WO2021110143A1 - 细胞判读方法及系统 - Google Patents

细胞判读方法及系统 Download PDF

Info

Publication number
WO2021110143A1
WO2021110143A1 PCT/CN2020/133949 CN2020133949W WO2021110143A1 WO 2021110143 A1 WO2021110143 A1 WO 2021110143A1 CN 2020133949 W CN2020133949 W CN 2020133949W WO 2021110143 A1 WO2021110143 A1 WO 2021110143A1
Authority
WO
WIPO (PCT)
Prior art keywords
interpretation
result
cell
cells
results
Prior art date
Application number
PCT/CN2020/133949
Other languages
English (en)
French (fr)
Inventor
黄仁斌
蓝兴杰
石剑峰
张京璐
Original Assignee
珠海圣美生物诊断技术有限公司
珠海横琴圣澳云智科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 珠海圣美生物诊断技术有限公司, 珠海横琴圣澳云智科技有限公司 filed Critical 珠海圣美生物诊断技术有限公司
Publication of WO2021110143A1 publication Critical patent/WO2021110143A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/60—Type of objects
    • G06V20/69—Microscopic objects, e.g. biological cells or cellular parts
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning

Definitions

  • the present disclosure relates to the technical field of biostatistical analysis, and in particular to a cell interpretation method and system.
  • the current cell interpretation method is to determine the corresponding cells through a machine learning algorithm.
  • this type of method uses machine learning algorithms to interpret a large number of cell data, and the number of interpretation results obtained is huge, and it is difficult to review the interpretation results of these cells from an artificial perspective. Therefore, the reliability of the interpretation result obtained by using a machine learning algorithm is low.
  • the accuracy rate of the current AI algorithm may decline with the application of the algorithm.
  • the reason is: First, the algorithm will encounter a variety of data in the actual use process, some of which may be training The algorithm cannot guarantee the accuracy of this type of data recognition, which will affect the overall accuracy; secondly, the AI algorithm may not have automatic learning due to the problem of its algorithm selection. Ability, or some parameters of the algorithm model cannot be automatically modified. Therefore, another one or more AI algorithms are needed to automatically supervise the accuracy of each algorithm, and then strengthen the training or optimization of the algorithm to prevent the algorithm from being low in accuracy during the application process, so as to achieve the effect of strengthening the training and further improvement of the algorithm .
  • the purpose of the present disclosure is to provide a method and system for cell interpretation that can discover algorithms with poor results in time, and then train them in time to ensure the reliability and accuracy of the algorithms, make the interpretation process more rigorous, and improve the accuracy and reliability of cell interpretation Sex.
  • the cell interpretation method provided by the present disclosure includes: acquiring cell data corresponding to a cell image, and using multiple AI algorithms to interpret the cells respectively based on the cell data to obtain multiple interpretation results of the cells; The multiple interpretation results of the cell are compared, so that multiple AI algorithms can supervise each other to obtain a comparison result; the final interpretation result of the cell is determined based on the comparison result.
  • determining the final interpretation result of the cell based on the comparison result includes: if the comparison result is that multiple interpretation results of the cell are consistent, then comparing the cell data with the target interpretation result of the cell Sent to the first client, where the target interpretation result is the result of the multiple interpretation results; receiving the manual review result sent by the first client based on the target interpretation result, wherein the manual review The result is used to characterize the correctness of the interpretation result; if it is determined that the interpretation of the multiple interpretation results is correct based on the manual review result, an analysis report of the interpretation result of the cell is generated.
  • the method further includes: sending the interpretation result analysis report to a second client, so that the user can confirm the interpretation result analysis report; and receiving the second The client confirms the result of the interpretation result analysis report; if the confirmation result is that the interpretation has no objection, then the cell data and the interpretation result analysis report of the cell are stored in the user no objection database.
  • the method further includes: if the confirmation result is that the interpretation is objectionable, marking the cell, and combining the marked cell with The confirmation result of the cell is sent to the third client, so that the staff can manually review the cell to obtain the confirmation result of the manual review; receive the confirmation result of the manual review sent by the third client; if said If the result of manual review confirms that the interpretation is objectionable, the labeled cell and multiple interpretation results of the cell are stored in the user objection database.
  • determining the final interpretation result of the cell based on the comparison result includes: if the comparison result is that the multiple interpretation results of the cell are inconsistent, then the cell data and/or the multiple interpretation results of the cell are inconsistent. Sending two interpretation results to the first client to obtain a manual review result; receiving the manual review result fed back by the first client, and sending the manual review result as an analysis report of the cell interpretation result To the second client, so that the user can confirm the interpretation result analysis report.
  • the method further includes: if the comparison result is that the multiple interpretation results of the cell are inconsistent, then comparing the cell data and/or the The multiple interpretation results of the cells are stored in the interpretation inconsistency database.
  • the method further includes: counting the number of cells in the inconsistent interpretation database, and when the number of cells in the inconsistent interpretation database reaches a preset number, retraining the multiple AI algorithms.
  • the method further includes: if the manual review result is an interpretation error, interpreting the cell and multiple interpretations of the cell The result is stored in the wrong interpretation database.
  • the present disclosure provides a cell interpretation system, which includes: an acquisition and interpretation module configured to acquire cell data corresponding to a cell image, and based on the cell data, various AI algorithms are used to interpret the cells respectively to obtain the cell data Multiple interpretation results; a comparison module configured to compare multiple interpretation results of the cell so that multiple AI algorithms can supervise each other to obtain a comparison result; a determination module configured to determine based on the comparison result The final interpretation of the cell.
  • the determining module includes: a sending unit configured to send the cell data and the target interpretation result of the cell to the first if the comparison result is that the multiple interpretation results of the cell are consistent.
  • a client wherein the target interpretation result is a result of the multiple interpretation results;
  • a receiving unit configured to receive a manual review result sent by the first client based on the target interpretation result, wherein the manual The review result is configured to characterize the correctness of the interpretation result;
  • the generating unit is configured to generate an interpretation result analysis report of the cell if it is determined that the multiple interpretation results are correctly interpreted based on the manual review result.
  • the cell interpretation method and system provided by the present disclosure first obtain cell data corresponding to cell images, and use multiple AI algorithms to interpret the cells based on the cell data to obtain multiple interpretation results of the cells; The interpretation results are compared; finally, the final interpretation results of the cells are determined based on the comparison results.
  • the present disclosure can find algorithms with poor results in time through mutual supervision of multiple algorithms, and then train them in time, thereby ensuring the reliability of the algorithm. Indirectly, it can form a supervising and advancing effect on the accuracy of the algorithm, and continuous training can be carried out. The accuracy can be improved to ensure the reliability of cell judgment, and the accuracy can be improved through training.
  • the present disclosure uses a variety of artificial intelligence algorithms to interpret cells, and determines the final interpretation result based on the comparison results of multiple interpretation results, making the interpretation process more rigorous, reducing the possibility of misjudgment and omission, thereby improving The accuracy and reliability of cell interpretation are improved.
  • FIG. 1 is a flowchart of a cell interpretation method provided by an embodiment of the disclosure
  • FIG. 2 is a flowchart of step S103 in FIG. 1;
  • FIG. 3 is a flowchart of another cell interpretation method provided by an embodiment of the disclosure.
  • FIG. 5 is a flowchart of another cell interpretation method provided by an embodiment of the disclosure.
  • FIG. 6 is a schematic structural diagram of a cell interpretation system provided by an embodiment of the disclosure.
  • FIG. 7 is a schematic diagram of the structure of the determining module in FIG. 6;
  • FIG. 8 is a schematic structural diagram of another cell interpretation system provided by an embodiment of the disclosure.
  • Icon 11-acquisition and interpretation module; 12-comparison module; 13-determination module; 14-first sending module; 15-first receiving module; 16-first storage module; 17-marking module; 18-second receiving Module; 19-Second storage module; 20-Second sending module; 21-Third receiving module; 22-Third storage module; 23-Statistics module; 24-Fourth storage module; 131-Sending unit; 132-Receiving Unit; 133-generating unit.
  • the current cell interpretation method is to determine the corresponding cell through a machine learning algorithm, and the reliability of the interpretation result obtained by a machine learning algorithm is low.
  • the cell interpretation method and system provided by the embodiments of the present disclosure utilize multiple artificial intelligence algorithms to interpret cells, and determine the final interpretation result based on the comparison results of multiple interpretation results, making the interpretation process more rigorous and reducing The possibility of misjudgment and miss-judgment is improved, thereby improving the accuracy and reliability of cell interpretation.
  • an embodiment of the present disclosure provides a cell interpretation method, which may include the following steps:
  • Step S101 Obtain cell data corresponding to the cell image, and use multiple AI algorithms to interpret the cells respectively based on the cell data to obtain multiple interpretation results of the cells.
  • a cell image contains multiple cells, so the cell data corresponding to the cell image is also data of multiple cells.
  • AI algorithms include, but are not limited to, deep learning algorithms (such as various image segmentation and recognition algorithms based on convolutional neural networks) and machine learning algorithms (such as random forests, support vector machines, neural networks, etc.) .
  • deep learning algorithms such as various image segmentation and recognition algorithms based on convolutional neural networks
  • machine learning algorithms such as random forests, support vector machines, neural networks, etc.
  • a variety of different AI algorithms are arbitrarily selected, and the process of a variety of different AI algorithms is completely independent when interpreting cells. Among them, the interpretation result depends on the interpretation content. For example, the interpretation result can be divided into two categories: the cell marker recognition result and the cell overall interpretation result.
  • the cell marker recognition result is specific to each cell and can refer to The type of fluorescent dots in the cell and/or the number of fluorescent dots of each type in the cell, and the overall interpretation result of the cell may include: the total number of cells and/or the result of cell classification, and the result of cell classification may include: cell type, and each cell type. The number of cells in this cell type.
  • the cell classification result is used as the interpretation result (that is, the multiple interpretation results of the above-mentioned cells)
  • the cell type and the number of cells in each type can be determined.
  • the first AI algorithm corresponds to the first interpretation result
  • the second AI algorithm corresponds to the second interpretation result.
  • the first interpretation result is that there are 3 cell types, namely type A, type B and type C, and the number of cells corresponding to type A is 10, the number of cells corresponding to type B is 6, and the number of cells corresponding to type C is 20.
  • the second interpretation result is that there are two types of cell types, namely type A and type B, and the number of cells corresponding to type A is 10, and the number of cells corresponding to type B is 6. It can be seen that using two different AI algorithms to interpret the same cell data can reduce misjudgments and missed judgments.
  • an AI algorithm is a combination of one or more deep learning algorithms or other non-artificial intelligence algorithms, and each AI algorithm can perform cell segmentation recognition, cell marker recognition, and cell classification. .
  • the cell segmentation operation Through the cell segmentation operation, the cell structure after segmentation can be recognized clearly, and the cell recognition can effectively exclude other non-cellular structures.
  • the recognition of cell markers Through the recognition of cell markers, the markers required for cell interpretation can be obtained. Through interpretation, cell classification results can be obtained.
  • two deep learning network algorithms may be used to interpret cells respectively. It should be noted that the two deep learning network algorithms, Mask R-CNN and Yolo, have the same interpretation process when interpreting cells in the same cell image.
  • Step 21 first identify the cell characteristics of the cells, including but not limited to: cell contour, cell staining signal and/or cell tissue; Step 22, then after identifying the cell characteristics Intracellular marker recognition is performed on the cells in the cell; Step 23: Based on the cell marker recognition result obtained in step 22, the interpretation result of Mask R-CNN on the cell is determined; among them, the above-mentioned cell marker recognition result can realize different types of fluorescence in the cell
  • the classification of points, and the interpretation of cells by Mask R-CNN can also be called the result of cell classification.
  • Mask R-CNN recognizes 10 cells, of which 2 cells have 2 red fluorescent dots and 3 green fluorescent dots; among them, fluorescent dots of different colors are used to indicate different types of fluorescent dots. There are 5 red fluorescent dots and 6 green fluorescent dots in the other 8 cells.
  • the preset interpretation standard is: the cell type containing 2 red fluorescent dots and 3 green fluorescent dots is type 1, including The cell type with 5 red fluorescent dots and 6 green fluorescent dots is type 2, so it can be seen that the interpretation result is that there are two types of cell types, namely type 1 and type 2, and the number of cells corresponding to type 1 is 2. The number of cells corresponding to type two is 8.
  • step S101 of this application in order to achieve the high efficiency of interpretation, the interpretation process is continuously occurring, that is, the two AI algorithms continuously interpret the cell data corresponding to each cell image, in order to ensure the interpretation of the two AI algorithms It is a cell in the same cell image, that is, in order to ensure the consistency of interpretation, it is necessary to obtain the cell data at the same location on the same path.
  • step S102 the multiple interpretation results of the cells are compared, so that multiple AI algorithms can supervise each other to obtain a comparison result.
  • multiple AI algorithms can be automatically supervised by each other, and algorithms with poor results can be found in time, and then trained in time, thereby ensuring the reliability of the algorithm, and indirectly forming a supervising and advancing effect on the accuracy of the algorithm. , Continuous training, so that the accuracy can be improved.
  • Step S103 Determine the final interpretation result of the cell based on the comparison result.
  • the comparison result includes, but is not limited to: multiple interpretation results of the cells are consistent, part of the interpretation results of the cells are consistent, another part of the interpretation results are inconsistent, and/or multiple interpretation results of the cells are inconsistent.
  • the final interpretation result can refer to the analysis report of the interpretation result of the cell.
  • the interpretation result analysis report includes but is not limited to cell data, multiple interpretation results of cells and/or manual review results.
  • the above-mentioned step S103 can determine the final judgment result in the following two ways: Method one, fully adopts manual review, that is, relying on the professional experience of professionals for interpretation and judgment; Method two: tool-assisted combined with manual review, that is, first The feature judgment tool developed for sex is automatically compared.
  • the above features include but are not limited to: cell morphology, cell staining signal and/or signal distribution, and then perform manual review.
  • the cell interpretation method provided by the embodiment of the present disclosure first obtains the cell data corresponding to the cell image, and uses a variety of AI algorithms to interpret the cells respectively based on the cell data to obtain multiple interpretation results of the cells; The interpretation results are compared; finally, the final interpretation results of the cells are determined based on the comparison results.
  • the embodiments of the present disclosure can detect algorithms with poor results in time through mutual supervision of multiple algorithms, and then train them in time, thereby ensuring the reliability of the algorithms. Indirectly, they can form a supervising and advancing effect on the accuracy of the algorithms, and continue training. , So that the accuracy can be improved, and finally used to ensure the reliability of cell judgment, and then the accuracy can be improved through training.
  • the embodiments of the present disclosure use a variety of artificial intelligence algorithms to interpret cells, and determine the final interpretation result based on the comparison results of multiple interpretation results, making the interpretation process more rigorous and reducing the possibility of misjudgment and omission. In turn, the accuracy and reliability of cell interpretation are improved.
  • step S103 may include the following steps:
  • step S201 if the comparison result is that the multiple interpretation results of the cell are consistent, the cell data and the target interpretation result of the cell are sent to the first client, where the target interpretation result is the result of the multiple interpretation results.
  • the embodiments of the present disclosure take the use of two AI algorithms for cell interpretation as an example. Since the first interpretation result of the cell is consistent with the second interpretation result of the cell, the first interpretation result sent to the first client is the same as the first interpretation result sent to the first client. The effect of sending the second interpretation result from the terminal is the same. Therefore, it is only necessary to send the first interpretation result to the first client, and the first interpretation result is the target interpretation result.
  • Step S202 Receive a manual review result sent by the first client based on the target interpretation result, where the manual review result is used to characterize the correctness of the interpretation result.
  • step S203 if it is determined that the multiple interpretation results are correctly interpreted based on the manual review result, a cell interpretation result analysis report is generated.
  • step S203 the method further includes:
  • Step S204 Send the interpretation result analysis report to the second client, so that the user can confirm the interpretation result analysis report.
  • Step S205 Receive a confirmation result of the interpretation result analysis report from the second client.
  • step S206 if the result of the confirmation is that the interpretation has no objection, the analysis report of the cell and the interpretation result of the cell is stored in the user no objection database.
  • the user receives the interpretation result analysis report of the cell through the second client, and can view the interpretation result analysis report of the cell after receiving it, and can confirm the interpretation result analysis report to obtain the confirmation result.
  • the confirmation result includes There are two types of interpretation without objection and interpretation with objection.
  • step S205 the method further includes:
  • step S207 if the confirmation result is that the interpretation is objectionable, the cells are marked, and the marked cells and the confirmation result of the cells are sent to the third client, so that the staff can manually review the cells and obtain the manual review confirmation result.
  • Step S208 Receive a manual review confirmation result sent by the third client.
  • the manual review confirmation result includes: determining that the interpretation has no objection and determining that the interpretation has objection. If the result of manual review confirms that there is no objection to the interpretation, the manual review confirmation result is added to the interpretation result analysis report to update the interpretation result analysis report.
  • step S209 if the result of the manual review is that the interpretation is determined to be disagreeable, the labeled cells and multiple interpretation results of the cells are stored in the user disagreement database.
  • the user can modify the interpretation result of the objectionable cell in the user objectionable database.
  • the above-mentioned multiple AI algorithms are relearned or retrained, and the data training set used during training can select cells from the user's objection database. Or regularly use cells in the user's objection database to relearn or retrain the above-mentioned multiple AI algorithms, and optimize the above-mentioned multiple AI algorithms, and then optimize the cell interpretation system.
  • step S102 the following steps may be further included:
  • step S301 if the comparison result is that the multiple interpretation results of the cell are inconsistent, the cell data and/or the multiple interpretation results of the cell are sent to the first client to obtain the manual review result.
  • Step S302 Receive the manual review result fed back by the first client, and send the manual review result as a cell interpretation result analysis report to the second client, so that the user can confirm the interpretation result analysis report.
  • Inconsistent interpretation errors can include three cases, case 1, the first interpretation result of the cell is correct, and the second interpretation result of the cell is wrong; case 2, the first interpretation result of the cell is wrong, and the second interpretation result of the cell is correct; case 3, The first interpretation result of the cell and the second interpretation result of the cell are both wrong.
  • the first interpretation result of the cell is that the number of type 1 cells is 5, the number of type 3 cells is 2, and the second interpretation result of the cell is The number of type 1 cells is 3, the number of type 3 cells is 4, and the actual result of the cells is that the number of type 1 cells is 2 and the number of type 3 cells is 6.
  • the user confirms the interpretation result analysis report, that is, confirms whether the two interpretation results are correct.
  • step S102 the method further includes:
  • step S303 if the comparison result is that the multiple interpretation results of the cells are inconsistent, the cell data and/or the multiple interpretation results of the cells are stored in the inconsistent interpretation database.
  • Step S304 Count and interpret the number of cells in the inconsistent database. When the number of cells in the inconsistent database reaches a preset number, retrain multiple AI algorithms.
  • this application is to conduct model optimization training on the second AI algorithm on the one hand, and also on the first A kind of AI algorithm for model intensive training. Therefore, this embodiment can supervise each other among several AI algorithms to prevent the low accuracy of each AI algorithm in the application, so as to achieve the effect of enhanced training and further improvement of the algorithm.
  • the data training set and the data test set need to be used on the previously trained model.
  • the data training set can also contain new training data. New training data can be generated after comparing the two interpretation results of cells. Specifically, the cell data corresponding to the cells in the three databases of inconsistent interpretation, incorrect interpretation and user objection database can be trained according to the preset Annotation is required to form new training data that is qualified and standardized.
  • the embodiment of the present disclosure takes two AI algorithms for cell interpretation as an example.
  • the two AI algorithms are relearned or retrained.
  • the used data training set can be used to interpret cells in the inconsistent database.
  • the purpose of retraining in the embodiments of the present disclosure is to optimize the above two AI algorithms until the accuracy of the two AI algorithms reaches a preset threshold.
  • the cells in the interpretation inconsistent database can be put into the data training set and the data test set in a certain proportion.
  • the conditions for retraining various AI algorithms in the embodiments of the present disclosure are not limited to the interpretation that the number of cells in the inconsistent database reaches the preset number. Therefore, the number of cells in the incorrect interpretation database reaches the preset number or the interpretation is objectionable. The number of cells in the database reaching the preset number can be used as a condition for retraining a variety of AI algorithms.
  • the number of cells in the incorrectly interpreted database is counted, and when the number of cells in the incorrectly interpreted database reaches a preset number, multiple AI algorithms are retrained.
  • step S202 the method further includes:
  • the cells and multiple interpretation results of the cells are stored in the interpretation error database.
  • the manual review result is an interpretation error.
  • the first interpretation result of the cell is 3 red fluorescent dots in the cell
  • the second interpretation result of the cell is 3 red fluorescent dots in the cell
  • the cells with incorrect interpretations and the first interpretation results of the cells are stored in the incorrect interpretation database.
  • the cells in the misinterpreted database can be used as a training set to train two AI algorithms.
  • Figure 5 provides a flowchart of another cell interpretation method.
  • the two AI algorithms are continuously trained and optimized to ensure the reliability of the algorithm.
  • the accuracy can be improved, the accuracy of the interpretation result can be guaranteed, and there are manual review operations and user confirmation operations, which ensure the rigorous interpretation process and reduce the possibility of misjudgment and omission, and improve the reliability of the cell interpretation results. .
  • an embodiment of the present disclosure provides a cell interpretation system, which may include the following modules:
  • the obtaining and interpretation module 11 is configured to obtain the cell data corresponding to the cell image, and use a variety of AI algorithms to respectively interpret the cells based on the cell data to obtain multiple interpretation results of the cells;
  • the comparison module 12 is configured to compare multiple interpretation results of cells to realize mutual automatic supervision of multiple AI algorithms
  • the determining module 13 is configured to determine the final interpretation result of the cell based on the comparison result
  • the determining module 13 includes the following units:
  • the sending unit 131 is configured to send the target interpretation result of the cell and the cell to the first client if the comparison result is that the multiple interpretation results of the cell are consistent, where the target interpretation result is the result of the multiple interpretation results;
  • the receiving unit 132 is configured to receive the manual review result sent by the first client based on the target interpretation result, where the manual review result is configured to characterize the correctness of the interpretation result;
  • the generating unit 133 is configured to generate a cell interpretation result analysis report if it is determined that the multiple interpretation results are correctly interpreted based on the manual review result.
  • the system also includes the following modules:
  • the first sending module 14 is configured to send the interpretation result analysis report to the second client, so that the user can confirm the interpretation result analysis report.
  • the first receiving module 15 is configured to receive the confirmation result of the interpretation result analysis report from the second client.
  • the first storage module 16 is configured to store the cell and the cell interpretation result analysis report in the user no objection database if the confirmation result is that the interpretation is no objection.
  • system further includes:
  • the marking module 17 is configured to mark the cell if the confirmation result is that the interpretation is objectionable, and send the marked cell and the confirmation result of the cell to the third client, so that the staff can manually review the cell and obtain the manual review confirmation result ;
  • the second receiving module 18 is configured to receive the manual review confirmation result sent by the third client;
  • the second storage module 19 is configured to store the marked cells and multiple interpretation results of the cells in the user's objection database if the manual review confirmation result is that the interpretation is determined to be objectionable.
  • system may also include the following modules:
  • the second sending module 20 is configured to send the cell data and/or the multiple interpretation results of the cell to the first client to obtain the manual review result if the comparison result is that the multiple interpretation results of the cell are inconsistent;
  • the third receiving module 21 is configured to receive the manual review result fed back by the first client, and send the manual review result as a cell interpretation result analysis report to the second client, so that the user can confirm the interpretation result analysis report.
  • the system further includes: a third storage module 22, configured to store the cell data and/or the multiple interpretation results of the cell in the inconsistent interpretation database if the comparison result is that the multiple interpretation results of the cells are inconsistent ;
  • the statistics module 23 is configured to count the number of cells in the inconsistent database, and when the number of cells in the inconsistent database reaches the preset number, retrain multiple AI algorithms;
  • the system further includes: a fourth storage module 24, if the manual review result is an interpretation error, the cells and multiple interpretation results of the cells are stored in the interpretation error database.
  • each block in the flowchart or block diagram may represent a module, program segment, or part of the code, and the module, program segment, or part of the code contains one or more Executable instructions.
  • the functions marked in the block may also occur in a different order from the order marked in the drawings. For example, two consecutive blocks can actually be executed substantially in parallel, or they can sometimes be executed in the reverse order, depending on the functions involved.
  • each block in the block diagram and/or flowchart and the combination of blocks in the block diagram and/or flowchart can be implemented by a dedicated hardware-based system that performs the specified functions or actions. Or it can be realized by a combination of dedicated hardware and computer instructions.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • General Physics & Mathematics (AREA)
  • Molecular Biology (AREA)
  • Data Mining & Analysis (AREA)
  • General Health & Medical Sciences (AREA)
  • Multimedia (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Investigating Or Analysing Biological Materials (AREA)

Abstract

一种细胞判读方法及系统，涉及生物统计分析的技术领域，包括：获取细胞图像对应的细胞数据，并基于细胞数据利用多种AI算法分别对细胞进行判读，得到细胞的多个判读结果（S101）；将细胞的多个判读结果进行比对，以使多种AI算法相互监督，得到比对结果（S102）；基于比对结果确定细胞的最终判读结果（S103）。一方面，本方法通过多算法相互监督，能够及时发现效果差的算法，进而及时训练，保证了算法的可靠性和准确度，另一方面，本方法利用多种人工智能算法对细胞进行判读，然后通过对多种判读结果的比对结果确定最终判读结果，使判读流程更加严谨，减少了误判漏判的可能性，进而提高了细胞判读的准确率和可靠性。

Description

细胞判读方法及系统
相关申请的交叉引用
本公开要求于2020年11月24日提交中国专利局的申请号为CN202011335162.2、名称为“细胞判读方法及系统”的中国专利申请的优先权，其全部内容通过引用结合在本公开中。
本公开要求于2019年12月06日提交中国专利局的申请号为CN201911246728.1、名称为“细胞判读方法及系统”的中国专利申请的优先权，其全部内容通过引用结合在本公开中。
技术领域
本公开涉及生物统计分析技术领域，尤其是涉及一种细胞判读方法及系统。
背景技术
目前细胞判读方法是通过一种机器学习的算法判读出相应的细胞。但是这类方法通过机器学习算法对大量的细胞数据进行判读，得到的判读结果数量庞大，很难从人工的角度对这些细胞的判读结果进行复核。因此对于利用一种机器学习算法得出的判读结果，可靠性较低。
再者，目前AI算法的精确率会随着算法应用可能出现下降的趋势，原因在于：第一方面，算法在实际使用的过程中，会遇到各种各样的数据，有些可能是在训练中没有遇到的，算法无法保证此类数据识别的精确度，这就会对整体的准确度带来影响；第二方面，AI算法会因其算法选型的问题，可能不具备自动学习的能力，或者算法模型部分参数无法自动修改。因此需要另外一个或多个AI算法相互自动监督各算法准确度，进而对算法强化训练或优化，防止算法在应用过程中出现准确度较低的情况，从而达到对算法加强训练和进一步提升的效果。
发明内容
本公开的目的在于提供一种细胞判读方法及系统，能够及时发现效果差的算法，进而及时训练，保证了算法的可靠性和准确度，使判读流程更加严谨，提高细胞判读的准确率和可靠性。
本公开提供的一种细胞判读方法，其中，包括：获取细胞图像对应的细胞数据，并基于所述细胞数据利用多种AI算法分别对所述细胞进行判读，得到细胞的多个判读结果；将所述细胞的多个判读结果进行比对，以使多种AI算法相互监督，得到比对结果；基于所述比对结果确定所述细胞的最终判读结果。
进一步的，基于所述比对结果确定所述细胞的最终判读结果包括：若所述比对结果为 所述细胞的多个判读结果一致，则将所述细胞数据和所述细胞的目标判读结果发送至第一客户端，其中，所述目标判读结果为所述多个判读结果中的结果；接收所述第一客户端基于所述目标判读结果发送的人工复核结果，其中，所述人工复核结果用于表征判读结果的正确性；若基于所述人工复核结果确定出所述多个判读结果判读正确，则生成所述细胞的判读结果分析报告。
进一步的，在生成所述细胞的判读结果分析报告之后，还包括：将所述判读结果分析报告发送至第二客户端，以使用户对所述判读结果分析报告进行确认；接收所述第二客户端对所述判读结果分析报告的确认结果；若所述确认结果为判读无异议，则将所述细胞数据和所述细胞的判读结果分析报告存储至用户无异议数据库。
进一步的，在接收所述第二客户端对所述判读结果分析报告的确认结果之后，还包括：若所述确认结果为判读有异议，则标记所述细胞，并将标记的所述细胞和所述细胞的确认结果发送至第三客户端，以使工作人员对所述细胞进行人工复核，得到人工复核确认结果；接收所述第三客户端发送的所述人工复核确认结果；若所述人工复核确认结果为确定判读有异议，则将标记的所述细胞和所述细胞的多个判读结果存储至用户有异议数据库。
进一步的，基于所述比对结果确定所述细胞的最终判读结果包括：若所述比对结果为所述细胞的多个判读结果不一致，则将所述细胞数据和/或所述细胞的多个判读结果发送至所述第一客户端，以得到人工复核结果；接收所述第一客户端反馈的所述人工复核结果，并将所述人工复核结果作为所述细胞的判读结果分析报告发送至第二客户端，以使用户对所述判读结果分析报告进行确认。
进一步的，在将所述细胞的多个判读结果进行比对之后，方法还包括：若所述比对结果为所述细胞的多个判读结果不一致，则将所述细胞数据和/或所述细胞的多个判读结果存储至判读不一致数据库。
进一步的，方法还包括：统计所述判读不一致数据库中细胞的个数，当所述判读不一致数据库中细胞的个数达到预设个数时，对所述多种AI算法重新进行训练。
进一步的，在接收所述第一客户端基于所述目标判读结果发送的人工复核结果之后，还包括：若所述人工复核结果为判读错误，则将所述细胞和所述细胞的多个判读结果存储至判读有误数据库。
本公开提供的一种细胞判读系统，其中，包括：获取判读模块，配置成获取细胞图像对应的细胞数据，并基于所述细胞数据利用多种AI算法分别对所述细胞进行判读，得到细胞的多个判读结果；比对模块，配置成将所述细胞的多个判读结果进行比对，以使多种AI算法相互监督，得到比对结果；确定模块，配置成基于所述比对结果确定所述细胞的最终 判读结果。
进一步的，所述确定模块，包括：发送单元，配置成若所述比对结果为所述细胞的多个判读结果一致，则将所述细胞数据和所述细胞的目标判读结果发送至第一客户端，其中，所述目标判读结果为所述多个判读结果中的结果；接收单元，配置成接收所述第一客户端基于所述目标判读结果发送的人工复核结果，其中，所述人工复核结果配置成表征判读结果的正确性；生成单元，配置成若基于所述人工复核结果确定出所述多个判读结果判读正确，则生成所述细胞的判读结果分析报告。
本公开提供的一种细胞判读方法及系统，先获取细胞图像对应的细胞数据，并基于细胞数据利用多种AI算法分别对细胞进行判读，得到细胞的多个判读结果；然后将细胞的多个判读结果进行比对；最后基于比对结果确定细胞的最终判读结果。一方面，本公开通过多算法相互监督，能够及时发现效果差的算法，进而及时训练，进而保证了算法的可靠性，间接可对算法准确率形成一个督促推进的效果，不断的进行训练，使得准确度得以提升，最终用以保证细胞判断的可靠性，再而通过训练提高准确度。另一方面，本公开利用多种人工智能算法对细胞进行判读，并基于多种判读结果的比对结果确定最终判读结果，使判读流程更加严谨，减少了误判漏判的可能性，进而提高了细胞判读的准确率和可靠性。
附图说明
为了更清楚地说明本公开具体实施方式或现有技术中的技术方案，下面将对具体实施方式或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图是本公开的一些实施方式，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1为本公开的实施例提供的一种细胞判读方法的流程图；
图2为图1中步骤S103的流程图；
图3为本公开的实施例提供的另一种细胞判读方法的流程图；
图4为本公开的实施例提供的另一种细胞判读方法的流程图；
图5为本公开的实施例提供的另一种细胞判读方法的流程图；
图6为本公开的实施例提供的一种细胞判读系统的结构示意图；
图7为图6中确定模块的结构示意图；
图8为本公开的实施例提供的另一种细胞判读系统的结构示意图。
图标：11-获取判读模块；12-比对模块；13-确定模块；14-第一发送模块；15-第一接收模块；16-第一存储模块；17-标记模块；18-第二接收模块；19-第二存储模块；20-第二发送模块；21-第三接收模块；22-第三存储模块；23-统计模块；24-第四存储模块；131-发送单 元；132-接收单元；133-生成单元。
具体实施方式
下面将结合实施例对本公开的技术方案进行清楚且完整地描述，显然，所描述的实施例是本公开一部分实施例，而不是全部的实施例。基于本公开中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本公开保护的范围。
目前细胞判读方法是通过一种机器学习的算法判读出相应的细胞，对于利用一种机器学习算法得出的判读结果，可靠性较低。基于此，本公开实施例提供的一种细胞判读方法及系统，利用多种人工智能算法对细胞进行判读，并基于多种判读结果的比对结果确定最终判读结果，使判读流程更加严谨，减少了误判漏判的可能性，进而提高了细胞判读的准确率和可靠性。
为便于对本实施例进行理解，首先对本公开实施例所公开的一种细胞判读方法进行详细介绍。
实施例一：
参照图1，本公开实施例提供了一种细胞判读方法，其中，可以包括以下步骤：
步骤S101，获取细胞图像对应的细胞数据，并基于细胞数据利用多种AI算法分别对细胞进行判读，得到细胞的多个判读结果。
在本公开实施例中，一个细胞图像中包含多个细胞，因此该细胞图像对应的细胞数据也是多个细胞的数据。AI算法(Artificial IntelligenceAlgorithm，人工智能算法)包括但不限于深度学习算法(如基于卷积神经网络的各类图像分割及识别算法)和机器学习算法(如随机森林，支持向量机及神经网络等)。任意选取多种不同的AI算法，且多种不同的AI算法在对细胞进行判读时过程完全独立。其中，判读结果依据判读内容而定，例如：判读结果可以分为两大类：细胞标志物识别结果和细胞整体判读结果，其中，细胞标志物识别结果是针对每个细胞来说的，可以指细胞内荧光点的类型和/或细胞内每种类型荧光点的数量，而细胞整体判读结果可以包括：细胞总数量和/或细胞分类结果，而细胞分类结果又可以包括：细胞类型，以及每种细胞类型下的细胞数量。将细胞分类结果作为判读结果(即上述细胞的多个判读结果)时，可以确定细胞类型以及每种类型下的细胞数量。示例性的，若AI算法为两种，则第一种AI算法对应第一判读结果，第二种AI算法对应第二判读结果。第一判读结果为细胞类型有3种，分别是类型A，类型B和类型C，且类型A对应的细胞数量为10个，类型B对应的细胞数量为6个，类型C对应的细胞数量为20个。第二判读结果为细胞类型有2种，分别是类型A，类型B，且类型A对应的细胞数量为10个，类型B对应的细胞数量为6个。由此可见，使用两个不同的AI算法对同一份细胞数据进行判读， 可以减少误判漏判。
需要注意的是，一种AI算法是由一个或多个深度学习算法或者其他非人工智能算法组合，且每一种AI算法均可以进行细胞分割识别、细胞标志物识别以及细胞分类这三步操作。通过细胞分割操作，可以使识别分割后的细胞结构清晰，细胞识别可以有效排除非细胞的其它结构。通过细胞标志物识别，可以得到细胞判读所需标志物。通过判读，可以得到细胞分类结果。
本公开实施例可以采用Mask R-CNN和/或Yolo这两种深度学习网络算法分别对细胞进行判读。需要注意的是，Mask R-CNN和Yolo这两种深度学习网络算法在对同一细胞图像中的细胞进行判读时，判读过程一致。以Mask R-CNN为例进行如下说明：步骤21，首先对细胞进行细胞特征识别，细胞特征包括但不限于：细胞轮廓、细胞染色信号和/或细胞组织；步骤22，然后对细胞特征识别后的细胞进行细胞内标志物识别；步骤23：基于步骤22得到的细胞标志物识别结果，确定Mask R-CNN对细胞的判读结果；其中，上述细胞标志物识别结果可以实现对细胞内不同类型荧光点的分类，而Mask R-CNN对细胞的判读结果又可以称为细胞分类结果。例如：Mask R-CNN识别到10个细胞，其中2个细胞内均有2个红色荧光点和3个绿色荧光点；其中，不同颜色的荧光点用于表示不同类型的荧光点。另外8个细胞内均有5个红色荧光点和6个绿色荧光点，预先设定好的判读标准为：包含有2个红色荧光点和3个绿色荧光点的细胞类型为类型一，包含有5个红色荧光点和6个绿色荧光点的细胞类型为类型二，因此可知该判读结果为，共有两种细胞类型，分别为类型一和类型二，且类型一对应的细胞数量为2个，类型二对应的细胞数量为8个。
在实施本申请上述步骤S101时，为了达到判读的高效性，判读过程是持续发生的，即两个AI算法持续地对每个细胞图像对应的细胞数据进行判读，为了保证两种AI算法判读的是同一细胞图像中的细胞，即为了保证判读的一致性，需要获取同一路径同一位置的细胞数据。
步骤S102，将细胞的多个判读结果进行比对，以使多种AI算法相互监督，得到比对结果。通过多个判读结果的比对可以实现多种AI算法相互自动监督，能够及时发现效果差的算法，进而及时训练，进而保证了算法的可靠性，间接可对算法准确率形成一个督促推进的效果，不断的进行训练，使得准确度得以提升。
步骤S103，基于比对结果确定细胞的最终判读结果。
在本公开实施例中，比对结果包括但不限于：细胞的多个判读结果一致、细胞的部分判读结果一致、另一部分判读结果不一致和/或细胞的多个判读结果不一致。最终判读结果可以指细胞的判读结果分析报告。判读结果分析报告包括但不限于细胞数据、细胞的多个 判读结果和/或人工复核结果。上述步骤S103可以通过以下两种方式确定最终判断结果：方式一，完全采用人工复核的方式，即依靠专业人员的专业经验进行判读判断；方式二：工具辅助结合人工复核的方式，即先利用针对性研发的特征判断工具进行自动化比对，上述特征包括但不限于：细胞形态、细胞染色信号和/或信号分布，然后再进行人工复核。
本公开实施例提供的一种细胞判读方法，先获取细胞图像对应的细胞数据，并基于细胞数据利用多种AI算法分别对细胞进行判读，得到细胞的多个判读结果；然后将细胞的多个判读结果进行比对；最后基于比对结果确定细胞的最终判读结果。一方面，本公开实施例通过多算法相互监督，能够及时发现效果差的算法，进而及时训练，进而保证了算法的可靠性，间接可对算法准确率形成一个督促推进的效果，不断的进行训练，使得准确度得以提升，最终用以保证细胞判断的可靠性，再而通过训练提高准确度。另一方面，本公开实施例利用多种人工智能算法对细胞进行判读，并基于多种判读结果的比对结果确定最终判读结果，使判读流程更加严谨，减少了误判漏判的可能性，进而提高了细胞判读的准确率和可靠性。
进一步的，参照图2，步骤S103可以包括以下步骤：
步骤S201，若比对结果为细胞的多个判读结果一致，则将细胞数据和细胞的目标判读结果发送至第一客户端，其中，目标判读结果为多个判读结果中的结果。
本公开实施例以采用两种AI算法对细胞进行判读为例，由于细胞的第一判读结果和细胞的第二判读结果一致，因此向第一客户端发送的第一判读结果和向第一客户端发送第二判读结果的效果一样。因此，仅向第一客户端发送第一判读结果即可，该第一判读结果为目标判读结果。
步骤S202，接收第一客户端基于目标判读结果发送的人工复核结果，其中，人工复核结果用于表征判读结果的正确性。
步骤S203，若基于人工复核结果确定出多个判读结果判读正确，则生成细胞的判读结果分析报告。
进一步的，参照图3，在步骤S203之后，方法还包括：
步骤S204，将判读结果分析报告发送至第二客户端，以使用户对判读结果分析报告进行确认。
步骤S205，接收第二客户端对判读结果分析报告的确认结果。
步骤S206，若确认结果为判读无异议，则将细胞和细胞的判读结果分析报告存储至用户无异议数据库。
在本公开实施例中，用户通过第二客户端接收细胞的判读结果分析报告，在接收之后 可以查看细胞的判读结果分析报告，并且可以对判读结果分析报告进行确认，得到确认结果，确认结果包括判读无异议和判读有异议两种。
进一步的，参照图3，在步骤S205之后，方法还包括：
步骤S207，若确认结果为判读有异议，则标记细胞，并将标记的细胞和细胞的确认结果发送至第三客户端，以使工作人员对细胞进行人工复核，得到人工复核确认结果。
步骤S208，接收第三客户端发送的人工复核确认结果。
在本公开实施例中，人工复核确认结果包括：确定判读无异议和确定判读有异议。若人工复核确认结果为确定判读无异议，则将人工复核确认结果添加到判读结果分析报告中，以更新判读结果分析报告。
步骤S209，若人工复核确认结果为确定判读有异议，则将标记的细胞和细胞的多个判读结果存储至用户有异议数据库。
在本公开实施例中，用户可以对用户有异议数据库中细胞有异议的判读结果进行修正。在用户有异议数据库中细胞的个数达到一定数量时，对上述多种AI算法进行再学习或再训练，在训练时所利用的数据训练集可以选用用户有异议数据库中的细胞。或者定期使用用户有异议数据库中细胞对上述多种AI算法进行再学习或再训练，将上述多种AI算法优化，进而优化细胞判读系统。
进一步的，参照图4，在步骤S102之后还可以包括以下步骤：
步骤S301，若比对结果为细胞的多个判读结果不一致，则将细胞数据和/或细胞的多个判读结果发送至第一客户端，以得到人工复核结果。
步骤S302，接收第一客户端反馈的人工复核结果，并将人工复核结果作为细胞的判读结果分析报告发送至第二客户端，以使用户对判读结果分析报告进行确认。
在本公开实施例中，以采用两种AI算法对细胞进行判读为例，细胞的第一判读结果和细胞的第二判读结果不一致时，人工复核结果为判读不一致错误。判读不一致错误可以包括三种情况，情况1，细胞的第一判读结果正确，细胞的第二判读结果错误；情况2，细胞的第一判读结果错误，细胞的第二判读结果正确；情况3，细胞的第一判读结果和细胞的第二判读结果均错误，例如：细胞的第一判读结果为类型一的细胞数量为5个，类型三的细胞数量为2个，细胞的第二判读结果为类型一的细胞数量为3个，类型三的细胞数量为4个，而细胞实际的结果为类型一的细胞数量为2个，类型三的细胞数量为6个。用户对判读结果分析报告进行确认，即对两种判读结果是否正确分别进行确认。
进一步的，参照图4，在步骤S102之后，方法还包括：
步骤S303，若比对结果为细胞的多个判读结果不一致，则将细胞数据和/或细胞的多个 判读结果存储至判读不一致数据库。
步骤S304，统计判读不一致数据库中细胞的个数，当判读不一致数据库中细胞的个数达到预设个数时，对多种AI算法重新进行训练。
通过统计判读不一致数据库中细胞的个数，可以确定多种AI算法的正确率是否达到预设阈值。判读不一致数据库中细胞的个数越少，说明多种AI算法的预测越准确。例如：当利用两种AI算法进行判读时，由于第一种AI算法的正确率达到了预设阈值，其判读效果较好，第二种AI算法的正确率未达到预设阈值，其判读效果较差，因此判读不一致数据库中细胞的个数在短时间内就可以达到预设个数，之后本申请一方面是要对第二种AI算法进行模型优化训练，另一方面还要对第一种AI算法进行模型强化训练。因此本实施例可以通过几个AI算法之间互相监督，防止每个AI算法在应用中出现准确度较低的情况，从而达到算法加强训练和进一步提升的效果。
无论是对第一种AI算法进行模型强化训练，还是对第二种AI算法进行模型优化训练，均需要在之前已训练好的模型上，用到数据训练集和数据测试集。数据训练集除了可以包含历史训练数据，还可以包含新的训练数据。新的训练数据可以在细胞的两个判读结果进行比对之后产生，具体的，判读不一致数据库、判读有误数据库以及用户有异议数据库三个库中的细胞对应的细胞数据均可以按预设训练要求进行标注，形成规范合格的新的训练数据。
本公开实施例以采用两种AI算法对细胞进行判读为例，在判读不一致数据库中细胞的个数达到预设个数时，对上述两种AI算法进行再学习或再训练，在训练时所利用的数据训练集可以选用判读不一致数据库中的细胞。本公开实施例重新训练的目的在于调优上述两种AI算法，直至两种AI算法的正确率达到预设阈值。具体的，可以将判读不一致数据库中的细胞按一定的比例放到数据训练集和数据测试集里。然后通过数据训练集中的细胞数据分别训练第一种AI算法和第二种AI算法，并根据数据测试集的测试结果分别调优第一种AI算法和第二种AI算法，直到第一种AI算法和第二种AI算法对数据测试集的测试结果一致为止。
本公开实施例对多种AI算法重新进行训练的条件不局限于判读不一致数据库中细胞的个数达到预设个数，因此判读有误数据库中细胞的个数达到预设个数或判读有异议数据库中细胞的个数达到预设个数均可以作为对多种AI算法重新进行训练的条件。
具体的，统计判读有误数据库中细胞的个数，当判读有误数据库中细胞的个数达到预设个数时，对多种AI算法重新进行训练。
统计判读有异议数据库中细胞的个数，当判读有异议数据库中细胞的个数达到预设个 数时，对多种AI算法重新进行训练。
进一步的，在步骤S202之后，方法还包括：
若人工复核结果为判读错误，则将细胞和细胞的多个判读结果存储至判读有误数据库。
在本公开实施例中，以采用两种AI算法对细胞进行判读为例，若细胞的第一判读结果和第二判读结果一致，但是上述两种判读结果错误，因此人工复核结果为判读错误。例如：细胞的第一判读结果为细胞内3个红色荧光点，细胞的第二判读结果为细胞内3个红色荧光点，而细胞实际为2个红色荧光点。则将判读错误的细胞以及细胞的第一判读结果存储至判读有误数据库。本公开实施例可以将判读有误数据库中的细胞作为训练集训练两种AI算法。
在本公开实施例中，以采用两种AI算法对细胞进行判读为例，图5提供了另一种细胞判读方法的流程图，两种AI算法不断的训练优化，保证了算法的可靠性，使得准确度得以提升，可以保证判读结果的准确性，且存在人工复核的操作和用户确认的操作，保证了判读流程严谨且减少了误判漏判的可能性，提高了细胞判读结果的可靠性。
实施例二：
参照图6，本公开实施例提供了一种细胞判读系统，其中，可以包括以下模块：
获取判读模块11，配置成获取细胞图像对应的细胞数据，并基于细胞数据利用多种AI算法分别对细胞进行判读，得到细胞的多个判读结果；
比对模块12，配置成将细胞的多个判读结果进行比对，以实现多种AI算法相互自动监督；
确定模块13，配置成基于比对结果确定细胞的最终判读结果；
进一步的，参照图7，确定模块13包括以下单元：
发送单元131，配置成若比对结果为细胞的多个判读结果一致，则将细胞和细胞的目标判读结果发送至第一客户端，其中，目标判读结果为多个判读结果中的结果；
接收单元132，配置成接收第一客户端基于目标判读结果发送的人工复核结果，其中，人工复核结果配置成表征判读结果的正确性；
生成单元133，配置成若基于人工复核结果确定出多个判读结果判读正确，则生成细胞的判读结果分析报告。
进一步的，参照图8，系统还包括以下模块：
第一发送模块14，配置成将判读结果分析报告发送至第二客户端，以使用户对判读结果分析报告进行确认。
第一接收模块15，配置成接收第二客户端对判读结果分析报告的确认结果。
第一存储模块16，配置成若确认结果为判读无异议，则将细胞和细胞的判读结果分析报告存储至用户无异议数据库。
进一步的，参照图8，系统还包括：
标记模块17，配置成若确认结果为判读有异议，则标记细胞，并将标记的细胞和细胞的确认结果发送至第三客户端，以使工作人员对细胞进行人工复核，得到人工复核确认结果；
第二接收模块18，配置成接收第三客户端发送的人工复核确认结果；
第二存储模块19，配置成若人工复核确认结果为确定判读有异议，则将标记的细胞和细胞的多个判读结果存储至用户有异议数据库。
进一步的，参照图8，系统还可以包括以下模块：
第二发送模块20，配置成若比对结果为细胞的多个判读结果不一致，则将细胞数据和/或细胞的多个判读结果发送至第一客户端，以得到人工复核结果；
第三接收模块21，配置成接收第一客户端反馈的人工复核结果，并将人工复核结果作为细胞的判读结果分析报告发送至第二客户端，以使用户对判读结果分析报告进行确认。
进一步的，参照图8，系统还包括：第三存储模块22，配置成若比对结果为细胞的多个判读结果不一致，则将细胞数据和/或细胞的多个判读结果存储至判读不一致数据库；
统计模块23，配置成统计判读不一致数据库中细胞的个数，当判读不一致数据库中细胞的个数达到预设个数时，对多种AI算法重新进行训练；
进一步的，参照图8，系统还包括：第四存储模块24，若人工复核结果为判读错误，则将细胞和细胞的多个判读结果存储至判读有误数据库。
附图中的流程图和框图显示了根据本公开的多个实施例的系统、方法和计算机程序产品的可能实现的体系架构、功能和操作。在这点上，流程图或框图中的每个方框可以代表一个模块、程序段或代码的一部分，所述模块、程序段或代码的一部分包含一个或多个配置成实现规定的逻辑功能的可执行指令。也应当注意，在有些作为替换的实现中，方框中所标注的功能也可以以不同于附图中所标注的顺序发生。例如，两个连续的方框实际上可以基本并行地执行，它们有时也可以按相反的顺序执行，这依所涉及的功能而定。也要注意的是，框图和/或流程图中的每个方框以及框图和/或流程图中的方框的组合，可以用执行规定的功能或动作的专用的基于硬件的系统来实现，或者可以用专用硬件与计算机指令的组合来实现。
此外，术语“第一”、“第二”、“第三”和“第四”仅用于描述目的，而不能理解为指示或暗示相对重要性。
最后应说明的是：以上各实施例仅用以说明本公开的技术方案，而非对其限制；尽管参照前述各实施例对本公开进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分或者全部技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本公开各实施例技术方案的范围。

Claims (10)

  1. 一种细胞判读方法，其特征在于，包括：
    获取细胞图像对应的细胞数据，并基于所述细胞数据利用多种AI算法分别对所述细胞进行判读，得到细胞的多个判读结果；
    将所述细胞的多个判读结果进行比对，以使多种AI算法相互监督，得到比对结果；
    基于所述比对结果确定所述细胞的最终判读结果。
  2. 根据权利要求1所述的方法，其特征在于，基于所述比对结果确定所述细胞的最终判读结果包括：
    若所述比对结果为所述细胞的多个判读结果一致，则将所述细胞数据和所述细胞的目标判读结果发送至第一客户端，其中，所述目标判读结果为所述多个判读结果中的结果；
    接收所述第一客户端基于所述目标判读结果发送的人工复核结果，其中，所述人工复核结果用于表征判读结果的正确性；
    若基于所述人工复核结果确定出所述多个判读结果判读正确，则生成所述细胞的判读结果分析报告。
  3. 根据权利要求2所述的方法，其特征在于，在生成所述细胞的判读结果分析报告之后，还包括：
    将所述判读结果分析报告发送至第二客户端，以使用户对所述判读结果分析报告进行确认；
    接收所述第二客户端对所述判读结果分析报告的确认结果；
    若所述确认结果为判读无异议，则将所述细胞数据和所述细胞的判读结果分析报告存储至用户无异议数据库。
  4. 根据权利要求3所述的方法，其特征在于，在接收所述第二客户端对所述判读结果分析报告的确认结果之后，还包括：
    若所述确认结果为判读有异议，则标记所述细胞，并将标记的所述细胞和所述细胞的确认结果发送至第三客户端，以使工作人员对所述细胞进行人工复核，得到人工复核确认结果；
    接收所述第三客户端发送的所述人工复核确认结果；
    若所述人工复核确认结果为确定判读有异议，则将标记的所述细胞和所述细胞的多个判读结果存储至用户有异议数据库。
  5. 根据权利要求2所述的方法，其特征在于，基于所述比对结果确定所述细胞的最终判读结果包括：
    若所述比对结果为所述细胞的多个判读结果不一致，则将所述细胞数据和/或所述细胞的多个判读结果发送至所述第一客户端，以得到人工复核结果；
    接收所述第一客户端反馈的所述人工复核结果，并将所述人工复核结果作为所述细胞的判读结果分析报告发送至第二客户端，以使用户对所述判读结果分析报告进行确认。
  6. 根据权利要求5所述的方法，其特征在于，在将所述细胞的多个判读结果进行比对之后，还包括：
    若所述比对结果为所述细胞的多个判读结果不一致，则将所述细胞数据和/或所述细胞的多个判读结果存储至判读不一致数据库。
  7. 根据权利要求6所述的方法，其特征在于，还包括：
    统计所述判读不一致数据库中细胞的个数，当所述判读不一致数据库中细胞的个数达到预设个数时，对所述多种AI算法重新进行训练。
  8. 根据权利要求2所述的方法，其特征在于，在接收所述第一客户端基于所述目标判读结果发送的人工复核结果之后，还包括：
    若所述人工复核结果为判读错误，则将所述细胞数据和所述细胞的多个判读结果存储至判读有误数据库。
  9. 一种细胞判读系统，其特征在于，包括：
    获取判读模块，配置成获取细胞图像对应的细胞数据，并基于所述细胞数据利用多种AI算法分别对所述细胞进行判读，得到细胞的多个判读结果；
    比对模块，配置成将所述细胞的多个判读结果进行比对，以使多种AI算法相互监督，得到比对结果；
    确定模块，配置成基于所述比对结果确定所述细胞的最终判读结果。
  10. 根据权利要求9所述的系统，其特征在于，所述确定模块，包括：
    发送单元，配置成若所述比对结果为所述细胞的多个判读结果一致，则将所述细胞和所述细胞的目标判读结果发送至第一客户端，其中，所述目标判读结果为所述多个判读结果中的结果；
    接收单元，配置成接收所述第一客户端基于所述目标判读结果发送的人工复核结果，其中，所述人工复核结果配置成表征判读结果的正确性；
    生成单元，配置成若基于所述人工复核结果确定出所述多个判读结果判读正确，则生成所述细胞的判读结果分析报告。
PCT/CN2020/133949 2019-12-06 2020-12-04 细胞判读方法及系统 WO2021110143A1 (zh)

Applications Claiming Priority (4)

Application Number Priority Date Filing Date Title
CN201911246728 2019-12-06
CN201911246728.1 2019-12-06
CN202011335162.2A CN112257681A (zh) 2019-12-06 2020-11-24 细胞判读方法及系统
CN202011335162.2 2020-11-24

Publications (1)

Publication Number Publication Date
WO2021110143A1 true WO2021110143A1 (zh) 2021-06-10

Family

ID=74225509

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/133949 WO2021110143A1 (zh) 2019-12-06 2020-12-04 细胞判读方法及系统

Country Status (2)

Country Link
CN (1) CN112257681A (zh)
WO (1) WO2021110143A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113658642A (zh) * 2021-08-16 2021-11-16 杭州凯曼健康科技有限公司 基于荧光免疫层析法的荧光曲线的检测方法与装置

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102682305A (zh) * 2012-04-25 2012-09-19 深圳市迈科龙医疗设备有限公司 宫颈液基细胞学自动筛查方法和系统
CN106991673A (zh) * 2017-05-18 2017-07-28 深思考人工智能机器人科技（北京）有限公司 一种可解释性的宫颈细胞图像快速分级识别方法及系统
CN107274386A (zh) * 2017-05-18 2017-10-20 深思考人工智能机器人科技（北京）有限公司 一种宫颈细胞液基涂片人工智能辅助阅片系统
US20180068083A1 (en) * 2014-12-08 2018-03-08 20/20 Gene Systems, Inc. Methods and machine learning systems for predicting the likelihood or risk of having cancer
CN108961242A (zh) * 2018-07-04 2018-12-07 北京临近空间飞行器系统工程研究所 一种荧光染色图像ctc智能识别方法

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109858386A (zh) * 2019-01-09 2019-06-07 深圳市趣方科技有限公司 一种基于荧光显微图像的微藻细胞识别方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102682305A (zh) * 2012-04-25 2012-09-19 深圳市迈科龙医疗设备有限公司 宫颈液基细胞学自动筛查方法和系统
US20180068083A1 (en) * 2014-12-08 2018-03-08 20/20 Gene Systems, Inc. Methods and machine learning systems for predicting the likelihood or risk of having cancer
CN106991673A (zh) * 2017-05-18 2017-07-28 深思考人工智能机器人科技（北京）有限公司 一种可解释性的宫颈细胞图像快速分级识别方法及系统
CN107274386A (zh) * 2017-05-18 2017-10-20 深思考人工智能机器人科技（北京）有限公司 一种宫颈细胞液基涂片人工智能辅助阅片系统
CN108961242A (zh) * 2018-07-04 2018-12-07 北京临近空间飞行器系统工程研究所 一种荧光染色图像ctc智能识别方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
CHE SHUANLONG, LIU DONG; LIU SI; LUO PIFU: "Applicational Progress and Challenges of the Artificial Intelligence-Aided Cervical Cancer Cytological Screening", CHINESE JOURNAL OF CLINICAL LABORATORY MANAGEMENT(ELECTRONIC EDITION), vol. 7, no. 4, 1 November 2019 (2019-11-01), pages 193 - 198, XP055818031, ISSN: 2095-5820, DOI: 10.3877/cma.j.issn.2095-5820.2019.04.001 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113658642A (zh) * 2021-08-16 2021-11-16 杭州凯曼健康科技有限公司 基于荧光免疫层析法的荧光曲线的检测方法与装置

Also Published As

Publication number Publication date
CN112257681A (zh) 2021-01-22

Similar Documents

Publication Publication Date Title
US20230119593A1 (en) Method and apparatus for training facial feature extraction model, method and apparatus for extracting facial features, device, and storage medium
CN110532880B (zh) 样本筛选及表情识别方法、神经网络、设备及存储介质
CN110693486A (zh) 一种心电图的异常标注方法及装置
WO2021082428A1 (zh) 基于半监督学习的图片分类方法、装置和计算机设备
WO2021110143A1 (zh) 细胞判读方法及系统
CN111176953B (zh) 一种异常检测及其模型训练方法、计算机设备和存储介质
CN109635643B (zh) 一种基于深度学习的快速人脸识别方法
CN108805196A (zh) 用于图像识别的自动增量学习方法
CN110188834A (zh) 一种电力通信网的故障诊断方法、装置及设备
CN112509583A (zh) 一种基于调度操作票系统的辅助监督方法及系统
CN109784343A (zh) 一种基于深度学习模型的资源分配方法及终端
CN110347407A (zh) 一种获取内存占用量的方法、装置、计算机设备及介质
CN104317908B (zh) 基于三支决策和距离的离群点检测方法
CN110909881A (zh) 一种面向跨媒体知识推理任务的知识表示方法
CN115272027A (zh) 一种用于双碳管理的低碳排放数据监测系统
CN110807402A (zh) 一种基于肤色检测的五官定位方法、系统和终端设备
CN112435651A (zh) 一种语音数据自动标注的质量评估方法
CN113452802A (zh) 设备型号的识别方法、装置及系统
CN110544047A (zh) 一种不良数据辨识方法
CN106778834A (zh) 一种基于距离测度学习的ap聚类图像标注方法
CN114998691A (zh) 半监督船舶分类模型训练方法及装置
US20220329504A1 (en) Network traffic classification method and system based on improved k-means algorithm
CN113242213B (zh) 一种电力通信骨干网节点脆弱性诊断方法
CN111178530A (zh) 一种基于数据标注的质检方法和装置
CN109657722A (zh) 基于深度学习算法的舌苔图像识别方法及系统

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20897548

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20897548

Country of ref document: EP

Kind code of ref document: A1