A medical cell detection method

Abstract

The application provides a medical cell detection method, comprising: S1: collecting a cell image for preprocessing; S2: performing structural processing and encoding processing on multi-modal data; S3: performing feature extraction on the cell image and the multi-modal data; S4: fusing the cell image and the multi-modal data; S5: inputting the fused feature vector into a segmentation head of FastSAM; S6: generating a segmentation mask of the image; S7: judging whether the cell is abnormal; S8: for the cell with the abnormality, resampling and culturing the cell; S9: collecting a cell image in a cell culture device in real time for preprocessing; and S10: inputting the cell image into a FastSAM model, and analyzing indexes of the cell based on the generated segmentation mask; the application combines the cell image and multi-modal data such as a case and a clinical examination result, more comprehensively understands characteristics of the cell, and improves detection accuracy.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of cell detection, and in particular to a medical cell detection method. BACKGROUND

[0002] In the field of medical cell detection, in order to train an accurate target detection model, a large number of training sample sets with accurate class labels are needed. The traditional manual labeling method is inefficient and prone to errors, especially the target position labeling error is large and the precision is insufficient, which is difficult to meet the actual demand. The SAM model (Segment Anything Model) has a milestone significance in the field of computer vision. SAM has zero sample generalization ability and can handle unseen image and object classes without fine-tuning for specific tasks, which significantly reduces the dependence on specific domain data sets. SAM is the first "base model" for image segmentation, similar to GPT in NLP, and has a wide range of downstream task adaptation capabilities, providing a new paradigm for subsequent research.

[0003] The SAM model is based on the ViT-Huge architecture, which has a slow inference speed, a large number of model parameters, and a high deployment cost. Due to the dependence on the distribution characteristics of the SA-1B data set, the generalization ability for some special fields, such as rare lesions in medical images, is limited. FastSAM (Fast Segment Anything Model) improves the shortcomings of SAM through lightweight design and algorithm optimization. Fast SAM uses YOLOv8 as a detector and replaces ViT with CNN. The model parameters are only about 1 / 9 of SAM, and the inference speed is increased by more than 50 times. FastSAM can run on low-end GPUs or even CPUs without the need for high-end hardware support, significantly reducing the deployment cost. However, FastSAM is trained on a large number of ordinary pictures, and the effect is not good for medical cell detection.

[0004] The applicant intends to combine FastSAM with the medical field to realize medical cell detection. After detailed search, no related technical solutions have been found. Therefore, a new technical solution is needed to solve the above technical problems. SUMMARY

[0005] The application provides a medical cell detection method, which comprises the following steps:

[0006] S1: collecting cell images and pre-processing the collected cell images;

[0007] S2: structuring and encoding the multi-modal data; wherein the multi-modal data includes cases and clinical examination results;

[0008] S3: feature extraction is performed on the cell image and multi-modal data;

[0009] S4: the cell image and multi-modal data are fused;

[0010] S5: the fused feature vector is input into the segmentation head of FastSAM, and the model will predict the probability of each pixel belonging to different categories according to the fused features;

[0011] S6: according to the predicted pixel category probability, the pixels are classified into the categories of normal cell area, abnormal cell area and background, and a segmentation mask of the image is generated;

[0012] S7: based on the segmentation mask, it is judged whether the cell is abnormal;

[0013] S8: for the cell with abnormality, resample for cell culture;

[0014] S9: real-time acquisition of cell images in the cell culture device, and pre-processing of the cell images;

[0015] S10: the pre-processed cell images in S9 are input into the FastSAM model for analysis of cell morphological indicators, cell number and density indicators, cell motion and migration indicators, cell internal structure indicators, and cell population feature indicators.

[0016] As a preferred scheme, the cell morphological indicators include cell size and area, cell shape, and cell perimeter.

[0017] As a preferred scheme, the cell shape includes circularity, and the calculation formula is: wherein A is the cell area, P is the cell perimeter, and the cell perimeter P is detected by an edge detection algorithm to obtain the number of edge pixels.

[0018] As a preferred scheme, the cell number and density indicators include cell count and cell density.

[0019] As a preferred scheme, the cell motion and migration indicators include cell displacement and cell migration speed.

[0020] As a preferred scheme, the cell internal structure indicators include cell nucleus features and organelle distribution.

[0021] As a preferred scheme, the cell population feature indicators include cell aggregation and distribution, and cell clonogenicity.

[0022] As a preferred scheme, a three-dimensional model of the cell is generated by three-dimensional reconstruction technology using the segmentation mask, the generated three-dimensional model of the cell is imported into a VR / AR environment, and a virtual cell observation scene is created.

[0023] As a preferred solution, the collected cell images are transmitted to a cloud server, the sensor device collects environmental parameters in the cell culture device in real time, and the environmental parameters are also transmitted to the cloud server. FastSAM performs cell segmentation and analysis on the images, and automatically adjusts the operating parameters of the cell culture device according to the analysis results of FastSAM and the real-time monitored environmental parameters.

[0024] As a preferred solution, the center position and area of each cell are calculated according to the segmented cell mask, the number of cells in a unit area is counted to obtain the density information of the cells, and whether the cell distribution is uneven is analyzed according to the center position and density information of the cells. If it exists, the unevenly distributed area and the reason are analyzed.

[0025] As a preferred solution, the inoculation method is adjusted or the mixing method is improved according to the unevenly distributed area and the reason.

[0026] As a preferred solution, the calculation method of the cell center position is as follows:

[0027] For each cell mask, the centroid is calculated as the center position of the cell. The cell mask is a binary image, where the cell area is the foreground with a value of 1, and the rest is the background with a value of 0. The coordinates (x c , y c ) of the centroid are calculated by the following formula:

[0028] ; wherein, M ( i , j ) is the pixel value of the cell mask image at the position of (x i , y j ), i and j represent the row and column indices of the image, respectively. The centroid coordinates of the cell are obtained by weighted average of the coordinates of all foreground pixels.

[0029] As a preferred solution, the calculation method of the cell area is to directly count the number of foreground pixels in the cell mask.

[0030] As a preferred solution, according to the position and density information of the cells, a visualization tool is used to generate a visualization map of the cell distribution.

[0031] As a preferred solution, the specific steps of generating the visualization map of the cell distribution are as follows:

[0032] Divide the entire image into grids of equal size;

[0033] For each grid, the number of cells contained therein is counted;

[0034] By judging whether the cell center position falls within the grid, it is determined whether the cell belongs to the grid, and then the ratio of the number of cells in each grid to the area of the grid is calculated to obtain the cell density of the grid region;

[0035] According to the cell density of each grid, a cell density map is drawn using a visualization method.

[0036] As a preferred solution, the medical image segmentation evaluation index is used to evaluate the difference between the segmentation results of the FastSAM model before and after the fusion of multi-modal information and the true label, and the experimental results of the model are evaluated.

[0037] As a preferred solution, according to the experimental results of the evaluation model, the model is optimized and adjusted.

[0038] The present application combines cell images and multi-modal data such as cases, clinical examination results, etc., can more comprehensively understand the characteristics of cells, improve the accuracy of detection; real-time image acquisition and analysis during cell culture can timely find the changes of cells, provide dynamic information for the research and treatment of diseases; and the FastSAM model is used for pixel-level segmentation, which can accurately identify normal cells and abnormal cells, provide accurate basis for subsequent analysis, and has high flexibility, and can realize arbitrary target segmentation. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 The logical block diagram of the present application is provided; DETAILED DESCRIPTION

[0040] The specific embodiments of the present application are described in detail below in conjunction with the accompanying drawings; it should be noted that the specific embodiments described herein are only used to illustrate and explain the present application, and are not used to limit the present application. Example 1

[0041] The present application provides a medical cell detection method, comprising the following steps:

[0042] S1: Collecting cell images and pre-processing the collected cell images;

[0043] Before this step, the FastSAM model needs to be trained. During the training process, cell images need to be collected. The cell images collected in this process cover different types of cells, such as blood cells, tumor cells, and nerve cells, and cover different imaging methods, such as optical microscopy and electron microscopy imaging. A rich and diverse cell image dataset is constructed. The collected images are sorted and classified according to cell type, imaging parameters, and other information to provide a clear data structure for subsequent model training and evaluation. The above belongs to the conventional model training method, and the present application does not make specific elaboration here.

[0044] This step collects the cells to be detected. The cells to be detected are pretreated, including but not limited to adjusting the image size, normalizing the pixel value, image enhancement, etc., to meet the requirements of subsequent processing. For example, adjusting the image size to the required size of the model input, such as 640x640 pixels, normalizing the pixel value to the [0, 1] interval, and performing data enhancement operations such as random flipping, rotation, brightness adjustment, etc. to enhance the generalization ability of the model.

[0045] S2: Structured processing and encoding processing of multi-modal data; wherein the multi-modal data includes case and clinical examination results;

[0046] Structured processing: For text information in the medical record, such as medical history description and diagnosis record, structured extraction is performed through natural language processing technology. Key entities such as disease name, symptoms, and time are extracted using named entity recognition (NER). Information extraction models are used to convert text into structured table data. For clinical examination results such as blood test indicators and biochemical data, they are directly sorted into numerical data tables. For text information that cannot be structured, such as detailed medical condition descriptions, pre-trained word embedding models (such as Word2Vec, BERT) are used to convert text into fixed-length vector representations. The average value of all word vectors in the text or the final text feature vector is obtained through a pooling operation.

[0047] Encoding processing: For numerical data, normalize the numerical values to the [0, 1] interval, and then directly convert them into one-dimensional vectors. For categorical data, use one-hot encoding or label encoding to convert vectors, such as gender (male / female) and disease categories. For example, the gender "male" is encoded as [1, 0] and the gender "female" is encoded as [0, 1]. If there are n disease categories, use one-hot encoding to generate an n-dimensional vector, with only the corresponding disease category position being 1 and the rest being 0.

[0048] S3: Feature extraction of cell images and multi-modal data;

[0049] The deep learning model extracts features from cell images and multi-modal data. For cell image feature extraction, convolutional neural networks (CNN) such as ResNet, VGG, Inception, etc. are used. The pre-processed cell images in S1 are input into the pre-trained CNN model to extract feature vectors from the intermediate or last layer. Taking ResNet as an example, its basic structure includes multiple residual blocks. Each residual block solves the gradient vanishing problem in deep neural networks through jump connection, enabling the model to train deeper network structures. When extracting features, the cell image is input into the ResNet model, and after a series of convolution and pooling operations, the feature vector is obtained from the global average pooling layer or the fully connected layer.

[0050] Regarding the extraction of multi-modal data features: 1. For unstructured text information converted into fixed-length vector representation, use the text feature vectors obtained by pre-training word embedding models (Word2Vec, BERT, etc.). These vectors can be directly input into pre-trained language models (such as BERT, GPT, etc.). The encoder layer in the model will further extract and abstract features from these vectors, learning semantic and syntactic features in the text through self-attention mechanisms, and output higher-level feature representations. For example, for a detailed description of a disease, after processing by the BERT model, a feature vector containing contextual semantic information can be obtained, better capturing key information in the disease description. 2. For structured table data text information converted by named entity recognition and information extraction models, such as disease names and symptoms, these text fields can be connected into a string, and then the above method for processing unstructured text information can be used to extract features using pre-trained language models. For example, connect "high blood pressure" and "headache, dizziness" into "high blood pressure, headache, dizziness" and input it into the pre-trained language model to obtain features. 3. For normalized numerical data converted into one-dimensional vectors, such as age and numerical indicators of clinical examinations, these vectors can be directly used as features input into subsequent models. If the data volume is large or the feature dimension is high, principal component analysis (PCA) or other dimension reduction methods can be considered to process the data, extract the main feature components, reduce the data dimension, and retain most of the information. 4. For category data vectors converted by one-hot encoding or label encoding, such as gender and disease categories, they can be directly used as features. However, one-hot encoding may result in high vector dimension, which may increase the model's computational load and risk of overfitting. At this time, an embedding layer can be used to map the one-hot encoded vector to a low-dimensional continuous vector space to obtain the embedding representation of the category data, which can be used as features input into the model. This way, the dimension can be reduced while learning the semantic relationship between categories.

[0051] S4: Fusion of cell image and multi-modal data;

[0052] The image feature vector and the multi-modal data feature vector are fused by simple splicing, weighted summation, etc., which facilitates subsequent use of the fused feature vector for final classification or prediction;

[0053] The feature vector extracted from the cell image is set as X image The feature vector extracted from the multi-modal data is X text They can be spliced into a new feature vector to obtain the fused feature vector X fusion [ X image ; X text ].

[0054] S5: The fused feature vector is input into the segmentation head of FastSAM, and the model will predict the probability of each pixel belonging to different categories according to the fused features;

[0055] S6: According to the predicted pixel category probability, use threshold method or other post-processing method to classify the pixels into different categories to generate the segmentation mask of the image; for example, if the cells are classified, the pixels in the image may be divided into normal cell area, abnormal cell area, background different categories, each category corresponds to a specific mask area, the segmentation mask is superimposed or mapped with the original image to obtain a segmented picture, different parts of the picture are accurately divided into corresponding category areas, achieving the purpose of image segmentation; by analyzing the segmentation mask, the geometric features such as the bounding box, area and perimeter of each cell can be calculated, which is helpful for further quantitative analysis of the cells.

[0056] The segmentation mask is a binary image or a multi-channel image with the same size as the original image. In the binary image, the value of each pixel represents whether the pixel belongs to a specific category. In this embodiment, the specific category is the foreground cell or the background. In the multi-channel image, each channel can represent a different category, such as channel 1 representing normal cells and channel 2 representing cancer cells, etc. Different color or gray value is used to distinguish different category areas.

[0057] S7: Based on the segmentation mask, determine whether the cell is abnormal;

[0058] Since the case and the clinical examination result information of the patient are fused, the segmentation result of the embodiment is combined with the specific situation of the patient; for example, according to the information such as the medical history and symptoms of the patient and the segmentation result of the cell image, whether the cells of the patient are abnormal, and the type and degree of the abnormality are judged, and a more comprehensive diagnosis basis is provided for the doctor.

[0059] More specifically, the following steps are included:

[0060] S71: judging whether the cells are abnormal according to the collected case and clinical examination result combined with the segmentation mask;

[0061] The case records in detail the past medical history of the patient, including whether the patient has suffered from cancer, autoimmune diseases, infectious diseases and the like, and family genetic history, and records the past treatment of the patient, such as whether the patient has received radiotherapy, chemotherapy and the like, which may have an impact on the cells; also used to record the current symptoms of the patient, including the occurrence time, duration and severity of the symptoms such as discomfort, pain, lump, fever and fatigue of each part of the body;

[0062] The segmentation mask obtained above clearly defines the boundary and internal structure of the cell and the like, and extracts the morphological features (such as shape, size, roundness and the like) of the cell, the texture features (such as gray scale distribution, texture complexity and the like) and the internal structure features (such as nucleus features, organelle distribution and the like) from the segmentation result; specific discussion is as follows:

[0063] Regarding shape: normal cells generally have a relatively regular shape, such as a circular shape, an oval shape and the like, if the segmentation mask shows that the shape of the cell is distorted, has a protrusion or is irregular polygonal, it may be an abnormal cell; for example, cancer cells often exhibit irregular shape, jagged boundary or pseudopod-like protrusion.

[0064] Regarding size: each cell type has its normal size range, the area, perimeter, diameter and the like of the cell are measured through the segmentation mask, if it is obviously larger or smaller than the normal range, there may be abnormality; for example, tumor cells may appear to have increased volume, while atrophic cells have smaller volume.

[0065] Regarding symmetry: normal cells mostly have good symmetry, if the cell shows obvious asymmetry in the segmentation mask, it may be an abnormality.

[0066] Regarding gray scale distribution: the gray scale distribution of normal cells is relatively uniform, while abnormal cells may have obvious changes in gray scale value due to internal structure changes, such as increased chromatin, abnormal organelles and the like, such as local gray scale value being too high or too low, uneven gray scale change and the like; by analyzing the gray scale histogram, mean, variance and the like of the cell region in the segmentation mask, whether the cell is abnormal can be assisted to be judged.

[0067] About texture complexity: the texture of normal cells is relatively simple and regular, while the texture of abnormal cells may be rough and chaotic. Methods such as gray level co-occurrence matrix can be used to quantify texture features, such as calculating energy, entropy, contrast, etc. If these parameters deviate from the normal range, it may mean that the cell is abnormal.

[0068] About cell nucleus features: the shape, size and location of the cell nucleus are important criteria for judgment. Normal cell nuclei are usually located in the center of the cell, have regular shapes, and have moderate sizes relative to the overall cell. If the segmentation mask shows that the cell nucleus deviates from the center, has an irregular shape, has a rough nuclear membrane, or has an abnormally increased nuclear-cytoplasmic ratio, it may indicate that the cell is abnormal. For example, cancer cells often have large and darkly stained nuclei, and an increased nuclear-cytoplasmic ratio.

[0069] About organelle distribution: although the segmentation mask generally cannot accurately display organelles, if it can be roughly distinguished, the distribution of organelles in normal cells has certain regularity. For example, mitochondria are uniformly distributed in the cytoplasm, and organelles such as endoplasmic reticulum have specific locations and shapes. If the distribution of organelles is disordered, aggregated or missing, it may indicate that the cell is abnormal.

[0070] When the segmentation mask determines that one of the shape, size, roundness, gray scale distribution, texture complexity, cell nucleus features, and organelle distribution is abnormal, a secondary determination is made. If one of the secondary indicators is abnormal, it is determined that the cell is abnormal. When at least two of the indicators are determined to be abnormal by the segmentation mask, it is directly determined that the cell is abnormal.

[0071] S8: For cells with abnormalities, perform cell culture;

[0072] For cells detected to be abnormal, locate the specific patient through patient case information, re-sample the patient, and place the re-sampled cells in a cell culture device for cell culture.

[0073] S9: Real-time acquisition of cell images in the cell culture device in S8 and preprocessing of the cell images; cell image preprocessing is the same as S1, and will not be described in detail here.

[0074] S10: Input the preprocessed cell images in S9 into the FastSAM model to generate segmentation masks for the images, which are used to analyze cell morphology indicators, cell number and density indicators, cell movement and migration indicators, cell internal structure indicators, and cell population feature indicators.

[0075] S11: Based on cell morphology indicators, cell number and density indicators, cell movement and migration indicators, cell internal structure indicators, and cell population feature indicators, conduct cell research, disease treatment, and provide more accurate treatment plans for patients.

[0076] The cell morphology indicators include cell size and cell area, cell shape, and cell perimeter. Specifically, the cell size and cell area can be calculated after the single cells or cell groups are accurately segmented by the FastSAM model. In tumor research, the size of cancer cells is usually larger than that of normal cells and varies. Real-time monitoring of the cell size change helps to determine the growth state and proliferation of cells, and the inhibition or promotion of cell growth by drugs. The cell shape can be analyzed by parameters such as circularity and aspect ratio. Normal cells usually have regular shapes, while cells that have undergone pathological changes, such as cancer cells, may have irregular shapes. By monitoring the dynamic changes of the cell shape, early pathological characteristics of cells can be found, and the effect of drugs on cell morphology can be evaluated. The cell perimeter can reflect the edge characteristics and membrane state of cells. When cells are stimulated by external factors or undergo pathological changes, the cell perimeter may change. For example, during the process of apoptosis, cells gradually shrink, and the perimeter also decreases accordingly.

[0077] The calculation method of the cell size is as follows: after the cell region is segmented by FastSAM, the number of pixels in the region is counted, and the pixel number is converted into the actual area size in combination with the pixel resolution of the image. The calculation method of the cell area is as follows: the segmented cell region is represented by a binary image, in which the foreground pixel value is 1 and the background pixel value is 0. The total number of pixels with a value of 1 in the binary image is calculated and denoted as N. If the actual physical size corresponding to each pixel is s (square microns), then the cell area A is N x s.

[0078] The cell shape includes circularity and aspect ratio. The calculation formula of the circularity is as follows: wherein A is the cell area, and P is the cell perimeter, which is detected by an edge detection algorithm and then counted to obtain the number of edge pixels. The Canny edge detection algorithm can be used to process the segmented cell image to obtain the edge image of the cell, and then the number of pixels with a value of 255 (indicating edge pixels) is counted to obtain the cell perimeter P. The circularity measures the closeness of the cell shape to a circle. The closer the circularity is to 1, the closer the cell shape is to a circle. The calculation method of the aspect ratio is as follows: the lengths of the long axis and the short axis of the cell are determined by the minimum circumscribed rectangle method, and are denoted as L and W, respectively. Then the aspect ratio is The aspect ratio describes the size ratio of the cell in the long axis and the short axis directions.

[0079] The cell quantity and density indicators include cell count and cell density. Specifically, regarding the cell count, the FastSAM model can accurately identify and segment each cell in the image, thereby achieving accurate statistics of the cell quantity, real-time monitoring of the change in the cell quantity during cell culture, understanding of the cell proliferation speed and growth curve, and provision of a basis for adjusting the culture conditions. In drug screening experiments, the cell quantity of different drug treatment groups is compared to evaluate the cytotoxicity or proliferation-promoting effect of the drugs. Regarding the cell density, the cell density can be calculated in combination with the cell quantity and the area of the culture region. The cell density has an important influence on the growth and function of cells. In tissue engineering, a suitable cell density helps the adhesion, proliferation, and differentiation of cells to form a functional tissue. Real-time monitoring of the cell density can timely adjust the culture strategy and optimize the cell culture effect.

[0080] The cell movement and migration indicators include cell displacement and cell migration speed. Regarding the cell displacement, the FastSAM model is used to track the same cell by continuously collecting cell images, so as to measure the position change of the cell at different time points and calculate the displacement of the cell. The migration ability of cells plays a key role in physiological and pathological processes such as tumor metastasis and wound healing. Monitoring the cell displacement can study the migration behavior and mechanism of cells and screen drugs that inhibit or promote cell migration. Regarding the cell migration speed, the migration speed of the cell can be calculated according to the cell displacement and the time interval. Different types of cells have different migration speeds. In a disease state, the migration speed of cells may change. For example, the migration speed of cancer cells is usually faster than that of normal cells. By monitoring the cell migration speed, the invasiveness and metastatic potential of tumors can be evaluated.

[0081] The internal structure of the cell includes the nucleus characteristics and the distribution of organelles. Regarding the nucleus characteristics, the FastSAM model can identify the nucleus and analyze its size, shape, position, and chromatin distribution. Abnormal changes in the nucleus are important markers of many diseases. For example, in tumor cells, the nucleus usually increases in size, has an irregular shape, and the distribution of chromatin changes. Real-time monitoring of the nucleus characteristics helps early diagnosis of diseases and evaluation of treatment effects. Regarding the distribution of organelles, the organelles in the cell, such as mitochondria and endoplasmic reticulum, can be segmented and analyzed to understand their quantity, size, morphology, and distribution. The functional state of the organelles is closely related to the physiological and pathological processes of cells. For example, mitochondria are the energy factories of cells. Changes in the quantity and morphology of mitochondria can reflect the energy metabolism state of cells.

[0082] The cell population characteristic indexes include cell aggregation and distribution, and cell clonogenicity. Specifically, as for cell aggregation and distribution, the aggregation degree and distribution pattern of cells in the culture environment are analyzed. Some cells will form aggregation under certain conditions, such as the clump growth of tumor cells. The study on the aggregation and distribution of cells helps to understand the interaction and signal transduction mechanism between cells, and the influence of the microenvironment on the behavior of cells. As for cell clonogenicity, the number and size of cell clones formed in the culture process are observed and analyzed to evaluate the clonogenicity of cells. The clonogenicity is an important indicator of the self-renewal and proliferation ability of cells, and is of great significance in stem cell research and tumor stem cell research.

[0083] The embodiment can improve the accuracy of cell detection by analyzing cells in combination with the information of the patient. After analyzing the abnormality, the cells in culture can be analyzed in real time, so that the doctor can draw a better treatment plan and improve the cure rate of the disease. Embodiment Two

[0084] In the embodiment, a three-dimensional model of the cell is generated by three-dimensional reconstruction technology using a segmentation mask, the generated three-dimensional model of the cell is imported into a VR / AR environment, and a virtual cell observation scene is created. The segmentation mask includes a cell image processed after S1 acquisition, and also includes a cell image processed after real-time detection of the cells in culture by S9.

[0085] An interactive interface is designed so that the user can interact with the cell model through gestures, voice, etc., such as rotating, scaling, and cutting the cell to observe the internal structure of the cell. This combination can enhance the understanding of the structure and function of the cell by researchers and doctors, and also provides a new teaching method for medical education, improving the learning interest and effect of students. Embodiment Three

[0086] The embodiment can adjust the operating parameters of the cell culture equipment according to the real-time acquisition of the cell image.

[0087] Specifically: the camera or microscope collects cell images of cells cultured in the cell culture device. The microscope uses phase contrast microscopy, live cell imaging system, etc. without sectioning treatment. The sensor device collects environmental parameters in the cell culture device in real time, including but not limited to temperature, humidity, pH value, etc. The collected cell images and environmental parameters are transmitted to the cloud server through the Internet of Things technology. FastSAM is used for cell segmentation and analysis of images. According to the analysis results of FastSAM combined with real-time monitoring of environmental parameters, the operating parameters of the cell culture device are automatically adjusted, such as adjusting the temperature, adding nutrients, etc. to optimize the cell culture conditions. This scheme can improve the automation level and quality control of cell culture, reduce manual intervention, reduce the cost of culture, and at the same time provide more reliable cell sources for the fields of cell therapy and biopharmaceuticals.

[0088] The specific principle is as follows:

[0089] Cell state reflects environmental adaptability: FastSAM analysis of cell images can obtain cell growth state, morphological characteristics, viability, etc. such as cell density, morphological changes, whether apoptosis or abnormality, etc. It can intuitively reflect the adaptability of the cell to the current culture environment. If the cell grows slowly, the morphology is abnormal, or a large number of apoptosis occurs, it may mean that the environmental parameters such as temperature, nutrient concentration, pH value, etc. are not suitable.

[0090] Environmental parameters affect cell state: temperature, nutrients, pH, gas environment, etc. Environmental parameters directly affect cell growth, metabolism and function. Different types of cells have specific requirements for these environmental parameters. Suitable environmental parameters can promote the normal growth and proliferation of cells, while unsuitable parameters can lead to cell growth inhibition or even death. For example, too high or too low temperature can affect the activity of enzymes in cells, thereby affecting the metabolic process of cells. Lack or excess of nutrients can affect the growth rate and physiological function of cells.

[0091] Analysis results guide environmental parameter adjustment: based on the analysis results of FastSAM, it can be judged whether the cell culture environment is suitable, and the environmental parameters are adjusted accordingly. If the analysis result shows that the cell grows slowly, it may be necessary to increase the supply of nutrients or adjust the temperature and other parameters to meet the needs of cell growth. By continuously monitoring the cell state (FastSAM analysis result) and adjusting the environmental parameters accordingly, a feedback regulation mechanism is formed, so that the cells are always in the most suitable culture conditions, thereby realizing the optimization of cell culture.

[0092] The specific control logic is:

[0093] Control strategy: According to the analysis results of FastSAM and environmental parameter monitoring data, an intelligent control strategy is formulated. If the analysis results show that the cell growth is slow, the temperature may need to be appropriately increased, and the amount of nutrient addition may need to be increased; if the pH value deviates from the normal range, the pH value of the culture solution needs to be adjusted.

[0094] Temperature control algorithm: PID (Proportional-Integral-Derivative) control algorithm is used to accurately control the temperature of the culture equipment; according to the real-time monitoring temperature data and the set target temperature, the temperature deviation is calculated, and the power of the heating element or the cooling element is adjusted through the PID algorithm, so that the temperature quickly and stably reaches the target value.

[0095] Heating and cooling system: The culture equipment is equipped with heating and cooling system, the heating system can use electric heating wire or heating plate, the cooling system can use compressor cooling or semiconductor cooling; according to the demand of temperature adjustment, the opening and closing of heating and cooling system is automatically controlled.

[0096] Nutrient addition: The culture equipment is equipped with nutrient solution storage tank and delivery pump, the nutrient solution storage tank stores various nutrients required for cell culture; the delivery pump can deliver nutrient solution to the culture chamber according to the control signal; according to the growth state of cells and analysis results, nutrient addition strategy is formulated, for example: if the number of cells increases rapidly, the addition frequency and amount of nutrients need to be increased; if the cell growth is slow, the amount of nutrient addition can be appropriately reduced.

[0097] pH value adjustment: The culture equipment is equipped with acid-base adjusting liquid storage tank and delivery pump, the storage tank stores acid and alkaline adjusting liquid respectively, the delivery pump can deliver adjusting liquid to the culture solution according to the control signal to adjust the pH value of the culture solution; according to the real-time monitoring pH value data and the set pH value range, the amount of acid-base adjusting liquid needed to be added is automatically calculated, and the pH value of the culture solution is accurately adjusted by controlling the running time and flow rate of the delivery pump.

[0098] Preferably, in order to facilitate system monitoring and maintenance, remote monitoring is set up, specifically, a monitoring platform based on Web or mobile application is developed, allowing users to remotely monitor the running state of the cell culture equipment, environmental parameters and cell images in real time; users can view data reports and analysis charts through the monitoring platform to timely understand the cell culture situation; more preferably, a fault diagnosis model is established on the cloud server to analyze the running data of the culture equipment in real time, and when abnormal conditions are detected in the running data, warning information is sent in time to notify maintenance personnel to handle; abnormal conditions include temperature exceeding upper or lower threshold, humidity exceeding upper or lower threshold, etc. Example Four

[0099] According to the uneven distribution of cells and the reasons, the seeding method is adjusted or the mixing method is improved; the above includes the cell image collected in S1 and the cell image in S9 culture;

[0100] Specifically: according to the segmented cell mask, the center position and area of each cell are calculated, the cell density information is obtained by counting the number of cells in a unit area, the image can be divided into multiple small areas, and the cell density of each area is calculated respectively to generate a cell density map; after calculating the cell density of each small area, the density difference between adjacent areas can be further calculated, and the standard deviation can be used to measure the dispersion degree of the cell density of each area in the whole image, the larger the standard deviation, the more obvious the difference in cell density; the uneven distribution of cells and the reasons are analyzed according to the center position and density information of the cells; the seeding method is adjusted or the mixing method is improved according to the uneven distribution of cells and the reasons.

[0101] By analyzing the distribution of cell positions, it is determined whether there is obvious directionality, such as calculating the principal component direction of the cell position, if the principal component direction is consistent with the operation direction during seeding, such as the spraying direction, it may be related to the seeding method; if it is dropwise seeding, check the cell distribution near the dropwise position, if the cells around the dropwise position are obviously gathered, and the cells in other areas are sparse, then the seeding method may be the cause of the uneven distribution. If it is caused by the problem of the seeding method, the seeding method can be adjusted, such as using uniform dropwise, spraying seeding and other methods.

[0102] If the cell suspension is not mixed uniformly, the mixing method can be improved, such as using a shaker or ultrasonic treatment, to ensure that the cells are uniformly distributed in the suspension.

[0103] Specifically, the calculation method of the cell center position is as follows:

[0104] For each cell mask, the centroid is calculated as the center position of the cell. The cell mask is a binary image, where the cell area is the foreground with a value of 1, and the rest is the background with a value of 0. The coordinates (x c , y c ) of the centroid are calculated by the following formula:

[0105] ; Wherein, M ( i , j ) is the pixel value of the cell mask image at the position of (x i , y j ), i and j represent the row and column indices of the image, respectively. The centroid coordinates of the cell are obtained by weighted average of the coordinates of all foreground pixels.

[0106] The calculation method of the cell area is to directly count the number of foreground pixels in the cell mask, that is, the area of the cell; because each pixel represents a basic unit in the image, the number of foreground pixels corresponds to the size of the area occupied by the cell; in the binary cell mask image, only the number of pixels with a value of 1 needs to be counted. Example Five

[0107] In this embodiment, according to the position and density information of the cells, a visualization tool is used to generate a visualization map of the cell distribution, such as the Matplotlib library of Python; on the map, different colors or markers can be used to represent the density of the cells, and the distribution of the cells can be intuitively displayed.

[0108] The specific steps of generating the visualization map of the cell distribution are as follows:

[0109] Divide the entire image into grids of equal size; determine the size of the grid according to the size of the image and actual requirements; for example, divide the image into n x m grids;

[0110] For each grid, count the number of cells contained therein;

[0111] Determine whether the cell belongs to the grid by judging whether the center position of the cell falls within the grid, and then calculate the ratio of the number of cells in each grid to the area of the grid to obtain the cell density of the grid area;

[0112] According to the cell density of each grid, use a visualization method to draw a cell density map; the visualization method uses color mapping or other methods, such as using different colors to represent different density ranges, with darker colors representing higher cell density and lighter colors representing lower cell density; commonly used color mapping schemes include heat maps such as jet and hot, and related drawing libraries such as Matplotlib can be used to implement the drawing of the density map. Example Six

[0113] In this embodiment, medical image segmentation evaluation indicators such as the Dice coefficient, the Jaccard coefficient, and the Hausdorff distance are used to compare the differences between the segmentation results of the FastSAM model before and after fusing multi-modal information and the true labels, and to evaluate the experimental results of the model.

[0114] Furthermore, according to the experimental results of the evaluation model, the model is optimized and adjusted; the fusion strategy, model structure, and hyperparameters are optimized and adjusted; if the fusion effect is not good, other fusion strategies are tried, if the model appears overfitting, the regularization parameter is adjusted or the intensity of data enhancement is increased, the model is continuously optimized to improve the accuracy and reliability of the segmentation. Example Seven

[0115] This embodiment provides a specific application scenario:

[0116] Cell image acquisition: Collect cell images from breast tissue samples, which can be derived from breast needle biopsy, surgical resection of breast tumor tissue, etc. Use high-resolution microscopes such as confocal laser scanning microscopes or phase contrast microscopes to image cells at different magnifications (e.g. 100x, 200x, 400x) to capture the overall morphology and internal details of cells. Image formats can be common JPEG, PNG, etc., with a resolution of at least 1440x960 pixels to ensure clear presentation of cell features.

[0117] Case data: Collect basic information of patients, including age, gender, menstrual history (age of menarche, menopausal age, etc.), reproductive history (whether to give birth, number of births, age of first birth, etc.), family history (whether there are breast cancer patients in the family), etc.

[0118] Clinical examination results: 1. Blood test indicators, such as the levels of tumor markers such as carcinoembryonic antigen (CEA) and carbohydrate antigen 15-3 (CA15-3). Under normal circumstances, the reference value of CA15-3 is generally less than 25 U / ml, and if it is higher than this value, it may indicate an increased risk of breast cancer; 2. Imaging examination results: breast ultrasound reports may contain information such as the size, shape, boundary, and internal echo of breast nodules, breast molybdenum target reports may describe the density of breast tissue and whether there are calcification points, and breast magnetic resonance imaging (MRI) reports can more detailedly show the structure of breast tissue and the characteristics of lesions.

[0119] S1: Collect cell images and preprocess the collected cell images

[0120] Prepare the breast tissue sample into a thin slice and place it on the microscope stage. Adjust the microscope parameters to obtain clear cell images. During the acquisition process, multiple shots of cells in different fields of view are required to ensure coverage of various types of cells. Preprocess the collected cell images, including the following operations:

[0121] Image enhancement: Use histogram equalization algorithm to enhance the contrast of the image; for a cell image with low contrast, this algorithm can make the outline and internal structure of the cell clearer, facilitating subsequent feature extraction;

[0122] De-noising: Use Gaussian filter algorithm to remove noise in the image; for random noise in the image due to the microscope optical system or environmental factors, Gaussian filter can smooth the image and reduce the interference of noise on cell features;

[0123] Normalization: Normalize the pixel values of the image to the interval [0, 1]; map the range of pixel values in the original image from 0-255 to 0-1, so that there is comparability between different images.

[0124] S2: Structuring and encoding processing of multi-modal data

[0125] Structuring: Organize case data and clinical examination results into table form, unify data format; list patient's age, menstrual history, tumor marker level, etc. information in different columns for subsequent analysis.

[0126] Encoding: For categorical data such as family history, use one-hot encoding to convert it to a numerical vector, with family history encoded as [1, 0] and no family history encoded as [0, 1]; for numerical data, age and tumor marker level, perform standardization, such as age mean 50 years old, standard deviation 10 years old, for a 60-year-old patient, the standardized age value is (60-50) / 10=1.

[0127] S3: Feature extraction of cell image and multi-modal data

[0128] Use a pre-trained ResNet-50 convolutional neural network to extract deep features of the image, extract cell shape features from the cell image, including circularity, aspect ratio, and texture features, including roughness, contrast, etc.; construct a fully connected neural network, input structured and encoded multi-modal data, extract features related to breast cancer risk from case data and clinical examination results.

[0129] S4: Fusion of cell image and multi-modal data

[0130] Concatenate the extracted cell image features and multi-modal data features at the feature level to form a fused feature vector; in this embodiment, the length of the cell image feature vector is 512, the length of the multi-modal data feature vector is 256, and the length of the fused feature vector is 768.

[0131] S5: Input the fused feature vector into the segmentation head of FastSAM, the model will predict the probability of each pixel belonging to different categories based on the fused features;

[0132] FastSAM model based on fused features, through convolutional layers and activation functions, predicts the probability of each pixel in the image belonging to cancer cells, normal breast cells or background; for example: for a pixel in the image, the model predicts that it belongs to cancer cells with a probability of 0.8, to normal breast cells with a probability of 0.1, and to background with a probability of 0.1.

[0133] S6: According to the predicted pixel class probability, the pixels are classified into different categories to generate the segmentation mask of the image;

[0134] Using threshold method, pixels with probability greater than a set threshold are classified into the corresponding category to generate the segmentation mask of cancer cells, normal breast cells and background; The threshold in this embodiment is set to 0.5, that is, for a certain pixel, if the probability of belonging to cancer cells is greater than 0.5, the pixel is marked as a cancer cell; For the breast cancer cell image described above, after threshold processing, the cancer cell region, normal cell region and background region are accurately segmented, and the segmentation mask clearly outlines the outline of the cancer cells.

[0135] S7: Based on the segmentation mask, it is judged whether the cells are abnormal or not;

[0136] By analyzing the proportion and morphology of cancer cells in the segmentation mask, it is judged whether the cells are abnormal or not; In this embodiment, if the proportion of cancer cells exceeds 10% or the morphology of cancer cells shows obvious atypia (such as enlarged nucleus, increased nucleolus, etc.), it is judged as abnormal.

[0137] S8: For cells with abnormalities, cell culture is performed;

[0138] When the cell abnormalities are detected, the cells are resampled and placed in the cell culture device, and the culture solution containing appropriate nutrients and growth factors is added, and the culture is carried out in a 37℃, 5% CO2 incubator. Preferably, the culture solution needs to be replaced regularly during the culture process to maintain the growth environment of the cells.

[0139] S9: Real-time acquisition of cell images in the cell culture device and pre-processing of the cell images;

[0140] Using the camera installed on the cell culture device, the cell images are acquired every 1 hour; The acquired images are pre-processed by the same method as step S1, including image enhancement, denoising and normalization; The quality of the processed images is significantly improved, and the growth state of the cells can be clearly observed.

[0141] S10: The pre-processed cell images are input into the FastSAM model for analysis of cell morphological indicators, cell number and density indicators, cell movement and migration indicators, cell internal structure indicators, and cell population characteristic indicators.

[0142] The cell morphology index includes the size, shape, aspect ratio, etc. of the cell. The change of the cell morphology is determined by measuring the diameter and the ratio of the long axis to the short axis. The cancer cell will show abnormal shape such as polygonal and long spindle shape, and the ratio of the long axis to the short axis will deviate from the normal range. The doctor can preliminarily determine the nature of the breast lesion by observing the cell morphology index combined with other clinical examination results. During the treatment, the change of the morphology index can also be used as a reference for evaluating the treatment effect. For example, if the cell morphology gradually returns to normal after a period of treatment, it indicates that the treatment may be effective. On the contrary, if the cell morphology becomes more abnormal, the treatment plan may need to be adjusted. Studies have shown that the average area of breast cancer cells is usually 80-150 μm², the average perimeter is about 30-60 μm, and the aspect ratio is between 1-2.

[0143] The cell number and density index is to count the number of cells and calculate the density of cells. In a certain area of the culture device, the number of cells is counted every certain period of time to observe the proliferation of cells. Breast cancer cells have the characteristics of strong proliferation, so the number of cells will increase rapidly during the culture process, and the cell density will also increase accordingly. By regularly counting the number of cells and calculating the density, it can be found that the proliferation rate of breast cancer cells is significantly faster than that of normal breast cells. The change of the number and density of cells can reflect the growth of the tumor. In the diagnosis aspect, if the number of cells increases rapidly and the density abnormally increases, it is highly suggestive of a malignant tumor. In the treatment monitoring, this index can be used to evaluate the inhibitory effect of anticancer drugs on the proliferation of cancer cells. For example, after using chemotherapy drugs, if the growth of the number of cells slows down and the density no longer continues to rise, it indicates that the drugs have a certain inhibitory effect on the proliferation of cancer cells, which can provide a basis for the adjustment of the subsequent treatment plan. Under in vitro culture conditions, the proliferation rate of breast cancer cells is relatively fast. On the 3rd-5th day of culture, the number of cells can increase from about 1000 to 5000-10000. In terms of cell density, when the area of the culture dish is 25 cm², the density of breast cancer cells can reach 10-20 / μm² in the later culture period.

[0144] Cell movement and migration indicators: Calculate the movement speed and migration distance of cells through continuously collected images; track the position changes of cells using image analysis software to calculate the moving distance and speed of cells within a certain time; breast cancer cells have strong movement and migration ability, which is an important basis for their invasion and metastasis; by tracking cells using image analysis software, it can be found that breast cancer cells have a longer moving distance and faster movement speed within a certain time, while normal breast cells have relatively weak movement and migration ability; cell movement and migration indicators are important for judging the malignancy and prognosis of breast cancer; cancer cells with strong movement and migration ability are more likely to metastasize, suggesting that the patient's condition may be more serious and the prognosis is relatively poor; in terms of treatment, targeted drugs developed for cell movement and migration-related signaling pathways can reduce the risk of metastasis by inhibiting the movement and migration of cancer cells; by monitoring changes in cell movement and migration indicators, the efficacy of these targeted drugs can be evaluated to support personalized treatment; the average movement speed of breast cancer cells is between 3-10 μm / h; in the in vitro scratch test, the migration distance of breast cancer cells within 24-48 hours can reach 20-50 μm.

[0145] Cell internal structure indicators: Observe the distribution and morphology of organelles within cells; observe the morphology and distribution of mitochondria, endoplasmic reticulum, and other organelles through fluorescence labeling technology; the morphology and distribution of organelles in breast cancer cells will differ from those in normal breast cells; the chromatin in the nucleus usually appears loose, with reduced heterochromatin and increased euchromatin; through image analysis, it can be found that the average gray value of chromatin decreases significantly (P<0.05) compared to normal breast cells; the number of mitochondria in breast cancer cells increases significantly, which can increase by 1-2 times compared to normal breast epithelial cells; in addition, the mitochondria of cancer cells may appear swollen and have broken cristae, and the endoplasmic reticulum may also appear expanded and arranged in disorder; these changes reflect the abnormal metabolism and function of cancer cells; changes in cell internal structure can provide more accurate diagnostic information; in terms of treatment, corresponding treatment strategies can be developed based on the characteristics of organelle abnormalities in cancer cells; some drugs can specifically target the mitochondria of cancer cells to induce apoptosis, thereby achieving the purpose of treating breast cancer; by observing changes in cell internal structure indicators, the effects of these drugs can be evaluated to provide a reference for optimizing treatment plans.

[0146] Cell population characteristics: analyze the aggregation degree and distribution of the cell population; observe whether the cells form cell clusters and the size and distribution of the cell clusters; breast cancer cells may form cell clusters during culture, and the size and distribution of the cell clusters have certain characteristics. Compared with normal breast cells, the cell clusters formed by breast cancer cells are usually larger, tighter, and more unevenly distributed. The aggregation degree of the breast cancer cell population can reach 60%-80%, and the size of the cell clusters also varies, generally with a diameter of 50-200 μm. Smaller cell clusters may consist of a few cells, while larger cell clusters may contain dozens or even hundreds of cells. The formation and characteristics of cell clusters help to distinguish breast cancer cells from normal breast cells, providing additional basis for pathological diagnosis. In treatment research, understanding the characteristics of the cell population is of great significance for developing treatment methods targeting the behavior of cancer cell populations. Research has found that there are some special microenvironments inside cancer cell clusters that may affect the drug resistance of cancer cells. Based on this characteristic, drugs can be designed to destroy the structure of cancer cell clusters, improving the killing effect of drugs on cancer cells, thereby providing new ideas and methods for the treatment of breast cancer.

[0147] The above indicators can reflect the biological characteristics of breast cancer cells from multiple angles, providing important basis and reference for early diagnosis, disease assessment, treatment plan selection, and efficacy monitoring of breast cancer, and helping to achieve precise treatment of breast cancer and improve the cure rate and survival rate of patients.

[0148] Through the above steps, it can be determined whether there are cancer cells in the breast tissue sample, as well as the proportion and distribution of cancer cells. For example, if the detection result shows that the proportion of cancer cells in the sample is 15%, mainly distributed in the central region of the image.

[0149] Through the above cancer cell detection implementation process based on specific data, we can comprehensively and accurately detect cancer cells and conduct in-depth analysis of various characteristics of cancer cells during culture, providing rich and reliable data support for cancer research and treatment, fully embodying the advantages of the detection method in multi-modal data fusion, real-time monitoring, and high-precision analysis.

[0150] This application combines cell images and case data, clinical examination results, and other multi-modal data to more comprehensively understand the characteristics of cells and improve the accuracy of detection. Real-time image acquisition and analysis during cell culture can timely detect changes in cells and provide dynamic information for disease research and treatment. The use of FastSAM model for pixel-level segmentation can accurately identify abnormal and normal cells, providing an accurate basis for subsequent analysis.

[0151] It should be noted that, in the present document, the terms "comprises / comprising" or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a" does not, without further constraints, exclude the presence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0152] From the above description of the embodiments, it can be clear to those skilled in the art that the above-mentioned example methods can be realized by means of software plus a necessary general hardware platform, and of course, can also be realized by hardware, but in many cases, the former is a better embodiment; based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which is stored in a storage medium (such as a ROM / RAM, a magnetic disk, an optical disk), and includes a plurality of instructions for causing a terminal (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present application.

[0153] The embodiments of the present application are described above in combination with the accompanying drawings, but the present application is not limited to the above-described specific embodiments, and the above-described specific embodiments are merely illustrative, rather than limiting, and those of ordinary skill in the art can make many forms under the inspiration of the present application without departing from the scope of the present application and the scope protected by the claims.

Claims

1. A medical cell detection method, characterized in that, Includes the following steps: S1: Acquire cell images and preprocess the collected cell images; S2: Perform structured and encoded processing on multimodal data; where multimodal data includes case reports and clinical examination results; S3: Feature extraction from cell images and multimodal data; S4: Fusion of cell images and multimodal data; S5: Input the fused feature vector into the segmentation head of FastSAM, and the model predicts the probability of each pixel belonging to different categories based on the fused features; S6: Based on the predicted pixel category probabilities, classify pixels into normal cell regions, abnormal cell regions, and background categories, and generate a segmentation mask for the image. S7: Determine whether there are abnormalities in cells based on the collected case data and clinical examination results combined with the segmentation mask; S8: For cells with abnormalities, resample and culture them. S9: Real-time acquisition of cell images in cell culture equipment and preprocessing of cell images; S10: The preprocessed cell images from S9 are input into the FastSAM model, and cell morphology indicators, cell number and density indicators, cell movement and migration indicators, cell internal structure indicators, and cell population characteristic indicators are analyzed based on the generated segmentation mask.

2. The medical cell detection method according to claim 1, characterized in that, By using segmentation masks and 3D reconstruction technology, a 3D model of a cell is generated. The generated 3D cell model is then imported into a VR / AR environment to create a virtual cell observation scene.

3. The medical cell detection method according to claim 1, characterized in that, Cell images acquired by the S9 are transmitted to the cloud server. Sensor devices collect environmental parameters in the cell culture equipment in real time and transmit these parameters to the cloud server as well. FastSAM performs cell segmentation and analysis on the images and automatically adjusts the operating parameters of the cell culture equipment based on the analysis results of FastSAM and the real-time monitored environmental parameters.

4. The medical cell detection method according to claim 1, characterized in that, Based on the cell mask obtained from the segmentation, the center position and area of ​​each cell are calculated. By counting the number of cells per unit area, the cell density information is obtained. Based on the cell center position and density information, it is analyzed whether there is uneven cell distribution. If so, the region and cause of uneven cell distribution are analyzed.

5. The medical cell detection method according to claim 4, characterized in that, Adjust the inoculation method or improve the mixing method according to the area and cause of uneven cell distribution.

6. The medical cell detection method according to claim 4, characterized in that, The method for calculating the cell center position is as follows: For each cell mask, its centroid is calculated as the cell center position. The cell mask is defined as a binary image, where the cell region is the foreground with a value of 1, and the rest is the background with a value of 0. The coordinates of the centroid (x, y, y) are... c y c ), calculated using the following formula: ; Where M(i,j) is the pixel value of the cell mask image at position (i,j), and i and j represent the row and column indices of the image, respectively. The centroid coordinates of the cell are obtained by weighted averaging of the coordinates of all foreground pixels.

7. The medical cell detection method according to claim 4, characterized in that, The area of ​​the cell is calculated by directly counting the number of foreground pixels in the cell mask.

8. The medical cell detection method according to claim 4, characterized in that, Based on cell location and density information, a visualization map of cell distribution is generated using visualization tools.

9. A medical cell detection method according to claim 8, characterized in that, The specific steps for generating a visual map of cell distribution are as follows: Divide the entire image into a grid of equal size; For each grid, count the number of cells it contains; A cell is determined to belong to a grid by determining whether its center falls within the grid. Then, the ratio of the number of cells in each grid to the grid area is calculated to obtain the cell density of that grid region. Based on the cell density of each grid, a cell density map is drawn using visualization methods.

10. A medical cell detection method according to claim 3, characterized in that, In S9, a camera or microscope acquires cell images of cells cultured in the cell culture device, and a sensor device acquires environmental parameters in the cell culture device in real time. The acquired cell images and environmental parameters are transmitted to a cloud server, where FastSAM is used to segment and analyze the images. Based on the FastSAM analysis results and the real-time monitored environmental parameters, the operating parameters of the cell culture device are automatically adjusted.

Images