Search

CN-122017216-A - Abnormal sample intelligent recognition and rechecking system of platelet function analyzer

CN122017216ACN 122017216 ACN122017216 ACN 122017216ACN-122017216-A

Abstract

The invention relates to the technical field of platelet function analysis, and discloses an abnormal sample intelligent recognition and rechecking system of a platelet function analyzer, which comprises a multi-physical-field cooperative sensing module, a sensor module and a control module, wherein the sensor module is connected with the sensor module; the intelligent recognition center, the automatic recheck execution module, the system self-evolution engine and the prospective sample quality screening module are used for synchronously collecting multidimensional heterogeneous response signals of physical, chemical, biomechanics and the like of samples through the integrated quantum dot coded magneto-electric sensor, the terahertz time-domain spectrum, the microfluidic acoustic tweezers, the multi-frequency impedance spectrum and other multi-physical-field sensing modules, so that the limitation of a single detection method is overcome, an unprecedented rich information basis is provided for abnormal recognition, a neural-symbol hybrid reasoning algorithm adopted by the intelligent recognition center is used for automatically mining complex modes from high-dimensional data by utilizing a deep transducer network, and logic verification and causal reasoning are performed through symbol reasoning of knowledge in the fusion field.

Inventors

  • Li Haichong

Assignees

  • 安徽君禾生物科技有限公司

Dates

Publication Date
20260512
Application Date
20260311

Claims (9)

  1. 1. The abnormal sample intelligent recognition and rechecking system of the platelet function analyzer is characterized by comprising the following components: The multi-physical-field collaborative sensing module is used for applying various physical-field stimuli to the blood sample and synchronously collecting heterogeneous response signals generated by the physical-field stimuli; The intelligent identification center receives heterogeneous response signals acquired by the multi-physical-field cooperative sensing module through a communication link and executes a neural-symbol hybrid reasoning algorithm to identify abnormal states of samples and infer root causes of the abnormal states; The automatic re-inspection execution module receives the re-inspection instruction issued by the intelligent identification center and controls the integrated microfluidic chip laboratory to automatically reprocess and re-inspect the original sample through a self-adaptive decision algorithm based on reinforcement learning; The system self-evolution engine continuously collects the reasoning result of the intelligent identification center, the verification result of the automatic reinspection execution module and external clinical feedback information, and performs iterative optimization on model parameters and a knowledge base in the intelligent identification center through a federal learning framework and an abnormal mode library dynamic updating algorithm; and the prospective sample quality screening module is used for acquiring a preliminary terahertz absorption spectrum, a low-frequency impedance baseline and microscopic image textures of the sample within 30 seconds by utilizing one rapid scanning sub-mode of the multi-physical-field collaborative sensing module before the sample enters a main detection flow.
  2. 2. The intelligent abnormal sample recognition and review system of the platelet function analyzer according to claim 1, wherein the multi-physical-field cooperative sensing module comprises: the magneto-electric coupling sensor array of the quantum dot code, it is made up of a plurality of magneto-electric nanometer particles with different biological molecular probes of surface modification, each particle encodes through the unique quantum dot fluorescence spectrum, is used for detecting the specific biological molecule binding event and electromagnetic characteristic change under the nanometer scale in the platelet aggregation process synchronously, in situ; The terahertz time-domain spectrum unit comprises a terahertz wave generating device and a detecting device, and is used for transmitting electromagnetic pulses with the frequency range of 0.1 to 10 terahertz to a sample, receiving transmission or reflection signals and analyzing characteristics of dielectric relaxation, hydrodynamics and biological macromolecule collective vibration modes in the sample; the microfluidic acoustic tweezer unit is integrated in a microfluidic channel, generates surface acoustic waves through an interdigital transducer, forms a stable acoustic potential well in the channel, and is used for capturing a single blood platelet and measuring the displacement response of the single blood platelet under the action of acoustic radiation force so as to calculate the mechanical characteristics of the blood platelet; a multi-frequency electrical impedance spectroscopy scanning unit that measures the complex impedance of a sample at a plurality of discrete frequency points from kilohertz to hundred megahertz and obtains a dispersion curve related to the conductivity of the intracellular and extracellular media and the capacitance of the cell membrane.
  3. 3. The intelligent recognition and review system for abnormal samples of the platelet function analyzer according to claim 2, wherein the heterogeneous response signals acquired by the multi-physical-field collaborative sensing module are modeled as a high-order tensor in the intelligent recognition center, different dimensions of the tensor respectively correspond to a time sequence, a physical-field stimulation type, a spatial sensing point location and a signal characteristic mode, and the intelligent recognition center adopts a fusion algorithm based on tensor decomposition and graph neural network to excavate the high-order tensor in association relation.
  4. 4. The intelligent recognition and review system for abnormal samples of the platelet function analyzer according to claim 1, wherein the neural-symbol hybrid reasoning algorithm comprises a neural network subsystem and a symbol reasoning subsystem; The neural network subsystem is realized by a hierarchical transducer architecture which comprises a time transducer encoder for processing time sequence signals, a frequency spectrum transducer encoder for processing spectrum signals and a cross-modal attention fusion layer, and is responsible for extracting high-level features from the heterogeneous response signals and generating initial symbolized propositions; The symbol reasoning subsystem comprises an extensible domain knowledge base which uses a probability logic programming language to represent the knowledge of platelet physiology, pathology and interference factors as logic rules with weights; the subsystem executes an inference algorithm combining a probability map model and a Markov logic network, takes symbolized propositions generated by the neural network subsystem as evidence input, performs uncertainty reasoning under the rule constraint of a knowledge base, calculates to obtain the posterior probability distribution of anomaly types and root cause assumptions thereof, and generates an interpretable reasoning path map; The key fusion step of the neuro-symbolic hybrid reasoning algorithm is characterized by the following formula for the final decision variables Prior probability whose probability is perceived by neural network Posterior probability with sign logical reasoning Weighted integration by dirichlet distribution, i.e. Wherein the method comprises the steps of For the original multi-physical field data, In order to be a knowledge base of the knowledge, For the purpose of evidence proposition, To reflect the hyper-parameters for the confidence of the two subsystems.
  5. 5. The intelligent recognition and review system for abnormal samples of a platelet function analyzer according to claim 1, wherein the adaptive decision algorithm based on reinforcement learning models the review process as a partially observable Markov decision process, wherein the state space is defined by the combination of an abnormal hypothesis confidence vector, a sample residual, a set of available review methods and a history of operation records output by the intelligent recognition center, wherein the action space is all executable atomic review operations and parameter combinations thereof, and wherein the reward function is defined as Wherein Is to perform an action The information gain brought later, calculated by the expected KL divergence, Is the cost of the reagent consumed by the action, Is the time it takes for the action to take, , , Is a weight coefficient.
  6. 6. The intelligent abnormal sample recognition and review system of the platelet function analyzer of claim 5, wherein the integrated microfluidic chip laboratory in the automated review execution module uses digital microfluidic technology as a core manipulation platform.
  7. 7. The intelligent recognition and review system for abnormal samples of a platelet function analyzer according to claim 1, wherein the federal learning framework in the self-evolution engine of the system adopts an asynchronous federal average algorithm based on contribution evaluation.
  8. 8. The intelligent recognition and review system for abnormal samples of the platelet function analyzer according to claim 1, wherein the dynamic update algorithm of the abnormal pattern library is specifically as follows: Feature extraction and representation, namely, for each newly input abnormal sample data, mapping the abnormal sample data to a low-dimensional semantic embedding space by utilizing a trained feature extractor in the intelligent recognition center to obtain feature vectors of the abnormal sample data ; Similarity calculation and search, calculation Prototype vectors associated with all known abnormal patterns in the abnormal pattern library Cosine similarity of (1), wherein the prototype vector From calculation of the mean of all sample feature vectors belonging to the class, i.e. , For the feature extraction function, Is of the category Is a support set of (2); novel determination if maximum similarity Below a preset threshold Determining that the sample may represent a new abnormal pattern, and temporarily storing the new abnormal pattern in the to-be-confirmed emerging pattern pool; The expert auditing and confirmation is that the cases in the emerging pattern pool to be confirmed, the original signal characteristics of multiple physical fields and the primary analysis of the intelligent identification center are submitted to the human expert for auditing and confirmation; And (3) knowledge base expansion, namely adding the feature vector of the new mode confirmed by the expert into an abnormal mode base as a new prototype, triggering a knowledge base expansion flow of a symbol reasoning subsystem in the neural symbol hybrid reasoning algorithm, and allowing the expert or adding logic rules describing the new mode through a natural language processing interface to complete forward expansion of system knowledge.
  9. 9. The intelligent abnormal sample recognition and review system of claim 1, further comprising: a multisource confidence fusion and conflict resolution module; the module receives a plurality of competitive hypotheses and confidence levels of the intelligent identification center on the same abnormal source, and simultaneously receives verification evidences which are possibly generated by different rechecking methods in the automatic rechecking execution module and are mutually independent or partially related; when collision exists among different evidence bodies, an improved combination rule considering evidence distance and reliability is adopted for synthesis, and finally a comprehensive credibility function distribution is output for guiding whether to terminate rechecking or start a higher-order expert arbitration flow; the detection process is carried out by monitoring and self-calibrating subsystem in real time, running a dynamic state space model based on recurrent neural network in parallel in each main detection and re-detection process, taking sensor reading and control instruction at the previous moment as input, predicting sensor reading at the current moment, comparing predicted value with residual sequence of actual measured value, carrying out chi-square test and sequential probability ratio test on the predicted value and the actual measured value, detecting abnormal processes such as sensor drift, reagent failure, micro-channel blockage or bubble interference in real time, triggering pointed self-calibrating routine or alarming to the system once significant deviation is detected, and ensuring reliability of data acquisition.

Description

Abnormal sample intelligent recognition and rechecking system of platelet function analyzer Technical Field The invention relates to the technical field of platelet function analysis, in particular to an abnormal sample intelligent identification and rechecking system of a platelet function analyzer. Background Platelet function detection is critical for assessing bleeding risk, monitoring anti-platelet drug efficacy and diagnosing related diseases; the existing platelet function analysis technology commonly used in clinic and laboratory mainly comprises an optical nephelometry, an impedance method, a rapid platelet function analysis method, a flow cytometry and the like, wherein the optical nephelometry is used as a traditional gold standard, the detection result is obviously influenced by plasma turbidity and a sample processing process and is complex in operation and difficult in standardization, the impedance method can realize whole blood detection and is closer to an in-vivo physiological environment, signals of the impedance method are easily interfered by factors such as hematocrit, platelet count, non-platelet particles (such as small red blood cells and giant platelets) and the like, the result interpretation is complex, the rapid analysis method improves the bedside detection convenience, but the provided information dimension is limited, deep differential diagnosis is difficult to perform, the flow cytometry can realize multiparameter and high-sensitivity analysis, but has high professional requirements on operators, high cost and difficulty in realizing rapid automatic detection, the traditional various analyzers generally place analysis key points on a normal aggregation curve and parameters in the detection process, and are not suitable (such as partial activation) for sample acquisition, storage condition identifiers, blood platelet interference substances (such as hemolysin, hemolysis protein, blood platelet) or other than the typical analysis algorithms are high in the detection time, the traditional analysis algorithm is not wasted, the traditional analysis algorithm is not limited by the prior art, the clinical analysis algorithm is only wasted, but the prior diagnosis and the clinical analysis algorithm is not limited by the fact that the conventional analysis is very high in the analysis results are very difficult to be very high, and the clinical analysis results are very difficult to be extracted, and the typical analysis results are very difficult to be easily extracted by the detection results are only can be easily extracted by the peak detection, the method has the advantages that the method fails to deeply mine the rich information contained in the complete dynamic curve, the identification and the subsequent treatment of the abnormal sample are seriously dependent on the experience of operators, the standardized and intelligent reinspection decision and execution flow are lacked, and the ever-increasing requirements of clinic on the accuracy, timeliness and interpretability of the detection result are difficult to meet. Therefore, we propose an abnormal sample intelligent recognition and rechecking system of the platelet function analyzer. Disclosure of Invention The invention aims to provide an abnormal sample intelligent recognition and rechecking system of a platelet function analyzer, so as to solve the problems in the background technology. In order to achieve the above purpose, the invention provides the following technical scheme that the abnormal sample intelligent identification and rechecking system of the platelet function analyzer comprises: The multi-physical-field collaborative sensing module is used for applying various physical-field stimuli to the blood sample and synchronously collecting heterogeneous response signals generated by the physical-field stimuli; The intelligent identification center receives heterogeneous response signals acquired by the multi-physical-field cooperative sensing module through a communication link and executes a neural-symbol hybrid reasoning algorithm to identify abnormal states of samples and infer root causes of the abnormal states; The automatic re-inspection execution module receives the re-inspection instruction issued by the intelligent identification center and controls the integrated microfluidic chip laboratory to automatically reprocess and re-inspect the original sample through a self-adaptive decision algorithm based on reinforcement learning; The system self-evolution engine continuously collects the reasoning result of the intelligent identification center, the verification result of the automatic reinspection execution module and external clinical feedback information, and performs iterative optimization on model parameters and a knowledge base in the intelligent identification center through a federal learning framework and an abnormal mode library dynamic updating algorithm; and the prospective sample quality scre