CN-120782904-B - Magnetic resonance image reconstruction method, device, electronic equipment and readable storage medium

CN120782904BCN 120782904 BCN120782904 BCN 120782904BCN-120782904-B

Abstract

The application provides a magnetic resonance image reconstruction method, a device, electronic equipment and a readable storage medium, and relates to the field of magnetic resonance data processing. The method comprises the steps of acquiring acquired k-space data, distributing sampling probability to points in the k-space data to obtain probability values corresponding to all the points in the k-space data, carrying out interpolation processing on the k-space data according to the probability values to obtain interpolation data, sampling the k-space data by using a preset importance graph to obtain sampling data, fusing the interpolation data with the sampling data to obtain fusion data, carrying out image reconstruction by using the fusion data to obtain a reconstructed image, evaluating the reconstructed image to obtain evaluation scores, adjusting the probability values corresponding to all the points in the k-space data when the evaluation scores are larger than a preset evaluation threshold, carrying out Fourier transformation on the reconstructed image to obtain reconstruction data, updating the importance graph by using the reconstruction data, and carrying out interpolation and sampling.

Inventors

WANG CHUNSHENG
MENG XIANPING
YAO LIJUN

Assignees

江阴万康医疗科技有限公司

Dates

Publication Date: 20260512
Application Date: 20250703

Claims (9)

1. A method of magnetic resonance image reconstruction, the method comprising: Acquiring acquired k-space data, and distributing sampling probability to points in the k-space data to obtain probability values corresponding to all the points in the k-space data; Performing interpolation processing on the k-space data according to the probability value to obtain interpolation data, and sampling the k-space data by using a preset importance graph to obtain sampling data; the importance map is obtained through the following steps: obtaining a plurality of training samples, wherein the training samples comprise a part name, a k-space sample corresponding to the part name and an important point mark graph corresponding to the k-space sample; inputting the part name and the k-space sample into a preset importance learning network to generate a prediction graph; Calculating the value of a loss function by using the predictive graph and the important point mark graph, and performing parameter adjustment on the importance learning network by using the value of the loss function until a preset training condition is met to obtain an importance graph; fusing the interpolation data and the sampling data to obtain fused data, and performing image reconstruction by using the fused data to obtain a reconstructed image; And evaluating the reconstructed image to obtain an evaluation score, adjusting probability values corresponding to all points in the k-space data under the condition that the evaluation score is larger than a preset evaluation threshold value to obtain an adjusted probability value, taking the adjusted probability value as the probability value, carrying out Fourier transformation on the reconstructed image to obtain reconstructed data, updating the importance map by utilizing the reconstructed data to obtain an updated importance map, taking the updated importance map as the importance map, executing interpolation processing on the k-space data according to the probability value to obtain interpolation data, and sampling the k-space data by utilizing the preset importance map to obtain sampling data.
2. The method of claim 1, wherein interpolating the k-space data according to the probability values to obtain interpolated data, comprising: performing key data sampling on the k-space data according to the probability value to obtain key sampling data; Generating mask center points for the key sampling data by using probability distribution corresponding to the probability values, and filling the mask center points to the periphery according to preset point intervals and the probability distribution to obtain a sampling mask; And carrying out interpolation calculation on the key sampling data according to the sampling mask to obtain the interpolation data.
3. The method according to claim 2, wherein the interpolating the key sample data according to the sample mask to obtain the interpolated data includes: Cross-fusing the sampling mask and the key sampling data to obtain data to be interpolated, wherein the cross-fusing is to perform product operation on the sampling mask and the key sampling data after moving according to preset displacement, and then integrate the sampling mask and the key sampling data; and interpolating the data to be interpolated by using a preset sliding window algorithm to obtain the interpolated data.
4. The method according to claim 3, wherein interpolating the data to be interpolated using a predetermined sliding window algorithm to obtain the interpolated data comprises: Dividing the data to be interpolated into a plurality of sub-block data according to the sliding window algorithm; Interpolation is carried out on each piece of sub-block data to obtain a plurality of pieces of sub-interpolation data, and data generation is carried out on each piece of sub-block data to obtain generated data corresponding to each piece of sub-block data; And carrying out weighted fusion on the sub interpolation data corresponding to each sub block data and the generated data to obtain interpolation fusion data, and carrying out low-pass filtering on the interpolation fusion data to obtain the interpolation data.
5. The method of claim 1, wherein updating the importance map with the reconstructed data results in an updated importance map, comprising: And marking the reconstruction data, adding the marked reconstruction data into the training sample, and retraining the importance learning network by using the training sample to obtain the updated importance graph.
6. The method of claim 1, wherein performing image reconstruction using the fused data results in a reconstructed image, comprising: Dividing the fusion data according to the probability value to obtain a central area and an edge area, carrying out first normalization processing on the central area to obtain first normalization data, carrying out second normalization processing on the edge area to obtain second normalization data, and splicing the first normalization data and the second normalization data to obtain third normalization data; and reconstructing the third normalized data by using a preset variation automatic encoder to obtain the reconstructed image.
7. A magnetic resonance image reconstruction apparatus, the apparatus comprising: A data acquisition and processing module (110) for acquiring acquired k-space data, and assigning sampling probabilities to points in the k-space data to obtain probability values corresponding to each point in the k-space data; the data sampling module (120) is used for carrying out interpolation processing on the k-space data according to the probability value to obtain interpolation data, and sampling the k-space data by utilizing a preset importance map to obtain sampling data; the importance map is obtained through the following steps: obtaining a plurality of training samples, wherein the training samples comprise a part name, a k-space sample corresponding to the part name and an important point mark graph corresponding to the k-space sample; inputting the part name and the k-space sample into a preset importance learning network to generate a prediction graph; Calculating the value of a loss function by using the predictive graph and the important point mark graph, and performing parameter adjustment on the importance learning network by using the value of the loss function until a preset training condition is met to obtain an importance graph; the data reconstruction module (130) is used for fusing the interpolation data and the sampling data to obtain fused data, and performing image reconstruction by utilizing the fused data to obtain a reconstructed image; And a data adjustment module (140) configured to evaluate the reconstructed image to obtain an evaluation score, adjust probability values corresponding to points in the k-space data when the evaluation score is greater than a preset evaluation threshold value, obtain an adjusted probability value, perform fourier transform on the reconstructed image with the adjusted probability value as the probability value to obtain reconstructed data, update the importance map with the reconstructed data to obtain an updated importance map, perform interpolation processing on the k-space data according to the probability value to obtain interpolation data, and sample the k-space data with the preset importance map to obtain sampling data.
8. An electronic device comprising a processor (501), a memory (505), a user interface (503), a communication bus (502) and a network interface (504), the processor (501), the memory (505), the user interface (503) and the network interface (504) being respectively connected to the communication bus (502), the memory (505) being for storing instructions, the user interface (503) and the network interface (504) being for communicating to other devices, the processor (501) being for executing the instructions stored in the memory (505) for causing the electronic device (500) to perform the method according to any of claims 1-6.
9. A computer readable storage medium storing instructions which, when executed, perform the method of any one of claims 1-6.

Description

Magnetic resonance image reconstruction method, device, electronic equipment and readable storage medium Technical Field The present application relates to the field of magnetic resonance data processing technology, and in particular, to a magnetic resonance image reconstruction method, apparatus, electronic device, and readable storage medium. Background Magnetic resonance imaging (Magnetic Resonance Imaging, MRI) is widely used as a non-invasive, non-radiative imaging technique in the fields of medical diagnosis, biological research, etc. The method comprises the steps of applying radio frequency pulses with specific frequency to a human body, exciting hydrogen nuclei in the human body to resonate, releasing electromagnetic signals, scanning and acquiring k-space data through MRI, and reconstructing an image by using the acquired data. Image reconstruction is one of the key links of MRI technology, and the reconstruction quality directly affects the examination of MRI. In the related art, the undersampling technology is generally adopted to collect k-space data, then the compressed sensing algorithm is utilized to reconstruct an image, the compressed sensing algorithm is based on strict data assumption, but actually collected k-space data cannot meet the strict data assumption, so that the accuracy of image reconstruction is low. Disclosure of Invention The application provides a magnetic resonance image reconstruction method, a device, electronic equipment and a readable storage medium, which can improve the accuracy of image reconstruction. The technical scheme of the embodiment of the application is as follows: in a first aspect, an embodiment of the present application provides a magnetic resonance image reconstruction method, including: Acquiring acquired k-space data, and distributing sampling probability to points in the k-space data to obtain probability values corresponding to all the points in the k-space data; Performing interpolation processing on the k-space data according to the probability value to obtain interpolation data, and sampling the k-space data by using a preset importance graph to obtain sampling data; fusing the interpolation data and the sampling data to obtain fused data, and performing image reconstruction by using the fused data to obtain a reconstructed image; And evaluating the reconstructed image to obtain an evaluation score, adjusting probability values corresponding to all points in the k-space data under the condition that the evaluation score is larger than a preset evaluation threshold value to obtain an adjusted probability value, taking the adjusted probability value as the probability value, carrying out Fourier transformation on the reconstructed image to obtain reconstructed data, updating the importance map by utilizing the reconstructed data to obtain an updated importance map, taking the updated importance map as the importance map, executing interpolation processing on the k-space data according to the probability value to obtain interpolation data, and sampling the k-space data by utilizing the preset importance map to obtain sampling data. In the technical scheme, firstly, acquired k-space data are acquired, data support is provided for processing the k-space data, sampling probability is allocated to points in the k-space data, probability values corresponding to all the points in the k-space data are obtained, and the probability values corresponding to all the points can reflect the importance degree of all the points in the k-space data; interpolation processing is carried out on k space data according to probability values to obtain interpolation data, the k space data is expanded through interpolation, the density of the k space data is increased, the resolution of a reconstructed image is improved under the condition that the k space data is not increased, the defect caused by undersampling the k space data is avoided, a preset importance map is utilized to sample the k space data to obtain sampling data, the importance of each position can be reflected by the obtained sampling data so as to carry out edge processing, the occurrence of reconstruction image artifacts is avoided, the interpolation data and the sampling data are fused to obtain fusion data, the fusion data can improve the expression capacity of the data, image reconstruction is carried out by utilizing the fusion data to obtain the reconstruction image so as to judge whether the reconstruction image meets the requirement or not, a more accurate image is obtained, evaluation is carried out on the reconstruction image to obtain an evaluation score, the condition that the evaluation score is larger than a preset evaluation threshold value indicates that the reconstruction image does not reach the image requirement, the probability value corresponding to each point in the k space data is adjusted, the adjusted probability value is taken as the probability value so as to carry out the adju