Blood sampling system and method based on a single-use lancet

Abstract

This application discloses a blood collection system and method based on disposable blood collection needles. The blood collection system includes: an image acquisition unit for acquiring real-time images of the blood collection operation area; an AI processing unit connected to the image acquisition unit for receiving the real-time images and using deep learning algorithms to identify and monitor the blood collection actions and sequence in the real-time images to determine the integrity of the blood collection process and the compliance of the blood collection operation, and generating detection results; and a human-computer interaction unit connected to the AI ​​processing unit for receiving the detection results and providing prompts for each operation step of the blood collection process based on the detection results. According to this application, it standardizes the blood collection operation process and reduces blood contamination.

Description

Technical Field

[0001] This application relates to the field of medical devices, and more particularly to a blood collection system and method based on a disposable blood collection needle. Background Technology

[0002] Disposable blood collection needles are essential tools for routine medical testing. Their main function is to connect a vein to a vacuum blood collection tube, using the negative pressure within the tube to draw blood from the vein. During blood collection, the needle hub must be screwed onto the needle holder. Hold the needle holder and insert it into the vein. After successful puncture, hold the needle holder in place with one hand, and insert the blood collection needle into the holder with the other. Hold the edge of the holder with your index and middle fingers, and press down on the bottom of the blood collection tube with your thumb. By squeezing the bottom of the tube, the rubber stopper passes through the blood-blocking sleeve and the collection needle, connecting with the puncture needle. Venous blood then enters the blood collection tube through the lumen of the puncture needle and the (soft tube) collection needle, thus collecting the blood sample. The specific usage procedure for disposable venous blood collection needles is as follows: 1. Tie the tourniquet, disinfect the puncture site, open the small package of the blood collection needle, and remove the needle tip sheath in preparation for puncture; 2. After the disinfectant solution dries, hold the needle handle and perform intravenous puncture. If it is a movable blood collection needle, tighten the connecting tip first. 3. After observing blood return, puncture the blood collection tube stopper with the puncture needle to collect blood; 4. When the blood collection volume reaches the rated volume, remove the tube stopper and puncture needle. (When collecting multiple tubes,) replace with other blood collection tubes and repeat the blood collection process. 5. After blood collection is complete, remove the venous puncture needle first, then remove the blood collection tube; 6. Dispose of used blood collection needles in the designated trash can.

[0003] Traditional manual blood collection methods require each step to be standardized and accurate; otherwise, improper blood collection and disinfection, or the reuse of blood collection needles, can easily lead to blood contamination. Summary of the Invention

[0004] This application provides a method and apparatus for testing the linearity of a touchscreen to standardize the blood collection process and reduce blood contamination.

[0005] To address the aforementioned technical problems, the embodiments of this application disclose the following technical solutions: On the one hand, a blood collection system based on a disposable blood collection needle is provided, including: The image acquisition unit is used to acquire real-time images of the blood collection operation area. The AI ​​processing unit is connected to the image acquisition unit and is used to receive the real-time image and use deep learning algorithms to identify and monitor the blood collection actions and blood collection sequence in the real-time image in order to determine the integrity of the blood collection process and the compliance of the blood collection operation, and generate detection results. The human-computer interaction unit, connected to the AI ​​processing unit, is used to receive the detection results and provide prompts for each operation step of the blood collection process based on the detection results.

[0006] In addition to one or more of the features disclosed above, or as an alternative, the AI ​​processing unit includes: The target detection module is equipped with a pre-trained neural network model for detecting target objects in the real-time image and outputting the category, location coordinates and confidence level of the target object. The target object includes: the outer packaging of the blood collection needle, the blood collection needle body, the puncture needle tip, and the vacuum blood collection tube cap. The temporal behavior analysis module is used to perform temporal matching between the target detection results of multiple consecutive frames and the preset blood collection process state machine to determine whether the current operation step conforms to the preset order. The compliance determination module is used to compare the target detection result with a preset rule base to generate the detection result.

[0007] In addition to one or more of the features disclosed above, or alternatively, the neural network model is a YOLOv3 model, which is pre-trained in the following manner: Collect videos of blood collection operations under different lighting, angles, and occlusion conditions in real blood collection environments to form a training dataset; Each frame of the training dataset is labeled with target objects, wherein the color of the vacuum blood collection tube cap is labeled according to clinical standards. The YOLOv3 network structure is trained under supervision using the labeled training dataset until the average accuracy of the model on the validation set reaches a preset threshold.

[0008] In addition to one or more of the features disclosed above, or as an alternative, the AI ​​processing unit is deployed on an embedded AI computing platform, and the resolution and frame rate of the real-time image meet the real-time inference requirements of the target detection module; the neural network model is deployed after optimization by the inference engine, and the single-frame inference time is less than the frame interval time of real-time blood sampling monitoring.

[0009] In addition to one or more of the features disclosed above, or alternatively, the time-series behavior analysis module is specifically used for: The puncture is considered complete when a sequence of images in consecutive frames is detected showing the needle tip approaching the skin area, the needle tip disappearing from the skin area, and blood return characteristics appearing in the skin area.

[0010] In addition to one or more of the features disclosed above, or alternatively, the AI ​​processing unit is also used for: If no key target object is detected for a consecutive preset number of frames, or if the confidence level of the key target object is lower than a preset threshold, it is determined to be an uncertain identification state. The human-computer interaction unit is also used to output prompt information in the uncertain recognition state, without performing blocking or alarm operations.

[0011] In addition to one or more of the features disclosed above, or alternatively, the AI ​​processing unit is also used for: The real-time image or video segment corresponding to the identified uncertain state is marked as a low-confidence segment and stored in the storage unit for subsequent manual review or model optimization.

[0012] In addition to one or more of the features disclosed above, or as an alternative, the AI ​​processing unit is also used to: generate an alarm signal when the detection result indicates that the blood collection process is incomplete or the blood collection operation is non-compliant; The human-computer interaction unit is also used to: execute an audible and visual alarm based on the alarm signal.

[0013] In addition to one or more of the features disclosed above, or as an alternative, it also includes: The blood collection needle recovery unit, connected to the AI ​​processing unit, is used to control the blood collection needle to move to the waste recovery bin when the detection result indicates that the blood collection process is in compliance with regulations and has been correctly completed.

[0014] On the other hand, a blood collection method is provided, applied to a blood collection system as described in any of the above claims, comprising the following steps: Real-time images of the blood collection area are acquired using the image acquisition unit; The AI ​​processing unit uses deep learning algorithms to identify and monitor the blood collection actions and sequence in the real-time images to determine the integrity of the blood collection process and the compliance of the blood collection operation, and generates detection results. The human-computer interaction unit provides prompts for each step of the blood collection process based on the test results.

[0015] One of the aforementioned technical solutions offers the following advantages or beneficial effects: The image acquisition unit captures visual information of the blood collection area in real time, and the AI ​​processing unit uses deep learning algorithms to automatically identify and monitor the blood collection actions and sequence. The human-computer interaction unit then converts the detection results into immediate prompts for each operational step, constructing a closed-loop workflow from visual perception to intelligent judgment and proactive guidance. This technical approach eliminates reliance on nurses' personal memory or post-operative review for the integrity and compliance of the blood collection process, instead providing real-time, step-by-step feedback and correction during the operation. Specifically, the AI ​​processing unit, through dual discrimination of blood collection actions and sequence, can promptly detect missed steps or sequential errors, providing prompts via voice or visual means through the human-computer interaction unit. Because the prompts are given at the moment of the error or just before it occurs, nurses can correct it immediately, thus preventing blood contamination, sample errors, or repeated punctures caused by non-standard procedures from the outset. Furthermore, the continuous operation of the system does not alter nurses' core operating habits, only intervening as an auxiliary role, thus possessing extremely high clinical acceptance. Therefore, this solution systematically reduces the operational risks of manual blood collection without requiring additional manpower, improves the standardization and traceability of the blood collection process, and ultimately results in improved patient safety and reduced nursing workload. Attached Figure Description

[0016] The technical solution and other beneficial effects of this application will become apparent from the following detailed description of specific embodiments in conjunction with the accompanying drawings.

[0017] Figure 1 This is a structural topology diagram of a blood collection system according to one embodiment of this application; Figure 2 This is a flowchart of a blood collection method according to one embodiment of this application; Figure 3 This is a flowchart of a blood collection method according to one embodiment of this application; Figure 4 This is a flowchart of a blood collection method according to one embodiment of this application; Detailed Implementation

[0018] To make the objectives, technical solutions, and beneficial effects of this application clearer, the following detailed description, in conjunction with the accompanying drawings and specific embodiments, further illustrates this application. It should be understood that the specific embodiments described in this specification are merely for explaining this application and are not intended to limit it.

[0019] In this application, unless otherwise expressly specified and limited, "above" or "below" the second feature can include direct contact between the first and second features, or contact between the first and second features through another feature between them. Furthermore, "above," "over," and "on top" of the second feature includes the first feature being directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature includes the first feature being directly above or diagonally above the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0020] In clinical blood collection practice, the procedure for using disposable blood collection needles appears mature and standardized: from disinfection, puncture, connection to disposal, to recycling, each step has clear operational guidelines. However, in actual work, problems such as blood contamination, reuse of blood collection needles, disordered blood collection sequence, and blood leakage frequently occur due to improper operation. For a long time, the industry has generally attributed the root cause of these problems to individual nurse negligence, insufficient training, or excessive workload, and has attempted to alleviate them by strengthening training, increasing double-checking, and improving management systems. While these methods have reduced errors to some extent, they have never completely eliminated the hidden dangers, especially during peak hours or in emergency situations, where human error is still difficult to avoid.

[0021] The inventors of this application discovered a common, often overlooked, flaw in the aforementioned traditional improvement paths through long-term clinical observation: they all rely on "post-procedure review" or "supervision by others," lacking an automated means that can be embedded in the blood collection process to identify actions and sequences in real time and provide immediate feedback. In other words, the industry's conventional thinking treats "people" as the core variable in quality control, with all improvement measures focused on making people more accurate and meticulous, without ever questioning whether "in the absence of objective, real-time monitoring, simply relying on human memory and a sense of responsibility is sufficient to guarantee absolute compliance of the process."

[0022] Furthermore, the inventors recognized that the integrity and compliance of the blood collection process is essentially a series of visually verifiable events with strict temporal logic (such as unpacking, removing the protective sheath, puncture, blood return, cannulation, needle removal, and recycling). These events are naturally suitable for automatic monitoring through image acquisition and pattern recognition. However, in existing technologies, although machine vision and deep learning have been widely used in industrial quality inspection, security monitoring, and other fields, almost no one has introduced them into the real-time compliance determination of the blood collection process. This is because the medical industry is accustomed to viewing blood collection as a "simple operation" and believes that it is not worthwhile to adopt complex intelligent systems; at the same time, algorithm engineers lack a deep understanding of clinical scenarios and find it difficult to convert blood collection actions into a computable sequence of visual features.

[0023] It is precisely based on this interdisciplinary cognitive gap that the inventors of this invention have taken the lead in proposing a core technical problem that has long been overlooked in the industry: how to construct an intelligent monitoring system that can automatically identify blood collection actions, verify the blood collection sequence, and provide real-time feedback during the operation, thereby transforming the compliance of the blood collection process from "human responsibility" to "system assurance." The formulation of this problem does not stem from a superficial summary of the shortcomings of existing technologies, but from an insight into the deep mechanism that "human error is essentially a lack of information feedback." Before this problem was raised, those skilled in the art generally believed that strengthening management would solve the problem, and would not have considered introducing AI visual monitoring; however, after the problem was clarified, although building a system using methods such as object detection, time series analysis, and state machine matching also required creative effort, the discovery of the problem itself had already crossed the conventional cognitive boundaries of ordinary people skilled in the art.

[0024] In view of this, this embodiment provides a blood collection system based on a disposable blood collection needle, such as... Figure 1 As shown, it mainly includes an image acquisition unit, an AI processing unit, and a human-computer interaction unit.

[0025] The image acquisition unit (e.g., an industrial camera or a webcam) is installed at the front of the nurse's workstation, covering the area where the nurse is drawing blood. This unit is used to acquire high-definition video streams and images of the blood collection area in real time, with a resolution of no less than 1280×720 and a frame rate of no less than 25fps, to ensure that key actions and object details during the blood collection process can be clearly captured.

[0026] The AI ​​processing unit is electrically or communicatively connected to the image acquisition unit, receives real-time image data, and uses deep learning algorithms to identify and monitor blood collection actions and sequences in the images. The AI ​​processing unit has a pre-set standard state machine and compliance rule base for the blood collection process. By comparing the real-time identification results with the pre-set standards, it determines whether the current blood collection process is complete and whether the blood collection operation is compliant, and generates detection results.

[0027] The human-machine interface unit is connected to the AI ​​processing unit and includes a display screen, a speaker, and / or indicator lights. The human-machine interface unit receives the detection results output by the AI ​​processing unit and provides real-time prompts for each step of the blood collection process based on these results. For example, after the nurse completes the puncture, the system will prompt "Please insert the blood collection tube"; when an operational error is detected, it will prompt "The cap color does not match the doctor's order, please check."

[0028] The system acquires real-time images of the blood collection area via an image acquisition unit, and an AI processing unit uses deep learning algorithms to identify and monitor the blood collection actions and sequence. A human-computer interaction unit then provides step-by-step prompts for each operation, forming a closed loop from visual perception to intelligent judgment and proactive guidance. This allows the system to automatically track the integrity and compliance of the blood collection process, providing real-time feedback to nurses during operations. Consequently, without altering existing operating habits, the system significantly reduces the risk of blood contamination and patient complaints caused by omissions in the process or improper actions.

[0029] To further clarify the technical implementation of the AI ​​processing unit, this embodiment specifically defines its internal modules. The AI ​​processing unit includes a target detection module, a temporal behavior analysis module, and a compliance determination module.

[0030] Target detection module: Deployed with a pre-trained neural network model (preferably YOLOv3 in this embodiment). This module performs target detection on each real-time image frame input from the image acquisition unit, outputting the target object's category, location coordinates (usually marked with a rectangle), and confidence score. The target object includes at least: the outer packaging of the blood collection needle, the blood collection needle body, the puncture needle tip, and the cap of the vacuum blood collection tube. The detection results are passed to subsequent modules in structured data format.

[0031] The temporal behavior analysis module receives continuous multi-frame detection results from the target detection module and performs temporal matching with a preset blood collection procedure. The blood collection procedure defines a complete state sequence from "removing the blood collection needle" → "unpacking" → "removing the protective sleeve" → "puncture" → "inserting the blood collection tube" → "removing the needle" → "retrieval," along with the transition conditions between each state. The module compares the actual detected object sequence with the blood collection procedure to determine whether the current operation step conforms to the preset order.

[0032] The compliance assessment module compares the specific test results output by the target detection module (such as cap color, whether the needle cap is detached, and whether the packaging is damaged) with a preset rule base. The rule base stores mandatory requirements for clinical blood collection, such as "the needle tip protective sheath must be intact before puncture" and "the color of the blood collection tube cap must match the patient's test items." If the comparison matches, a "compliant" test result is generated; if they do not match, a "non-compliant" test result is generated along with the specific violations.

[0033] The AI ​​processing unit is further divided into an object detection module, a temporal behavior analysis module, and a compliance determination module. The object detection module outputs the object category and location; the temporal behavior analysis module matches the detection results from multiple consecutive frames with a preset process state machine; and the compliance determination module compares the results with a rule base. This modular architecture allows the blood collection process to move beyond a single black-box model. Instead, it achieves dual verification of the operation sequence and procedures through a clear causal chain of "object detection → temporal matching → rule comparison," improving the interpretability and debuggability of the system's judgments.

[0034] This embodiment specifically illustrates the training process of the neural network model in the target detection module, taking YOLOv3 as an example.

[0035] First, a training dataset was collected: In a real blood collection environment, multiple cameras at different angles and under different lighting conditions (such as sunlight and LED lights), with varying degrees of hand obstruction, recorded a large number of videos of nurses performing blood collection procedures. Keyframes were extracted from the videos, resulting in no fewer than 50,000 frames.

[0036] Secondly, each image frame was manually labeled: using a labeling tool (such as LabelImg), rectangular boxes were used to label the outer packaging of the blood collection needle, the blood collection needle itself, the puncture needle tip, and the vacuum blood collection tube cap in the image. For vacuum blood collection tube caps, they were classified and labeled according to clinical standards: purple (EDTA anticoagulant tube), blue (sodium citrate anticoagulant tube), red (ordinary serum tube), green (heparin anticoagulant tube), etc. At least 5,000 instances were labeled for each category.

[0037] Next, model training is performed: a YOLOv3 network structure is used, and supervised training is conducted using the labeled training dataset mentioned above. The loss function is the weighted sum of YOLOv3's default coordinate error, confidence error, and classification error. The network weights are iteratively updated using the backpropagation algorithm, and the mean average precision (mAP) is calculated on the validation set after each training round. Training is stopped when the mAP reaches 0.95 or higher, and the model weight file is saved.

[0038] Finally, the trained model is deployed to the object detection module.

[0039] By collecting operation videos under different lighting, angles, and occlusion conditions in real blood collection environments, and labeling the outer packaging of the blood collection needle, the blood collection needle itself, the puncture needle tip, and the color of the cap according to clinical standards in each frame of the image, the YOLOv3 model was then trained under supervised supervision using the labeled dataset until the average precision reached a threshold. The neural network model trained in this way can accurately adapt to the complex visual environment of clinical blood collection. In particular, its fine-grained classification ability of cap color ensures the reliability of subsequent blood collection information verification, solving the problem of insufficient recognition accuracy of general object detection models in medical scenarios from the source of the model.

[0040] To ensure that the system can respond in real time at the blood collection site, this embodiment limits the hardware deployment and inference acceleration of the AI ​​processing unit.

[0041] The AI ​​processing unit is deployed on an embedded AI computing platform, such as an NVIDIA Jetson Xavier NX or a domestically produced edge computing device equipped with a neural network processing unit (NPU). The real-time image resolution input from the image acquisition unit is set to 1280×720, with a frame rate of 25fps, i.e., a frame interval of 40ms.

[0042] To improve inference speed, the trained YOLOv3 model is optimized using TensorRT or OpenVINO inference engines, including model quantization (converting from FP32 to FP16 or INT8), layer fusion, and operator optimization. The optimized model achieves a measured single-frame inference time of no more than 30ms on the target hardware, which is less than the frame interval of 40ms, ensuring that the system can process each frame of image in real time without any accumulated latency.

[0043] By deploying the AI ​​processing unit on an embedded AI computing platform and limiting the resolution and frame rate of real-time images to meet real-time inference requirements, while optimizing the neural network model using an inference engine to ensure that the inference time per frame is less than the frame interval, this technique ensures real-time synchronization between image acquisition and AI analysis. This avoids delayed prompts or missed actions due to processing latency, thus maintaining a smooth interactive experience even in high-frequency blood collection operations and guaranteeing the system's usability in real-world outpatient environments.

[0044] To avoid the uncertainty caused by directly identifying the complex act of "puncture", this embodiment adopts the method of time-series image sequence analysis.

[0045] The temporal behavior analysis module does not directly identify the "puncture" action category, but rather makes a logical judgment based on the target detection results of multiple consecutive frames. Specifically, the judgment condition is: when the following three features are detected sequentially in consecutive frames, the puncture action is determined to be complete: Feature 1: The needle tip is detected to be near the skin area (distance is less than a preset pixel threshold). Feature 2: In the following frames, the needle tip disappears from the skin area (indicating that it has penetrated the subcutaneous tissue). Feature 3: In the frame after the needle tip disappears, an expanded red pixel area is detected near the puncture point (blood return feature, which can be achieved through color segmentation or deep learning semantic segmentation).

[0046] The temporal behavior analysis module determines the completion of the puncture procedure by detecting a continuous image sequence of "needle tip approaching the skin area → needle tip disappearing from the skin area → blood return characteristics appearing in the skin area." This approach transforms the abstract "puncture action" into quantifiable and sequentially matchable visual feature changes, avoiding the large number of training samples and difficult-to-converge model design required for directly recognizing complex actions. Furthermore, it utilizes frame continuity to eliminate the impact of single-frame detection errors on the determination result, thus achieving accurate capture of the puncture completion moment with lower computational cost.

[0047] When the above three features appear consecutively in chronological order, the temporal behavior analysis module outputs the "puncture action completed" event.

[0048] In real-world blood collection environments, factors such as sudden changes in light, nurses' rapid hand movements obstructing the camera, and reflections from the target object can temporarily lower the confidence level of the AI ​​model's recognition. This embodiment incorporates a specific contingency mechanism to address this issue.

[0049] The AI ​​processing unit is also used to determine an uncertain identification state when no key target object corresponding to the current blood collection process stage is detected for a consecutive preset number of frames (e.g., 3 consecutive frames), or when the confidence level of the detected key target object is lower than a preset threshold (e.g., 0.6). The key target object is different in different stages: it is the blood collection needle body in the unpacking stage, the puncture needle tip in the puncture stage, and the vacuum blood collection tube cap in the blood collection stage.

[0050] Once the identification is deemed uncertain, the human-machine interface unit only outputs a prompt, such as a voice prompt "Please confirm whether the operation is correct" or a screen display "Identification is temporarily unclear, please continue," but does not trigger an audible or visual alarm, nor does it issue any blocking command to the needle retrieval unit. This design is to avoid false alarms due to brief uncertainties in the AI, which could disrupt the nurse's normal operating rhythm or even cause the nurse to be distracted and lead to a puncture accident.

[0051] The confidence level of AI model recognition is affected by a variety of factors such as lighting, occlusion, angle, and rapid hand movement, and false positives are inevitable (i.e. the actual operation is correct, but the AI ​​fails to recognize it due to image blur or occlusion).

[0052] If the system directly triggers an alarm or blocks the process (such as locking the recycling bin or forcibly stopping the process) when the identification is uncertain, it will result in: 1. If a nurse is performing a puncture and the camera is briefly blocked by her arm, the system will not detect the needle tip and will suddenly sound an alarm. This will distract the nurse and may lead to puncture errors or patient injury.

[0053] 2. If the sensor and visual recognition are briefly out of sync the moment the blood collection needle is inserted into the recycling bin, the system will misjudge it as "not properly recycled", trigger an alarm, and interfere with subsequent operations.

[0054] 3. The color of the tube cap is low due to light reflection, causing the system to alarm "tube cap mismatch". In fact, the nurse is using the correct blood collection tube.

[0055] When a key target object is not detected for a preset number of consecutive frames or its confidence level is below a threshold, the system determines it to be in an uncertain identification state. In this case, the human-computer interaction unit only outputs a prompt message without blocking or alarming. This processing mechanism is based on the prediction of objective interferences such as sudden changes in light and hand obstruction in the real blood collection environment. It adopts a "low-confidence conservative response" strategy, prioritizing the continuity of operation rather than forcibly intervening when identification is unreliable. This avoids distracting nurses or interrupting the process due to AI misdetection, reflecting the design principle that auxiliary systems should serve people rather than replace them.

[0056] To further utilize the data from identifying uncertain states, this embodiment adds a data feedback mechanism.

[0057] The AI ​​processing unit is also used to: automatically label real-time image or video clips corresponding to uncertain states as "low-confidence clips" and store them in a storage unit (such as a local hard drive or cloud storage). These clips can have two subsequent uses: Manual review: Head nurses or quality control personnel can periodically check these low-confidence segments to confirm whether there are any actual operational errors. If manual review finds that an error does exist but the system does not alert (false negative), it can be used to improve the rule base or prompting strategy.

[0058] Model optimization: Add these low-confidence segments to the training dataset, retrain or fine-tune the YOLOv3 model, and gradually improve the model's recognition accuracy and robustness in complex environments.

[0059] Image or video clips corresponding to uncertain states are marked as low-confidence segments and stored for subsequent manual review or model optimization. This method transforms the "uncertain data" generated during system operation into traceable and reusable resources. On the one hand, it provides objective review samples for nursing quality control; on the other hand, it accumulates labeled data for continuous model iteration, forming a closed-loop evolutionary path of "operation-collection-optimization," enabling the system's recognition capabilities to continuously improve with usage time.

[0060] When the AI ​​processing unit clearly determines that the blood collection process is incomplete or the operation is non-compliant (i.e., violation under high confidence), the system needs to take stronger warning measures.

[0061] The AI ​​processing unit is also used to generate an alarm signal when the detection result indicates that the blood collection process is incomplete (e.g., puncture was performed without scanning the patient's barcode) or that the blood collection operation is non-compliant (e.g., using a damaged lancet, incorrect cap color, and a confidence level higher than 0.9). Based on this alarm signal, the human-machine interaction unit immediately activates an audible and visual alarm: a red indicator light flashes, and the speaker emits a warning voice message such as "Operation error, please stop immediately."

[0062] When the test results indicate an incomplete blood collection process or non-compliant operation, the AI ​​processing unit generates an alarm signal, which is then activated by the human-computer interaction unit with both sound and light alarms. Compared to "just a prompt" in uncertain situations, this approach employs a strong warning for high-confidence violations, forming a tiered response mechanism: restraint in ambiguous situations and proactive intervention in clear errors. This tiered design prevents excessive alarms from interfering with nurses while ensuring that truly dangerous procedures are stopped promptly, achieving a balance between safety and user-friendliness.

[0063] This embodiment introduces a blood collection needle retrieval unit to achieve automatic retrieval control of used blood collection needles.

[0064] The system also includes a lancet recycling unit connected to the AI ​​processing unit. When the AI ​​processing unit determines that the blood collection process is standard and has been correctly completed (e.g., the needle has been removed and all blood collection tubes have been collected), the human-computer interaction unit provides a voice prompt: "Please put the lancet into the recycling slot." After the nurse places the used lancet into the recycling location, the recycling unit automatically transfers it to the waste recycling bin, avoiding cross-infection caused by the reuse of lancets or indiscriminate disposal.

[0065] A needle retrieval unit is introduced, which controls the movement of the needle to the waste collection bin when the blood collection process is deemed compliant and correctly completed. This solution directly links the final judgment result of visual recognition to the physical retrieval action, creating a complete closed-loop management system for the needles from retrieval to disposal. Since the retrieval action is automatically triggered only when the process is compliant and completed, the possibility of needle reuse is physically eliminated, while also reducing the number of manual sharps handling steps for nurses, thus lowering the risk of occupational exposure to needlestick injuries.

[0066] This embodiment further defines the structure of the recycling unit.

[0067] The blood collection needle retrieval unit includes: Collection location: Located on the side of the workbench, it is a recess or box with an opening for temporarily placing blood collection needles to be collected.

[0068] Motor: Connected to the AI ​​processing unit, when a recycling command is received, the motor drives the conveyor belt or flipping mechanism to move the blood collection needle placed at the recycling location to the closed waste recycling bin.

[0069] Sensors: Array-type infrared beam sensors are installed at the entrance of the lancing station. When a nurse inserts a lancet into the lancing station, the sensor detects an obstruction, immediately generates a feedback signal, and sends it to the AI ​​processing unit.

[0070] The AI ​​processing unit records information such as the retrieval time and operator of the blood collection needle based on the feedback signal, and associates it with previous blood collection records to form a complete closed-loop data chain of "needle retrieval → use → retrieval".

[0071] The lancet retrieval unit is further defined by including a retrieval location, a motor, and a sensor. The sensor monitors the insertion of the lancet and generates a feedback signal, while the AI ​​processing unit records the retrieval status based on the feedback signal. Through the mechanical and electronic coordination of "sensor confirmation—motor drive—data recording," precise perception and tamper-proof recording of the retrieval action of each lancet are achieved, forming a traceable evidence chain from lancet removal and use to retrieval, providing objective data support for medical waste management and operational performance evaluation.

[0072] This embodiment extends the quality inspection function of the AI ​​processing unit before blood collection.

[0073] The AI ​​processing unit is also used for: Identify whether the outer packaging of the blood collection needle is damaged: The target detection module identifies features such as tears and holes on the outer packaging. If it is determined to be damaged, the human-computer interaction unit will prompt "Packaging is damaged, please replace the blood collection needle".

[0074] To check if the protective sleeve of the blood collection needle is intact and not detached: After the nurse removes the packaging, check if the protective sleeve of the blood collection needle tip is still in place. If the protective sleeve is found to be detached or missing, a message will be displayed: "Protective sleeve detached, please replace the blood collection needle".

[0075] Identify the color of the vacuum blood collection tube cap and verify it against the patient's blood collection information: The target detection module identifies the color of the blood collection tube cap held by the nurse and compares it with the standard cap color corresponding to the patient's test item obtained through barcode scanning. If they do not match, the system will prompt "Cap color is incorrect, please check".

[0076] The AI ​​processing unit is also used to identify whether the outer packaging of the blood collection needle is damaged, whether the protective sleeve has been intact and detached, and to identify the color of the tube cap and verify it with the patient's blood collection information. These detections are all completed automatically before or during the blood collection operation, moving the quality inspection point forward. This allows potential hazards such as damaged packaging, detached protective sleeves, and incorrect blood collection tubes to be intercepted before they cause actual harm, ensuring blood collection safety from the source and reducing puncture failures or sample errors caused by consumable quality issues.

[0077] To achieve complete traceability of the blood collection process, this embodiment limits the recording function.

[0078] In addition to real-time recognition, the image acquisition unit is used to record video and audio of the entire blood collection process. The system also includes a high-capacity storage unit (such as a solid-state drive or network storage server) to store this video and audio data. Each record is associated with the patient ID, nurse ID, and blood collection timestamp. The storage period is at least 3 months for retrieval in case of subsequent medical disputes or adverse event investigations.

[0079] This embodiment details the functions of the human-computer interaction unit.

[0080] The human-computer interaction unit is also used to: provide voice prompts for non-compliant actions or missing key steps based on the detection results. For example: When the system detects that a nurse picks up a blood collection needle without scanning the patient's barcode, it prompts "Please scan the patient's barcode first"; When the system detects that the nurse has not inserted a blood collection tube for an extended period of time after puncture, it will prompt "Please insert a blood collection tube promptly"; When the system detects that the blood collection needle has not been put back into the recycling port after blood collection is completed, it will prompt "Please put the blood collection needle back into the recycling port".

[0081] All prompts are in voice format, combined with on-screen text display, ensuring that nurses can receive key information without looking down at the screen.

[0082] Based on the detection results, the human-computer interaction unit provides voice prompts for non-compliant actions or missing key steps. This design utilizes a voice channel to transmit instructions, allowing nurses to obtain error correction information without shifting their gaze, thus maintaining continuous focus on the task at hand. Compared to screen text prompts, voice prompts are more in line with the actual scenario of busy hands during blood collection, effectively reducing secondary errors caused by poor information access.

[0083] Reference Figure 2 This embodiment provides a blood collection method applicable to any of the above-mentioned blood collection systems, including the following steps: Step S1: Acquire real-time images of the blood collection operation area using the image acquisition unit; Step S2: The AI ​​processing unit uses deep learning algorithms to identify and monitor the blood collection actions and sequence in the real-time image to determine the integrity of the blood collection process and the compliance of the blood collection operation, and generates detection results. Step S3: The human-computer interaction unit provides prompts for each step of the blood collection process based on the detection results.

[0084] The method runs automatically after the device is started and the nurse scans the patient's barcode, until the blood collection needle is retrieved.

[0085] The system captures real-time visual information of the blood collection area using an image acquisition unit, and the AI ​​processing unit uses deep learning algorithms to automatically identify and monitor the blood collection actions and sequence. The human-computer interaction unit then converts the detection results into real-time prompts for each step, creating a closed-loop workflow from visual perception to intelligent judgment and proactive guidance. This technological approach eliminates reliance on nurses' personal memories or post-operative review for the integrity and compliance of the blood collection process, instead providing real-time, step-by-step feedback and correction during the procedure. Specifically, the AI ​​processing unit, through dual discrimination of blood collection actions and sequence, can promptly detect missed steps or sequence errors, providing prompts via voice or visual means through the human-computer interaction unit. Because the prompts are given at the moment of error or just before it occurs, nurses can correct them immediately, thus preventing blood contamination, sample errors, or duplicate punctures caused by non-standard procedures. Furthermore, the system's continuous operation does not alter nurses' core operating habits, acting only as an auxiliary role, thus enjoying high clinical acceptance. Therefore, this solution systematically reduces the operational risks of manual blood collection without requiring additional manpower, improves the standardization and traceability of the blood collection process, and ultimately results in improved patient safety and reduced nursing workload.

[0086] The above steps are provided only to help understand the method, structure, and core ideas of this application. Those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the scope of protection of the claims.

Claims

1. A blood sampling system based on a single-use lancet, characterized in that, include: The image acquisition unit is used to acquire real-time images of the blood collection operation area. The AI ​​processing unit is connected to the image acquisition unit and is used to receive the real-time image and use deep learning algorithms to identify and monitor the blood collection actions and blood collection sequence in the real-time image in order to determine the integrity of the blood collection process and the compliance of the blood collection operation, and generate detection results. The human-computer interaction unit, connected to the AI ​​processing unit, is used to receive the detection results and provide prompts for each operation step of the blood collection process based on the detection results.

2. The blood sampling system of claim 1, wherein The AI ​​processing unit includes: The target detection module is equipped with a pre-trained neural network model for detecting target objects in the real-time image and outputting the category, location coordinates and confidence level of the target object. The target object includes: the outer packaging of the blood collection needle, the blood collection needle body, the puncture needle tip, and the vacuum blood collection tube cap. The temporal behavior analysis module is used to perform temporal matching between the target detection results of multiple consecutive frames and the preset blood collection process state machine to determine whether the current operation step conforms to the preset order. The compliance determination module is used to compare the target detection result with a preset rule base to generate the detection result.

3. The blood collection system according to claim 2, characterized in that, The neural network model is a YOLOv3 model, which is pre-trained using the following methods: Collect videos of blood collection operations under different lighting, angles, and occlusion conditions in real blood collection environments to form a training dataset; Each frame of the training dataset is labeled with target objects, wherein the color of the vacuum blood collection tube cap is labeled according to clinical standards. The YOLOv3 network structure is trained under supervision using the labeled training dataset until the average accuracy of the model on the validation set reaches a preset threshold.

4. The blood collection system according to claim 2, characterized in that, The AI ​​processing unit is deployed on an embedded AI computing platform, and the resolution and frame rate of the real-time image meet the real-time inference requirements of the target detection module; the neural network model is deployed after optimization by the inference engine, and the single-frame inference time is less than the frame interval time of real-time blood sampling monitoring.

5. The blood collection system according to claim 2, characterized in that, The time-series behavior analysis module is specifically used for: The puncture is considered complete when a sequence of images in consecutive frames is detected showing the needle tip approaching the skin area, the needle tip disappearing from the skin area, and blood return characteristics appearing in the skin area.

6. The blood collection system according to claim 2, characterized in that, The AI ​​processing unit is also used for: If no key target object is detected for a consecutive preset number of frames, or if the confidence level of the key target object is lower than a preset threshold, it is determined to be an uncertain identification state. The human-computer interaction unit is also used to output prompt information in the uncertain recognition state, without performing blocking or alarm operations.

7. The blood collection system according to claim 6, characterized in that, The AI ​​processing unit is also used for: The real-time image or video segment corresponding to the identified uncertain state is marked as a low-confidence segment and stored in the storage unit for subsequent manual review or model optimization.

8. The disposable blood collection system according to claim 1, characterized in that, The AI ​​processing unit is also used to generate an alarm signal when the detection result indicates that the blood collection process is incomplete or the blood collection operation is non-compliant. The human-computer interaction unit is also used to: execute an audible and visual alarm based on the alarm signal.

9. The blood collection system according to claim 1, characterized in that, Also includes: The blood collection needle recovery unit, connected to the AI ​​processing unit, is used to control the blood collection needle to move to the waste recovery bin when the detection result indicates that the blood collection process is in compliance with regulations and has been correctly completed.

10. A blood collection method, applied to the blood collection system as described in any one of claims 1 to 9, characterized in that, Includes the following steps: Real-time images of the blood collection area are acquired using the image acquisition unit; The AI ​​processing unit uses deep learning algorithms to identify and monitor the blood collection actions and sequence in the real-time images to determine the integrity of the blood collection process and the compliance of the blood collection operation, and generates detection results. The human-computer interaction unit provides prompts for each step of the blood collection process based on the test results.

Images