red blood cells and platelets is integrated upstream of the main membrane: Microcolumn array pre-separation: A microcolumn array with a spacing of 2-4 μm is set upstream of the main membrane. The deformation ability of red blood cells and platelets is used to allow them to pass through first and be diverted to the bypass or waste liquid area, while white blood cells are retained and enter the main membrane filtration area due to their larger size.

[0047] Cross-flow filtration design: The flow path is parallel to the membrane surface rather than perpendicular to it. The sample flows tangentially along the membrane surface. The inertia of red blood cells and platelets makes it difficult for them to be vertically embedded in the membrane pores. The filtration driving force is from the vertical direction, drawing out plasma and some deformed red blood cells. The ratio of tangential flow velocity to normal osmotic flow velocity is optimized to 3:1 to 8:1. In actual tests, a 5:1 ratio resulted in a retention rate of >96% and optimal filtration time. This design can delay membrane clogging time by 5-10 times.

[0048] 3. Optimization of membrane structure parameters Membrane porosity: optimized to 10 5 -10 7 pores / cm 2 Within this range, sufficient flow resistance is ensured to drive the separation of white blood cells, red blood cells, and platelets, while avoiding excessively high local flow rates due to low porosity that could cause cell shearing damage.

[0049] Film thickness: 10-25 μm. If the thickness is too low (<6 μm), the mechanical strength is insufficient and it is easy to break during the bonding process; if the thickness is too high (>25 μm), it will increase the background signal and light scattering in fluorescence imaging.

[0050] 4. Anti-clogging dynamic structural design Micro-piezoelectric vibration structures or magnetic stirring microstructures are integrated upstream of the membrane or on the membrane surface to generate micro-vibrations or localized turbulence during operation, preventing cell sedimentation and accumulation on the membrane surface. Here, the acoustic waves / vibrations are only used for anti-clogging; the separation driving force is still achieved by the physical sieving of the membrane pores, which is fundamentally different from ultrasonic separation methods.

[0051] The specific implementation of the intelligent auxiliary diagnostic system of the present invention is as follows: Example 1: Home self-testing system.

[0052] The microfluidic chip uses cyclic olefin copolymer (COC) as the substrate material, a PET membrane as the porous membrane, and an imaging cavity depth of 20 μm. It is pre-encapsulated with dry acridine orange fluorescent dye. The chip employs a capillary-driven structure, enabling whole blood sample injection, leukocyte filtration, and washing without an external pump source.

[0053] Furthermore, the chip can also be pressurized by an external syringe, a design that can improve the efficiency of whole blood sample introduction and filtration.

[0054] The imaging module is integrated into the phone's back cover accessory, using a miniature objective lens and the phone's camera to complete fluorescence / bright field imaging. The accessory has a built-in infrared ear temperature sensor, which measures body temperature upon wearing. The phone app / mini-program asks "Where do you feel unwell?" via voice, and the user verbally replies with their complaint. The app / mini-program then performs voice recognition and generates a structured text of the complaint.

[0055] The multimodal data fusion processing unit is deployed on the mobile NPU or in the cloud. When a user complains of "sore throat and fever for two days," with a temperature of 38.5℃, and cell images show a significant increase in neutrophils accompanied by toxic granulation, the model comprehensively judges and outputs: "High probability of acute purulent tonsillitis (bacterial infection). It is recommended to drink plenty of water and rest. If the sore throat worsens or the high fever persists, please seek medical attention." Simultaneously, the app / mini-program displays the counts and historical trends of various white blood cell subclasses.

[0056] All data is encrypted on the mobile device, and after the diagnosis is completed, it is not uploaded to the cloud unless the user actively shares it, thus protecting user privacy.

[0057] Example 2: Primary care clinic type system.

[0058] The device is a diagnostic instrument the size of a tablet computer, integrating a chip slot, a press-to-pump micro-infrared device, a lensless holographic imaging module, a forehead thermometer module, and a medical record entry interface that can be connected to a keyboard.

[0059] The operating procedure is as follows: the patient provides blood from a fingertip, the nurse inserts a chip into the instrument, and after pressing it several times, an image is automatically formed. Simultaneously, the nurse enters the patient's chief complaint and physical signs via a drop-down menu and free text input. The diagnostic instrument's local multimodal data fusion processing unit integrates blood counts, temperature, symptoms, and medical history to output a preliminary diagnosis and referral suggestions, which can be printed or uploaded to the township health center's HIS system.

[0060] Example 3: Wilderness rescue system.

[0061] The system is equipped with a rugged shell, a solar-charged portable case, and contains a microfluidic chip, a lensless imaging unit, a Bluetooth vital signs monitoring vest (capable of measuring heart rate, blood oxygen, and body temperature), and a satellite communication module.

[0062] The workflow is as follows: Blood is drawn from the wounded soldier's fingertip for white blood cell analysis, while the vest automatically acquires continuous body temperature and vital signs. A multimodal model considers the trauma context and temperature trends to assess infection risk and generate decision information such as "suspected intestinal perforation with early sepsis; emergency evacuation and broad-spectrum antibiotics recommended." Encrypted reports are then transmitted to the rear command via satellite link.

[0063] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. An intelligent auxiliary diagnostic system integrating peripheral blood images and multi-dimensional clinical information, characterized in that, include: Microfluidic chips are used to physically separate white blood cells from whole blood samples and arrange the separated white blood cells within their imaging chambers; The multimodal information acquisition unit includes: a computational optical microscopy imaging module, used for multimodal imaging of white blood cells in the imaging cavity; Physiological parameter sensors are used to acquire the user's physiological parameters; The clinical information acquisition module is used to obtain users' medical record text information; The multimodal data fusion processing unit is used to receive and fuse image data, physiological parameter data and clinical information text data output by the multimodal information acquisition unit, and perform joint reasoning based on the fused data to output a diagnostic report containing white blood cell classification results and / or auxiliary diagnostic opinions.

2. The intelligent auxiliary diagnostic system for integrating peripheral blood images and multi-dimensional clinical information according to claim 1, characterized in that, The microfluidic chip includes: A lower substrate, the surface of which is provided with a flow channel system, the flow channel system including a sample injection area, a filtration area and a waste liquid collection area; A porous membrane is placed over the filtration zone. The average pore size of the porous membrane is smaller than the average diameter of the target white blood cells to be retained but larger than the average diameter of red blood cells and platelets. The porous membrane allows visible light to pass through. The upper substrate, located above the porous membrane, is made of an optically transparent material and together with the lower substrate and the porous membrane, forms an imaging cavity located upstream of the porous membrane. The imaging cavity provides an optical path to an external optical imaging module through the upper substrate, and the outer surface of the upper substrate is provided with a standardized optical interface for docking with the computational optical microscopy imaging module.

3. The intelligent auxiliary diagnostic system for integrating peripheral blood images and multi-dimensional clinical information according to claim 1, characterized in that, The multimodal data fusion processing unit is a multimodal fusion diagnostic engine based on a large language model architecture, including: The image coding branch is used to extract image features from cell image data; The numerical coding branch is used to map physiological parameter data into numerical feature vectors; The text encoding branch is used for semantic understanding and medical terminology standardization of medical record text information. The cross-modal fusion layer is used to perform cross-modal attention fusion on the outputs of the image encoding branch, numerical encoding branch and text encoding branch through an attention mechanism to generate diagnostic reasoning text.

4. The intelligent auxiliary diagnostic system for integrating peripheral blood images and multi-dimensional clinical information according to claim 1, characterized in that, The physiological parameter sensor includes an infrared body temperature sensor for acquiring the user's body temperature value; the clinical information acquisition module includes a voice interaction or structured questionnaire interface for acquiring the user's chief complaint, present illness, and past medical history text information.

5. The intelligent auxiliary diagnostic system for integrating peripheral blood images and multi-dimensional clinical information according to claim 1, characterized in that, The multi-mode imaging includes at least two of bright field, flat field, fluorescence, phase difference / phase, and polarization images. The computational optical microscopy imaging module uses a lensless holographic scheme or a miniature objective scheme for imaging.

6. The intelligent auxiliary diagnostic system for integrating peripheral blood images and multi-dimensional clinical information according to claim 1, characterized in that, The microfluidic chip has a capillary-driven structure, which enables blood samples to complete the entire process of injection, filtration and washing entirely by relying on capillary forces.

7. The intelligent auxiliary diagnostic system for integrating peripheral blood images and multi-dimensional clinical information according to claim 1, characterized in that, The diagnostic report includes a list of differential diagnoses, an assessment of disease severity, and recommendations for medical treatment.

8. The intelligent auxiliary diagnostic system for integrating peripheral blood images and multi-dimensional clinical information according to any one of claims 1-7, characterized in that, Also includes: The remote review module automatically sends anonymized diagnostic data to the cloud for professional physician review when the output confidence level of the multimodal data fusion processing unit is lower than the threshold or abnormal cells are detected.

9. A method for assisted diagnosis using the system according to any one of claims 1 to 8, comprising: Step 1: Physically separate white blood cells from a whole blood sample using a microfluidic chip and arrange them in the imaging chamber; Step 2: Obtain multi-mode image data of white blood cells within the imaging cavity using a computational optical microscopy imaging module; Step 3: Acquire the user's physiological parameter data through the physiological parameter sensor, and acquire the user's medical record text information through the clinical information acquisition module; Step 4: Input the multimodal image data, physiological parameter data and medical record text information into the multimodal data fusion processing unit, and output a diagnostic report containing white blood cell classification results and / or auxiliary diagnostic suggestions through cross-modal joint reasoning.

10. The intelligent assisted diagnostic method according to claim 9, characterized in that, The multimodal data fusion processing unit continuously optimizes the model using federated learning technology while protecting user data privacy.

Images