A medical image analysis method and system based on plain scan CT data

Abstract

The present invention discloses a medical image analysis method and system based on plain scan CT data, and relates to the technical field of medical image analysis. The medical image analysis method based on plain scan CT data obtains chest plain scan CT data of a patient, wherein the chest plain scan CT data includes chest cross-sectional image data of several scanning positions; a three-dimensional model of the patient's chest is established based on the chest plain scan CT data of the patient; the three-dimensional model of the patient's chest is segmented to obtain a three-dimensional model of the patient's lungs. The present invention automatically reconstructs a three-dimensional model of healthy lungs by using a deep learning model, and compares it with the patient's actual lung model, thereby providing a more accurate disease diagnosis. This automated comparison reduces human errors and improves the efficiency of analysis, so that doctors can quickly and accurately identify the lesion area and the degree of the disease, thereby formulating a more effective treatment plan.

Description

Technical Field

[0001] The present invention relates to the technical field of image analysis, and in particular to a medical image analysis method and system based on plain scan CT data. Background Art

[0002] Medical imaging technology is an indispensable part of the modern medical system. It provides necessary visual information for clinical diagnosis and treatment. In particular, computed tomography (CT) technology has become an important tool for evaluating various diseases, especially lung diseases, because it can provide high-resolution internal images of the body. However, traditional CT image analysis mostly relies on the experience of radiologists for visual interpretation, which is not only time-consuming, but also easily limited by the experience differences of individual doctors, which may lead to diagnostic inconsistency and misdiagnosis.

[0003] With the development of digital image processing technology, especially the rise of deep learning technology, computer-aided diagnosis systems (CAD) have begun to show great potential in the field of medical imaging. These systems can assist doctors in identifying pathological features faster and more accurately by automatically processing and analyzing medical image data, thereby improving diagnostic efficiency and accuracy. However, existing systems still face challenges such as processing complex data, improving system generalization capabilities, and seamlessly integrating with clinical processes.

[0004] This is especially true for the diagnosis of lung diseases. The structure of the lungs is complex and the contrast with surrounding tissues is low, which makes it particularly difficult to accurately identify and quantify lung lesions from traditional two-dimensional CT images. In addition, two-dimensional images cannot effectively represent the three-dimensional morphology of the lung structure, which limits the comprehensive assessment of the disease. Therefore, developing a system that can automatically reconstruct a three-dimensional lung model based on plain scan CT data and perform high-precision analysis is the key to improving the efficiency of lung disease diagnosis and treatment.

[0005] The limitations of existing technologies include at least the following problems. First, traditional technologies usually rely on two-dimensional images or simple three-dimensional reconstructions, and fail to fully utilize the data richness provided by modern medical imaging technology, which easily leads to difficulty in fully understanding the spatial distribution of the disease and the detailed changes in tissues during the analysis process, thereby affecting the accuracy of diagnosis and the personalization of treatment. Second, existing analysis methods often rely on the experience of doctors to interpret images, and lack tools that can systematically and automatically extract image features. This not only increases the workload of doctors, but also easily leads to differences in diagnostic results due to different subjective judgments. In addition, traditional methods often lack clear quantitative standards when evaluating lung health, which makes disease monitoring and effect evaluation across time points lack precision, making it difficult to systematically track disease progression or evaluate treatment effects. Summary of the invention

[0006] In view of the deficiencies in the prior art, the present invention provides a medical image analysis method and system based on plain scan CT data, which solves the problem that traditional technologies usually rely on two-dimensional images or simple three-dimensional reconstructions, and fail to fully utilize the data richness provided by modern medical imaging technology, which easily leads to difficulty in fully understanding the spatial distribution of the disease and the detailed changes in tissues during the analysis process, thereby affecting the accuracy of diagnosis and the personalization of treatment. Secondly, existing analysis methods often rely on the experience of doctors to interpret images, and lack tools that can systematically and automatically extract image features, which not only increases the workload of doctors, but also easily leads to differences in diagnostic results due to different subjective judgments. In addition, traditional methods often lack clear quantitative standards when evaluating lung health, which makes it difficult to systematically track the course of the disease or evaluate the effectiveness of treatment in terms of precision.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: a medical image analysis method based on plain scan CT data, comprising the following steps: obtaining a patient's chest plain scan CT data, wherein the chest plain scan CT data includes chest cross-sectional image data at several scanning positions; establishing a three-dimensional model of the patient's chest based on the patient's chest plain scan CT data; segmenting the patient's chest three-dimensional model to obtain a three-dimensional model of the patient's lungs; inputting the patient's lung three-dimensional model into a pre-trained deep learning model for restoration analysis to obtain a three-dimensional model of the patient's lung health; analyzing the patient's lung health scoring index based on the patient's lung three-dimensional model and the patient's lung health three-dimensional model, and dividing the patient's lung health level according to preset scoring rules.

[0008] Furthermore, the chest cross-sectional image data includes pixel values ​​of a plurality of chest cross-sectional pixel points.

[0009] Furthermore, the specific steps of establishing a three-dimensional model of the patient's chest based on the patient's chest plain scan CT data are as follows: reading the chest cross-sectional image data of each scanning position, and performing scanning position sorting and stacking processing to obtain the patient's chest initial three-dimensional model; performing interpolation judgment processing on the patient's chest initial three-dimensional model to obtain the patient's chest three-dimensional model, wherein the patient's chest three-dimensional model includes voxel values ​​of several chest voxel points.

[0010] Furthermore, the specific steps for obtaining the three-dimensional model of the patient's lungs are as follows: based on a preset voxel threshold range, a preliminary screening and analysis of several chest voxel points in the patient's chest three-dimensional model is performed to obtain several initial lung voxel points; the LBP value of each initial lung voxel point is analyzed separately; based on the preset LBP threshold, each initial lung voxel point is screened and analyzed to obtain several lung voxel points, and connection processing is performed to obtain the patient's lung three-dimensional model.

[0011] Furthermore, the specific steps of analyzing the LBP value of each initial lung voxel point are as follows: for each initial lung voxel point, adjacent initial lung voxel points within a preset neighborhood are identified respectively; each initial lung voxel point is comprehensively analyzed with adjacent initial lung voxel points within a preset neighborhood to obtain the LBP value of each initial lung voxel point, and the calculation formula is as follows: ;in, For the The LBP value of the initial lung voxel point, is the symbolic function, For the The first The voxel values ​​of the adjacent initial lung voxel points, For the The voxel value of the initial lung voxel point, , is the number of initial lung voxel points, , is the number of adjacent initial lung voxel points within the preset neighborhood of the initial lung voxel point.

[0012] Furthermore, the pre-trained deep learning model is specifically a variational autoencoder, and the specific steps to obtain the patient's healthy lung three-dimensional model are as follows: loading the pre-trained variational autoencoder; inputting the patient's lung three-dimensional model into the encoder of the variational autoencoder for encoding processing; after the encoding processing is completed, the decoder in the variational autoencoder performs decoding and reconstruction processing to obtain the patient's healthy lung three-dimensional model, and the patient's healthy lung three-dimensional model includes healthy voxel values ​​of several chest healthy voxel points.

[0013] Furthermore, the specific steps of analyzing the patient's lung health score index based on the patient's lung three-dimensional model and the patient's lung health three-dimensional model are as follows: the voxel value of each chest voxel point in the patient's lung three-dimensional model is comprehensively analyzed with the healthy voxel value of each chest healthy voxel point in the patient's lung health three-dimensional model to obtain the patient's lung voxel intensity deviation score index and lung voxel variability score index; the patient's lung voxel intensity deviation score index and lung voxel variability score index are comprehensively analyzed to obtain the patient's lung health score index; wherein, the specific formula for calculating the patient's lung health score index is as follows: ;in, Scoring indicators for patients' lung health, is the patient's lung voxel intensity deviation score indicator, is the weighting coefficient of the patient's lung voxel intensity deviation score index, is the patient's lung voxel variability scoring index, is the weighted coefficient of the patient's lung voxel variability score, , is a natural constant.

[0014] Furthermore, the specific steps for obtaining the patient's lung voxel intensity deviation score index and lung voxel variability score index are as follows: read the voxel value of each chest voxel point in the patient's lung three-dimensional model, and perform difference mean analysis with the healthy voxel value of each chest healthy voxel point in the patient's lung three-dimensional model, to obtain the patient's lung voxel difference mean; perform standard deviation analysis on the voxel value of each chest voxel point in the patient's lung three-dimensional model and the healthy voxel value of each chest healthy voxel point in the patient's lung three-dimensional model, to obtain the patient's lung voxel difference standard deviation; comprehensively analyze the patient's lung voxel difference mean and lung voxel difference standard deviation to obtain the patient's lung voxel intensity deviation score index and lung voxel variability score index.

[0015] Furthermore, the specific formulas for calculating the patient's lung voxel difference mean, lung voxel difference standard deviation, lung voxel intensity deviation score index, and lung voxel variability score index are as follows: ;in, is the mean lung voxel difference of the patient, The first 3D model of the patient's lungs The voxel value of the chest voxel point, The first 3D model of the healthy lung of the patient corresponds to the 3D model of the patient's lung. The health voxel value of the chest health voxel point, The first 3D model of the patient's lungs The weighting coefficient of the voxel value of the chest voxel point, The first 3D model of the patient's lungs The voxel value of the chest voxel point, The first 3D model of the healthy lung of the patient corresponds to the 3D model of the patient's lung. The health voxel value of the chest health voxel point, The first 3D model of the patient's lungs The weighting coefficient of the voxel value of the chest voxel point, , is the standard deviation of the patient’s lung voxel difference, is the patient's lung voxel intensity deviation score indicator, is the patient's lung voxel variability scoring index, is the weight coefficient of the mean difference of the patient’s lung voxels, is the weight coefficient of the standard deviation of the patient’s lung voxel difference, , , The number of chest voxel points.

[0016] A medical image analysis system based on plain scan CT data comprises: a data acquisition module, a three-dimensional model building module, a segmentation module, a restoration module, and a health analysis module; the data acquisition module is used to acquire a patient's chest plain scan CT data, wherein the chest plain scan CT data comprises chest cross-sectional image data at a plurality of scanning positions; the three-dimensional model building module is used to build a patient's chest three-dimensional model based on the patient's chest plain scan CT data; the segmentation module is used to segment the patient's chest three-dimensional model to obtain a patient's lung three-dimensional model; the restoration module is used to input the patient's lung three-dimensional model into a pre-trained deep learning model for restoration analysis to obtain a patient's lung health three-dimensional model; the health analysis module is used to analyze the patient's lung health scoring index based on the patient's lung three-dimensional model and the patient's lung health three-dimensional model, and divide the patient's lung health level according to a preset scoring rule.

[0017] The present invention has the following beneficial effects:

[0018] (1) This medical image analysis method based on plain scan CT data automatically reconstructs a three-dimensional model of healthy lungs using a deep learning model and compares it with the patient's actual lung model, thereby providing a more accurate disease diagnosis. This automated comparison reduces human errors and improves analysis efficiency, allowing doctors to quickly and accurately identify the lesion area and the extent of the disease, thereby developing a more effective treatment plan.

[0019] (2) This medical image analysis method based on plain scan CT data can generate a personalized healthy lung model based on the patient's specific lung condition by utilizing deep learning technologies such as variational autoencoders. This model can not only be used for current health assessment, but also as a benchmark for evaluating treatment effects. As treatment progresses, through regular scanning and model comparison, doctors can monitor the treatment effect in real time and adjust the treatment plan.

[0020] (3) This medical image analysis method based on plain scan CT data converts complex medical image data into easy-to-understand three-dimensional models and intuitive health scores, enabling patients to better understand their health conditions, thereby helping to enhance patients' trust and satisfaction with the treatment process.

[0021] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a flow chart of a medical image analysis method based on plain scan CT data of the present invention.

[0023] Figure 2 The present invention is a flowchart of the specific steps of establishing a three-dimensional model of a patient's chest based on the patient's chest plain scan CT data in a medical image analysis method based on plain scan CT data of the present invention.

[0024] Figure 3 The block diagram of a medical image analysis system based on plain scan CT data of the present invention. DETAILED DESCRIPTION

[0025] The embodiment of the present application solves the problem that traditional technologies usually rely on two-dimensional images or simple three-dimensional reconstructions and fail to fully utilize the data richness provided by modern medical imaging technology, which easily leads to difficulty in fully understanding the spatial distribution of the disease and the detailed changes in tissues during the analysis process, thereby affecting the accuracy of diagnosis and the personalization of treatment. Secondly, existing analysis methods often rely on the experience of doctors to interpret images, and lack tools that can systematically and automatically extract image features, which not only increases the workload of doctors, but also easily leads to differences in diagnostic results due to different subjective judgments. In addition, traditional methods often lack clear quantitative standards when evaluating lung health, which makes disease monitoring and effect evaluation across time points lack precision, and it is difficult to systematically track the course of the disease or evaluate the effect of treatment.

[0026] The overall idea of ​​the problem in the embodiment of this application is as follows:

[0027] First, the cross-sectional image data of the patient's chest is collected using a CT scanner. Then, based on these two-dimensional image data, an initial three-dimensional model of the patient's chest is constructed by sorting the scan positions and stacking them. The initial three-dimensional model is segmented in detail to extract the three-dimensional structure of the lungs. Then, this lung model is input into a pre-trained deep learning model (such as a variational autoencoder) to generate a three-dimensional lung model that reflects the health status. Based on the patient's three-dimensional lung model and the health model, the voxel intensity deviation score and voxel variability score are calculated. These scores comprehensively reflect the health status of the patient's lungs. Based on these scores, the lung health status is divided into different health levels using preset rules.

[0028] See also Figure 1The embodiment of the present invention provides a technical solution: a medical image analysis method based on plain scan CT data, comprising the following steps: obtaining a patient's chest plain scan CT data, the chest plain scan CT data including chest cross-sectional image data at several scanning positions (the specific process is: using a CT scanning device to perform multi-angle scanning on the patient's chest according to a predetermined protocol to generate multiple cross-sectional images); establishing a three-dimensional model of the patient's chest based on the patient's chest plain scan CT data; segmenting the patient's chest three-dimensional model to obtain a three-dimensional model of the patient's lungs; inputting the patient's lung three-dimensional model into a pre-trained deep learning model for restoration analysis to obtain a healthy three-dimensional model of the patient's lungs (i.e., a three-dimensional model of the lungs in a healthy state); analyzing the patient's lung health scoring index based on the patient's lung three-dimensional model and the patient's lung health three-dimensional model, and dividing the patient's lung health level according to preset scoring rules.

[0029] The preset scoring rules include but are not limited to the following examples:

[0030] Patients with lung health scores between 90 and 100 are considered "excellent."

[0031] A patient's lung health score of 80-89 is considered "good".

[0032] Patients with lung health scores between 70 and 79 are considered "fair."

[0033] Patients with lung health scores between 60 and 69 are considered "poor."

[0034] A patient's lung health score of less than 60 is considered "unhealthy".

[0035] The chest cross-sectional image data includes pixel values ​​of a number of chest cross-sectional pixel points (this value reflects the X-ray absorption rate of the point, and these pixel values ​​are usually quantified and expressed in Hounsfield units (HU) to quantitatively describe the absorption intensity of X-rays by different tissues).

[0036] Specifically, Figure 2As shown, the specific steps of establishing a patient's chest three-dimensional model based on the patient's chest plain scan CT data are as follows: read the chest cross-sectional image data of each scanning position, and perform scanning position sorting and stacking processing (specifically, stacking in descending order according to the change of the Z value of the scanning position) to obtain the patient's initial three-dimensional chest model, and the scanning position in this embodiment is specifically: the head angle direction of the patient standing vertically on the ground is recorded as the Z axis, wherein, when the patient is undergoing a CT scan, the patient is lying on the CT scanning device to perform a head angle movement scan, that is, a Z-axis movement scan, and X and Y axes are perpendicular to the Z-axis direction and perpendicular to each other on the ground, and the X, Y, and Z axes are recorded. as the origin direction; performing interpolation judgment processing on the initial three-dimensional model of the patient's chest to obtain the three-dimensional model of the patient's chest, the three-dimensional model of the patient's chest includes voxel values ​​of several chest voxel points, wherein the interpolation judgment processing is specifically as follows: judging one by one whether the Z value difference of any adjacent chest cross-sectional image data in the initial three-dimensional model of the patient's chest is higher than a preset Z value difference threshold; if there are adjacent chest cross-sectional image data with a Z value difference higher than the preset Z value difference threshold, then establishing chest cross-sectional interpolation image data between the adjacent chest cross-sectional image data, and the chest cross-sectional interpolation image data includes interpolation pixel values ​​of several chest cross-sectional interpolation pixel points.

[0037] The specific formula for calculating the interpolation pixel value of each chest cross-section interpolation pixel point of the chest cross-section interpolation image data is: ;in, The first image in the chest cross-section interpolation image data The interpolated pixel values ​​of the interpolated pixels of the chest cross section, The first adjacent chest cross-sectional image data of the chest cross-sectional interpolation image data corresponds to the first The pixel value of the chest cross-section pixel, The first adjacent chest cross-sectional image data of the chest cross-sectional interpolation image data corresponds to the first The weight coefficient of the pixel value of the chest cross-section pixel, The first adjacent chest cross-sectional image data corresponding to the second adjacent chest cross-sectional image data of the chest cross-sectional interpolation image data is The pixel value of the chest cross-section pixel, The first adjacent chest cross-sectional image data corresponding to the second adjacent chest cross-sectional image data of the chest cross-sectional interpolation image data is The weight coefficient of the pixel value of the chest cross-section pixel, , , The number of chest cross-section interpolation pixels of the chest cross-section interpolation image data.

[0038] It needs to be explained that , The specific calculation steps are: the first and second adjacent chest cross-sectional image data of the chest cross-sectional interpolation image data corresponding to the The pixel values ​​of the chest cross-section pixels are summed and analyzed to obtain the first chest cross-section interpolation image data. The pixel sum value of the first and second adjacent chest cross-sectional image data of the chest cross-sectional interpolation image data is then calculated. The pixel values ​​of the first chest cross-sectional pixel points are respectively related to the first The pixel and value of each chest cross-section pixel point are analyzed for proportion, and the analysis result is the corresponding weight coefficient.

[0039] In this embodiment, interpolation processing is particularly important in medical imaging. It increases the continuity and smoothness of image data by creating intermediate image layers between adjacent image layers. This process not only improves the spatial resolution of the image, but also helps to more clearly identify and locate lesions, such as tumors or inflammation. High-quality image reconstruction directly affects the accuracy of diagnosis and helps doctors better assess the scope and severity of the disease. By performing meticulous interpolation processing on chest cross-sectional image data, a more accurate three-dimensional model can be obtained, which enables doctors to observe the patient's lung structure from multiple angles and dimensions, thereby performing more accurate lesion localization and lesion assessment. This detailed three-dimensional view provides important auxiliary information for surgical planning, radiotherapy planning, and other treatment methods, especially in cases involving In the case of complex anatomical structures, interpolation technology helps doctors avoid misdiagnosis or missed diagnosis due to discontinuous or incomplete image information by filling in the data gaps between scan layers. This is especially important for the identification of early diseases, such as early lung cancer, which may only show abnormalities in a small area in the early stages. More continuous and comprehensive image data can help doctors detect these lesions earlier, so that treatment can be started in time and the success rate of treatment can be improved. Accurate diagnosis and effective treatment plans directly affect the patient's treatment experience and ultimate health outcomes. By using advanced image processing technology, patients can obtain faster diagnostic results, more accurate disease assessments, and more personalized treatment plans, so that they can feel more confident throughout the treatment process and improve overall satisfaction and treatment compliance.

[0040] Specifically, the specific steps for obtaining the three-dimensional model of the patient's lungs are as follows: based on a preset voxel threshold range, a preliminary screening and analysis of several chest voxel points in the patient's chest three-dimensional model is performed (the chest voxel points with voxel values ​​outside the preset voxel threshold range are eliminated), and several initial lung voxel points are obtained; the LBP value of each initial lung voxel point is analyzed separately; based on a preset LBP threshold, each initial lung voxel point is screened and analyzed (the initial lung voxel points with LBP values ​​less than the preset LBP threshold are eliminated), and several lung voxel points are obtained, and connection processing is performed to obtain the patient's lung three-dimensional model.

[0041] The specific steps of analyzing the LBP value of each initial lung voxel point are as follows: for each initial lung voxel point, identify the adjacent initial lung voxel points within a preset neighborhood range (such as within a 3x3x3 cube); perform a comprehensive analysis on each initial lung voxel point and the adjacent initial lung voxel points within the preset neighborhood range to obtain the LBP value of each initial lung voxel point, and the calculation formula is as follows: ;in, For the The LBP value of the initial lung voxel point, is a symbolic function, which is hour, , hour, , For the The first The voxel values ​​of the adjacent initial lung voxel points, For the The voxel value of the initial lung voxel point, , is the number of initial lung voxel points, , is the number of adjacent initial lung voxel points within the preset neighborhood of the initial lung voxel point.

[0042] In this embodiment, the LBP method can effectively capture the local texture information of the lung tissue by comparing each voxel point with the voxel points in its neighborhood. This high degree of detail capture makes it possible to more accurately identify and distinguish normal lung tissue from diseased lung tissue. For example, tumors, inflammation or other pathological changes usually show texture characteristics that are significantly different from those of the surrounding normal tissues. LBP can convert these differences into numerical values ​​for further analysis. By setting a threshold, the LBP method can automatically eliminate voxel points that do not meet specific texture characteristics, which not only reduces the amount of data that needs to be further processed, but also improves the efficiency of the overall analysis. This method is particularly suitable for processing large-scale medical scan data, because Because it can quickly reduce the computational burden while retaining the most critical information, in the process of establishing a three-dimensional lung model, LBP analysis can ensure that the model contains voxel points that are representative and of diagnostic value. Such screening not only improves the quality of the model, but also ensures that subsequent analysis or diagnosis based on these models is more reliable. In addition, by accurately depicting the texture differences in different areas of the lungs, it helps doctors better understand the nature and scope of the disease. Using LBP for the establishment and analysis of lung models can provide strong support for clinical decision-making. Doctors can judge the severity and type of lesions based on the LBP values ​​of voxel points and their distribution characteristics, thereby formulating more personalized and accurate treatment plans.

[0043] Specifically, the pre-trained deep learning model is a variational autoencoder, and the specific steps to obtain the patient's healthy three-dimensional lung model are as follows: load the pre-trained variational autoencoder; input the patient's three-dimensional lung model into the encoder of the variational autoencoder for encoding; after the encoding process is completed, the decoder in the variational autoencoder performs decoding and reconstruction processing to obtain the patient's healthy three-dimensional lung model, which includes healthy voxel values ​​of several healthy chest voxel points.

[0044] The specific implementation process is as follows:

[0045] The input lung 3D model is preprocessed, including normalization, to ensure that the data range is consistent with the training data.

[0046] Input the 3D lung model to the encoder.

[0047] The encoder processes the data through its network structure and outputs parameters (logarithm of mean and variance) that describe the distribution in the latent space.

[0048] According to the parameters output by the encoder, the latent vector is sampled from the Gaussian distribution using the re-parameterization technique. This step is the core of VAE, which enables the model to be optimized by gradient descent during training.

[0049] The latent vector is fed into the decoder.

[0050] The decoder processes the latent vector through its network structure, increasing the data size layer by layer, and finally reconstructs a three-dimensional model of the lungs in a healthy state.

[0051] Among them, the variational autoencoder includes: an encoder and a decoder.

[0052] The encoder is used to convert high-dimensional input data (such as a 3D CT scan image) into a lower-dimensional latent representation. This latent space is usually a continuous vector with fewer dimensions that contains compressed knowledge of the input data.

[0053] The encoder includes an input layer for receiving raw 3D CT data, a hidden layer consisting of multiple layers of 3D convolutional layers (each layer includes an activation layer (such as ReLU) and a pooling layer (such as MaxPooling) to gradually extract features and reduce data dimensions), and an output layer for outputting the logarithm of the mean (μ) and variance (σ) (these parameters define the Gaussian distribution in the latent space).

[0054] The decoder is used to reconstruct the input data from randomly sampled points in the latent space, as close as possible to the original healthy lung data.

[0055] The decoder consists of an input layer for receiving vectors sampled from the latent space, a hidden layer consisting of multiple layers of 3D deconvolution layers (transposed convolutions) (each layer includes an activation layer, which gradually increases the data dimension and restores the shape and size of the original data), and an output layer for outputting reconstructed 3D data (whose structure and dimensions should match the original CT scan data).

[0056] The training process of the variational autoencoder is as follows:

[0057] Collect a large training dataset consisting of lung scans in healthy and diseased states.

[0058] Normalize the image and resize it to fit the model input.

[0059] Initialize the network parameters, typically using small random values.

[0060] Choose an optimizer, such as Adam, and set an appropriate learning rate and other hyperparameters.

[0061] The input data passes through the encoder, which outputs the latent space parameters (mean and variance) for each data point.

[0062] Sampling from a parameterized latent distribution (using the reparameterization trick) to obtain a latent representation.

[0063] The latent representation is passed through the decoder to generate the reconstructed image.

[0064] Calculate the reconstruction loss (such as MSE) and KL divergence loss, and combine the two into the total loss.

[0065] Compute the gradients of the network parameters with respect to the total loss.

[0066] Use the optimizer to update the network parameters of the encoder and decoder.

[0067] Model performance is evaluated on an independent validation set, monitoring reconstruction quality and continuity of the latent space.

[0068] Adjust model hyperparameters, such as adjusting the trade-off coefficient (β) between reconstruction loss and KL divergence in the loss function.

[0069] After training is complete, save the model parameters.

[0070] In this implementation, the variational autoencoder can learn and extract rich features from the CT scan data of the lungs through advanced neural network technology. This includes not only surface structural information, but also in-depth micro-texture changes in lung tissue, so that the model can accurately reflect the health and pathological status of the lungs. This precise model reconstruction helps doctors better understand the pathological changes of patients and improve the accuracy and reliability of diagnosis. Through the encoding and decoding process of the VAE model, a lung health model specific to each patient can be generated, which provides a basis for the implementation of personalized medicine. This model can not only be used to diagnose the current lung condition, but also simulate the potential impact of different treatment options on the patient's lung health, provide decision support for doctors, and formulate more accurate and effective treatment plans. Through automated deep learning model processing and Analyzing large amounts of image data can significantly reduce the time doctors spend on image analysis, allowing them to focus more on the clinical treatment and care of their patients. In addition, automated and standardized image processing processes can reduce errors caused by human factors and improve overall diagnosis and treatment efficiency. The high-quality reconstruction models provided by VAE make continuous monitoring and long-term management possible. Doctors can effectively track the progression or improvement of lung lesions during treatment through regular scans and model comparisons, adjust treatment strategies in a timely manner, and improve the adaptability and responsiveness of treatment. Accurate and visual three-dimensional lung models can help patients better understand their condition and treatment plan. This transparent and educational information exchange helps to increase patients' sense of trust and increase their participation in the treatment process, thereby improving overall treatment outcomes and patient satisfaction.

[0071] Specifically, the specific steps of analyzing the patient's lung health score index based on the patient's lung three-dimensional model and the patient's lung health three-dimensional model are as follows: the voxel value of each chest voxel point in the patient's lung three-dimensional model is respectively analyzed with the healthy voxel value of each chest healthy voxel point in the patient's lung health three-dimensional model to obtain the patient's lung voxel intensity deviation score index and lung voxel variability score index; the patient's lung voxel intensity deviation score index and lung voxel variability score index are comprehensively analyzed to obtain the patient's lung health score index; wherein, the specific formula for calculating the patient's lung health score index is as follows: ;in, Scoring indicators for patients' lung health, is the patient's lung voxel intensity deviation score indicator, is the weighting coefficient of the patient's lung voxel intensity deviation score index, is the patient's lung voxel variability scoring index, is the weighted coefficient of the patient's lung voxel variability score, , is a natural constant.

[0072] It needs to be explained that , The specific calculation steps are: summing and analyzing the patient's lung voxel intensity deviation score index and the lung voxel variability score index to obtain the patient's health score and value, and then performing a proportion analysis of the patient's lung voxel intensity deviation score index and the lung voxel variability score index with the patient's health score and value, and the analysis result is the corresponding weighting coefficient.

[0073] In this embodiment, by comparing the healthy and actual states of each voxel point, this method can accurately quantify the impact of the disease on the lung structure. This voxel-level analysis provides very detailed data, which helps to reveal small but potentially clinically important changes in lung tissue. For example, early tumor activity or subtle fibrosis changes may not be easily perceived at the macro level, but can be effectively identified and scored at the voxel level. The use of voxel intensity deviation scores and variability scores allows doctors to quantitatively monitor the progression of the disease or the response to treatment. This quantitative method can accurately track changes in the disease state and provide data support for adjusting treatment plans. For example, by regularly comparing the variability and intensity deviation in the lung model, doctors can It can determine whether lung disease is worsening or improving, so as to adjust drug dosage or treatment strategy in time. The deviation and variability scores of voxel intensity can help doctors classify patients' conditions into different risk levels. This classification helps determine which patients need more aggressive treatment or closer monitoring. It also provides a basis for screening and patient management in clinical trials. For example, high voxel variability scores may indicate higher heterogeneity of lesions, which may be associated with poorer prognosis in certain types of cancer. By automatically calculating and analyzing voxel values, subjective judgment errors in the diagnosis and treatment process can be reduced. This method provides doctors with a reliable, data-based decision-making tool, which helps to improve the consistency and standardization of treatment decisions.

[0074] Specifically, the specific steps for obtaining the patient's lung voxel intensity deviation score index and lung voxel variability score index are as follows: read the voxel value of each chest voxel point in the patient's lung three-dimensional model, and perform difference mean analysis with the healthy voxel value of each chest healthy voxel point in the patient's lung three-dimensional model, to obtain the patient's lung voxel difference mean; perform standard deviation analysis on the voxel value of each chest voxel point in the patient's lung three-dimensional model and the healthy voxel value of each chest healthy voxel point in the patient's lung three-dimensional model, to obtain the patient's lung voxel difference standard deviation; conduct a comprehensive analysis on the patient's lung voxel difference mean and lung voxel difference standard deviation, to obtain the patient's lung voxel intensity deviation score index and lung voxel variability score index.

[0075] The specific formulas for calculating the mean lung voxel difference, standard deviation of lung voxel difference, lung voxel intensity deviation score index, and lung voxel variability score index of the patient are as follows: ;in, is the mean lung voxel difference of the patient, The first 3D model of the patient's lungs The voxel value of the chest voxel point, The first 3D model of the healthy lung of the patient corresponds to the 3D model of the patient's lung. The health voxel value of the chest health voxel point, The first 3D model of the patient's lungs The weighting coefficient of the voxel value of the chest voxel point, The first 3D model of the patient's lungs The voxel value of the chest voxel point, The first 3D model of the healthy lung of the patient corresponds to the 3D model of the patient's lung. The health voxel value of the chest health voxel point, The first 3D model of the patient's lungs The weighting coefficient of the voxel value of the chest voxel point, , is the standard deviation of the patient’s lung voxel difference, is the patient's lung voxel intensity deviation score indicator, is the patient's lung voxel variability scoring index, is the weight coefficient of the mean difference of the patient’s lung voxels, is the weight coefficient of the standard deviation of the patient’s lung voxel difference, , , The number of chest voxel points.

[0076] It needs to be explained that , The specific calculation steps are as follows: The voxel value of the chest voxel point is consistent with the corresponding patient's lung health 3D model. The healthy voxel values ​​of the chest healthy voxel points are analyzed by difference analysis to obtain the first healthy voxel point in the patient's lung 3D model. The voxel difference of each chest voxel point.

[0077] The patient's lungs are modeled The voxel value of the chest voxel point is consistent with the corresponding patient's lung health 3D model. The healthy voxel values ​​of the chest healthy voxel points are analyzed by difference analysis to obtain the first healthy voxel point in the patient's lung 3D model. The voxel difference of each chest voxel point.

[0078] The patient's lungs are modeled , The voxel differences of the chest voxel points are summed and analyzed to obtain the The voxel difference and value are then calculated for the first , The voxel differences of the chest voxel points are The voxel difference and value of each group are analyzed by ratio, and the analysis result is the corresponding weighting coefficient.

[0079] , The specific calculation process is: the absolute value of the patient's lung voxel difference mean and the patient's lung voxel difference standard deviation are summed and analyzed to obtain the patient's voxel difference score and value, and then the absolute value of the patient's lung voxel difference mean and the patient's lung voxel difference standard deviation are respectively analyzed with the patient's voxel difference score and value, and the analysis result is the corresponding weight coefficient.

[0080] In this embodiment, by accurately calculating the voxel intensity deviation and variability, the method can map in detail the specific differences between the healthy state of the lungs and the diseased areas. This detailed quantitative analysis enables doctors to more accurately identify and locate lung diseases, especially small lesions in the early stages that may not be easily identified by standard imaging technology. These scoring indicators provide doctors with a powerful tool to monitor changes in the disease or the effectiveness of treatment. By regularly evaluating voxel differences and variability, doctors can objectively evaluate whether the disease has improved or worsened, so as to adjust the treatment strategy in time. Detailed health scores based on voxel level make treatment more personalized. According to the specific health status of the patient's lungs, doctors can design specific treatment plans to treat the affected areas in a targeted manner rather than using a wide range of treatment methods, which may bring better treatment effects and lower risks of side effects. Through an automated and standardized health scoring system, medical institutions can allocate resources more effectively. For example, for those patients whose voxel deviation and variability indicate low risk, less monitoring and intervention can be selected, and more resources can be concentrated on high-risk patients, thereby improving the efficiency and effectiveness of resource use.

[0081] See also Figure 3 The embodiment of the present invention provides a technical solution: a medical image analysis system based on plain scan CT data, comprising: a data acquisition module, a three-dimensional model building module, a segmentation module, a restoration module, and a health analysis module; the data acquisition module is used to acquire the chest plain scan CT data of the patient, the chest plain scan CT data including chest cross-sectional image data at several scanning positions; the three-dimensional model building module is used to establish a three-dimensional model of the patient's chest based on the chest plain scan CT data of the patient; the segmentation module is used to segment the three-dimensional model of the patient's chest to obtain a three-dimensional model of the patient's lungs; the restoration module is used to input the three-dimensional model of the patient's lungs into a pre-trained deep learning model for restoration analysis to obtain a three-dimensional model of the patient's lung health; the health analysis module is used to analyze the patient's lung health scoring index based on the patient's lung three-dimensional model and the patient's lung health three-dimensional model, and divide the patient's lung health level according to a preset scoring rule.

[0082] In summary, this application has at least the following effects:

[0083] By using a deep learning model to automatically reconstruct a 3D model of a healthy lung and compare it with the patient's actual lung model, a more accurate disease diagnosis can be provided. This automated comparison reduces human errors and improves analysis efficiency, allowing doctors to quickly and accurately identify the lesion area and the extent of the disease, thereby developing a more effective treatment plan.

[0084] By utilizing deep learning techniques such as variational autoencoders, it is possible to generate a personalized healthy lung model based on the patient's specific lung condition. This model can not only be used for current health assessments, but also as a benchmark for evaluating treatment effectiveness. As treatment progresses, through regular scans and model comparisons, doctors can monitor treatment effectiveness in real time and adjust treatment plans.

[0085] By converting complex medical image data into easy-to-understand 3D models and intuitive health scores, patients can better understand their health status, which helps enhance their trust and satisfaction with the treatment process.

[0086] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0087] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A medical image analysis method based on plain scan CT data, characterized in that: The following steps are involved: Acquire chest plain scan CT data of the patient, wherein the chest plain scan CT data includes chest cross-sectional image data at a plurality of scanning positions; Establish a three-dimensional model of the patient's chest based on the patient's chest plain scan CT data; Segmenting the patient's chest 3D model to obtain the patient's lung 3D model; The patient's lung 3D model is input into a pre-trained deep learning model for restoration analysis to obtain a healthy 3D model of the patient's lungs; Analyze the patient's lung health scoring index based on the patient's lung 3D model and the patient's lung health 3D model, and divide the patient's lung health level according to preset scoring rules; The specific steps of analyzing the patient's lung health score index based on the patient's lung three-dimensional model and the patient's lung health three-dimensional model are as follows: The voxel value of each chest voxel point in the patient's lung three-dimensional model is comprehensively analyzed with the healthy voxel value of each chest healthy voxel point in the patient's lung healthy three-dimensional model to obtain the patient's lung voxel intensity deviation score index and lung voxel variability score index; Comprehensively analyze the patient's lung voxel intensity deviation score index and lung voxel variability score index to obtain the patient's lung health score index; Among them, the specific formula for calculating the patient's lung health score index is as follows: ； in, Scoring indicators for patients' lung health, is the patient's lung voxel intensity deviation score indicator, is the weighting coefficient of the patient's lung voxel intensity deviation score index, is the patient's lung voxel variability scoring index, is the weighted coefficient of the patient's lung voxel variability score, , is a natural constant; The specific steps for obtaining the patient's lung voxel intensity deviation score index and lung voxel variability score index are as follows: Read the voxel value of each chest voxel point in the patient's lung three-dimensional model, and perform difference mean analysis with the healthy voxel value of each chest healthy voxel point in the patient's lung healthy three-dimensional model to obtain the patient's lung voxel difference mean; Performing standard deviation analysis on the voxel value of each chest voxel point in the patient's lung three-dimensional model and the healthy voxel value of each chest healthy voxel point in the patient's lung healthy three-dimensional model, to obtain the standard deviation of the patient's lung voxel difference; The mean and standard deviation of the lung voxel differences of the patients were comprehensively analyzed to obtain the lung voxel intensity deviation score index and the lung voxel variability score index of the patients; The specific formulas for calculating the mean lung voxel difference, the standard deviation of the lung voxel difference, the lung voxel intensity deviation score index, and the lung voxel variability score index of the patient are as follows: ； in, is the mean lung voxel difference of the patient, The first 3D model of the patient's lungs The voxel value of the chest voxel point, The first 3D model of the healthy lung of the patient corresponds to the 3D model of the patient's lung. The health voxel value of the chest health voxel point, The first 3D model of the patient's lungs The weighting coefficient of the voxel value of the chest voxel point, The first 3D model of the patient's lungs The voxel value of the chest voxel point, The first 3D model of the healthy lung of the patient corresponds to the 3D model of the patient's lung. The health voxel value of the chest health voxel point, The first 3D model of the patient's lungs The weighting coefficient of the voxel value of the chest voxel point, , is the standard deviation of the patient’s lung voxel difference, is the patient's lung voxel intensity deviation score indicator, is the patient's lung voxel variability scoring index, is the weight coefficient of the mean difference of the patient’s lung voxels, is the weight coefficient of the standard deviation of the patient’s lung voxel difference, , , The number of chest voxel points.

2. The medical image analysis method based on plain scan CT data according to claim 1, characterized in that: The chest cross-sectional image data includes pixel values ​​of a plurality of chest cross-sectional pixel points.

3. The medical image analysis method based on plain scan CT data according to claim 2, characterized in that: The specific steps for establishing a 3D model of the patient's chest based on the patient's chest plain scan CT data are as follows: Read the chest cross-sectional image data at each scanning position, and perform scanning position sorting and stacking processing to obtain an initial three-dimensional model of the patient's chest; An interpolation judgment process is performed on the initial three-dimensional model of the patient's chest to obtain a three-dimensional model of the patient's chest, wherein the three-dimensional model of the patient's chest includes voxel values ​​of a plurality of chest voxel points.

4. The medical image analysis method based on plain scan CT data according to claim 3, characterized in that: The specific steps to obtain the 3D model of the patient's lungs are as follows: Based on a preset voxel threshold range, a plurality of chest voxel points in the patient's chest three-dimensional model are preliminarily screened and analyzed to obtain a plurality of initial lung voxel points; The LBP value of each initial lung voxel point was analyzed separately; Based on the preset LBP threshold, each initial lung voxel point is screened and analyzed to obtain a number of lung voxel points, which are then connected to obtain a three-dimensional model of the patient's lungs.

5. The medical image analysis method based on plain scan CT data according to claim 4, characterized in that: The specific steps for analyzing the LBP value of each initial lung voxel point are as follows: For each initial lung voxel point, adjacent initial lung voxel points within a preset neighborhood are identified respectively; Each initial lung voxel point is comprehensively analyzed with the adjacent initial lung voxel points within the preset neighborhood range to obtain the LBP value of each initial lung voxel point, and the calculation formula is as follows: ； in, For the The LBP value of the initial lung voxel point, is the symbolic function, For the The first The voxel values ​​of the adjacent initial lung voxel points, For the The voxel value of the initial lung voxel point, , is the number of initial lung voxel points, , is the number of adjacent initial lung voxel points within the preset neighborhood of the initial lung voxel point.

6. The medical image analysis method based on plain scan CT data according to claim 4, characterized in that: The pre-trained deep learning model is specifically a variational autoencoder. The specific steps to obtain the patient's lung health 3D model are as follows: Load a pre-trained variational autoencoder; Inputting the patient's lung three-dimensional model into the encoder of the variational autoencoder for encoding processing; After the encoding process is completed, the decoder in the variational autoencoder performs decoding and reconstruction processing to obtain a healthy three-dimensional lung model of the patient, wherein the healthy three-dimensional lung model of the patient includes healthy voxel values ​​of several healthy chest voxel points.

7. A medical image analysis system based on plain scan CT data, using the medical image analysis method based on plain scan CT data according to any one of claims 1 to 6, characterized in that: include: Data acquisition module, 3D model building module, segmentation module, restoration module, health analysis module; The data acquisition module is used to acquire chest plain scan CT data of the patient, wherein the chest plain scan CT data includes chest cross-sectional image data at a plurality of scanning positions; The three-dimensional model building module is used to build a three-dimensional model of the patient's chest based on the patient's chest plain scan CT data; The segmentation module is used to segment the three-dimensional model of the patient's chest to obtain a three-dimensional model of the patient's lungs; The restoration module is used to input the patient's lung three-dimensional model into a pre-trained deep learning model for restoration analysis to obtain a healthy three-dimensional lung model of the patient; The health analysis module is used to analyze the patient's lung health scoring index based on the patient's lung three-dimensional model and the patient's lung health three-dimensional model, and to classify the patient's lung health level according to preset scoring rules.

Images