Prostate cancer grading method based on multi-parameter nuclear magnetic imaging

Through the deep learning method of multi-parameter nuclear magnetic imaging, combined with multimodal fusion segmentation network and PIRADS scoring standards, the problem of insufficient feature utilization in prostate cancer segmentation and grading is solved, the segmentation accuracy and grading accuracy are improved, and efficient assistance is provided Diagnostic tools.

CN119006358APending Publication Date: 2024-11-22NANJING TECH UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410748376.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-06-11
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

In the prior art, in the process of segmentation and grading of prostate cancer, it is difficult to effectively utilize multimodal nuclear magnetic imaging features, resulting in low efficiency in segmentation and grading of prostate cancerous areas, which brings a great burden to doctors.

Method used

Using a deep learning method of multi-parameter nuclear magnetic images, the features of T2, ADC and DWI images are fused through a multi-modal fusion segmentation network, and feature extraction is performed using intra-modal and inter-modal attention mechanisms, and combined with PIRADS scoring standards to train segmentation and hierarchical networks to improve the accuracy of lesion segmentation and grading.

Benefits of technology

It improves the segmentation accuracy and grading accuracy of prostate cancerous areas, reduces the work burden of doctors, and provides more reliable auxiliary diagnostic tools.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119006358A_ABST
    Figure CN119006358A_ABST
Patent Text Reader

Abstract

The invention relates to a canceration grading method of a multi-parameter nuclear magnetic image, belongs to the field of images, and provides a prostate lesion area grading method based on a deep learning network of multi-modal feature fusion. According to the method, modeling is mainly carried out according to different nuclear magnetic images needing to be checked by lesions at different positions in PIRADS scoring rules, T2, ADC and DWI are input into a network, features are fused by using a designed multi-modal feature fusion module, and a more accurate segmentation result is obtained. And inputting the segmentation result, the T2, the ADC and the DWI into a grading network to obtain a grading result of the focus. The model can more accurately position the area where the focus is located, and therefore classification of the focus is promoted.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to a canceration grading method of multi-parameter nuclear magnetic resonance imaging, and belongs to the field of images. Background Art

[0002] With the rapid development and popularization of medical imaging equipment, imaging technology has been widely used in clinical practice and has become an indispensable auxiliary means for disease diagnosis, surgical planning, prognosis assessment, and follow-up. Medical images often play a vital role in the diagnosis and treatment process. Therefore, medical images have become one of the most important sources of evidence for clinical analysis and medical intervention.

[0003] Medical image segmentation can extract key information from specific tissue images and is a key step in achieving medical image visualization. The segmented images are provided to doctors for different tasks such as quantitative analysis of tissue volume, diagnosis, location of pathologically changed tissues, depiction of anatomical structures, and treatment plans. Medical images contain a huge amount of information. Manually outlining the target area of ​​medical images in clinical practice is a time-consuming and laborious task, which adds a great burden to the daily work of clinicians. Therefore, medical image segmentation in clinical practice remains a difficult problem. At the same time, grading the tumors marked by doctors also increases the workload of doctors. Through deep learning methods, tumors can be segmented and then classified, which is important for doctors to make diagnoses.

[0004] In the segmentation and grading of prostate cancer, different images need to be viewed for cancers in different locations. Lesions in the peripheral zone of the prostate need to be viewed with ADC / DWI images, while lesions in the transition zone focus more on T2 images. The literature currently searched does not make a good modeling of the features between prostate lesions and images. The multimodal feature fusion method proposed in this patent can model the lesion area and modality, give higher weights to more important modalities, and thus promote the segmentation and grading of prostate cancerous areas. Summary of the invention

[0005] In the first aspect, the present invention is a prostate cancer grading method based on multi-parameter nuclear magnetic resonance imaging. The deep learning method is first used to segment the lesion area. The network model is as follows: Figure 1 shown.

[0006] Secondly, for the lesion area processed by deep learning segmentation, the original image and segmentation results are input into the deep learning network to rate the lesion area. The network model is as follows: Figure 2 shown.

[0007] In the examples disclosed herein, the prostate cancer grading method based on multi-parameter nuclear magnetic resonance imaging comprises the following steps:

[0008] Step 1: Data preparation and data processing:

[0009] The public online datasets Prosate158 and Picai were used to have experienced doctors perform PIRADS scoring on the lesions.

[0010] Step 2: Segmentation model training

[0011] Input the processed data into the multimodal fusion segmentation network, and save the training parameters after training is completed.

[0012] Step 3: Rating Network Training

[0013] The training parameters in step 2 are applied to the data set to obtain the segmentation results of each example. The multi-parameter MRI image and the segmentation results obtained by the network are sent to the grading network together to obtain the grading results of the lesions. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for describing the embodiments:

[0015] Figure 1 Schematic diagram of multimodal fusion network

[0016] Figure 2 Hierarchical network diagram

[0017] Figure 3 Method flow chart DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention; it is obvious that the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments, and all other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making creative work are within the scope of protection of the present invention.

[0019] Example:

[0020] See attached Figure 3 The present application provides a method for grading prostate cancer based on multi-parameter nuclear magnetic resonance imaging, the method comprising the following steps:

[0021] Data preparation and preprocessing: Download Prostate158 and Picai, and give the dataset to doctors with sufficient experience to perform PIRADS scoring on the lesion area. For the dataset in Picai, project the ADC and DWI images into the T2 image space to make the spatial dimensions of the three images the same. The data is divided into training set, validation set and test set in a ratio of 6:2:2.

[0022] Model training: In order to improve the generalization of the model, data augmentation is performed on the data before the data input network, including random rotation, random contrast enhancement, random Gaussian noise, etc., and then input into Figure 1 In the network shown in Figure 1, the network is mainly divided into an encoder and a decoder. In the encoder part, a three-branch network structure is used to input T2, ADC and DWI into one of the branches respectively. After the feature extraction is completed, a multimodal feature fusion module is used to fuse the features using intra-modal attention and inter-modal attention, layer by layer until the bottom of the model. In the decoder part, for the features provided by the encoder, the decoder part will upsample the features until the size is the same as the original image.

[0023] After obtaining the model output, the loss is compared with the actual segmentation result, including DICELoss and FocaLoss, and reverse gradient propagation is performed to update the parameters of the feature extraction network until the loss reaches the minimum value and the model converges. The final model is saved for the next step.

[0024] Based on the model parameters obtained in the previous step, the lesion segmentation results of all the data in the two data sets are predicted, and the minimum enclosing rectangle of the segmentation results is found. T2, ADC, DWI and segmentation results are input into the grading network, and finally the grading results of prostate cancer are obtained.

[0025] The present invention builds models according to the scoring criteria in PIPADS, models different modalities and different positions, so that the obtained features are more conducive to lesion segmentation, lays a foundation for prostate cancer grading, improves the accuracy of lesion segmentation, and thus better solves the problem of lesion grading.

[0026] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A prostate cancer grading method based on multi-parameter nuclear magnetic resonance imaging, characterized in that: The method comprises the following steps: Step 1: Organize the public datasets Prostate158 and Picai, map the ADC and DWI images in the Picai data into the T2 space, and give the two datasets to experienced doctors for prostate cancer rating. Step 2: Put the processed data into the multimodal feature fusion network for training, and save the parameters after the training is completed. Step 3: Use the trained model to predict the lesion segmentation area in the two data and save the segmentation results. Step 4: Put the predicted segmentation results, T2, ADC and DWI images into the grading network for training to obtain the grading results.

2. The prostate cancer grading method based on multi-parameter nuclear magnetic resonance imaging according to claim 1, characterized in that: The step 1 includes the following steps: Step 1: Map the ADC and DWI images into the space of the T2 image so that the spatial dimensions of the three images are the same and the spatial positions correspond one to one. Step 2: The lesion areas in the data need to be marked and the grading results of each lesion area need to be evaluated.

3. The prostate cancer grading method based on multi-parameter nuclear magnetic resonance imaging according to claim 1, characterized in that: The step 2 includes the following steps: Step 1: Put the processed data into the multimodal feature fusion network for training. Step 2: Use the network output and the actual segmentation result as the loss function, perform directional propagation, and optimize the network parameters. Step 3: Select the parameters that perform best on the validation set.

4. The method for grading prostate cancer based on multi-parameter nuclear magnetic resonance imaging according to claim 1, characterized in that: The step 3 includes the following steps: Step 1: Load the trained parameters into the network. Step 2: Input the data into the network and save the segmentation results of the network output.

5. The method for grading prostate cancer based on multi-parameter nuclear magnetic resonance imaging according to claim 1, characterized in that: The step 4 includes the following steps: Step 1: Input the saved segmentation results, ADC, DWI and T2 into the network for training. Step 2: Use the network output and the actual segmentation result as the loss function, perform directional propagation, and optimize the network parameters. Step 3: Select the parameters that perform best on the validation set.