Search

CN-122017274-A - Machine vision and intelligent decision-making-based cytological specimen pretreatment system and method

CN122017274ACN 122017274 ACN122017274 ACN 122017274ACN-122017274-A

Abstract

The invention discloses a cytological specimen pretreatment system and method based on machine vision and intelligent decision, which relates to the technical field of biological specimen treatment, and comprises an operation platform, wherein the operation platform is provided with a first oscillation module, a centrifugal module, a visual identification module and a specimen tube operation assembly, the labor cost and the infection probability of medical staff in the operation process are reduced, the working efficiency of the early treatment of the cytological specimen is improved, the experience requirement on the operator is lower, the risk of polluting the specimen is avoided, in addition, the cytological specimen is fully prepared, the cytological wax block, the thin-layer sheet and the supernatant are reserved, the opportunity of the subsequent grafting of the cytological specimen on other detection platforms is increased, and the value of the cytological specimen in pathological diagnosis is improved.

Inventors

  • WANG ZHENG
  • LIU DONGGE
  • PENG YANKE

Assignees

  • 北京医院

Dates

Publication Date
20260512
Application Date
20260228

Claims (10)

  1. 1. The cytological specimen pretreatment system based on machine vision and intelligent decision comprises an operation platform (1), and is characterized in that a first oscillation module (2), a centrifugal module (3), a vision identification module (41) and a sample tube operation assembly (5) are arranged on the operation platform (1); The first oscillation module (2) is used for mixing the sample liquid and the reagent in the sample tube (100), the centrifugal module (3) is used for carrying out centrifugal separation on the mixed sample liquid, the visual identification module (41) is used for carrying out image acquisition on the sample tube (100) in the processing process, and the sample tube operation assembly (5) is used for executing the operations of transferring, uncovering, covering and sucking and discharging the sample tube (100); the first oscillation module (2) and the sample tube operation assembly (5) are respectively provided with two or more groups, each group can independently work, wherein when one group is in a working state, the other group can be operated by an operator to take and put the sample tube (100); The centrifugal module (3) comprises a rotary table (31) and a base (32), wherein a rotary motor is fixedly connected to the base (32), the rotary table (31) is sleeved on an output rotating shaft of the rotary motor, a plurality of jacks which are arranged at intervals around the output rotating shaft are arranged on the rotary table (31), and each sample tube (100) is reliably positioned in each jack.
  2. 2. The machine vision and intelligent decision-based cytological specimen pretreatment system according to claim 1, wherein the first oscillation module (2) comprises an oscillation plate (21) and a bottom plate (25), a plurality of fixed sliding rails (211) are arranged on the top surface of the oscillation plate (21), the oscillation plate (21) is supported on the bottom plate (25) through springs (23), an oscillation motor (24) is arranged on the bottom plate (25), an eccentric wheel is sleeved on an output shaft of the oscillation motor (24), a sample tube (100) is arranged on a sample slot seat (101), and a sliding groove which forms a clamping limit with the fixed sliding rails (211) is arranged on the bottom surface of the sample slot seat (101); The first oscillation module (2) further comprises a telescopic rod (22), and the telescopic rod (22) can apply force on each sample slot seat (101) so that the sample slot seat (101) slides out from the fixed sliding rail (211).
  3. 3. The machine vision and intelligent decision-based cytological specimen pretreatment system according to claim 1, wherein the sample tube operation assembly (5) comprises a clamping jaw assembly (51) and a pipetting assembly (52), the clamping jaw assembly (51) is provided with a clamping jaw lifting driving assembly (511) for clamping or screwing a tube cover of the sample tube (100) so as to realize opening and closing of the tube cover, and the pipetting assembly (52) is provided with a pipetting pump lifting driving assembly (521) for driving a pipetting pump to move downwards to insert and take a suction head for sucking and discharging liquid in the sample tube (100); the operation platform (1) is provided with a suction head module (61) and a suction head recovery bin (62), the suction head module (61) is internally provided with a plurality of suction heads for inserting and taking the pipetting assembly (52), and the suction head recovery bin (62) is used for accommodating the suction heads discarded by the pipetting assembly (52).
  4. 4. The machine vision and intelligent decision-based cytological specimen pretreatment system according to claim 1, wherein the operation platform (1) further comprises a motion driving device composed of a transverse driving module (71) and a longitudinal driving module (72), and the sample tube operation assembly (5) is positioned on the transverse driving module (71); The operation platform (1) is also provided with a light source (42), the light source (42) and the visual identification module (41) are arranged at intervals, and a sample is arranged between the light source (42) and the visual identification module (41).
  5. 5. The cytological specimen pretreatment system based on machine vision and intelligent decision according to claim 1, wherein a sample reserving module (81) is further arranged on the operation platform (1), an empty sample tube (100) is arranged in the sample reserving module (81), and the sample reserving module (81) is arranged adjacent to the first oscillation module (2); The operation platform (1) is also provided with a second oscillation module (82), the second oscillation module (82) is arranged adjacent to the centrifugal module (3), and the second oscillation module (82) is provided with an oscillation effect with higher frequency and larger amplitude than the first oscillation module (2); the operation platform (1) further comprises a cell collection device (9), wherein the cell collection device (9) can be placed at the bottom of the inner side of the sample tube (100), and after the collected cells are subjected to liquid adding and fixing, the cells centrifuged in the sample tube (100) can be transferred to a target position; the cell collection device (9) comprises a filter pocket (91) and a lifting lug (92) arranged on the top side of the filter pocket (91), wherein the aperture of a filter hole of the filter pocket (91) is 0.05-1mm.
  6. 6. A machine vision and intelligent decision based cytological specimen pretreatment method for treating a machine vision and intelligent decision based cytological specimen pretreatment system according to any of claims 1-5, comprising the steps of: Step one, multi-mode data acquisition and preprocessing; Step two, multi-mode feature extraction and fusion models; Step three, calculating dynamic quantization parameters; And step four, strengthening learning-driven intelligent decision.
  7. 7. The method for preprocessing cytological specimen based on machine vision and intelligent decision as recited in claim 6, wherein said step one, collect the multisource data in the course of sample processing comprehensively, and guarantee the data quality through preprocessing; When acquiring images, using a high resolution multispectral camera system, continuously shooting multi-angle images of a sample tube (100) at a frequency of 5 frames per second in a centrifugation process, wherein the camera covers visible light and near infrared spectrum; in the image acquisition process, the acquired position is positioned at the outlet side of the centrifugal module (3), so that the image is ensured to cover the full field of view of the sample tube (100), and meanwhile, the centrifugal time point is recorded; In the process of sensor data acquisition, physical parameters of the centrifugal module (3) and the frequency and amplitude of the first oscillating module (2) are acquired in real time through sensing equipment, wherein the physical parameters comprise rotating speed, centrifugal force G value and acceleration curve, meanwhile, sample type metadata are recorded, the data are transmitted through an Internet of things sensor, and the sampling frequency is 100Hz; in a specific preprocessing process, image preprocessing and sensor data calibration are included.
  8. 8. The machine vision and intelligent decision-based cytological specimen pretreatment method of claim 6, wherein the step two is to extract and fuse multi-source features by a hybrid deep learning model; the model architecture design comprises a vision TransformerViT branch and a CNN branch, wherein the vision TransformerViT branch is input into a spliced tensor of a multispectral image, the image is firstly segmented into segments of 16 multiplied by 16 pixels, the segments are input into a ViT coder, viT is used for globally analyzing cell layer morphology, mucus distribution and impurity boundaries by using a self-attention mechanism, attention weight visualization can identify key areas, the CNN branch is used for extracting local texture features by using a lightweight CNN in parallel, the microstructure of a red cell mass and the edge information of solid impurities are mainly captured, the CNN outputs a local feature map, and the local feature map and the global features of ViT are fused through a cross-modal attention module.
  9. 9. The method for pretreating a cytological specimen based on machine vision and intelligent decision-making according to claim 6, wherein said step three, calculating quantization parameters based on fusion characteristics, introducing dynamic correction factors to adaptively change the parameters with the treatment process, including mucus residual index M, red blood cell residual degree R, cell layer morphology definition C and centrifugal efficiency score E; the mucus residual index M is calculated as follows: M=(A m /A c )×K m ×F g ; A m and A c are respectively mucus and cell layer pixel areas, semantic segmentation output is carried out through ViT branches, K m is a sample type correction coefficient, and when a sensor detects that the fluctuation of centrifugal force is more than 10%, K m is automatically multiplied by a correction factor, wherein the correction factor is specifically 1.1-1.3; F g is a centrifugal force correction factor, and the calculation formula is F g =1/(1+0.01xg), wherein G is a real-time centrifugal force; the calculation formula of the erythrocyte residue R is as follows: R=(ΣA ri /A fov )×(1+α···T); Sigma A ri is the pixel area of the erythrocyte pellet and the target detection output through CNN branches, T is the centrifugation time, A fov is the total pixel area of the image field of view, and alpha is the time attenuation factor; the calculation formula of the cell layer morphology definition C is as follows: C=(G max -G min )/G avg ; g max is the maximum value of the pixel gray scale in the cell layer region, G min is the minimum value of the pixel gray scale in the cell layer region, and G avg is the average value of the pixel gray scale in the cell layer region; the centrifugal efficiency score E is calculated as follows: E=β 1 ·C+β 2 ·(1-R)-β 3 ·M; The weights β 1 ,β 2 ,β 3 are initially set to 0.4, 0.3.
  10. 10. The method for preprocessing cytological specimen based on machine vision and intelligent decision as recited in claim 6, wherein said step four, substituting fixed rules with reinforcement learning, implementing dynamic decision generation, and including state space, action space and rewarding function when setting decision engine; the state space contains real-time values of parameters M, R, C, E, as well as sensor data, and the action space defines a variety of operations including adjusting centrifugation parameters, adding reagents, enabling filtration, starting secondary oscillations; Reward =w 1 ·E-w 2 ·Time-w 3 ·cost, wherein Time is processing Time consumption, cost is reagent consumption, weight w 1 、w 2 、w 3 is learned through training, and state is evaluated every 5 seconds in the decision process; The decision instruction is issued through the intelligent decision control system to drive the sample tube operation assembly (5) to execute, after each decision execution, new data needs to be acquired immediately, parameters M, R, C, E are recalculated, if the parameters are not improved, an abnormal processing flow is triggered, and after manual intervention data labeling, the model reinforcement learning is realized, so that a feedback loop is realized.

Description

Machine vision and intelligent decision-making-based cytological specimen pretreatment system and method Technical Field The invention relates to the technical field of biological sample processing, in particular to a cytological specimen pretreatment system and method based on machine vision and intelligent decision. Background The early treatment of the cytological specimen is a key link of pathological diagnosis and molecular detection, the treatment quality directly determines the accuracy of subsequent film-making, diagnosis and detection results, the Cytological Specimen (CS) is an important pathological diagnosis and molecular pathological detection specimen type, and the CS early treatment is a key step for determining the accuracy of pathological diagnosis and molecular detection; At present, the CS early treatment has some clinical pain points, the cytological specimen quantity is gradually increased, the early treatment work of the cell room specimen of the pathology department is heavy, the requirement on a high-quality automatic specimen treatment integrated machine is urgent, the specimen is manually treated, the probability of possible errors exists, an automatic identification system is urgently needed, the artificial caused specimen errors are stopped, the manual operation is difficult to realize flow, standardization and standardization, thus influencing the cytological tabletting and wax block manufacturing quality, causing misdiagnosis, most CS is not provided with a fixing and sterilizing program at present, the risk of infection can be brought to operation technicians in the early treatment process, and the air pathogen transmission and underground water resource pollution can be caused; Although the quality of the sheet preparation of the cytological specimen such as sputum and hydrothorax and ascites specimen is improved by the thin-layer liquid-based cytological processing instrument on the market, an automatic pre-processing platform which is in butt joint with thin-layer sheet preparation equipment is not available at present, if an automatic integrated machine platform is directly grafted, the design and the equipment for automatically evaluating the specimen processing state are absent, in an actual scene, the experience of an operator is depended, and the sample with more impurities is filtered, centrifuged for multiple times and the manner of reducing phlegm/cracking red for multiple times is adopted to reach the optimal processing state of the sample so as to determine whether the next sheet preparation can be carried out, so that the automatic assumption cannot be realized; The traditional cytological pathological diagnosis only needs to make cytological smears, throwing tablets or liquid-based cytological tablets, but the current cytological wax block making also becomes an important work for CS pretreatment, the cytological wax block is made manually, the time and the labor are wasted, and the wax block making effect needs to be improved. Disclosure of Invention The invention provides a cytological specimen pretreatment system and a cytological specimen pretreatment method based on machine vision and intelligent decision, which can effectively solve the problems that samples with more impurities need to be manually treated, experience requirements on operators are high, a quick and convenient evaluation method is not available, and sample pollution risks exist in the background art. In order to achieve the aim, the invention provides the technical scheme that the cytological specimen pretreatment system based on machine vision and intelligent decision comprises an operation platform, wherein the operation platform is provided with a first oscillation module, a centrifugal module, a visual identification module and a sample tube operation assembly; The first oscillation module is used for uniformly mixing the sample liquid and the reagent in the sample tube, the centrifugal module is used for carrying out centrifugal separation on the uniformly mixed sample liquid, the visual identification module is used for carrying out image acquisition on the sample tube in the processing process, and the sample tube operation assembly is used for executing the transfer, uncovering, covering and liquid suction and discharge operations of the sample tube; The first oscillation module and the sample tube operation assembly are respectively provided with two or more groups, each group can independently work, and when one group is in a working state, the other group can be operated by an operator to take and put the sample tube; the centrifugal module comprises a rotary table and a base, wherein a rotary motor is fixedly connected to the base, the rotary table is sleeved on an output rotating shaft of the rotary motor, a plurality of jacks which are arranged at intervals around the output rotating shaft are arranged on the rotary table, and each sample tube is reliably positioned in each ja