Porta hepatis anatomy and hepatic parenchyma dissection identification method and system based on laparoscope video deep learning

Abstract

The invention discloses a porta hepatis anatomy and hepatic parenchyma dissection identification method and system based on laparoscope video deep learning. On one hand, the porta hepatis anatomy and hepatic parenchyma dissection identification method based on laparoscope video deep learning comprises the following steps: collecting a recorded video stream data set; performing annotation processing on the video stream data set to obtain a preprocessed video stream data set; performing video clip segmentation and overlapping segment editing processing on the preprocessed video stream data set to obtain target video stream data; and a training operation step identification algorithm is adopted to carry out identification processing on the steps of porta hepatis anatomy and hepatic parenchyma dissection operation in the target video stream data, so that an operation step identification auxiliary result is obtained. On the other hand, the invention provides a porta hepatis anatomy and hepatic parenchyma dissection recognition system based on laparoscope video deep learning. The method has the advantage of being capable of assisting a clinician in formulating an operation scheme so as to reduce the complexity and risk degree of an operation.

Description

Technical Field

[0001] The present invention relates to the field of medical technology, and more specifically, to a method and system for identifying hilar anatomy and liver parenchyma separation based on laparoscopic video deep learning. Background Art

[0002] According to clinical data, primary liver cancer ranks fifth among common cancer diseases in my country, and its mortality rate ranks second among common cancer diseases. It is one of the high-risk diseases that seriously endangers the health of Chinese people. At present, surgical treatment of primary liver cancer is one of the important ways to effectively prolong the survival period of patients. Among them, right hepatectomy (RH) is often used in clinical practice to perform liver cancer resection treatment. Right hepatectomy involves the removal of 3 / 5 of the total liver volume, which is a more complicated surgical method in liver cancer radical resection. In the prior art, the process of right hemihepatectomy usually includes two important surgical steps: hilar dissection and liver separation. Hilar dissection involves the dissection and protection of important human tissues such as the bile duct, hepatic artery, and portal vein. Its purpose is to cut off the blood flow into the liver and prevent bile leakage, thereby paving the way for the core step of liver separation. Liver separation is to gradually cut off the liver tissue and ligate the blood vessel branches at all levels to facilitate the removal of cancerous tissue. In clinical practice, this operation is prone to bleeding and residual liver function deficiency.

[0003] However, with the rise of modern digital medicine, in the field of artificial intelligence-assisted navigation surgery, based on the successful application of accurate minimally invasive navigation liver resection under 3D laparoscopy in accurately radically curing cancer lesions, artificial intelligence-assisted navigation surgery technology has been widely used in surgical operations by using deep learning to use convolutional neural networks to extract more visual information from surgical videos to form key frames to identify surgical steps; therefore, in the field of surgical treatment of primary liver cancer, how to apply artificial intelligence-assisted navigation surgery technology to the two key surgical steps of hilar dissection and liver resection, which have complicated steps and difficult operations, to solve the problems of complex and high-risk operations, thereby making surgical operations more standardized and safe.

[0004] How to solve the above problems has become a technical problem that needs to be solved urgently. Summary of the invention

[0005] One of the purposes of the present invention is to provide a method for identifying hilar anatomy and liver parenchyma separation based on laparoscopic video deep learning, which can assist clinicians in formulating surgical plans to reduce the complexity and risk of the operation.

[0006] Another object of the present invention is to provide a system for identifying hilar anatomy and hepatic parenchyma separation based on laparoscopic video deep learning, which adopts the above-mentioned method for identifying hilar anatomy and hepatic parenchyma separation based on laparoscopic video deep learning.

[0007] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0008] On the one hand, the present invention provides a method for identifying hilar anatomy and liver parenchyma separation based on laparoscopic video deep learning, the method comprising the following steps:

[0009] S1, a video stream dataset based on the steps of the clinician performing liver hilar dissection and liver parenchyma separation surgery;

[0010] S2, labeling the collected video stream data set in a predetermined time period according to a predetermined standard, thereby obtaining a preprocessed video stream data set;

[0011] S3, performing video segment segmentation and overlapping segment editing processing on the pre-processed video stream data set according to a predetermined standard and in a predetermined time period to obtain target video stream data of several different surgical steps;

[0012] S4. Use a trained surgical step recognition algorithm to perform recognition processing on the hilar anatomy and liver parenchyma separation surgical steps in the target video stream data according to preset standards, thereby obtaining surgical step recognition auxiliary results to assist clinicians in formulating surgical plans.

[0013] Preferably, the method further comprises the following steps:

[0014] S5. Synchronously display the auxiliary results of surgical step identification to locate the surgical position to assist clinicians in evaluating the surgical progress and provide a reference for clinicians to perform surgical operations.

[0015] Preferably, in step S3, the target video stream data is target video stream data with a resolution of 1080P, a frame rate of 30fps and a duration of 1 minute.

[0016] Preferably, in step S4,

[0017] The steps of the hilar dissection start and end with the identification and exposure of the hepatoduodenal ligament, including freeing the hepatoduodenal ligament, dissecting the common bile duct, dissecting the proper hepatic artery, dissecting the portal vein trunk and branches, and ending with the complete freeing of blood vessels and cleaning of surrounding tissues.

[0018] The starting and ending points of the liver parenchyma separation steps are: starting with the transection of the liver parenchyma and hepatic vessels and ending with the disappearance of the surgical instruments from the screen.

[0019] Preferably, in step S4, the steps of training the surgical procedure recognition algorithm to recognize the target video stream data are as follows:

[0020] S4.1, dividing the acquired target video stream data into a training set, a validation set and a test set according to a predetermined standard and in a predetermined surgical segment form;

[0021] S4.2. Use the predetermined TimeSformer algorithm to process the training set, the validation set, and the test set. The TimeSformer algorithm extracts a number of non-overlapping target video stream image patches from the training set, the validation set, and the test set, respectively, and performs calculation and linear projection processing according to predetermined standards, thereby obtaining a one-dimensional feature vector z of the target video stream data. The calculation formula of the one-dimensional feature vector z is as follows:

[0022] z=[z cls ,Ex1,Ex2,...,Ex N ]+p

[0023] Where z is the token, E is equivalent to 2D convolution, x1,x2,...,x N is the image patch, p is the position code, z cls Represents a vector that can be deeply learned;

[0024] S4.3. Input the one-dimensional feature vector z into the TimeSformer algorithm, and use the encoder preset in the TimeSformer algorithm, which is composed of multiple multi-head self-attention, layer normalization, and MLP blocks, to implement spatiotemporal feature modeling on the one-dimensional feature vector z to implement time series data classification for the target video stream data. The calculation formula for implementing spatiotemporal feature modeling is as follows:

[0025] y l ,Attention_score l =MSA(LN(z l ))+z l

[0026] z l+1 =MLP(LN(y l ))+y l

[0027] Among them, y l For multi-head self-attention output, Attention_score l is the attention score matrix, MAS is multi-head self-attention, LN is layer normalization, z l For tokens, the MLP consists of two linear projections separated by a GELU nonlinearity;

[0028] S4.4. In the process of implementing spatiotemporal feature modeling, the predetermined Attention_score is used l Implement redundant feature data filtering, the calculation formula is as follows:

[0029] z drop =z l .index(Attention_score 1 Attention_score 2 ·....·Attention_score l )

[0030] Among them, z drop The features are input into MAS after filtering features;

[0031] Attention_score 1 -Attention_score l Matrix multiplication is used between them;

[0032] S4.5. Use a preset cross entropy model to process the time series data of the target video stream data according to preset standards, calculate the difference between the predicted distribution of the time series data and the predetermined gold standard according to the preset standards to update the correction model parameters until the loss value converges, thereby obtaining a trained surgical step recognition algorithm L to perform recognition processing on the hilar anatomy and liver parenchyma separation surgical steps according to the preset standards, and then obtain surgical step recognition auxiliary results to assist clinicians in formulating surgical plans; wherein, the formula of the surgical step recognition algorithm L is as follows:

[0033]

[0034] Among them, L is the loss value, C′ i is the gold standard of the sample for category i, C i is the model's predicted value for the sample for category i.

[0035] Preferably, in step S4.3, it also includes the following steps:

[0036] Calculate the last layer to get z in z cls The feature vector is extracted and input into the predetermined calculation model to implement data processing to obtain the network's distribution prediction for a certain sample classification, thereby obtaining highly accurate acquisition data; among them, for z cls The calculation formula for feature vector data processing is as follows:

[0037] C=Head(z cls )

[0038] Among them, C is the final multi-category prediction distribution probability of the model.

[0039] Preferably, in step S1, the video stream dataset is a video stream dataset with a resolution of 1080P and a frame rate of 30fps based on the steps of the hepatic hilar dissection and liver parenchyma separation surgery performed by a clinician.

[0040] On the other hand, the present invention provides a system for identifying hilar anatomy and hepatic parenchyma separation based on laparoscopic video deep learning, which adopts the above-mentioned hilar anatomy and hepatic parenchyma separation identification method based on laparoscopic video deep learning, comprising a surgical laparoscope device, on which a collection module electrically connected to the surgical laparoscope device is embedded; and also comprising an identification module with a display device and electrically connected to the surgical laparoscope device; wherein:

[0041] The surgical laparoscope device is used to collect and synchronously transmit the video stream data set of the recorded liver hilar dissection and liver parenchyma separation surgical steps to the acquisition module;

[0042] The acquisition module is used to obtain the surgical video stream data set transmitted from the surgical laparoscope device according to the preset standard, and implement the labeling and video segmentation and overlapping segmentation editing in the form of a predetermined time period according to the preset standard, so as to obtain the target video stream data;

[0043] The recognition module is used to obtain the target video stream data from the acquisition module through the surgical laparoscope device, and to perform recognition processing on the hilar anatomy and liver parenchyma separation surgical steps in the target video stream data according to preset standards through a preset training surgical step recognition algorithm, so as to obtain surgical step recognition auxiliary results to assist clinicians in formulating surgical plans;

[0044] The display device is used to receive the surgical step identification auxiliary results transmitted from the identification module and perform real-time display processing to locate the surgical position, thereby assisting clinicians in evaluating the surgical progress and providing a reference basis for clinicians to perform surgical operations.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] In the present invention, when right hemihepatectomy is used to perform liver cancer resection treatment by dissecting the hilar part and separating the hepatic parenchyma on the patient, a video stream data set based on the steps of the clinician performing the dissecting the hilar part and separating the hepatic parenchyma is collected and recorded by a predetermined surgical laparoscope device, and the acquired video stream data set is synchronously transmitted to a predetermined acquisition module, and the predetermined acquisition module is used to perform labeling processing on the collected video stream data set in the form of a predetermined time period according to a preset standard, so as to obtain a preprocessed video stream data set; and then the predetermined acquisition module is used to perform video segment segmentation and overlap segmentation editing processing on the preprocessed video stream data set in the form of a predetermined time period according to a predetermined standard. Target video stream data of several different surgical steps are obtained; then the target video stream data is transmitted from a preset acquisition module to a predetermined recognition module through a preset surgical laparoscope device, and the preset training surgical step recognition algorithm in the predetermined recognition module is used to perform recognition processing on the hilar anatomy and liver parenchyma separation surgical steps in the target video stream data according to preset standards, thereby obtaining surgical step recognition auxiliary results to assist clinicians in formulating surgical plans, thereby improving the success rate and safety of the operation to effectively reduce the complexity and risk of the surgical operation; therefore, the present invention has the advantage of being able to assist clinicians in formulating surgical plans to reduce the complexity and risk of the operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0048] Figure 1 It is a flowchart of the method and system for identifying hilar anatomy and hepatic parenchyma separation based on laparoscopic video deep learning according to the present invention;

[0049] Figure 2 It is a structural block diagram of the hilar anatomy and liver parenchyma separation identification method and system based on laparoscopic video deep learning described in the present invention.

[0050] Explanation of the accompanying drawings: 1. Surgical laparoscopic equipment, 2. Acquisition module, 3. Display device, 4. Identification module. DETAILED DESCRIPTION

[0051] See also Figure 1 to Figure 2 As shown, a method for identifying hilar anatomy and liver parenchyma separation based on laparoscopic video deep learning comprises the following steps:

[0052] S1, a video stream dataset based on the steps of the clinician performing liver hilar dissection and liver parenchyma separation surgery;

[0053] S2, labeling the collected video stream data set in a predetermined time period according to a predetermined standard, thereby obtaining a preprocessed video stream data set;

[0054] S3, performing video segment segmentation and overlapping segment editing processing on the pre-processed video stream data set according to a predetermined standard and in a predetermined time period to obtain target video stream data of several different surgical steps;

[0055] S4. Use a trained surgical step recognition algorithm to perform recognition processing on the hilar anatomy and liver parenchyma separation surgical steps in the target video stream data according to preset standards, thereby obtaining surgical step recognition auxiliary results to assist clinicians in formulating surgical plans.

[0056] In this embodiment, the following steps are also included:

[0057] S5. Synchronously display the auxiliary results of surgical step identification to locate the surgical position to assist clinicians in evaluating the surgical progress and provide a reference for clinicians to perform surgical operations.

[0058] In this embodiment, in step S3, the target video stream data is target video stream data with a resolution of 1080P, a frame rate of 30fps and a duration of 1 minute, so as to facilitate sample input and data enhancement for subsequent model training.

[0059] In this embodiment, in step S4,

[0060] The steps of the hilar dissection start and end with the identification and exposure of the hepatoduodenal ligament, including freeing the hepatoduodenal ligament, dissecting the common bile duct, dissecting the proper hepatic artery, dissecting the portal vein trunk and branches, and ending with the complete freeing of blood vessels and cleaning of surrounding tissues.

[0061] The starting and ending points of the liver parenchyma separation steps are: starting with the transection of the liver parenchyma and hepatic vessels and ending with the disappearance of the surgical instruments from the screen.

[0062] In this embodiment, in step S4, the steps of training the surgical procedure recognition algorithm to recognize the target video stream data are as follows:

[0063] S4.1, dividing the acquired target video stream data into a training set, a validation set and a test set according to a predetermined standard and in a predetermined surgical segment form;

[0064] S4.2. Use the predetermined TimeSformer algorithm to process the training set, the validation set, and the test set. The TimeSformer algorithm extracts a number of non-overlapping target video stream image patches from the training set, the validation set, and the test set, respectively, and performs calculation and linear projection processing according to predetermined standards, thereby obtaining a one-dimensional feature vector z of the target video stream data. The calculation formula of the one-dimensional feature vector z is as follows:

[0065] z=[z cls,Ex1,Ex2,...,Ex N ]+p

[0066] Where z is the token, E is equivalent to 2D convolution, x1,x2,...,x N is the image patch, p is the position code, z cls Represents a vector that can be deeply learned;

[0067] S4.3. Input the one-dimensional feature vector z into the TimeSformer algorithm, and use the encoder preset in the TimeSformer algorithm, which is composed of multiple multi-head self-attention, layer normalization, and MLP blocks, to implement spatiotemporal feature modeling on the one-dimensional feature vector z to implement time series data classification for the target video stream data. The calculation formula for implementing spatiotemporal feature modeling is as follows:

[0068] y l ,Attention_score l =MSA(LN(z l ))+z l

[0069] z l+1 =MLP(LN(y l ))+y l

[0070] Among them, y l For multi-head self-attention output, Attention_score l is the attention score matrix, MAS is multi-head self-attention, LN is layer normalization, z l For tokens, the MLP consists of two linear projections separated by a GELU nonlinearity;

[0071] S4.4. In the process of implementing spatiotemporal feature modeling, the predetermined Attention_score is used l Implement redundant feature data filtering, the calculation formula is as follows:

[0072] z drop =z l .index(Attention_score 1 Attention_score 2 ·....·Attention_score l )

[0073] Among them, z drop The features are input into MAS after filtering features;

[0074] Attention_score1 -Attention_score l Matrix multiplication is used between them;

[0075] S4.5. Use a preset cross entropy model to process the time series data of the target video stream data according to preset standards, calculate the difference between the predicted distribution of the time series data and the predetermined gold standard according to the preset standards to update the correction model parameters until the loss value converges, thereby obtaining a trained surgical step recognition algorithm L to perform recognition processing on the hilar anatomy and liver parenchyma separation surgical steps according to the preset standards, and then obtain surgical step recognition auxiliary results to assist clinicians in formulating surgical plans; wherein, the formula of the surgical step recognition algorithm L is as follows:

[0076]

[0077] Among them, L is the loss value, C′ i is the gold standard of the sample for category i, C i is the model's predicted value for the sample for category i.

[0078] In this embodiment, in step S4.3, it also includes the following steps:

[0079] Calculate the last layer to get z in z cls The feature vector is extracted and input into the predetermined calculation model to implement data processing to obtain the network's distribution prediction for a certain sample classification, thereby obtaining highly accurate acquisition data; among them, for z cls The calculation formula for feature vector data processing is as follows:

[0080] C=Head(z cls )

[0081] Among them, C is the final multi-category prediction distribution probability of the model.

[0082] In this embodiment, in step S1, the video stream dataset is a video stream dataset with a resolution of 1080P and a frame rate of 30fps based on the steps of the hepatic hilar dissection and liver parenchyma separation surgery performed by a clinician.

[0083] In this embodiment, a system for identifying hilar anatomy and hepatic parenchyma separation based on laparoscopic video deep learning, which adopts the above-mentioned hilar anatomy and hepatic parenchyma separation identification method based on laparoscopic video deep learning, comprises a surgical laparoscope device 1, on which a collection module 2 electrically connected to the surgical laparoscope device 1 is embedded; and also comprises an identification module 4 having a display device 3 and electrically connected to the surgical laparoscope device 1; wherein:

[0084] The surgical laparoscopic device 1 is used to collect and record video stream data sets of the hepatic hilar dissection and liver parenchyma separation surgical steps and synchronously transmit them to the acquisition module 2;

[0085] The acquisition module 2 is used to obtain the surgical video stream data set transmitted from the surgical laparoscope device 1 according to the preset standard, and implement the labeling and video segment segmentation and overlapping segmentation editing in the form of a predetermined time period according to the preset standard, so as to obtain the target video stream data;

[0086] The recognition module 4 is used to obtain the target video stream data from the acquisition module 2 through the surgical laparoscope device 1, and perform recognition processing on the hilar anatomy and liver parenchyma separation surgical steps in the target video stream data according to preset standards through a preset training surgical step recognition algorithm, so as to obtain surgical step recognition auxiliary results to assist clinicians in formulating surgical plans;

[0087] The display device 3 is used to receive the surgical step identification auxiliary results transmitted from the identification module 4 and perform real-time display processing to locate the surgical position, thereby assisting clinicians in evaluating the surgical progress and providing a reference basis for clinicians to perform surgical operations.

[0088] When this embodiment is used specifically, when right hemihepatectomy is used to perform liver cancer resection treatment for patients with hilar dissection and liver parenchyma separation, first, a video stream data set based on the steps of the clinician performing the hilar dissection and liver parenchyma separation surgery is collected and recorded by the surgical laparoscope device 1, and the acquired video stream data set is synchronously transmitted to the acquisition module 2, and the acquisition module 2 is used to perform labeling processing on the collected video stream data set in the form of a predetermined time period according to a preset standard, so as to obtain a pre-processed video stream data set; secondly, the acquisition module 2 is used to perform video segment segmentation and overlapping segmentation editing processing on the pre-processed video stream data set in the form of a predetermined time period according to a predetermined standard to obtain target video stream data of several different surgical steps; then, the surgical laparoscope device Preparation 1 transmits the target video stream data from the acquisition module 2 to the recognition module 4, and uses the preset training surgical step recognition algorithm in the recognition module 4 to perform recognition processing on the hilar anatomy and liver parenchyma separation surgical steps in the target video stream data according to the preset standards, so as to obtain the surgical step recognition auxiliary results to assist clinicians in formulating surgical plans, thereby improving the success rate and safety of the operation to effectively reduce the complexity and risk of the surgical operation; finally, the surgical step recognition auxiliary results are synchronously transmitted to the display device 3 through the recognition module 4 to perform display processing in real time to locate the surgical position, thereby facilitating clinicians to understand the image information of the surgical site more accurately and intuitively, thereby assisting clinicians in evaluating the progress of the operation and providing a reference basis for clinicians to perform surgical operations.

[0089] In summary, the present invention adopts the above-mentioned method and system, which has the advantage of being able to assist clinicians in formulating surgical plans to reduce the complexity and risk of the operation.

[0090] For those skilled in the art, various other corresponding changes and deformations can be made according to the technical solutions and concepts described above, and all of these changes and deformations should fall within the protection scope of the claims of the present invention.

Claims

1. A method for identifying hilar anatomy and liver parenchyma separation based on deep learning of laparoscopic video, characterized by: The method comprises the following steps: S1, a video stream dataset based on the steps of the clinician performing liver hilar dissection and liver parenchyma separation surgery; S2, labeling the collected video stream data set in a predetermined time period according to a predetermined standard, thereby obtaining a preprocessed video stream data set; S3, performing video segment segmentation and overlapping segment editing processing on the pre-processed video stream data set according to a predetermined standard and in a predetermined time period to obtain target video stream data of several different surgical steps; S4. Use a trained surgical step recognition algorithm to perform recognition processing on the hilar anatomy and liver parenchyma separation surgical steps in the target video stream data according to preset standards, thereby obtaining surgical step recognition auxiliary results to assist clinicians in formulating surgical plans.

2. The method for identifying hilar anatomy and liver parenchyma separation based on laparoscopic video deep learning according to claim 1, characterized in that; The following steps are also included: S5. Synchronously display the auxiliary results of surgical step identification to locate the surgical position to assist clinicians in evaluating the surgical progress and provide a reference for clinicians to perform surgical operations.

3. The method for identifying hilar anatomy and liver parenchyma separation based on laparoscopic video deep learning according to claim 1, characterized in that; In step S3, the target video stream data is target video stream data with a resolution of 1080P, a frame rate of 30fps, and a duration of 1 minute.

4. The method for identifying hilar anatomy and liver parenchyma separation based on laparoscopic video deep learning according to claim 3, characterized in that; In step S4, The steps of the hilar dissection start and end with the identification and exposure of the hepatoduodenal ligament, including freeing the hepatoduodenal ligament, dissecting the common bile duct, dissecting the proper hepatic artery, dissecting the portal vein trunk and branches, and ending with the complete freeing of blood vessels and cleaning of surrounding tissues. The starting and ending points of the liver parenchyma separation steps are: starting with the transection of the liver parenchyma and hepatic vessels and ending with the disappearance of the surgical instruments from the screen.

5. The method for identifying hilar anatomy and liver parenchyma separation based on laparoscopic video deep learning according to claim 4, characterized in that; In step S4, the steps of training the surgical procedure recognition algorithm to recognize the target video stream data are as follows: S4.1, dividing the acquired target video stream data into a training set, a validation set and a test set according to a predetermined standard and in a predetermined surgical segment form; S4.

2. Use the predetermined TimeSformer algorithm to process the training set, the validation set, and the test set. The TimeSformer algorithm extracts a number of non-overlapping target video stream image patches from the training set, the validation set, and the test set, respectively, and performs calculation and linear projection processing according to predetermined standards, thereby obtaining a one-dimensional feature vector z of the target video stream data. The calculation formula of the one-dimensional feature vector z is as follows: z=[z cls ,Ex1,Ex2,...,Ex N ]+p Where z is the token, E is equivalent to 2D convolution, x1, x2, ..., x N is the image patch, p is the position code, z cls Represents a vector that can be deeply learned; S4.

3. Input the one-dimensional feature vector z into the TimeSformer algorithm, and use the encoder preset in the TimeSformer algorithm, which is composed of multiple multi-head self-attention, layer normalization, and MLP blocks, to implement spatiotemporal feature modeling on the one-dimensional feature vector z to implement time series data classification for the target video stream data. The calculation formula for implementing spatiotemporal feature modeling is as follows: y l ,Attention_score l =MSA(LN(z l ))+z l z l+1 =MLP(LN(and l ))+y l Among them, y l For multi-head self-attention output, Attention_score l is the attention score matrix, MAS is multi-head self-attention, LN is layer normalization, z l For tokens, the MLP consists of two linear projections separated by a GELU nonlinearity; S4.

4. In the process of implementing spatiotemporal feature modeling, the predetermined Attention_score is used l Implement redundant feature data filtering, the calculation formula is as follows: z drop =z l .index(Attention_score 1 ·Attention_score 2 ·....·Attention_score l ) Among them, z drop The features are input into MAS after filtering features; Attention_score 1 -Attention_score l Matrix multiplication is used between them; S4.

5. Use a preset cross entropy model to process the time series data of the target video stream data according to preset standards, calculate the difference between the predicted distribution of the time series data and the predetermined gold standard according to the preset standards to update the correction model parameters until the loss value converges, thereby obtaining a trained surgical step recognition algorithm L to perform recognition processing on the hilar anatomy and liver parenchyma separation surgical steps according to the preset standards, and then obtain surgical step recognition auxiliary results to assist clinicians in formulating surgical plans; wherein, the formula of the surgical step recognition algorithm L is as follows: Among them, L is the loss value, C′ i is the gold standard of the sample for category i, C i is the model's prediction value for the sample for category i.

6. The method for identifying hilar anatomy and liver parenchyma separation based on laparoscopic video deep learning according to claim 5, characterized in that; In step S4.3, it also includes the following steps: Calculate the last layer to get z in z cls The feature vector is extracted and input into the predetermined calculation model to implement data processing to obtain the network's distribution prediction for a certain sample classification, thereby obtaining highly accurate acquisition data; among them, for z cls The calculation formula for feature vector data processing is as follows: C=Head(z cls ) Among them, C is the final multi-category prediction distribution probability of the model.

7. The method for identifying hilar anatomy and liver parenchyma separation based on laparoscopic video deep learning according to claim 1, characterized in that; In step S1 , the video stream dataset is a video stream dataset with a resolution of 1080P and a frame rate of 30 fps based on the steps of a clinician performing liver hilar dissection and liver parenchyma separation surgery.

8. A system for identifying hilar anatomy and hepatic parenchyma separation based on laparoscopic video deep learning, which adopts the method for identifying hilar anatomy and hepatic parenchyma separation based on laparoscopic video deep learning according to any one of claims 1 to 7, characterized in that; The invention comprises a surgical laparoscope device (1), on which a collection module (2) electrically connected to the surgical laparoscope device (1) is embedded; and an identification module (4) having a display device (3) and electrically connected to the surgical laparoscope device (1); wherein: The surgical laparoscope device (1) is used to collect and record video stream data sets of liver hilar dissection and liver parenchyma separation surgical steps and synchronously transmit them to the acquisition module (2); The acquisition module (2) is used to obtain the surgical video stream data set transmitted from the surgical laparoscope device (1) according to a preset standard, and to perform labeling, video segment segmentation, and overlapping segment editing processing in a predetermined time period according to the preset standard, so as to obtain target video stream data; The recognition module (4) is used to obtain target video stream data from the acquisition module (2) through the surgical laparoscope device (1), and to perform recognition processing on the hilar anatomy and liver parenchyma separation surgical steps in the target video stream data according to preset standards through a preset training surgical step recognition algorithm, thereby obtaining surgical step recognition auxiliary results to assist clinicians in formulating surgical plans; The display device (3) is used to receive the surgical step identification auxiliary results transmitted from the identification module (4) and perform real-time display processing to locate the surgical position, thereby assisting clinicians in evaluating the surgical progress and providing a reference basis for clinicians to perform surgical operations.

Images