Primary hepatocellular carcinoma lesion screening and segmentation method and system based on MRI images

Abstract

The present invention relates to a primary hepatocellular carcinoma lesion screening and segmentation method and system based on MRI images. The present invention obtains and preprocesses liver multi-phase MRI data, inputs the data into a first pre-trained multi-channel convolutional neural network for liver segmentation, and generates a multi-phase MRI liver segmentation image; then inputs the data into a second pre-trained multi-channel convolutional neural network for lesion screening, and obtains a multi-phase MRI liver segmentation image containing primary hepatocellular carcinoma lesions; then inputs the data into a third pre-trained multi-channel convolutional neural network for lesion segmentation, and generates a multi-phase MRI primary hepatocellular carcinoma lesion segmentation result; finally, the lesion segmentation result and the liver multi-phase MRI data are visualized, and primary hepatocellular carcinoma lesion statistical information is obtained.

Description

Technical Field

[0001] The present invention belongs to the field of medical image processing and segmentation, and in particular relates to a primary hepatocellular carcinoma lesion screening and segmentation method and system based on MRI images. Background Art

[0002] Liver space-occupying lesions refer to lesions in the liver area of ​​the human body, which appear as abnormal echo signals in medical imaging data, showing uneven texture density. In some medical images, the segmentation of liver lesions is extremely difficult, and the liver lesion image may be covered by other parts, which increases the challenge of lesion segmentation. In recent years, medical imaging technology has developed rapidly and has become an important routine examination method, playing an important role in clinical examination, diagnosis, and surgical planning. The rapid development of medical imaging technology provides a detection method for lesion detection that is different from traditional biopsy. Doctors can observe from medical images whether there are lesion feature areas on living tissues, and then make auxiliary diagnosis based on the judgment results of the image. Traditional medical imaging technology has a strong correlation with the operator's experience and technology, which will be affected by the operator's subjectivity, and is labor-intensive, time-consuming and inefficient.

[0003] In the existing technology, a research paper by Takenaga T et al. on automatic focal liver lesion detection and classification based on MRI (Takenaga T, Hanaoka S, Nomura Y, et al. Multichannel three-dimensional fully convolutional residual network-based focal liver lesion detection and classification in Gd-EOB-DTPA-enhanced MRI[J]. International Journal of Computer Assisted Radiology and Surgery, 2021, 12: 1527–1536) shows that the sensitivity of hepatocellular carcinoma lesion detection obtained by using enhanced T1-weighted MRI image data in the non-enhanced phase, arterial phase, portal venous phase, delayed phase and liver-specific phase is 56%, and the accuracy rate is 79%. Bousabarah K et al.'s study on non-automatic detection and segmentation of hepatocellular carcinoma lesions (Bousabarah K, Letzen B, Tefera J, et al. Automated detection and delineation of hepatocellular carcinoma on multiphasic contrast-enhanced MRI using deep learning [J]. Abdominal Radiology, 2021, 46 (1): 216-225) showed that the detection sensitivity obtained by training the model with enhanced T1-weighted MRI images in the late arterial, portal and delayed phases was 75%, and the Dice coefficient was 0.68. Although these models have played a certain role in promoting the screening of hepatocellular carcinoma lesions, they have certain shortcomings: (a) In terms of process, they do not include automatic screening steps, and the system has a low degree of automation, resulting in insufficient release of medical resources. (b) The accuracy and sensitivity of lesion segmentation or classification of the models proposed in existing studies are not sufficient to support clinical applications.

[0004] The existing patent with application number 202110414395.X discloses "a method for liver lesion segmentation based on convolutional neural network". The proposed model is for the segmentation of liver space-occupying lesions, and its limitation is that it does not distinguish between benign lesions and malignant lesions. Hepatocellular carcinoma lesions are malignant lesions among liver space-occupying lesions, and it is necessary to detect and segment them separately in clinical practice.

[0005] Therefore, there is an urgent need for a rapid process for detecting and segmenting hepatocellular carcinoma lesions to solve the problem of labor-intensive and time-consuming visual inspection of medical images. The present invention replaces the mechanical reading task of radiological imaging physicians with an accurate auxiliary diagnosis method. By using multi-phase MRI data of the liver, the trained convolutional neural network model is more robust, and more stable and reliable detection results are obtained, greatly shortening the time of early medical image processing. Summary of the invention

[0006] The purpose of the present invention is to provide a method and system for screening and segmenting primary hepatocellular carcinoma lesions based on MRI images in response to the problems existing in the prior art.

[0007] The present invention relates to a primary hepatocellular carcinoma lesion screening and segmentation method and system based on MRI images. The present invention obtains and preprocesses liver multi-phase MRI data, inputs the data into a first pre-trained multi-channel convolutional neural network for liver segmentation, and generates a multi-phase MRI liver segmentation image; then inputs the data into a second pre-trained multi-channel convolutional neural network for lesion screening, and obtains a multi-phase MRI liver segmentation image containing primary hepatocellular carcinoma lesions; then inputs the data into a third pre-trained multi-channel convolutional neural network for lesion segmentation, and generates a multi-phase MRI primary hepatocellular carcinoma lesion segmentation result; finally, the lesion segmentation result and the liver multi-phase MRI data are visualized, and primary hepatocellular carcinoma lesion statistical information is obtained.

[0008] The purpose of the present invention can be achieved by the following scheme:

[0009] The present invention provides a method and system for screening and segmenting primary hepatocellular carcinoma lesions based on MRI images, the method comprising the following steps:

[0010] S1. Acquiring multi-phase MRI data of the liver I original ;

[0011] S2, for I original Data preprocessing was performed to obtain liver multi-phase MRI data of uniform standard. preprocessed ;

[0012] S3, I preprocessed Input the first pre-trained multi-channel convolutional neural network N1 for liver segmentation to obtain multi-phase MRI data I of the liver area liver ;

[0013] S4. I liver The second pre-trained multi-channel convolutional neural network N2 is input to perform primary hepatocellular carcinoma lesion screening and classification, and multi-phase MRI data I containing primary hepatocellular carcinoma lesions is obtained. liver,HCCand multi-phase MRI data I without the lesion liver,noHCC ;

[0014] S5. I liver,HCC The third pre-trained multi-channel convolutional neural network N3 is input to perform primary hepatocellular carcinoma lesion segmentation, and the multi-phase segmentation result R of primary hepatocellular carcinoma lesions is obtained. HCC ;

[0015] S6. I preprocessed and R HCC After data post-processing, multi-phase MRI liver and primary hepatocellular carcinoma lesion combined image I was obtained. combine , and statistical information of primary hepatocellular carcinoma lesions.

[0016] As an embodiment of the present invention, the liver multiphase phases in step S1 include an arterial phase and a portal venous phase. In the arterial phase, the hepatocellular carcinoma lesion is enhanced in whole or in part above the background liver parenchyma; in the portal venous phase, the hepatocellular carcinoma lesion is surrounded by a mostly or completely smooth, uniform, sharp-edged annular structure, i.e., thicker fibers around the lesion, compressed liver sinusoids and other tissue components - capsule.

[0017] As an embodiment of the present invention, the liver multi-phase phase in step S1 further includes one or more of a non-enhanced phase, a delayed phase, and a liver-specific phase.

[0018] As an embodiment of the present invention, the MRI data I in step S1 original The data are abdominal liver magnetic resonance imaging data enhanced by the contrast agent disodium gadoxetate. T1-weighted imaging is performed within the first 10 seconds, 25 seconds, 55 seconds, 5 minutes later, and 10 to 20 minutes later, and imaging data of the liver non-enhanced phase, arterial phase, portal venous phase, delayed phase, and liver-specific phase can be obtained respectively. T1-weighted images are MRI images that reflect the density differences between tissues, and the specific form of the data is data saved in DICOM format.

[0019] As an embodiment of the present invention, the preprocessing described in step S2 is to align the MRI data of different phases to the image of the specified phase to generate multi-phase MRI data of unified standards. The specified phase selects the arterial phase or the portal venous phase. The data of unified standards refers to the data in which the position information of the liver tissue in other phases is consistent with the image of the specified phase after the alignment step. The unified standard refers to the NIfTI image saved after data sorting, cleaning, and desensitization.

[0020] As an embodiment of the present invention, the pre-trained multi-channel convolutional neural networks N1, N2, and N3 include CNN series or U-Net series. Among them, CNN series include: k-CNN, FCN, DCNN or VGG-Net. U-Net series include 2DU-Net, 2.5D U-Net, 3D U-Net, Res-Unet or U-Net with improved loss function. N1, N2, and N3 are different from each other.

[0021] As an embodiment of the present invention, N1 and N3 are selected from the U-Net series; N2 is selected from the CNN series. N1, N2, and N3 can achieve the specified effect by any model in the corresponding series.

[0022] As an embodiment of the present invention, the pre-training process includes:

[0023] 1) Obtain multi-phase MRI training set, validation set data and corresponding labels;

[0024] 2) Input each phase MRI training set data into the network as a channel;

[0025] 3) Calculate the network output results of the training set data and the loss function value of the corresponding label;

[0026] 4) Adjust the parameter values ​​in the network model according to the loss function value;

[0027] 5) After all the training set data is input and trained, input the validation set data and corresponding labels;

[0028] 6) Calculate the network output results of the validation set data and the loss function value of the corresponding label;

[0029] 7) Adjust the hyperparameter values ​​of the network model according to the loss function results;

[0030] 8) Repeat 2)-7) until the number of iterations reaches the set value, the loss function value reaches the set value, or the segmentation (classification) accuracy reaches the established indicator, and then the training ends.

[0031] All three models follow this step, and the differences are in the training set, validation set data, and corresponding labels. Among them, the data of N1 is the multi-phase MRI data of the unified standard, and the label is the liver segmentation result; the data of N2 is the multi-phase MRI data of the liver area, and the label is whether it contains primary hepatocellular carcinoma lesions; the data of N3 is the multi-phase MRI data containing primary hepatocellular carcinoma lesions, and the label is the multi-phase segmentation result of primary hepatocellular carcinoma lesions.

[0032] As an embodiment of the present invention, the post-processing described in step S6 comprises: HCCInverse registration, restore to the MRI data of each phase I original The matching space is then superimposed on the I original The combined images of multi-phase MRI of the liver and the segmented lesions are obtained. By combining the images, the steps and time of counting the number, volume or other information of primary hepatocellular carcinoma lesions can be greatly simplified. Other information refers to the prediction results of tumor genotype, pathological stage, survival period, recurrence rate after resection or recurrence rate after liver transplantation.

[0033] The present invention also provides a method and system for screening and segmenting primary hepatocellular carcinoma lesions based on MRI images, the system comprising:

[0034] Module A1: Acquisition of multi-phase MRI data of the liver I original ;

[0035] Module A2, I original Data preprocessing was performed to obtain liver multi-phase MRI data of uniform standard. preprocessed ;

[0036] Module A3, I preprocessed Input the first pre-trained multi-channel convolutional neural network N1 for liver segmentation to obtain multi-phase MRI data I of the liver area liver ;

[0037] Module A4, I liver The second pre-trained multi-channel convolutional neural network N2 is input to screen primary hepatocellular carcinoma lesions to obtain multi-phase MRI data I containing primary hepatocellular carcinoma lesions. liver,HCC and multi-phase MRI data I without the lesion liver,noHCC ;

[0038] Module A5, I liver,HCC The third pre-trained multi-channel convolutional neural network N3 is input to perform primary hepatocellular carcinoma lesion segmentation, and the multi-phase segmentation result R of primary hepatocellular carcinoma lesions is obtained. HCC ;

[0039] Module A6, I preprocessed and R HCC After data post-processing, multi-phase MRI liver and primary hepatocellular carcinoma lesion combined image I was obtained. combine , and statistical information of primary hepatocellular carcinoma lesions.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] (1) By using multi-phase MRI data of the liver, the trained convolutional neural network model is more robust and obtains more stable and reliable detection results;

[0042] (2) By using convolutional neural network training data to learn and capture features at different levels, and by finding certain connections within the data, a prediction model is constructed to map the input to the output (label or predicted value), which can obtain detection results with higher detection rates and lower false positives.

[0043] (3) By adopting the method of the present invention, a structured report on hepatocellular carcinoma lesion detection and segmentation that meets clinical needs is obtained, which greatly shortens the time for early medical image processing and solves the problem that medical image visual inspection is labor-intensive and time-costly. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0045] Figure 1 is a flow chart of the screening and segmentation method of the present invention;

[0046] Figure 2 It is a schematic diagram of the multi-channel convolutional neural network U-Net model;

[0047] Figure 3 Schematic diagram of the multi-channel convolutional neural network VGG-Net model. DETAILED DESCRIPTION

[0048] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. The following examples are implemented on the premise of the technical solution of the present invention, provide detailed implementation methods and specific operation processes, and will help those skilled in the art to further understand the present invention. It should be pointed out that the protection scope of the present invention is not limited to the following embodiments, and several adjustments and improvements made under the premise of the concept of the present invention all belong to the protection scope of the present invention.

[0049] The present invention provides a method for intelligent screening and segmentation of primary hepatocellular carcinoma lesions, such as Figure 1 As shown, including:

[0050] Step 1: Acquire multi-phase MRI data of the liver;

[0051] Step 2: preprocessing of liver multi-phase MRI data;

[0052] Step 3: Multi-channel convolutional neural network multi-phase MRI liver segmentation;

[0053] Step 4: Multi-channel convolutional neural network multi-phase MRI primary hepatocellular carcinoma lesion screening;

[0054] Step 5: Multi-channel convolutional neural network multi-phase MRI primary hepatocellular carcinoma lesion segmentation;

[0055] Step 6: Post-process the multi-phase MRI data of the liver and primary hepatocellular carcinoma lesions and obtain lesion statistical information.

[0056] The present invention also provides a primary hepatocellular carcinoma lesion intelligent screening and segmentation system, such as Figure 1 As shown, including:

[0057] Step 1: Acquire multi-phase MRI data of the liver;

[0058] Step 2: preprocessing of liver multi-phase MRI data;

[0059] Step 3: Multi-channel convolutional neural network multi-phase MRI liver segmentation;

[0060] Step 4: Multi-channel convolutional neural network multi-phase MRI primary hepatocellular carcinoma lesion screening;

[0061] Step 5: Multi-channel convolutional neural network multi-phase MRI primary hepatocellular carcinoma lesion segmentation;

[0062] Step 6: Post-process the multi-phase MRI data of the liver and primary hepatocellular carcinoma lesions and obtain lesion statistical information.

[0063] Example 1

[0064] This embodiment provides a method for intelligent screening and segmentation of primary hepatocellular carcinoma lesions, comprising the following steps:

[0065] S1. Obtain MRI data of the liver arterial phase, portal venous phase, non-enhanced phase, delayed phase, and liver-specific phase I original The data are from abdominal liver magnetic resonance imaging data enhanced by the contrast agent gadoxetate disodium. T1-weighted imaging was performed within the first 10 seconds, 25 seconds, 55 seconds, 5 minutes, and 10 to 20 minutes to obtain imaging data of the liver non-enhanced phase, arterial phase, portal venous phase, delayed phase, and liver-specific phase, respectively.

[0066] S2, for I original The data are processed, and the MRI data of different phases are registered to the image of the specified phase, and the multi-phase MRI data of the liver with unified standards are obtained. preprocessed ; The designated phase is selected as the arterial phase or the portal venous phase. The unified standard data refers to the data in which the position information of the liver tissue in other phases is consistent with the image of the designated phase after the registration step.

[0067] S3, I preprocessedInput the first pre-trained multi-channel convolutional neural network N1 for liver segmentation to obtain multi-phase MRI data I of the liver area liver ; The pre-training is specifically as follows: 1. Obtain multi-phase MRI training sets, validation set data and corresponding labels; 2. Input each phase MRI training set data as a channel into the network; 3. Calculate the network output results of the training set data and the loss function value of the corresponding label; 4. Adjust the parameter values ​​in the network model according to the loss function value; 5. After a part of the training set data is input and trained, input the validation set data and the corresponding label; 6. Calculate the network output results of the validation set data and the loss function value of the corresponding label; 7. Adjust the hyperparameter values ​​of the network model according to the loss function result; Repeat 2-7 until the number of iterations reaches the set value, the loss function value reaches the set value, or the segmentation (classification) accuracy reaches the established indicator, and the training is terminated. . The pre-trained multi-channel convolutional neural network N1 is U-Net, such as Figure 2 As shown in the figure, all three models follow this step, and the differences are in the training set, validation set data, and corresponding labels. Among them, the data of N1 is the multi-phase MRI data of the unified standard, and the label is the liver segmentation result; the data of N2 is the multi-phase MRI data of the liver area, and the label is whether it contains primary hepatocellular carcinoma lesions; the data of N3 is the multi-phase MRI data containing primary hepatocellular carcinoma lesions, and the label is the multi-phase segmentation result of primary hepatocellular carcinoma lesions.

[0068] S4. I liver The second pre-trained multi-channel convolutional neural network N2 is input to screen primary hepatocellular carcinoma lesions to obtain multi-phase MRI data I containing primary hepatocellular carcinoma lesions. liver,HCC and multi-phase MRI data I without the lesion liver,noHCC ; The pre-trained multi-channel convolutional neural network N2 is VGG-Net, such as Figure 3 As shown;

[0069] S5. I liver,HCC The third pre-trained multi-channel convolutional neural network N3 is input to perform primary hepatocellular carcinoma lesion segmentation, and the multi-phase segmentation result R of primary hepatocellular carcinoma lesions is obtained. HCC ; The pre-trained multi-channel convolutional neural network N3 is U-Net, such as Figure 2 As shown;

[0070] S6. I preprocessed and R HCC Post-processing of data: HCC Inverse registration, restore to the MRI data of each phase I original The matching space is then superimposed on the I originalThe liver multi-phase MRI and the combined image of the segmented lesions are obtained; and the number, volume or other information of the primary hepatocellular carcinoma lesions are counted. combine .

[0071] Example 2

[0072] This embodiment provides a primary hepatocellular carcinoma lesion intelligent screening and segmentation system, comprising the following steps:

[0073] S1. Obtain MRI data of the liver arterial phase, portal venous phase, non-enhanced phase, delayed phase, and liver-specific phase I original ; Data are from abdominal liver MRI data enhanced by the contrast agent gadoxetate disodium. T1-weighted imaging was performed within the first 10 seconds, 25 seconds, 55 seconds, 5 minutes later, and 10 to 20 minutes to obtain imaging data of the liver's non-enhanced phase, arterial phase, portal venous phase, delayed phase, and liver-specific phase, respectively.

[0074] S2, for I original The data are processed, and the MRI data of different phases are registered to the image of the specified phase, and the multi-phase MRI data of the liver with unified standards are obtained. preprocessed ; The designated phase is selected as the arterial phase or the portal venous phase. The unified standard data refers to the data in which the position information of the liver tissue in other phases is consistent with the image of the designated phase after the registration step.

[0075] S3, I preprocessed Input the first pre-trained multi-channel convolutional neural network N1 for liver segmentation to obtain multi-phase MRI data I of the liver area liver ; The pre-training is specifically as follows: 1. Obtain multi-phase MRI training sets, validation set data and corresponding labels; 2. Input each phase MRI training set data as a channel into the network; 3. Calculate the network output results of the training set data and the loss function value of the corresponding label; 4. Adjust the parameter values ​​in the network model according to the loss function value; 5. After a part of the training set data is input and trained, input the validation set data and the corresponding label; 6. Calculate the network output results of the validation set data and the loss function value of the corresponding label; 7. Adjust the hyperparameter values ​​of the network model according to the loss function result; Repeat 2-7 until the number of iterations reaches the set value, the loss function value reaches the set value, or the segmentation (classification) accuracy reaches the established indicator, and the training is terminated. . The pre-trained multi-channel convolutional neural network N1 is U-Net, such as Figure 2As shown in the figure, all three models follow this step, and the differences are in the training set, validation set data, and corresponding labels. Among them, the data of N1 is the multi-phase MRI data of the unified standard, and the label is the liver segmentation result; the data of N2 is the multi-phase MRI data of the liver area, and the label is whether it contains primary hepatocellular carcinoma lesions; the data of N3 is the multi-phase MRI data containing primary hepatocellular carcinoma lesions, and the label is the multi-phase segmentation result of primary hepatocellular carcinoma lesions.

[0076] S4. I liver The second pre-trained multi-channel convolutional neural network N2 is input to screen primary hepatocellular carcinoma lesions to obtain multi-phase MRI data I containing primary hepatocellular carcinoma lesions. liver,HCC and multi-phase MRI data I without the lesion liver,no HCC ; The pre-trained multi-channel convolutional neural network N2 is VGG-Net, such as Figure 3 As shown;

[0077] S5. I liver,HCC The third pre-trained multi-channel convolutional neural network N3 is input to perform primary hepatocellular carcinoma lesion segmentation, and the multi-phase segmentation result R of primary hepatocellular carcinoma lesions is obtained. HCC ; The pre-trained multi-channel convolutional neural network N3 is U-Net, such as Figure 2 As shown;

[0078] S6. I preprocessed and R HCC Post-processing of data: HCC Inverse registration, restore to the MRI data of each phase I original The matching space is then superimposed on the I original The liver multi-phase MRI and the combined image of the segmented lesions are obtained; and the number, volume or other information of the primary hepatocellular carcinoma lesions are counted. combine .

[0079] Those skilled in the art know that, in addition to realizing the system and its various devices, modules, and units provided by the present invention in a purely computer-readable program code, it is entirely possible to realize the same functions in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system and its various devices, modules, and units provided by the present invention can be considered as a hardware component, and the devices, modules, and units included therein for realizing various functions can also be regarded as structures within the hardware component; the devices, modules, and units for realizing various functions can also be regarded as both software modules for realizing the method and structures within the hardware component.

[0080] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art may make various modifications or variations within the scope of the claims, which do not affect the essence of the present invention.

Claims

1. A method for intelligent screening and segmentation of primary hepatocellular carcinoma lesions based on MRI images, characterized in that: The method comprises the following steps: S1. Acquiring multi-phase MRI data of the liver I original ; S2, for I original Data preprocessing was performed to obtain liver multi-phase MRI data of uniform standard. preprocessed ; S3, I preprocessed Input the first pre-trained multi-channel convolutional neural network N1 for liver segmentation to obtain multi-phase MRI data I of the liver area liver ; S4. I liver The second pre-trained multi-channel convolutional neural network N2 is input to screen primary hepatocellular carcinoma lesions to obtain multi-phase MRI data I containing primary hepatocellular carcinoma lesions. liver，HCC and multi-phase MRI data I without the lesion liver，noHCC ; S5. I liver,HCC The third pre-trained multi-channel convolutional neural network N3 is input to perform primary hepatocellular carcinoma lesion segmentation, and the multi-phase segmentation result R of primary hepatocellular carcinoma lesions is obtained. HCC ; S6. I preprocessed and R HCC After data post-processing, multi-phase MRI liver and primary hepatocellular carcinoma lesion combined image I was obtained. combine ; The liver multi-phase phase in step S1 includes an arterial phase and a portal venous phase; The preprocessing described in step S2 is to register the MRI data of different phases to the image of the specified phase to generate multi-phase MRI data of unified standard; The post-processing described in step S6 includes converting R HCC Inverse registration, restore to the MRI data of each phase I original The matching space is then superimposed on the I original Multi-phase MRI of the liver and combined images of segmented lesions were obtained.

2. The method for intelligent screening and segmentation of primary hepatocellular carcinoma lesions according to claim 1, characterized in that: The liver multi-phase phase in step S1 also includes one or more of a non-enhanced phase, a delayed phase, and a liver-specific phase.

3. The method for intelligent screening and segmentation of primary hepatocellular carcinoma lesions according to claim 1, characterized in that: The pre-trained multi-channel convolutional neural networks N1, N2, and N3 include CNN series or U-Net series.

4. The method for intelligent screening and segmentation of primary hepatocellular carcinoma lesions according to claim 3, characterized in that: The N1 and N3 are selected from the U-Net series.

5. The method for intelligent screening and segmentation of primary hepatocellular carcinoma lesions according to claim 3, characterized in that: The N2 selects the CNN series.

6. The method for intelligent screening and segmentation of primary hepatocellular carcinoma lesions according to claim 1, characterized in that: The pre-training process includes: 1) Obtain multi-phase MRI training set, validation set data and corresponding labels; 2) Input each phase MRI training set data into the network as a channel; 3) Calculate the network output results of the training set data and the loss function value of the corresponding label; 4) Adjust the parameter values ​​in the network model according to the loss function value; 5) After a portion of the training set data is input and trained, input the validation set data and corresponding labels; 6) Calculate the network output results of the validation set data and the loss function value of the corresponding label; 7) Adjust the hyperparameter values ​​of the network model according to the loss function results; 8) Repeat 2)-7) until the number of iterations reaches the set value, the loss function value reaches the set value, or the segmentation accuracy reaches the set indicator, and then the training ends.

7. A primary hepatocellular carcinoma lesion intelligent screening and segmentation system, characterized in that: The system comprises: Module A1: Acquisition of multi-phase MRI data of the liver I original ; Module A2, I original Data preprocessing was performed to obtain liver multi-phase MRI data of uniform standard. preprocessed ; Module A3, I preprocessed Input the first pre-trained multi-channel convolutional neural network N1 for liver segmentation to obtain multi-phase MRI data I of the liver area liver ; Module A4, I liver The second pre-trained multi-channel convolutional neural network N2 is input to screen primary hepatocellular carcinoma lesions to obtain multi-phase MRI data I containing primary hepatocellular carcinoma lesions. liver，HCC and multi-phase MRI data I without the lesion liver，noHCC ; Module A5, I liver,HCC The third pre-trained multi-channel convolutional neural network N3 is input to perform primary hepatocellular carcinoma lesion segmentation, and the multi-phase segmentation result R of primary hepatocellular carcinoma lesions is obtained. HCC ; Module A6, I preprocessed and R HCC After data post-processing, multi-phase MRI liver and primary hepatocellular carcinoma lesion combined image I was obtained. combine , and statistical information of primary hepatocellular carcinoma lesions; The liver multiphase phases described in module A1 include the arterial phase and the portal venous phase; The preprocessing described in module A2 is to register the MRI data of different phases to the image of the specified phase to generate multi-phase MRI data of unified standard; Post-processing described in Module A6 involves converting R HCC Inverse registration, restore to the MRI data of each phase I original The matching space is then superimposed on the I original Multi-phase MRI of the liver and combined images of segmented lesions were obtained.

Images