CN-121661029-B - ECM-stem cell construct quality assessment method and system based on machine learning

CN121661029BCN 121661029 BCN121661029 BCN 121661029BCN-121661029-B

Abstract

The application relates to the technical field of image analysis and discloses a machine learning-based ECM-stem cell construct evaluation method and system. The method comprises the steps of obtaining an image sequence of an ECM-stem cell construct through equidistant defocusing shooting, calculating a projection area as a basic morphological index, determining a structure stability distance based on a change rate, extracting image texture features and spatial distribution of the image texture features under the distance, dividing a complete region of the construct into a plurality of subareas, calculating structural features of texture statistics among the subareas, inputting the structural features into a pre-trained machine learning model, and outputting quality scores of the construct. The application can carry out rapid, nondestructive and objective quality assessment on the ECM-stem cell construct.

Inventors

LIU XU
RAO YIWEI
WANG CHENGZHEN

Assignees

巴山泓(北京)医药科技有限公司
北京帝康医药投资管理有限公司

Dates

Publication Date: 20260512
Application Date: 20251211

Claims (9)

1. A machine learning-based ECM-stem cell construct quality assessment method, the method comprising: Step S1, continuously shooting an ECM-stem cell construct at a plurality of defocus distances with equal intervals to obtain a defocus image sequence, and calculating the projection area of the ECM-stem cell construct as a basic morphological index for each image of the defocus image sequence; s2, calculating the change rate of the basic morphological index relative to the defocus distance, and taking the defocus distance with the change rate smaller than a preset threshold value for the first time as the structure stability distance; S3, acquiring defocused images of an ECM-stem cell construct photographed under a structure stability distance, extracting a complete area of the ECM-stem cell construct from the defocused images, calculating texture features based on the complete area, acquiring spatial distribution of the texture features in the complete area, dividing the complete area into two subareas, calculating statistics of the texture features for each subarea, calculating structural features representing structural relations among the subareas based on the statistics among different subareas, wherein the complete area is divided into two subareas, and comprises the steps of mapping the complete area into a two-dimensional coordinate system, calculating an average value of a horizontal coordinate and an average value of a vertical coordinate of each pixel in the complete area, rounding up the average value of the horizontal coordinate and the average value of the vertical coordinate to obtain two coordinate values, taking the pixel corresponding to the two coordinate values as a geometric center of the complete area, calculating a distance from each pixel to the geometric center in the complete area, normalizing all distance values, dividing the complete area into two different subareas based on the normalized distance, setting a distance threshold, dividing the normalized distance into a first subarea and a second subarea as a large distance; step S4, training a machine learning model in advance, inputting the structural characteristics into the machine learning model, and outputting the quality scores of the ECM-stem cell constructs by the machine learning model.
2. The method of claim 1, wherein calculating the projected area of the ECM-stem cell construct comprises: Performing edge enhancement on an original defocused image to obtain a boundary enhanced image, performing self-adaptive threshold segmentation on the boundary enhanced image to generate a binary image, performing morphological closing operation on the binary image to obtain an optimized binary image, performing communication analysis on the optimized binary image, filtering a noise area, identifying all interconnected white pixel areas, counting the total number of pixels of the white pixel areas, and taking the total number of pixels as a projection area.
3. The method according to claim 1, wherein calculating the rate of change of the basic morphology index with respect to the defocus distance, taking the defocus distance whose rate of change is smaller than a preset threshold for the first time as the structurally stable distance, comprises: And obtaining basic morphological indexes of each pair of adjacent images of the defocused image sequence, calculating the change rates of the two basic morphological indexes to obtain a change rate sequence, comparing each change rate in the change rate sequence with a preset change rate threshold value, obtaining a change rate which is smaller than the change rate threshold value for the first time, namely a target change rate, obtaining two images corresponding to the target change rate, obtaining two defocusing distances corresponding to the two images, and selecting the minimum defocusing distance as a structural stability distance.
4. The method of claim 1, wherein computing texture features based on the complete region, obtaining a spatial distribution of the texture features in the complete region, comprises: Setting a sliding window, sliding the sliding window in a complete area according to a preset step length, performing first calculation and second calculation on the area block corresponding to each sliding to obtain a plurality of local texture characteristic values, and associating the plurality of local texture characteristic values with a central pixel of the corresponding area block to obtain the spatial distribution of the complete area.
5. The method of claim 4, wherein performing a first calculation comprises: For the current region block, a gray level co-occurrence matrix of the region block is calculated based on the gray level value of each pixel in the region block, and contrast, energy, entropy, and correlation of the region block are calculated based on the gray level co-occurrence matrix.
6. The method of claim 4, wherein performing a second calculation comprises: and carrying out two-dimensional fast Fourier transform on the current area block to obtain a power spectrum, and calculating the average energy of the power spectrum on different radius annular bands around the center and the average energy of the power spectrum on different angle sector areas.
7. The method of claim 1, wherein computing structural features representing structural relationships between sub-regions based on statistics between different sub-regions comprises: Extracting texture characteristic values of all pixels belonging to each subregion, and calculating statistics of all texture characteristic values, wherein the statistics comprise an average value and a standard deviation; For each statistic of texture feature values, the difference between the standard deviation and the ratio of the corresponding average values of the two sub-regions is calculated as the structural feature.
8. The method of claim 1, wherein pre-training the machine learning model comprises: Obtaining training images of ECM-stem cell constructs comprising a plurality of known quality score tags, the training images being captured at a structurally stable distance, calculating texture features and corresponding structural features based on the training images, training a machine learning model by a supervised machine learning algorithm using the structural features of all training images and their corresponding quality score tags.
9. A machine learning based ECM-stem cell construct quality assessment system for implementing a machine learning based ECM-stem cell construct quality assessment method according to any one of claims 1-8, said system comprising: The first calculation module continuously shoots the ECM-stem cell construct at a plurality of defocusing distances with equal intervals to obtain a defocusing image sequence, and calculates the projection area of the ECM-stem cell construct as a basic morphological index for each image of the defocusing image sequence; The second calculation module is used for calculating the change rate of the basic morphological index relative to the defocusing distance, and taking the defocusing distance with the change rate smaller than a preset threshold value for the first time as the structure stable distance; The feature extraction module is used for acquiring defocused images of an ECM-stem cell construct photographed under a structure stability distance, extracting a complete area of the ECM-stem cell construct from the defocused images, calculating texture features based on the complete area, acquiring spatial distribution of the texture features in the complete area, dividing the complete area into two subareas, calculating statistics of the texture features for each subarea, calculating structural features representing structural relations among the subareas based on the statistics among different subareas, wherein the complete area is divided into two subareas, and comprises the steps of mapping the complete area into a two-dimensional coordinate system, calculating an average value of a horizontal coordinate and an average value of a vertical coordinate of each pixel in the complete area, rounding up the average value of the horizontal coordinate and the average value of the vertical coordinate to obtain two coordinate values, taking the pixel corresponding to the two coordinate values as a geometric center of the complete area, calculating a distance from each pixel to the geometric center in the complete area, normalizing all distance values, dividing the complete area into two different subareas based on the normalized distance, setting a distance threshold, dividing the normalized distance into a first subarea and a subarea as a first subarea, and dividing the normalized distance into a first subarea and a second subarea; The quality assessment module pre-trains a machine learning model, inputs structural features into the machine learning model, and outputs a quality score of the ECM-stem cell construct from the machine learning model.

Description

ECM-stem cell construct quality assessment method and system based on machine learning Technical Field The application relates to the technical field of medical image analysis, in particular to a machine learning-based ECM-stem cell construct quality assessment method and system. Background With the rapid development of regenerative medicine and tissue engineering technology, the ECM-stem cell construct as a three-dimensional functional structure simulating the natural tissue microenvironment has wide application prospect in the fields of disease treatment, tissue repair, drug screening and the like. However, its quality assessment still faces a number of technical challenges. Currently, the mainstream methods mainly rely on traditional biological detection technologies, such as histochemical analysis, biochemical component detection and the like, and the methods generally require destructive sampling of the construct, so that the same construct cannot be continuously monitored or used later, which severely limits the feasibility of the same construct in long-term quality tracking and clinical application. The ECM-stem cell construct is a key product in tissue engineering and regenerative medicine, and the quality evaluation is a core link for guaranteeing the safety and effectiveness of clinical application. Currently, this field relies mainly on traditional biological detection methods, which have significant limitations. First, methods such as histochemical analysis often require destructive sampling of the construct, rendering continuous monitoring and subsequent use of the same construct impossible, and making long-term quality tracking difficult. Second, existing microscopic observation methods are highly dependent on subjective experience of operators, resulting in lack of objectivity and repeatability of the evaluation results. In addition, the conventional method can only acquire morphological indexes of single dimensions such as projection area, and key information such as internal texture and structural heterogeneity of the morphological indexes cannot be deeply mined and comprehensively quantized from the image. Although machine learning techniques have been introduced into medical image analysis, in applications in the art, how to stably and automatically extract features from conventional bright field images that reliably reflect the complexity of the internal structure of a three-dimensional construct and build an accurate quality assessment model remains a technical challenge to be solved. Thus, the invention provides a machine learning-based ECM-stem cell construct quality assessment method and system. Disclosure of Invention The embodiment of the specification provides the following technical scheme: Step S1, continuously shooting an ECM-stem cell construct at a plurality of defocus distances with equal intervals to obtain a defocus image sequence, and calculating the projection area of the ECM-stem cell construct as a basic morphological index for each image of the defocus image sequence; s2, calculating the change rate of the basic morphological index relative to the defocus distance, and taking the defocus distance with the change rate smaller than a preset threshold value for the first time as the structure stability distance; S3, acquiring defocused images of the ECM-stem cell construct photographed at a structure stability distance, extracting complete areas of the ECM-stem cell construct from the defocused images, calculating texture features based on the complete areas, acquiring spatial distribution of the texture features in the complete areas, dividing the complete areas into two sub-areas, calculating statistics of the texture features for each sub-area, and calculating structural features representing structural relations among the sub-areas based on the statistics among different sub-areas; step S4, training a machine learning model in advance, inputting the structural characteristics into the machine learning model, and outputting the quality scores of the ECM-stem cell constructs by the machine learning model. Compared with the prior art, the invention has the following beneficial effects: According to the technical scheme, quality assessment can be completed by analyzing image sequences shot at different defocus distances without destructive processing of an ECM-stem cell construct, continuous monitoring and subsequent use of the construct are supported, morphology and texture features are automatically extracted and analyzed by adopting image processing and a machine learning model, differences in artificial subjective judgment are effectively reduced, consistency and repeatability of assessment results are guaranteed, comprehensive and deeper quantitative assessment of the quality of the construct is realized by integrating projection areas, spatial distribution of the texture features and structural relations among subareas, structural stability and distance are determi