Brain organ quality evaluation method and system based on artificial intelligence, and medium

Abstract

The invention belongs to the technical field of biomedical engineering, medical image and artificial intelligence crossing, and particularly relates to a brain organ quality evaluation method and system based on artificial intelligence and a medium. The method comprises the following steps: acquiring three-dimensional high-field-intensity original MRI data of a brain organ; performing three-dimensional voxel-level segmentation on the MRI data through a deep learning model, and identifying organoid tissues, backgrounds and organoid internal structures; and calculating a plurality of quality indexes according to segmentation and identification results, and synthesizing the quality indexes to generate a standardized quality score. The invention also provides a system for realizing the method. The method can ensure that the score can objectively and accurately reflect the structural integrity and the biological function of the organoid. Compared with the existing destructive or artificial qualitative method, the invention provides a non-invasive, objective, automatic, high-throughput and standardized evaluation tool, and provides key technical support for quality control, drug screening and regenerative medicine application of organoids.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of biomedical engineering, medical imaging and artificial intelligence, and particularly relates to a brain organoid quality assessment method and system based on artificial intelligence and a medium. BACKGROUND

[0002] Brain organoids, as an advanced biological model derived from stem cells, can self-organize into structures and functions similar to the brain in a three-dimensional environment in vitro, and have shown great application potential in the fields of neural development research, disease modeling, drug screening and personalized medicine. However, the current development and application of organoid technology faces a core bottleneck, that is, the universal heterogeneity of brain organoids. Due to the inherent randomness of stem cell differentiation, uneven penetration of nutrients and oxygen in three-dimensional structures, brain organoids have significant intra-batch and inter-batch differences in size, shape, cell composition and functional maturity. This high heterogeneity seriously affects the reproducibility of scientific research results and becomes a major obstacle to its transformation into clinical and industrial applications.

[0003] In order to solve this problem, various quality assessment methods have been developed in the field, but all have fundamental limitations. The traditional gold standard method is histological analysis, such as hematoxylin-eosin staining or immunofluorescence staining, to evaluate cell stratification, necrotic areas or expression of specific neuronal markers on two-dimensional sections of organoids. Although this method can provide accurate cell-level information, it is essentially terminal and destructive, and the sample after evaluation cannot be used for subsequent in vivo experiments.

[0004] Non-invasive assessment methods mainly rely on optical microscopic imaging. Brightfield microscopy is the most common method, but it can only provide two-dimensional projection information of the organoid, and cannot observe its internal three-dimensional structure, and the evaluation results are highly dependent on the subjective experience of the observer.

[0005] Although advanced optical imaging techniques such as confocal or light sheet microscopy can achieve three-dimensional imaging, their penetration depth is usually limited to a few hundred microns in dense and opaque brain organoids, making it difficult to perform complete and in-depth imaging of millimeter-sized organoids. In addition, long-term laser irradiation may cause phototoxicity, and sample transparentization treatment and other pretreatment steps may change the physiological state of the organoid.

[0006] MRI technology has good soft tissue contrast and deep imaging capability, and can provide more three-dimensional information than traditional optical imaging. In particular, in the application of small animals and organoids, ultra-high field (such as 9.4T) MRI can achieve high voxel resolution (about 40 microns) by using special pulse sequences and gradient systems, so as to capture the fine features of organoids. In recent years, some studies have begun to explore the application of images and AI to organoid analysis. For example, some studies have first applied high-field MRI combined with neural networks to image analysis of human brain organoids, extracting organoid volume and structural features. However, existing studies mostly focus on single indicators or single targets (such as only classifying cysts), and there is no complete solution that integrates multiple MRI feature indicators into a unified scoring system. The existing technology cannot provide a comprehensive and objective evaluation of brain organoids based on MRI data, which hinders the application of MRI data in brain organoid culture and related research.

[0007] Therefore, there is an urgent need in the art to develop an MRI+AI-based automated system that can non-destructively extract and quantify multiple features of organoids and convert them into standardized quality scores for guiding the quality evaluation and quality control of brain organoids. Thus, a non-invasive, three-dimensional, objective, quantitative and high-throughput quality evaluation technology for living brain organoids is realized to establish a unified quality control standard. SUMMARY

[0008] The purpose of the present application is to provide an artificial intelligence-based brain organoid quality evaluation method, system and medium.

[0009] One aspect of the present application provides an artificial intelligence-based brain organoid quality evaluation method, comprising the following steps: Step 1: obtaining raw MRI data; Step 2: performing three-dimensional voxel-level segmentation on the raw MRI data through a deep learning model to identify organoid tissue, background and internal structure of the organoid; Step 3: calculating a plurality of quality indicators according to the segmentation result of Step 2, and generating a standardized quality score by integrating the quality indicators.

[0010] Preferably, in Step 1, the raw MRI data is collected at a three-dimensional spatial resolution of less than 100 microns under an ultra-high field strength of greater than or equal to 7 tesla.

[0011] Preferably, the deep learning model is a hybrid encoder-decoder architecture, which comprises: a) a 3D convolutional network encoder for extracting multi-scale hierarchical features from the raw MRI data; b) a patch embedding layer to convert the multi-scale hierarchical features into a series of spatial tokens; c) a Transformer encoder with multi-head self-attention mechanism inside to model the global dependencies among the spatial tokens; d) a 3D convolutional network decoder with skip connections to fuse the hierarchical features from the 3D convolutional network encoder and the global context from the Transformer encoder, to generate the voxel-level segmentation mask.

[0012] Preferably, the quality indicators include at least one of the following indicators: brain organoid volume, surface area, surface-to-volume ratio, sphericity, maximum Feret diameter, cystic lumen number, cystic lumen volume, signal intensity distribution characteristics.

[0013] Preferably, the surface area is calculated by applying a 3D surface reconstruction algorithm to the segmentation result of step 2; The surface-to-volume ratio is the ratio of the surface area to the volume, which is used to quantify the morphological complexity of the organoid; The sphericity is a dimensionless indicator calculated based on the volume and surface area, which is used to quantify the degree of organoid approaching a perfect sphere; The cystic lumen number is the result of counting the independent low-signal or high-signal connected regions segmented within the organoid; The cystic lumen volume fraction is the result of dividing the total volume of all cystic lumen regions by the total volume of the organoid; The signal intensity distribution characteristics are the distribution characteristics of the MRI signal intensity values within the organoid tissue, selected from at least one of the following indicators: mean value, standard deviation, histogram morphology.

[0014] Preferably, in step 3, the method for synthesizing the quality indicators to generate a standardized quality score is: Normalizing each quality indicator; Generating a quality score using a scoring model, which is a weighted linear combination or a supervised learning regression algorithm; Mapping the quality score to a score value of 0-100.

[0015] Preferably, the parameter tuning process of the scoring model includes the following steps: correlating the generated quality score results with histological analysis and electrophysiological function test results, and accordingly tuning the algorithm parameters.

[0016] Another aspect of the present application provides an artificial intelligence-based brain organoid quality evaluation system for implementing the above-mentioned artificial intelligence-based brain organoid quality evaluation method, comprising: The input module is configured to obtain original MRI data; The segmentation and recognition module is configured to perform three-dimensional voxel-level segmentation on the original MRI data through a deep learning model to identify organoid tissues, backgrounds, and internal structures of the organoids. The quality score module is configured to calculate a plurality of quality indicators according to the segmentation result and generate a standardized quality score by comprehensively combining the quality indicators.

[0017] Preferably, the method further comprises: The output module is configured to output the quality score and an evaluation result based on the quality score, wherein the evaluation result comprises key images of each organoid sample, a plurality of quantitative indicators, the quality score, and a quality level evaluation based on the quality score.

[0018] The third aspect of the application also provides a computer-readable storage medium having stored thereon a computer program for implementing the above-mentioned brain organoid quality evaluation method based on artificial intelligence.

[0019] The application provides a brain organoid quality evaluation method and system based on MRI+AI, which can provide a standardized and objective brain organoid quality score report. For example, for each evaluated organoid, the system outputs its main indicator values (such as organoid volume, surface area, SA / V ratio, sphericity, etc.) and a comprehensive quality score. The score is represented by 0-100, and a quality level threshold can be set. Researchers can thus screen high-quality organoids for key experiments, such as selecting only organoids with a score of “good” or above in drug screening tests to reduce experimental result variability. At the same time, the score also provides quantitative feedback for improving culture conditions. For example, if a batch of scores are generally low, it can be analyzed which indicator, such as volume or cyst cavity, deviates from the expectation, thereby inferring the problem with the culture (such as poor morphological complexity development, etc.). In summary, the application establishes a complete system from MRI imaging to AI analysis to quality scoring, fills the gap of the lack of existing brain organoid quality evaluation standards, solves the problem of difficulty in comprehensively and objectively reflecting brain organoid quality, and has important significance for improving the reliability of organoid research and application.

[0020] The technical scheme of the application can achieve the following beneficial technical effects: (1) Non-invasive and fully automatic quality evaluation: without any staining or destructive treatment of the organoids, the quality evaluation result can be obtained by only acquiring images through MRI scanning and automatically analyzing by an AI algorithm. This avoids the practice of destroying samples in traditional histological verification, so that valuable organoids can be preserved for subsequent experiments. At the same time, full automation of the process reduces human subjective factors, ensuring consistency and objectivity of the evaluation.

[0021] (2) Multi-parameter comprehensive analysis, more comprehensive and accurate: The present application considers multiple aspects such as the size, structural integrity, and tissue uniformity of organoids, and can more accurately distinguish quality differences. In particular, the hybrid CNN-Transformer model and its self-attention mechanism can effectively capture the global features of irregular organoid morphology, thereby improving the accuracy of segmentation and analysis.

[0022] (3) High sensitivity for early screening: MRI can detect internal abnormalities (such as small cyst formation) in the early stages of organoid culture. Combined with AI quantitative analysis, the system can assess the quality trend early in the organoid culture process and timely screen out potential low-quality samples or adjust the culture conditions. Compared with relying on naked eye observation to observe visible differences or end-point histology, the present system provides more sensitive quality monitoring.

[0023] (4) Quantitative standard uniformity, results easy to compare: The introduction of a unified score of 0-100 makes it possible to directly compare the quality of organoids from different batches, different experiments, or even different laboratories, providing a basis for establishing industry standards and cross-study comparisons.

[0024] (5) Verification of correlation with functional results: The scoring results output by the present system have been verified by independent biological experiments and have actual biological significance. High-scoring organoids often show more mature and healthy structures histologically and exhibit more active and regular firing patterns in MEA functional assays. Low-scoring organoids are often accompanied by tissue necrosis, poor neural network development, and weak and disordered electrical activity. The high correlation between the score and these key quality indicators indicates the reliability and effectiveness of the scoring system of the present application. Users can significantly improve the success rate and consistency of subsequent experiments (such as drug screening or disease model research) by screening organoids based on the score.

[0025] (6) Improve research and application efficiency: Using the present system, researchers can objectively evaluate the quality of a large number of organoids in a short period of time without having to spend days on individual slicing and microscopic analysis. In particular, in the case of high-throughput drug screening and other situations requiring batch processing of organoids, low-scoring samples can be removed using the present system, and high-quality samples can be focused on for subsequent testing, thereby avoiding wasting resources on low-quality samples. This will significantly improve overall experimental efficiency. For industrialized organoid production, the present system can also be used as one of the quality release standards, with each batch of products being delivered after passing the MRI+AI evaluation, greatly improving product consistency and reliability.

[0026] In summary, the present application combines ultra-high field MRI imaging technology with advanced AI image analysis to achieve accurate evaluation and scoring of brain organoid quality, providing rich quantitative information while maintaining sample integrity, and has important scientific research and application value.

[0027] Obviously, according to the above content of the present application, other various forms of modifications, substitutions or changes can be made according to the ordinary technical knowledge and common practice in the art without departing from the above basic technical idea of the present application.

[0028] The above content of the present application will be further illustrated in detail by way of specific embodiments. However, it should not be understood that the scope of the above subject matter of the present application is limited to the following examples. Any technology realized based on the above content of the present application falls within the scope of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0029] Figure 1 The figure is a high-throughput workflow functional diagram of the brain organoid quality evaluation and scoring system.

[0030] Figure 2 The figure is a physical diagram of key components of the high-throughput MRI imaging system. The left figure shows a customized multi-well plate carrier for high-throughput scanning. The upper right figure shows a microscopic view of brain organoids cultured in a single U-shaped well. The lower right figure shows a scenario of loading the organoid-loaded well plate into a 9.4 Tesla MRI scanner for data acquisition.

[0031] Figure 3 The figure is a schematic diagram of the hybrid CNN-Transformer network structure.

[0032] Figure 4 The figure is an example of brain organoid segmentation accuracy (left) and MRI scan images, segmentation results and true annotation results (right).

[0033] Figure 5 The figure is an example of midbrain organoid segmentation accuracy (left) and MRI scan images, segmentation results and true annotation results (right).

[0034] Figure 6 The figure is a schematic diagram of computer output quality score results, which shows the comprehensive evaluation report generated by the system for a single organoid.

[0035] Figure 7 The figure is a result graph of quality score and tissue immunofluorescence apoptosis-related indicator cleaved-caspase3 in the example.

[0036] Figure 8 The figure is a result graph of quality score and MEA electrophysiological function indicator firing rate in the example (p<0.0001). DETAILED DESCRIPTION

[0037] In the following examples and experimental examples, reagents and materials not specifically named are commercially available. The algorithms for data acquisition, transmission, storage and processing, and the hardware structure, circuit connection, etc. not specifically described in the examples can be realized through the existing technology disclosed.

[0038] Example 1 Artificial intelligence-based brain organoid quality assessment system and method The system of the present embodiment comprises: An input module configured to obtain raw MRI data; A segmentation and recognition module configured to perform three-dimensional voxel-level segmentation on the raw MRI data through a deep learning model to identify organoid tissues, background, and internal structures of the organoids; A quality score module configured to calculate a plurality of quality indicators according to the segmentation results and to generate a standardized quality score by synthesizing the quality indicators; An output module configured to output the quality score and an evaluation result based on the quality score, the evaluation result including key images of each organoid sample, quantitative indicators, quality scores, and quality level evaluations based on the quality scores.

[0039] The following describes a method for brain organoid quality assessment through the above system by means of a specific practical application case. A flowchart is shown in Figure 1 The figure shows that starting from 3D printing of organoid preparation and high-throughput well plates, human-induced pluripotent stem cells are cultured and differentiated to form brain organoids, then the well plates loaded with organoids are placed in a high-field MRI for scanning, and finally AI models are used for automated segmentation and quality assessment. The specific steps are as follows: 1. Organoid culture The evaluation method of the present application is applicable to brain organoids of various sources and types. In one specific embodiment, the brain organoids are differentiated from the H9 human embryonic stem cell line (hESC). According to different differentiation protocols, organoids of specific brain regions or whole brain organoids can be prepared.

[0040] (1) Preparation of whole brain organoids: A guide-free differentiation protocol based on Lancaster et al. is used. The brief process is as follows: First, the H9 cells maintained in culture are dissociated into single cells or small cell clusters to form embryoid bodies in low-attachment V-shaped or U-shaped bottom 96-well plates. After about 5-7 days of embryoid body formation and neural induction, healthy embryoid bodies are embedded in a basement membrane matrix (such as Matrigel) to promote the outgrowth of neural epithelium and tissue polarization. Subsequently, the embedded organoids are transferred to a dynamic culture system, such as an orbital shaker or a small-scale rotary bioreactor, for long-term culture to improve nutrient and oxygen exchange and promote further growth and maturation.

[0041] (2) Preparation of midbrain organoids: The preparation process is similar to that of whole-brain organoids, but specific pattern-inducing factors are introduced in the early differentiation stage to guide stem cells to differentiate into the midbrain neural lineage. Specifically, during the neural induction stage, small molecule compounds such as agonists of the Sonic hedgehog signaling pathway and fibroblast growth factor 8 are added to the culture medium. These factors mimic the signaling environment of the midbrain region during embryonic development, thereby specifically inducing midbrain organoids rich in dopaminergic neurons. Brain organoids prepared by the above method can be used in the quality assessment process of this invention after being cultured for a specific number of days (e.g., 40 days or longer).

[0042] 2. MRI image data acquisition The first step of the method of this invention is to acquire three-dimensional MRI image data of brain organoids. In a preferred embodiment, scanning is performed using a 9.4 Tesla ultra-high field preclinical MRI device (Bruker). To achieve high-throughput assessment, multiple brain organoids can be placed in a customized multi-well plate carrier for simultaneous batch scanning (a key physical component for data acquisition and high-throughput assessment in this invention, such as...). Figure 2 (As shown). The preferred imaging sequence is a T2-weighted three-dimensional RARE sequence, which is sensitive to the contrast between soft tissue and fluid and can clearly distinguish structures such as cysts inside organoids. To achieve fine imaging, the three-dimensional spatial resolution is preferably set to less than 100 micrometers.

[0043] This invention is not limited to specific field strengths or sequences. The method of this invention is applicable to ultra-high field magnetic resonance imaging (MRI) in a broad sense, where the magnetic field strength is typically defined as greater than or equal to 7T. In principle, devices with other field strengths are also applicable, provided the MRI equipment can provide sufficient spatial resolution to resolve internal organoid microstructures (such as cavities and tissue layers). Furthermore, in addition to T2 / T2* weighted imaging, other MRI sequences such as T1-weighted imaging and diffusion-weighted imaging (DWI) can also be used to extract supplementary structural or microstructural parameters (such as apparent diffusion coefficients). In practical applications, a trade-off can be made based on requirements: for quality scoring, a single high-contrast resolution structural image is sufficient; for research purposes, other sequences can be added, but the scan time needs to be extended.

[0044] 3. The raw MRI data is segmented into three dimensions at the voxel level using a deep learning model to identify organoid tissue, background, and internal structures of the organoid: After acquiring MRI data, this invention utilizes a deep learning segmentation model to process the images and automatically segment the tissue regions of organoids. For example... Figure 3As shown, in a preferred embodiment, the model employs a hybrid CNN-Transformer architecture. This model effectively combines the advantages of CNN and Transformer, specifically: (1) 3D CNN encoder: Extracts multi-level local spatial features from the input 3D MRI image. The encoder uses successive 3D convolution and straddle convolution operations to progressively compress the spatial dimension while increasing the feature channels. It specifically uses Instance Normalization to independently normalize each sample, making it more robust to possible signal intensity differences between different culture batches.

[0045] (2) Patch Embedding and Transformer Encoder: The deep feature map output by the encoder is fed into a patch embedding layer, which converts it into a one-dimensional serialized "tag" representation. These tags are then fed into an encoder consisting of multiple Transformer Blocks. The core of the Transformer is the multi-head self-attention mechanism, which allows each tag in the model to pay attention to and integrate information from all other tags in the sequence, thereby effectively capturing long-range, global contextual information. This is particularly effective for recognizing organoid structures with irregular shapes or diffuse distribution.

[0046] (3) 3D CNN decoder: The decoder receives a sequence of features rich in global context from the Transformer, upsamples them step by step, and fuses them with the skip connection features from the CNN encoder to generate a voxel-level segmentation mask.

[0047] (4) Model Training: Model training requires labeled MRI image data. First, a certain number of brain organoid MRI scans were obtained from different sources, and experienced biology experts manually delineated and marked the organoid outlines on these MRI 3D images. These manually labeled voxels were used as ground truth values ​​to train the deep learning model. The loss function was a weighted sum of binary cross-entropy (BCE) and Dice coefficient loss to balance overall and detail segmentation accuracy. The optimizer used was AdamW with an initial learning rate of 1e-4, coupled with a cosine annealing learning rate scheduler.

[0048] (5) Model performance verification: Figure 4 The model's Dice coefficient quantization distribution on a brain organoid test set (n=60) is shown, with an average Dice coefficient of 0.946, demonstrating the high accuracy of the model's segmentation. The triptych on the right visually compares the original MRI images from the 15-well plate, the AI ​​model's predicted segmentation results, and the manually annotated ground truth results, labeling the Dice values ​​for any organoid and showing a high degree of consistency between the AI ​​predictions and ground truth annotations.Figure 5 As shown, the violin plot on the left illustrates the Dice coefficient distribution of the model on the midbrain organoid test set (n=58), with an average Dice coefficient of 0.906. The triptych on the right compares the original image, AI prediction, and ground truth annotation. This figure demonstrates that the AI ​​model of this invention has good generalization ability and can accurately segment different types of brain organoids. The above results show that the deep learning segmentation model of this embodiment combines high accuracy and good generalization ability.

[0049] 4. Calculation of quality indicators (including structural quantitative indicators) After the model completes the accurate segmentation of the organoid tissue region, the system will automatically calculate and extract multiple quality indicators.

[0050] The indicators include: (1) Organoid volume: The volume (number of voxels × volume of individual voxels) is calculated based on the segmented organoid body regions. The volume reflects the growth size of the organoid and is one of the basic quality indicators.

[0051] (2) Surface area: A three-dimensional mesh is generated by applying a three-dimensional surface reconstruction algorithm to the three-dimensional segmentation mask, and the total surface area of ​​the mesh is calculated.

[0052] (3) Surface area to volume ratio: that is, the surface area divided by the volume. This indicator is a key indicator reflecting the morphological complexity of organoids (such as surface wrinkles). In healthy development, the SA / V ratio usually increases as organoids mature and become more complex.

[0053] (4) Sphericity: A dimensionless index calculated based on volume and surface area. The value is between 0 and 1, where 1 represents a perfect sphere. The lower the value, the more irregular or complex the morphology. In organoid development, this is usually associated with higher maturity and complexity.

[0054] (5) Other indicators: may also include signal intensity distribution characteristics (reflecting the distribution characteristics of MRI signal intensity values ​​inside organoid tissue), maximum Ferris diameter (reflecting the maximum spatial span of organoids), etc.

[0055] The quality indicators selected in this embodiment include: surface area to volume ratio, sphericity, and volume and signal strength distribution characteristics.

[0056] 5. Calculation and output of quality scores After acquiring indicators such as volume, surface area, SA / V ratio, and sphericity, the system needs to integrate them into a final quality score. First, each indicator is standardized. Then, different weights are assigned to each indicator, and the total score is calculated. The weights can be determined based on expert experience or through data training. For example, supervised learning regression algorithms (such as multiple linear regression or random forest regression) can be used to automatically learn the weights and mapping functions of each indicator based on training data labeled with quality levels, ensuring that the output score closely matches the expert evaluation results.

[0057] In this embodiment, the specific method for determining the weights is a supervised learning method based on biological calibration: a) Establishing the "gold standard" in biology: Organoid models were established for a "healthy" group and a "damaged" (low-quality) group. First, baseline high-field MRI scans were performed on brain organoid samples on day 0. Subsequently, CoCl2 was applied to the organoids (100 nM for 24 h followed by 50 nM for 72 h) to induce chemical hypoxia, simulating a low-quality state. After treatment (day 21), the same batch of organoids was scanned again on MRI. The use of CoCl2 to simulate chronic hypoxia is one of the main causes of low organoid quality (such as core necrosis).

[0058] b) Feature extraction and model training: MRI scans were performed on organoids from both the "healthy" group and the "hypoxic" group treated with CoCl2, and biological validation data were collected simultaneously.

[0059] Based on the characteristic differences between the chemically hypoxic group and the normal group, feature extraction and analysis were performed. The morphological complexity index, normalized size, and signal intensity distribution extracted from MRI were used as the model input features (X), and the "biological health index" was used as the prediction target (Y).

[0060] c) Determine the weights and scoring model: A supervised learning regression algorithm (XGBoos) was used to train the model to learn the mapping relationship between MRI features and biological health indices.

[0061] d) Once the quality score is calculated, it can be displayed on the user interface and stored in the database. The system generates a quality report for each organoid, such as... Figure 6 As shown in the image. The report includes key visualizations, detailed quantitative metrics, and concludes with a comprehensive quality score.

[0062] 6. System calibration and verification To ensure the scoring results have practical biological significance, the system of this invention underwent extensive verification and calibration during development. The obtained quality indicators and scores were correlated with traditional biological indicators. Based on the analysis results, the parameters of the scoring model used for quality scoring could be fine-tuned.

[0063] In this embodiment, the verification results are as follows: (1) Histological verification: such as Figure 7 As shown, a batch of organoids was re-scanned and divided into "high" and "low" groups based on their AI scores. Immunofluorescence staining was used to compare the apoptosis marker cleaved-caspase 3 (a marker of cell death). The results showed that the "high" group exhibited extremely low apoptosis signals, while the "low" group showed strong and widespread apoptosis signals. This confirms that the AI ​​score of this invention is highly correlated with the biological health status (cell viability) of organoids, with high scores corresponding to low apoptosis (greater health).

[0064] (2) Electrophysiological function verification: such as Figure 8 As shown, organoids with high and low scores were placed on a microelectrode array (MEA) for functional recording. The results showed that the average firing frequency of the "high-scoring group" was significantly higher than that of the "low-scoring group." This confirms that organoids with high AI scores possess more active and mature neural network functions. These independent experimental results confirm that the system's score does indeed represent the biological quality of organoids to a considerable extent.

[0065] Based on the above verification, different operators used the system to score the same batch of organoids, and the results were highly consistent; when the same organoid was scanned twice, the score difference was within 1-2 points. These all demonstrate the reliability and robustness of the system.

[0066] 7. Examples of using quality scoring to guide research Application in Culture Methodology Research: In a study aimed at optimizing culture protocols, researchers used this system to compare the effects of two different culture medium formulations on organoid quality. The system scoring results clearly showed that organoids cultured using formulation A scored approximately 15 points higher on average than those cultured using formulation B. This quantitative result was validated by subsequent histological analysis, which revealed clearer cortical lamellar structures and higher neuronal viability in group A organoids. This example demonstrates that this system can provide objective and quantitative feedback for optimizing culture conditions, helping researchers quickly identify superior culture protocols.

[0067] As can be seen from the above embodiments, this invention provides a brain organoid quality assessment and scoring system based on MRI and AI. It acquires brain organoid image data through high-resolution MRI imaging and analyzes the images using a deep learning model to extract various objective indicators and calculate a comprehensive quality score. The system and method provided by this invention have the advantages of being non-invasive, objective, automated, high-throughput, and standardized, providing key technical support for organoid quality control, drug screening, and regenerative medicine applications, and has excellent application prospects.

Claims

1. An artificial intelligence-based brain organoid quality assessment method, characterized by, The method comprises the following steps: Step 1, obtaining original MRI data; Step 2, performing three-dimensional voxel-level segmentation on the original MRI data by a deep learning model to identify organoid tissue, background and internal structure of the organoid; Step 3, calculating a plurality of quality indicators according to the segmentation result of step 2, and generating a standardized quality score by synthesizing the quality indicators.

2. The artificial intelligence-based brain organoid quality assessment method of claim 1, wherein, In step 1, the original MRI data is collected at a three-dimensional spatial resolution of less than 100 microns under an ultra-high field strength of greater than or equal to 7 tesla.

3. The artificial intelligence-based brain organoid quality assessment method of claim 1, wherein, The deep learning model is a hybrid encoder-decoder architecture, which comprises: a) a 3D convolutional network encoder for extracting multi-scale hierarchical features from the original MRI data; b) a patch embedding layer for converting the multi-scale hierarchical features into a series of spatial tokens; c) a Transformer encoder comprising a multi-head self-attention mechanism for modeling global dependencies between the spatial tokens; d) a 3D convolutional network decoder with skip connections to fuse hierarchical features from the 3D convolutional network encoder and global context from the Transformer encoder, thereby generating a voxel-level segmentation mask.

4. The artificial intelligence-based brain organoid quality assessment method of claim 1, wherein, The quality indicators include at least one of the following indicators: brain organoid volume, surface area, surface area to volume ratio, sphericity, maximum Feret diameter, cystic cavity number, cystic cavity volume, signal intensity distribution characteristics.

5. The artificial intelligence-based brain organoid quality assessment method of claim 4, wherein, The surface area is calculated by applying a three-dimensional surface reconstruction algorithm to the segmentation result of step 2; The surface area to volume ratio is the ratio of surface area to volume, which quantifies the morphological complexity of the organoid; The sphericity is a dimensionless indicator calculated based on volume and surface area to quantify the degree to which the organoid approximates a perfect sphere; The cystic cavity number is the result of counting independent low-signal or high-signal connected regions segmented within the organoid; The cystic cavity volume fraction is the result of dividing the total volume of all cystic regions by the total volume of the organoid; The signal intensity distribution characteristics are the distribution characteristics of the MRI signal intensity values within the organoid tissue, selected from at least one of the following indicators: mean, standard deviation, histogram shape.

6. The artificial intelligence-based brain organoid quality assessment method of claim 1, wherein, In step 3, the method for generating a standardized quality score by synthesizing the quality indicators is: Normalizing each quality indicator; Generating a quality score using a scoring model, which is a weighted linear combination or a supervised learning regression algorithm; Mapping the quality score to a score value of 0-100.

7. The artificial intelligence-based brain organoid quality assessment method of claim 6, wherein, The parameter tuning process of the scoring model includes the following steps: correlating the generated quality score results with histological analysis and electrophysiological function test results, and tuning the algorithm parameters accordingly. 8.A system for artificial intelligence-based quality assessment of brain organoids, characterized by, The method for implementing the artificial intelligence-based brain organoid quality evaluation method of any one of claims 1-7 comprises: an input module configured to obtain original MRI data; a segmentation and identification module configured to perform three-dimensional voxel-level segmentation on the original MRI data by a deep learning model to identify organoid tissue, background and internal structure of the organoid; The quality score module is configured to calculate a plurality of quality indexes according to the segmentation result, and to generate a standardized quality score by synthesizing the quality indexes.

9. The artificial intelligence-based brain organoid quality assessment system according to claim 8, characterized by, Further comprising: The output module is configured to output the quality score and an evaluation result based on the quality score, the evaluation result including a key image of each organoid sample, a plurality of quantitative indexes, the quality score, and a quality grade evaluation based on the quality score.

10. A computer-readable storage medium, characterized in that, A computer program for implementing the artificial intelligence-based brain organoid quality evaluation method of any one of claims 1-7 is stored thereon.

Images