CN-122023971-A - Method and system for monitoring survival rate of magnetic labeling stem cells of joint cavity based on improved contrast learning

CN122023971ACN 122023971 ACN122023971 ACN 122023971ACN-122023971-A

Abstract

The application discloses a method and a system for monitoring survival rate of magnetic labeling stem cells of a joint cavity based on improved contrast learning, and relates to the technical fields of biomedical engineering, medical image analysis and artificial intelligence intersection. The method comprises the steps of obtaining multi-sequence magnetic resonance imaging data of a stem cell transplanting region loaded with magnetic nano particles at a joint cavity part, preprocessing the data, inputting the data into a physical perception dual-domain contrast learning network, respectively extracting features by a parallel spatial domain encoder and a frequency domain encoder, innovatively introducing physical consistency constraint based on a Buuloch equation, constructing a joint loss function, driving the network to distinguish microscopic susceptibility differences caused by intracellular iron aggregation and extracellular iron dispersion in learning, and finally carrying out regression prediction by utilizing the optimized features to output the survival rate of the stem cells. The application reduces the dependence on a large amount of labeling data and realizes nondestructive and high-precision longitudinal monitoring of the survival state of the transplanted stem cells.

Inventors

CHEN RUNZHI
WANG JIAN
YIN WEIZHONG
JIAO KUN
LI KUN
QIN SHAOJIE

Assignees

上海市浦东新区人民医院

Dates

Publication Date: 20260512
Application Date: 20260129

Claims (9)

1. The method for monitoring the survival rate of the magnetic labeling stem cells of the joint cavity based on improved contrast learning is characterized by comprising the following steps of: S1, acquiring multi-sequence magnetic resonance imaging data of a joint cavity part of a subject, wherein the data comprise T2 weighted imaging and T2 of a stem cell transplantation area loaded with a magnetic nano material Weighted imaging and magnetically sensitive weighted imaging data; s2, preprocessing the multi-sequence magnetic resonance imaging data, including motion artifact correction and magnetic field non-uniformity correction, so as to obtain standardized three-dimensional voxel data; S3, constructing a physical perception double-domain contrast learning network, wherein the network comprises a parallel spatial domain encoder, a parallel frequency domain encoder and a parallel projection head; s4, inputting the standardized three-dimensional voxel data into the space domain encoder and the frequency domain encoder respectively, and extracting a space domain feature vector and a frequency domain feature vector; s5, constructing a physical consistency constraint term based on the physical difference of the transverse relaxation rate caused by the aggregation state of the magnetic nano material in the living cells and the dispersion state of the magnetic nano material after dead cells, and training the physical perception double-domain contrast learning network by combining a multi-view contrast loss function to obtain an optimized characteristic representation with physical discriminant; S6, inputting the optimized characteristic representation to a survival rate regression prediction component, and outputting the current survival rate value of the stem cells in the joint cavity.
2. The method of claim 1, wherein in step S5, the total loss function of the physically aware two-domain contrast learning network The definition is as follows: Wherein, the The loss is estimated for multi-view noise contrast, For the purpose of physical consistency regularization loss, In order to be a loss of consistency in the frequency domain, Is an adjustable super-parameter weight coefficient.
3. The method of claim 2, wherein the physical consistency regularization loss The calculation formula of (2) is as follows: Wherein, the For training sample number; Potential feature vectors for encoder output; Mapping functions from features to physical parameters; And Respectively the first Actual observed apparent transverse relaxation rate and transverse relaxation rate of individual sample voxels; is a theoretical relaxation difference function based on a static dephasing physical model.
4. The method of claim 1, wherein the workflow of the frequency domain encoder comprises: S41, performing three-dimensional fast Fourier transform on input standardized three-dimensional voxel data, and converting the input standardized three-dimensional voxel data from a space domain to a frequency domain; s42, decomposing the frequency domain data into an amplitude spectrum and a phase spectrum, and applying high-pass filtering to the phase spectrum to enhance local susceptibility mutation high-frequency signals caused by aggregation of magnetic nano materials; S43, performing self-adaptive weight distribution on the filtered spectrum characteristics through a spectrum attention mechanism; s44, carrying out inverse Fourier transform on the weighted spectrum characteristics, reconstructing the spectrum characteristics into spatial domain characteristics, and fusing the spatial domain characteristics with the output of the spatial domain encoder.
5. The method of claim 1, wherein the survivability regression prediction component employs a transform architecture based self-attention aggregation module, the prediction process of which is expressed as: Wherein, the In order to obtain the fused multi-modal feature vector, Is a multi-layer sensing machine, which is a multi-layer sensing machine, Is the percent survival predictor of the output.
6. The method according to claim 1, wherein the magnetic nanomaterial is a superparamagnetic iron oxide nanoparticle with a polylysine or polydopamine modified surface, the particle size of the magnetic nanomaterial ranges from 10 nanometers to 200 nanometers, the positive sample pair is selected from at least one of a multi-sequence image pair of the same region of interest under different echo times and an image pair generated by enhancing random data of the same region of interest in contrast learning training, and the negative sample pair is selected from at least one of a voxel pair of different regions with a spatial distance exceeding a preset threshold and a reference sample pair with a significant difference in survival rate based on priori knowledge.
7. A system for monitoring survival of magnetically labeled stem cells of a joint cavity based on improved contrast learning, characterized in that it is used to implement the monitoring method according to any one of claims 1 to 6, the system comprising: The data acquisition module is used for acquiring multi-sequence imaging data of the joint cavity from the magnetic resonance imaging equipment; the preprocessing module is used for carrying out standardization, denoising and correction processing on the imaging data; A physical perception comparison learning module, which is internally provided with the physical perception double-domain comparison learning network as claimed in claim 1 and is used for extracting and optimizing characteristic representation from the preprocessed data; And the survival rate analysis module is used for receiving the optimized characteristic representation, obtaining the survival rate of the stem cells through regression prediction and generating a visualized survival rate distribution report.
8. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, which when executed by the processor, implements the method of any one of claims 1 to 6.
9. A computer readable storage medium, characterized in that the storage medium has stored thereon computer program instructions, which when executed by a computer, cause the computer to perform the method according to any of claims 1 to 6.

Description

Method and system for monitoring survival rate of magnetic labeling stem cells of joint cavity based on improved contrast learning Technical Field The invention relates to the technical fields of biomedical engineering, medical image analysis and artificial intelligence intersection, in particular to a method and a system for monitoring survival rate of magnetic labeling stem cells of a joint cavity based on improved contrast learning. Background Osteoarthritis (Osteoarthritis, OA) is a common degenerative joint disease that severely affects the quality of life of the patient. Mesenchymal Stem Cells (MSCs) intra-articular injection therapy is a potential means for treating OA due to its cartilage regeneration and immunomodulatory capacity. However, stem cells tend to have low viability in the luminal microenvironment (e.g., hypoxia, inflammation, mechanical shear), and the viability status directly determines the therapeutic effect. Therefore, how to monitor the survival of stem cells after transplantation in a nondestructive, real-time and accurate way is a key technical bottleneck in clinical transformation. Currently, the common monitoring methods mainly include: 1. Invasive biopsy, the pathological analysis of tissue removed by arthroscopy or puncture. This method is accurate but invasive, cannot be continuously monitored longitudinally, and has sampling errors. 2. Optical imaging (fluorescence/bioluminescence) is highly sensitive, but has limited penetration depth, is difficult to apply to deep joints of the human body (e.g., knee joints, hip joints), and the optical signal is severely attenuated over time. 3. Magnetic Resonance Imaging (MRI) tracking, labelling stem cells with superparamagnetic iron oxide nanoparticles (SPIOs). SPIOs as negative contrast agent at T2/T2Low signals (dark areas) are generated on the weighted image. This is a very potential non-destructive monitoring means. However, existing MRI-based SPIO-labeled cell monitoring techniques present a significant "dead-living resolution challenge" in that when labeled stem cells die, the cell membrane breaks, and intracellular SPIOs is released into the extracellular matrix or phagocytized by macrophages within the joint cavity. On conventional MRI images, either live labeled stem cells, or post-mortem SPIOs released, or macrophages phagocytosed SPIOs, appear as low signal regions (SignalVoid). Traditional image processing methods or conventional supervised learning models (such as U-Net and the like) mainly rely on pixel gray values for segmentation, and it is difficult to morphologically distinguish fine signal differences caused by such microscopic magnetic environment changes, so that the survival rate of stem cells is overestimated or misjudged. In addition, the existing deep learning method generally needs a large amount of labeled data (i.e. needs to know the real death-activity ratio corresponding to each voxel), but the gold standard label can hardly be obtained in a living experiment, which limits the application of the supervised learning method. Based on the above, the present invention provides a method and a system for monitoring survival rate of magnetic labeling stem cells of joint cavity based on improved contrast learning to solve the above problems. Disclosure of Invention In order to overcome the defects in the prior art, the invention provides a method and a system for monitoring the survival rate of magnetic labeling stem cells of a joint cavity based on improved contrast learning, which are used for solving the problems in the prior art. The invention provides a joint cavity magnetic labeling stem cell survival rate monitoring method based on improved contrast learning, which comprises the following steps: S1, acquiring multi-sequence magnetic resonance imaging data of a joint cavity part of a subject, wherein the data comprise T2 weighted imaging and T2 of a stem cell transplantation area loaded with a magnetic nano material Weighted imaging and magnetically sensitive weighted imaging data; s2, preprocessing the multi-sequence magnetic resonance imaging data, including motion artifact correction and magnetic field non-uniformity correction, so as to obtain standardized three-dimensional voxel data; S3, constructing a physical perception double-domain contrast learning network, wherein the network comprises a parallel spatial domain encoder, a parallel frequency domain encoder and a parallel projection head; s4, inputting the standardized three-dimensional voxel data into the space domain encoder and the frequency domain encoder respectively, and extracting a space domain feature vector and a frequency domain feature vector; s5, constructing a physical consistency constraint term based on the physical difference of the transverse relaxation rate caused by the aggregation state of the magnetic nano material in the living cells and the dispersion state of the magnetic nano material after dead cells, and training the ph