Noninvasive blood separation process monitoring method and system based on infrared imaging

Abstract

The invention provides a non-invasive blood separation process monitoring method and system based on infrared imaging, and relates to the technical field of medical detection and biological separation.The non-invasive blood separation process monitoring method comprises the steps that an infrared image sequence of a blood separation device is collected through an infrared camera, and heat radiation distribution characteristics of blood components are captured; performing time domain noise reduction on the infrared image sequence by adopting a sliding window Kalman filtering algorithm, adaptively adjusting the length of a sliding window, dynamically updating a noise covariance, and outputting a noise-reduced infrared image; a convolutional neural network architecture based on multi-model switching processes the infrared image after noise reduction, and dynamically selects and loads a convolutional neural network model matched with the current blood separation stage to extract morphological characteristics of a blood separation interface; key separation parameters are calculated based on the separation interface morphological characteristics, and an early warning signal is triggered when the parameters are abnormal. According to the invention, noninvasive dynamic monitoring of clinical blood separation is realized.

Description

Technical Field

[0001] This invention relates to the field of medical detection and bioseparation technology, and in particular to a method and system for monitoring the non-invasive blood separation process based on infrared imaging. Background Technology

[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.

[0003] Blood separation is a core process in clinical treatment (such as plasma exchange and component transfusion) and biopharmaceuticals (such as plasma protein purification), and its separation effect directly affects treatment safety and product efficacy. Current mainstream monitoring technologies have significant limitations: invasive sampling requires repeated punctures to collect separated samples for biochemical analysis, which not only increases the risk of infection for patients but also disrupts the integrity of blood layering; radiation imaging technologies (such as low-dose X-ray fluoroscopy) can achieve limited visualization, but there are cumulative radiation hazards, making them unsuitable for long-term continuous monitoring; manual judgment relies on the operator's subjective assessment of the clarity of the layered interface, making it impossible to quantify key parameters (such as plasma layer height and hematocrit), resulting in an error rate as high as 15%–20%.

[0004] Infrared thermal imaging technology, with its non-invasive and radiation-free characteristics, has shown great potential in the field of medical monitoring. However, it faces the following technical challenges in blood separation scenarios: (1) Insufficient sensitivity of features: The difference in thermal radiation between plasma, red blood cells and platelets is weak (<0.5℃), and the signal-to-noise ratio (SNR) is less than 8dB due to fluctuations in ambient temperature and thermal noise from equipment; (2) Dynamic response lag: The cell sedimentation rate changes nonlinearly during centrifugation, and traditional static image processing methods (such as fixed threshold segmentation) cannot track the migration of the layered interface in real time; (3) Quantitative accuracy defects: Existing algorithms (such as edge detection) have difficulty distinguishing the spectral overlap area between plasma and buffer solution, and the interface positioning error exceeds ±3mm, which cannot meet the clinical accuracy requirements; (4) Lack of process monitoring: Existing technologies focus on the measurement and evaluation of static results (volume, purity) at the separation endpoint, lacking the ability to monitor the separation process dynamically, in real time, and quantitatively, and thus cannot detect abnormalities and issue early warnings in a timely manner during the separation process. Summary of the Invention

[0005] To address at least one of the technical problems mentioned in the background, particularly the inability of traditional methods to non-invasively, in real-time, and accurately monitor the dynamic process of blood separation, this invention utilizes infrared thermal radiation feature capture and dynamic data processing technology to achieve precise quantitative evaluation of the separation effect. This solves the technical problems of high operational risk, poor real-time performance, and weak anti-interference capability of traditional invasive sampling and radiation imaging methods, providing safe and efficient technical support for clinical blood separation.

[0006] The first aspect of this invention provides a method for monitoring a non-invasive blood separation process based on infrared imaging, comprising: Infrared image sequences of the blood separation device are acquired using an infrared camera to capture the thermal radiation distribution characteristics of blood components. The infrared image sequence is denoised in the temporal domain by using a sliding window Kalman filter algorithm. The noise covariance is dynamically updated and the length of the sliding window is adaptively adjusted to output the denoised infrared image. The convolutional neural network architecture based on multi-model switching processes the denoised infrared image, dynamically selects and loads the convolutional neural network model that matches the current blood separation stage, in order to extract the morphological features of the blood separation interface; the convolutional neural network model that matches the current blood separation stage includes a first model optimized for the rapid sedimentation phase of red blood cells, a second model optimized for the formation phase of white blood cells and platelets, and a third model optimized for the blood clarification phase. Key separation parameters are calculated based on the morphological characteristics of the separation interface, and an early warning signal is triggered when the parameters are abnormal.

[0007] Furthermore, the dynamic update of the noise covariance includes: Initialize a sliding window of length L and calculate the observation noise variance in real time: ; in, For the i-th frame of raw infrared image data, These are the Kalman filter predictions. The current frame number. This represents the length of the sliding window.

[0008] Furthermore, the time-domain noise reduction process includes dynamically switching filtering modes: When the signal-to-noise ratio is greater than or equal to the first threshold, the high-gain mode is enabled to increase the Kalman gain coefficient to accelerate the tracking of the separation interface abrupt change. When the signal-to-noise ratio is less than the first threshold, the low-gain mode is enabled, and the Kalman gain coefficient is reduced to suppress ambient noise.

[0009] Furthermore, the key separation parameters include plasma layer height. and hematocrit ; The height of the plasma layer The calculation formula is: ; hematocrit The calculation formula is: ; in, The pixel area of ​​the plasma region. This represents the total pixel area of ​​the container's cross-section. This refers to the physical height of the container.

[0010] Furthermore, the abnormal parameter triggering condition is as follows: Or the standard deviation of the separation interface fluctuation ; in, This is the warning threshold for hematocrit. This is the interface stability threshold.

[0011] Furthermore, the first model is used for the rapid sedimentation phase of erythrocytes, and its architecture includes: three convolutional layers with 32, 64, and 64 filters respectively, and sizes of 5×5 pixels, 3×3 pixels, and 3×3 pixels respectively; a max pooling layer; and two fully connected layers with 128 and 64 neurons respectively. The second model is used for the leukocyte-platelet formation phase. Its architecture includes: four convolutional layers with 32, 64, 128, and 128 filters, and sizes of 5×5 pixels, 5×5 pixels, 3×3 pixels, and 3×3 pixels, respectively; a max pooling layer; and two fully connected layers with 256 and 128 neurons, respectively. The third model is used for the blood clarification period. Its architecture includes: four convolutional layers with 64, 128, 256 and 256 filters respectively, all with a size of 3×3 pixels; a max pooling layer; and two fully connected layers with 512 and 256 neurons respectively. The convolutional neural network models matched to the current blood separation stage all employ the ReLU activation function and output the coordinates of the separation interface and the pixel area of ​​the plasma region. .

[0012] A second aspect of the present invention provides a non-invasive blood separation process monitoring system based on infrared imaging, comprising: Infrared image acquisition module: used to acquire infrared image sequences of the blood separation device through an infrared camera, capturing the thermal radiation distribution characteristics of blood components; Dynamic noise suppression module: used to perform temporal noise reduction on the infrared image sequence using a sliding window Kalman filter algorithm, dynamically update the noise covariance and adaptively adjust the sliding window length, and output the noise-reduced infrared image; Feature enhancement and extraction module: This module processes the denoised infrared image based on a multi-model switching convolutional neural network architecture, dynamically selects and loads a convolutional neural network model that matches the current blood separation stage, and extracts the morphological features of the blood separation interface. The convolutional neural network model that matches the current blood separation stage includes a first model optimized for the rapid sedimentation phase of red blood cells, a second model optimized for the formation phase of white blood cells and platelets, and a third model optimized for the blood clarification phase. Parameter generation and early warning module: used to calculate key separation parameters based on the morphological characteristics of the separation interface, and to trigger an early warning signal when the parameters are abnormal.

[0013] A third aspect of the present invention provides an electronic device including a memory, a processor, and a program stored in the memory and running on the processor, wherein the processor executes the program to implement the steps of the infrared imaging-based non-invasive blood separation process monitoring method described in the first aspect of the present invention.

[0014] A fourth aspect of the present invention provides a computer-readable storage medium having a program stored thereon that, when executed by a processor, implements the steps of the non-invasive blood separation process monitoring method based on infrared imaging as described in the first aspect of the present invention.

[0015] A fifth aspect of the present invention provides a computer program product comprising software code, wherein the program in the software code performs the steps of the infrared imaging-based non-invasive blood separation process monitoring method as described in the first aspect of the present invention.

[0016] Compared with existing technologies, the non-invasive blood separation process monitoring method and system based on infrared imaging provided by this invention has the following advantages: (1) The present invention acquires infrared image sequences by using an infrared camera and uses infrared thermal radiation imaging (non-contact, radiation-free) to replace invasive puncture or radiation imaging, thereby avoiding the risk of puncture infection and radiation accumulation hazards, and realizing non-invasive safety monitoring.

[0017] (2) The SW-KF provided by this invention calculates the noise variance through a sliding window, which can improve the signal-to-noise ratio by 12dB in a weak signal environment. By introducing a dynamic multi-model CNN architecture designed specifically for the physical characteristics of each stage of blood separation, the optimal balance between monitoring accuracy and efficiency is achieved: during the rapid sedimentation period, the first model achieves robust localization of significant interfaces with high efficiency; during the buffer layer formation period, the second model captures fine interfaces with stronger feature extraction capabilities; and during the clarification period, the third model achieves accurate judgment of the final state with the strongest abstraction capabilities. This targeted design overcomes the performance limitations of a single model in dealing with changes throughout the entire process. Combined with labeled sample training, it can accurately distinguish the spectral overlap area between plasma and buffer, reduce the interface localization error to <3%, and achieve accurate quantification of plasma layer height and hematocrit by combining key parameter calculations. (3) The present invention automatically triggers early warning (such as sound and light prompts) by preset hematocrit threshold and interface fluctuation threshold, replacing manual subjective judgment, reducing operation error and ensuring the stability of the separation process. Attached Figure Description

[0018] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.

[0019] Figure 1 This is a flowchart of the non-invasive blood separation process monitoring method based on infrared imaging provided in Embodiment 1 of the present invention; Figure 2 This is an example infrared image of blood separation during the rapid sedimentation phase of erythrocytes provided in Embodiment 1 of the present invention; Figure 3 This is an example infrared image of blood separation during the buffer layer formation period provided in Embodiment 1 of the present invention; Figure 4 This is an example infrared image of blood separation during the plasma clarification period provided in Embodiment 1 of the present invention; Figure 5 This is a schematic diagram of the anomaly separation early warning interface provided in Embodiment 1 of the present invention; Figure 6 This is a schematic diagram of a non-invasive blood separation process monitoring system based on infrared imaging provided in Embodiment 2 of the present invention. Detailed Implementation

[0020] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0021] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments of the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. Furthermore, it should be understood that the terms “comprising” and “having”, and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0022] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0023] All data acquisition in this embodiment is carried out in accordance with laws and regulations and with user consent, and the data is used legally.

[0024] Example 1 like Figure 1 This embodiment provides a method for monitoring a non-invasive blood separation process based on infrared imaging, including: Infrared image sequences of the blood separation device are acquired using an infrared camera to capture the thermal radiation distribution characteristics of blood components. The infrared image sequence is denoised in the temporal domain by using a sliding window Kalman filter algorithm. The noise covariance is dynamically updated and the length of the sliding window is adaptively adjusted to output the denoised infrared image. The convolutional neural network architecture based on multi-model switching processes the denoised infrared image, dynamically selects and loads the convolutional neural network model that matches the current blood separation stage, in order to extract the morphological features of the blood separation interface; the convolutional neural network model that matches the current blood separation stage includes a first model optimized for the rapid sedimentation phase of red blood cells, a second model optimized for the formation phase of white blood cells and platelets, and a third model optimized for the blood clarification phase. Key separation parameters are calculated based on the morphological characteristics of the separation interface, and an early warning signal is triggered when the parameters are abnormal.

[0025] The convolutional neural network model provided by this invention employs a multi-model switching architecture, comprising three convolutional neural network models optimized for the physical characteristics and visual features of different stages of blood separation, each matched to the current blood separation stage. This design is based on the understanding that blood separation is a dynamic, multi-stage process, with significant differences in interface morphology, feature scale, and recognition challenges at each stage; using a single model is unlikely to achieve optimal accuracy across all stages. The multi-model switching mechanism dynamically selects and loads the convolutional neural network model best suited to the current blood separation stage based on the time information of the centrifugation process or the interface features of the previous frame.

[0026] Specifically, the dynamic update of the noise covariance includes: Initial length is L A sliding window method is used to calculate the variance of observed noise in real time. Noise caused by centrifuge vibration and changes in cell sedimentation velocity during blood separation exhibits time-varying and non-stationary characteristics, which fixed-parameter filtering cannot adapt to. This sliding window method, used for real-time noise estimation, enables the Kalman filter to have adaptive capabilities, providing high-quality infrared image data for subsequent morphological feature extraction.

[0027] The formula for calculating the variance of observation noise is as follows: ; in, To observe the noise variance, For the first Frame of raw infrared image data, These are the Kalman filter predictions. The current frame number. This represents the length of the sliding window.

[0028] By initializing the sliding window and calculating the observation noise variance in real time, the noise suppression effect of Kalman filtering is optimized to adapt to the dynamically changing noise environment during blood separation. This solves the problem of low signal-to-noise ratio caused by ambient temperature fluctuations and equipment thermal noise in infrared images, improves the identification of weak signals (such as the slight thermal radiation differences between plasma and blood cells), and lays a high-quality data foundation for subsequent feature extraction.

[0029] Specifically, the time-domain noise reduction process includes dynamically switching filtering modes: When the signal-to-noise ratio is greater than or equal to the first threshold (e.g., 20dB), it indicates that the system is in the rapid cell sedimentation stage or the signal quality is good. At this time, the high-gain mode is enabled to increase the Kalman gain coefficient to accelerate the tracking of the separation interface change. When the signal-to-noise ratio (SNR) is less than the first threshold, it indicates that the system is in the final stage of sedimentation or in a strong noise interference environment. At this time, the low-gain mode is activated to reduce the Kalman gain coefficient and suppress environmental noise. This dual-mode adaptive mechanism ensures the real-time performance and stability of monitoring throughout the nonlinear separation process. It adapts to the nonlinear changes in cell sedimentation rate during centrifugation (such as rapid sedimentation in the early stage and slow stabilization in the later stage), avoiding tracking lag or excessive smoothing caused by a single filtering mode, and ensuring the real-time performance and stability of interface monitoring.

[0030] Specifically, the key separation parameters include plasma layer height. and hematocrit ; The height of the plasma layer The calculation formula is: ; hematocrit The calculation formula is: ; in, The pixel area of ​​the plasma region. This represents the total pixel area of ​​the container's cross-section. This refers to the physical height of the container.

[0031] This enables quantitative evaluation of separation effects, replacing subjective human judgment, and allows for precise calculation of key indicators such as plasma layer height and hematocrit (error <3%), providing data-driven support for clinical treatment and biopharmaceuticals.

[0032] Specifically, the abnormal parameter triggering condition is as follows: Or the standard deviation of the separation interface fluctuation ; in, This is the warning threshold for hematocrit. This is the interface stability threshold.

[0033] Establish an automated risk control mechanism to promptly detect separation anomalies (such as insufficient separation or interface disorder), avoid oversights in manual monitoring, and ensure the safety of the treatment or production process.

[0034] Specifically, the first model is used for the rapid sedimentation phase of red blood cells. Its architecture includes: three convolutional layers with 32, 64, and 64 filters respectively, and sizes of 5×5 pixels, 3×3 pixels, and 3×3 pixels respectively; a max pooling layer; and two fully connected layers with 128 and 64 neurons respectively. The second model is used for the leukocyte-platelet formation phase. Its architecture includes: four convolutional layers with 32, 64, 128, and 128 filters, and sizes of 5×5 pixels, 5×5 pixels, 3×3 pixels, and 3×3 pixels, respectively; a max pooling layer; and two fully connected layers with 256 and 128 neurons, respectively. The third model is used for the blood clarification period. Its architecture includes: four convolutional layers with 64, 128, 256 and 256 filters respectively, all with a size of 3×3 pixels; a max pooling layer; and two fully connected layers with 512 and 256 neurons respectively. The convolutional neural network models matched to the current blood separation stage all employ the ReLU activation function and output the coordinates of the separation interface and the pixel area of ​​the plasma region. .

[0035] The working principle of the first model (used for the rapid sedimentation phase of erythrocytes) 1. Model Architecture: The first model is specifically designed for the rapid sedimentation phase of red blood cells. Its architecture consists of the following layers in sequence: input layer (128×128 pixel grayscale image), convolutional layer 1 (using 32 5×5 pixel filters), convolutional layer 2 (using 64 3×3 pixel filters), convolutional layer 3 (using 64 3×3 pixel filters), max pooling layer (2×2 pixel window with a stride of 2), fully connected layer 1 (128 neurons), fully connected layer 2 (64 neurons), and output layer (3 neurons for coordinates and area).

[0036] 2. Model characteristics and beneficial effects: The first model employs a shallow network depth (3 convolutional layers) and a large initial convolutional kernel (5×5 pixels). This design allows the model to have a large receptive field even at a shallow level, enabling it to quickly capture a wide range of continuous interface features between the red blood cell layer and the components above it, without the need for deep and complex feature combinations.

[0037] The relatively small number of filters (32->64->64) and the smaller number of fully connected layers (128 neurons->64 neurons) result in fewer model parameters and higher computational efficiency, which can meet the high-frequency requirements for real-time monitoring during rapid settlement.

[0038] The first model focuses on solving the technical problem of salient targets that need to be located quickly during blood sedimentation, avoiding the computational redundancy and overfitting risks that may result from using complex models.

[0039] 3. Relationship with settlement characteristics: like Figure 2 As shown in the infrared image of blood separation during the rapid sedimentation phase of erythrocytes, the core characteristics of the rapid sedimentation phase of erythrocytes are a large proportion of sediment (erythrocytes), a fast sedimentation rate, and a macroscopic and significant interface formed.

[0040] The architecture of the first model was designed to suit the characteristics of this period: First, large convolutional kernels correspond to large targets: 5×5 pixel convolutional kernels can effectively perceive large areas of red cells and obvious horizontal edges, quickly locking in the approximate boundary lines, just like using a wide-angle lens for rapid scanning.

[0041] Secondly, shallow networks correspond to simple features: identifying a dominant red blood cell layer does not require deep feature layers like distinguishing buffer layers. Three convolutional layers are sufficient to extract basic features of edge -> texture -> object parts to complete the localization task.

[0042] Finally, the small model achieves high real-time performance: during the fastest settling phase, the model's processing speed can keep up with the rapid changes in the interface, ensuring real-time monitoring.

[0043] How the second model (for the leukocyte-platelet formation phase) works 1. Model Architecture: The second model is specifically designed for the leukocyte-platelet formation phase (brownish-yellow layer formation phase). Its architecture consists of the following layers in sequence: input layer (128×128 pixel grayscale image), convolutional layer 1 (using 32 5×5 pixel filters), convolutional layer 2 (using 64 5×5 pixel filters), convolutional layer 3 (using 128 3×3 pixel filters), convolutional layer 4 (using 128 3×3 pixel filters), max pooling layer (2×2 pixel window, stride 2), fully connected layer 1 (256 neurons), fully connected layer 2 (128 neurons), and output layer (3 neurons).

[0044] 2. Model characteristics and beneficial effects: Employing a deeper 4-layer convolutional structure allows the network to construct more complex feature layers. From simple edge and color features to complex textures and patterns, it can ultimately identify subtle differences between buffer layer structures such as the top layer of red blood cells, the white blood cell layer, and the platelet layer.

[0045] The first two layers use 5×5 pixel convolutional kernels for initial macroscopic localization, while the last two layers use 3×3 pixel convolutional kernels for fine feature extraction. This coarse-to-fine approach balances receptive field and parameter efficiency.

[0046] The progressively increasing number of filters (32 -> 64 -> 128 -> 128) and larger fully connected layers (256 -> 128 neurons) provide the model with sufficient capacity to learn and distinguish a variety of subtle features.

[0047] 3. Relationship with settlement characteristics: like Figure 3 The infrared image of blood separation during the buffer layer (leukocyte-platelet) formation period shows that the core characteristics of the leukocyte-platelet formation period are fine targets (the brown-yellow layer is very thin) and complex features (multi-layer structure and weak contrast).

[0048] In response, the second model uses a deeper network to handle complex features. The recognition of the brown-yellow layer requires understanding more abstract feature combinations. The deeper network can combine the edge information of the lower layer into higher-level texture concepts (such as the dense texture of the top of the red blood cell and the milky white flocculent texture of the brown-yellow layer).

[0049] Subtle and diverse visual features require a greater number of filters to capture them in parallel, ensuring that no critical information is missed.

[0050] The second model first roughly locates the cell layer region, and then focuses on the interface for detailed analysis, which is consistent with the logic of operators first finding the general area and then carefully observing the buffer layer.

[0051] How the third model (for blood clarification) works 1. Model Architecture: The third model is specifically designed for the clear blood phase. Its architecture consists of the following layers: input layer (128×128 pixel grayscale image), convolutional layer 1 (using 64 3×3 pixel filters), convolutional layer 2 (using 128 3×3 pixel filters), convolutional layer 3 (using 256 3×3 pixel filters), convolutional layer 4 (using 256 3×3 pixel filters), max pooling layer (2×2 pixel window with a stride of 2), fully connected layer 1 (512 neurons), fully connected layer 2 (256 neurons), and output layer (3 neurons).

[0052] 2. Model characteristics and beneficial effects: The third model has the deepest and widest architecture among the three convolutional neural network models. It has 4 convolutional layers and the most filters (64 -> 128 -> 256 -> 256), combined with fully connected layers (512 -> 256 neurons), providing the strongest learning and feature representation capabilities.

[0053] All convolutional kernels are stacked using 3×3 pixels. By increasing the network depth, the non-linear expressive power is enhanced, thereby learning very abstract and high-level features (such as blood uniformity and clarity), while the parameter efficiency is higher than that of using large convolutional kernels.

[0054] This model can go beyond specific edges and textures to determine the essential characteristics of plasma from a global and holistic perspective, such as identifying minute changes in spectral properties or minor turbidity at interfaces caused by mild hemolysis.

[0055] 3. Relationship with settlement characteristics: like Figure 4 As shown in the infrared image of blood separation during the plasma clarification period, the core characteristics of the blood clarification period are feature abstraction (the plasma layer itself has no obvious texture) and judgment depends on global information and high-level semantics (such as color uniformity and transparency).

[0056] In the design of the third model, considering the characteristics of the blood clarification period: Using the deepest and widest networks to process the most abstract features, determining whether blood plasma is completely clear is a high-level semantic task. This is similar to experienced experts relying on intuition, which stems from a deep synthesis of numerous low-level features. The third model's depth and breadth enable it to perform this complex feature abstraction process.

[0057] Stacked small convolutional kernels enable a layer-by-layer abstraction process. The network forces the signal through multiple nonlinear transformations, with each layer summarizing and combining the features of the previous layer, ultimately forming an abstract feature representation related to the ideal clarified state at the top layer.

[0058] For high-dimensional abstract features from deep convolutional layers, a large-capacity decision layer is needed to accurately map them to the final interface coordinates and area, ensuring the highest accuracy at the final stage.

[0059] Specifically, the training of the convolutional neural network model includes: Obtain a set of infrared image samples labeled with the boundaries of plasma regions and the location of separation interfaces; All three convolutional neural network models were trained independently using a large number of infrared images labeled with features corresponding to their respective stages, and the model parameters were optimized using the mean squared error loss function. Training with large-scale labeled samples enhances the model's generalization ability, ensuring stable feature extraction even in complex scenarios (such as different blood samples or equipment noise fluctuations), and reducing errors caused by sample differences.

[0060] In one specific embodiment, the present invention discloses an infrared monitoring data processing method for blood separation, wherein the method includes: The infrared image acquisition step is configured to: acquire a sequence of infrared images of the blood separation tube in the wavelength range of 8-14 micrometers (μm) using an infrared camera at a rate of 30 frames per second (fps) to capture the thermal radiation distribution characteristics of the blood components; The dynamic noise suppression step is configured as follows: the image sequence is denoised in the temporal domain using the sliding window Kalman filter (SW-KF) algorithm, the observation noise variance is calculated in real time by initializing a sliding window of 50 frames, and the high gain mode (SNR≥20dB) or low gain mode (SNR<20dB) is dynamically switched according to the signal-to-noise ratio (SNR) to balance the tracking speed and noise suppression. The feature enhancement and extraction steps are configured as follows: the denoised image is processed based on a multi-model switching convolutional neural network architecture; the system dynamically selects and loads a convolutional neural network model matching the current blood separation stage according to the centrifugation time or identified interface features to extract the coordinates of the blood separation interface and the area of ​​the plasma region; the convolutional neural network model matching the current blood separation stage is independently trained using more than 1,500 sets (500 sets per model) of infrared images containing normal and abnormal samples with targeted annotation; The parameter generation and early warning steps are configured to: calculate key separation parameters based on extracted features, including plasma layer height (…). ) and hematocrit ( ),when or standard deviation of interface fluctuation It triggers an early warning signal and outputs the dynamic curve of the separation interface and real-time parameter values.

[0061] In this embodiment, the specific implementation steps include: S1. Infrared Image Acquisition (1) An infrared camera acquires images of blood separation tubes at 30fps, with a wavelength range of 8–14µm; (2) The image sequence reflects the dynamic process of blood stratification.

[0062] S2. Dynamic Noise Suppression (SW-KF) (1) Initialize the window (L=50 frames) and calculate the observation noise variance in real time: ; (2) Dynamically switch filtering modes: ① High gain mode (SNR≥20dB): Fast tracking of abrupt changes at the separation interface; ②Low gain mode (SNR<20dB): Suppresses ambient noise interference.

[0063] S3. CNN Feature Extraction Based on Multi-Model Switching (1) Model switching mechanism: The system dynamically switches and loads three convolutional neural network models that match the current blood separation stage based on the time threshold of the centrifugation process or the interface morphology features identified in the previous frame (such as the number of interfaces and the clarity). In the early stages of separation (e.g., the first 2 minutes), or when the interface is simple and significant, load the first model (for the rapid sedimentation phase of erythrocytes).

[0064] During the mid-separation phase (e.g., around 5 minutes), or when multiple closely adjacent interfaces are identified, a second model (for the leukocyte-platelet formation phase) is loaded.

[0065] At the end of the separation phase (e.g., around 8 minutes), or when the upper region is homogeneous and clear, load the third model (for the blood clarification phase).

[0066] (2) Model architectures and functions matching the current blood separation stage: ① First Model (Rapid Sedimentation Phase): Employing a shallow and wide initial design, its architecture is as follows: Convolutional Layer 1 (32 5×5 pixel filters) → Convolutional Layer 2 (64 3×3 pixel filters) → Convolutional Layer 3 (64 3×3 pixel filters) → Max Pooling Layer → Fully Connected Layer 1 (128 neurons) → Fully Connected Layer 2 (64 neurons) → Output Layer. This model utilizes a large initial receptive field (5×5 pixel convolutional kernels) to quickly capture the significant, macroscopic interface between the erythrocyte layer and plasma, achieving highly efficient initial localization.

[0067] ② Second Model (Buffer Layer Formation Period): Employing a deeper and more refined design, its architecture is as follows: Convolutional Layer 1 (32 5×5 pixel filters) → Convolutional Layer 2 (64 5×5 pixel filters) → Convolutional Layer 3 (128 3×3 pixel filters) → Convolutional Layer 4 (128 3×3 pixel filters) → Max Pooling Layer → Fully Connected Layer 1 (256 neurons) → Fully Connected Layer 2 (128 neurons) → Output Layer. This model, through deeper network layers and progressively increasing filters, focuses on extracting subtle interface features between the red blood cell layer and the brown-yellow layer, and between the brown-yellow layer and the plasma layer, thus solving the buffer layer recognition challenge.

[0068] ③ The third model (blood clarification period): Employing the deepest and widest design, its architecture is as follows: Convolutional Layer 1 (64 3×3 pixel filters) → Convolutional Layer 2 (128 3×3 pixel filters) → Convolutional Layer 3 (256 3×3 pixel filters) → Convolutional Layer 4 (256 3×3 pixel filters) → Max Pooling Layer → Fully Connected Layer 1 (512 neurons) → Fully Connected Layer 2 (256 neurons) → Output Layer. This model builds a deep network by stacking a large number of small convolutional kernels, improving its ability to judge abstract features (such as plasma clarity and final interface clarity) and ensuring the highest accuracy in endpoint monitoring.

[0069] (3) Training and Output: All the above models use the ReLU activation function and are optimized using the mean squared error loss function. They receive denoised infrared images and ultimately output accurate separation interface coordinates and plasma region area. .

[0070] S4. Parameter Generation and Early Warning (1) Calculate key parameters: ①Plasma layer height: ; In the formula, This represents the pixel area of ​​the plasma region in the infrared image, specifically the pixel area occupied by the plasma portion in the denoised image output by the CNN feature extraction step.

[0071] This represents the total pixel area of ​​the test tube's cross-section, specifically the total pixel area of ​​the blood separation test tube in the infrared image. This value is extracted from the test tube outline using image segmentation technology and is correlated with the plasma area. Use the same pixel benchmark.

[0072] This indicates the actual physical height of the test tube (unit: mm), referring to the true physical height of the standard blood separation test tube used in the experiment. The conversion logic is as follows: This represents the percentage of the plasma layer's area in the test tube's cross-section, multiplied by... This allows the area ratio to be converted into the actual plasma layer height. (Unit: mm)

[0073] ②Hematocrit: ; In the formula This represents the pixel area of ​​the plasma region in the infrared image, and... The meanings are exactly the same, and they are used consistently in the hematocrit formula. Naming, and in the formula for plasma layer height Consistent.

[0074] The total pixel area of ​​the test tube cross-section is represented by the formula for the height of the plasma layer. The definitions are exactly the same.

[0075] This represents the percentage of the plasma layer area. This is the area ratio of the blood cell layer (mainly red blood cells). Since the blood cells precipitate densely after centrifugation, the cross-sectional area ratio can directly reflect the volume concentration. Therefore, this ratio is the hematocrit.

[0076] (2) Anomaly detection: If or standard deviation of interface fluctuation An alarm is triggered, and the alarm interface is as follows: Figure 5 As shown.

[0077] Example 2 like Figure 6 As shown, this embodiment provides a non-invasive blood separation process monitoring system based on infrared imaging, including: Infrared image acquisition module: used to acquire infrared image sequences of the blood separation device through an infrared camera, capturing the thermal radiation distribution characteristics of blood components; Dynamic noise suppression module: used to perform temporal noise reduction on the infrared image sequence using a sliding window Kalman filter algorithm, dynamically update the noise covariance and adaptively adjust the sliding window length, and output the noise-reduced infrared image; Feature enhancement and extraction module: This module processes the denoised infrared image based on a multi-model switching convolutional neural network architecture, dynamically selects and loads a convolutional neural network model that matches the current blood separation stage, and extracts the morphological features of the blood separation interface. The convolutional neural network model that matches the current blood separation stage includes a first model optimized for the rapid sedimentation phase of red blood cells, a second model optimized for the formation phase of white blood cells and platelets, and a third model optimized for the blood clarification phase. Parameter generation and early warning module: used to calculate key separation parameters based on the morphological characteristics of the separation interface, and to trigger an early warning signal when the parameters are abnormal.

[0078] Example 3 Embodiment 3 of the present invention provides an electronic device.

[0079] An electronic device includes a memory, a processor, and a program stored in the memory and running on the processor. When the processor executes the program, it implements the steps in the non-invasive blood separation process monitoring method based on infrared imaging as described in Embodiment 1 of the present invention.

[0080] The detailed steps are the same as those of the non-invasive blood separation process monitoring method based on infrared imaging provided in Example 1, and will not be repeated here.

[0081] Example 4 Embodiment 4 of the present invention provides a computer-readable storage medium.

[0082] A computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps in the infrared imaging-based non-invasive blood separation process monitoring method as described in Embodiment 1 of the present invention.

[0083] The detailed steps are the same as those of the non-invasive blood separation process monitoring method based on infrared imaging provided in Example 1, and will not be repeated here.

[0084] Example 5 Embodiment 5 of the present invention provides a computer program product.

[0085] A computer program product includes software code, wherein the program in the software code performs the steps of the non-invasive blood separation process monitoring method based on infrared imaging as described in Embodiment 1 of the present invention.

[0086] The detailed steps are the same as those of the non-invasive blood separation process monitoring method based on infrared imaging provided in Example 1, and will not be repeated here.

[0087] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention can be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0088] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0089] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0090] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0091] The above description is merely a preferred embodiment of this practice and is not intended to limit the scope of this practice. Various modifications and variations can be made to this practice by those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this practice should be included within the protection scope of this practice.

Claims

1. A method for monitoring a non-invasive blood separation process based on infrared imaging, characterized in that, include: Infrared image sequences of the blood separation device are acquired using an infrared camera to capture the thermal radiation distribution characteristics of blood components. The infrared image sequence is denoised in the temporal domain by using a sliding window Kalman filter algorithm. The noise covariance is dynamically updated and the length of the sliding window is adaptively adjusted to output the denoised infrared image. The convolutional neural network architecture based on multi-model switching processes the denoised infrared image, dynamically selects and loads the convolutional neural network model that matches the current blood separation stage, in order to extract the morphological features of the blood separation interface. The convolutional neural network model matched to the current blood separation stage includes a first model optimized for the rapid sedimentation phase of erythrocytes, a second model optimized for the leukocyte-platelet formation phase, and a third model optimized for the blood clarification phase. Key separation parameters are calculated based on the morphological characteristics of the separation interface, and an early warning signal is triggered when the parameters are abnormal.

2. The method as described in claim 1, characterized in that, The dynamically updated noise covariance includes: Initialize a sliding window of length L and calculate the observation noise variance in real time: ； in, For the i-th frame of raw infrared image data, These are the Kalman filter predictions. The current frame number. This represents the length of the sliding window.

3. The method as described in claim 1, characterized in that, The time-domain noise reduction process includes dynamically switching filtering modes: When the signal-to-noise ratio is greater than or equal to the first threshold, the high-gain mode is enabled to increase the Kalman gain coefficient to accelerate the tracking of the separation interface abrupt change. When the signal-to-noise ratio is less than the first threshold, the low-gain mode is enabled, and the Kalman gain coefficient is reduced to suppress ambient noise.

4. The method as described in claim 1, characterized in that, The key separation parameters include plasma layer height. and hematocrit ; The height of the plasma layer The calculation formula is: ； hematocrit The calculation formula is: ； in, The pixel area of ​​the plasma region. This represents the total pixel area of ​​the container's cross-section. This refers to the physical height of the container.

5. The method as described in claim 4, characterized in that, The abnormal triggering condition for the parameter is: Or the standard deviation of the separation interface fluctuation ; in, This is the warning threshold for hematocrit. This is the interface stability threshold.

6. The method as described in claim 1, characterized in that, The first model is used for the rapid sedimentation phase of erythrocytes. Its architecture includes: three convolutional layers with 32, 64, and 64 filters, and sizes of 5×5 pixels, 3×3 pixels, and 3×3 pixels, respectively; a max pooling layer; and two fully connected layers with 128 and 64 neurons, respectively. The second model is used for the leukocyte-platelet formation phase. Its architecture includes: four convolutional layers with 32, 64, 128, and 128 filters, and sizes of 5×5 pixels, 5×5 pixels, 3×3 pixels, and 3×3 pixels, respectively; a max pooling layer; and two fully connected layers with 256 and 128 neurons, respectively. The third model is used for the blood clarification period. Its architecture includes: four convolutional layers with 64, 128, 256 and 256 filters respectively, all with a size of 3×3 pixels; a max pooling layer; and two fully connected layers with 512 and 256 neurons respectively. The convolutional neural network models matched to the current blood separation stage all employ the ReLU activation function and output the coordinates of the separation interface and the pixel area of ​​the plasma region. .

7. A non-invasive blood separation process monitoring system based on infrared imaging, characterized in that, include: Infrared image acquisition module: used to acquire infrared image sequences of the blood separation device through an infrared camera, capturing the thermal radiation distribution characteristics of blood components; Dynamic noise suppression module: used to perform temporal noise reduction on the infrared image sequence using a sliding window Kalman filter algorithm, dynamically update the noise covariance and adaptively adjust the sliding window length, and output the noise-reduced infrared image; Feature enhancement and extraction module: This module is used to process the denoised infrared image based on a multi-model switching convolutional neural network architecture, dynamically select and load the convolutional neural network model that matches the current blood separation stage, and extract the morphological features of the blood separation interface. The convolutional neural network model matched to the current blood separation stage includes a first model optimized for the rapid sedimentation phase of erythrocytes, a second model optimized for the leukocyte-platelet formation phase, and a third model optimized for the blood clarification phase. Parameter generation and early warning module: used to calculate key separation parameters based on the morphological characteristics of the separation interface, and to trigger an early warning signal when the parameters are abnormal.

8. A non-invasive blood separation process monitoring device based on infrared imaging, characterized in that, The device includes a memory and a processor; the memory is used to store a computer program; the processor is used to implement, when executing the computer program, the non-invasive blood separation process monitoring method based on infrared imaging as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the non-invasive blood separation process monitoring method based on infrared imaging as described in any one of claims 1 to 6.

10. A computer program product, comprising software code, characterized in that, The program in the software code performs the steps of the non-invasive blood separation process monitoring method based on infrared imaging as described in any one of claims 1 to 6.

Images