Medical image scanning image processing method
Through deep learning technology, the features in medical images are extracted and feature matching algorithms are combined, the problem of insufficient accuracy of micro lesions recognition and tracking in the existing technology is solved, and more efficient micro lesions detection and tracking is achieved, and the accuracy of diagnosis is improved. and therapeutic effects.
Patent Information
- Application Number
- CN202411036380.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-31
- Publication Date
- 2024-12-03
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing medical imaging processing technology has problems such as low accuracy in intelligent identification and tracking of micro lesions, susceptible to noise and image quality, and difficulty in adapting to lesion morphology changes, and cannot meet the accuracy and reliability requirements of clinical diagnosis.
Deep learning technology is used to extract feature, learn discriminant features from medical images, and combine the tracking algorithm based on feature matching to extract and update the shape, texture and motion information of micro lesions to stably track the lesion area in a continuous image sequence. .
It significantly improves the detection accuracy and tracking effect of micro lesions, helps doctors to understand the lesions more comprehensively, provide more reasonable treatment plans, and provide valuable data support for medical research.
Smart Images

Figure CN119068032A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of medical image processing, and in particular to a medical image scanning image processing method. Background Art
[0002] As an important means of modern medical diagnosis, medical imaging provides doctors with intuitive, non-invasive information on the internal structure and physiological functions of the human body. From X-rays, CT scans to MRI, PET and other advanced technologies, the quality and diversity of medical imaging have been continuously improved, greatly promoting the development of medicine. These images not only help in the early detection of diseases, but also provide doctors with accurate disease assessment and treatment effect monitoring during the treatment process.
[0003] However, the existing medical image processing technology still has many shortcomings in the treatment of tiny lesions. In the intelligent identification of tiny lesions, traditional algorithms often rely on manually set features, and have limited ability to capture the complex morphology and subtle differences of lesions, resulting in low accuracy. Moreover, since tiny lesions are relatively weak in images, existing feature extraction methods are easily affected by noise and image quality, and misjudgments and missed judgments often occur. In terms of tracking algorithms, existing technologies are difficult to adapt to the morphological changes and complex movement patterns of lesions, and are prone to losing targets or generating tracking deviations in continuous image sequences. They are also unable to fully utilize the advantages of various images to improve the detection and tracking effects of tiny lesions.
[0004] In summary, the existing medical image processing methods have obvious defects in the intelligent recognition and tracking of tiny lesions and cannot meet the requirements of clinical diagnosis for accuracy and reliability. Therefore, the development of a more efficient and accurate medical image scanning image processing method, especially an intelligent recognition and tracking algorithm for tiny lesions, has important clinical significance and application value. Summary of the invention
[0005] The purpose of the present invention is to make up for the shortcomings of the existing technology and provide a medical image scanning image processing method, which can extract features through deep learning technology and learn highly discriminative features from medical images. Compared with traditional image processing technology, this intelligent recognition method can more accurately detect tiny abnormal areas, thereby improving the accuracy of diagnosis, which is crucial for early detection of diseases and providing timely treatment opportunities.
[0006] 3. In order to solve the above technical problems, the present invention provides the following technical solutions: a medical image scanning image processing method, the image processing method comprises the following steps:
[0007] S100, medical image data acquisition and preprocessing: performing a medical image scan on the patient, acquiring a scanned image of the target part, and preprocessing the acquired image data;
[0008] S200, Intelligent Identification of Minor Lesions: Based on deep learning technology, it is used to extract the features of minor lesions and learn discriminative features from the input medical images;
[0009] S300, tracking of micro-lesions: for the identified micro-lesion area, extract its shape, texture, and motion information, and use a tracking algorithm based on feature matching to track the lesion area in a continuous image sequence, and update its position and morphology information;
[0010] S400, result output and evaluation: The identification and tracking results are output to the doctor in a visual form, that is, an image and a report with the lesion area marked, and the doctor further evaluates and analyzes the lesion based on these results.
[0011] Furthermore, the specific steps of S100 are:
[0012] S101, medical imaging scanning: use medical imaging equipment to scan the target area, and adjust the scanning parameters according to the type of lesion, i.e., scanning resolution, scanning time, and contrast agent usage;
[0013] S102, data reception: after scanning is completed, the generated medical image data is transmitted from the device to the image processing workstation and server for subsequent processing;
[0014] S103, image segmentation: separate the target area from the background to reduce the interference of non-target areas on feature extraction.
[0015] Furthermore, the S101 medical imaging scanning equipment includes: high-resolution magnetic resonance imaging MRI, multi-slice spiral computed tomography CT, and positron emission tomography PET. For the detection of tiny brain lesions, high-field strength MRI equipment is preferred, and for tiny lung nodules, multi-slice spiral CT is preferred. When performing image scanning, the scanning parameters are accurately set, including scanning range, layer thickness, layer spacing, and scanning time. For brain scanning, the scanning coverage area is determined according to the location and size of the lesion. For lung scanning, the appropriate layer thickness and layer spacing are selected according to the size and distribution of the nodules to obtain sufficiently detailed image information.
[0016] Furthermore, the specific steps of S200 are:
[0017] S201, data preparation: collect medical image datasets containing microlesions, annotate them, and mark the location and category of the lesion area;
[0018] S202, Model selection and training: Automatically learn complex features in images based on deep learning models;
[0019] S203, feature extraction: extract low-level to high-level features in the image through multiple convolutional layers, and obtain a feature map through weighted summation and activation function;
[0020] S204, lesion area identification: The model finally outputs a probability map of each pixel belonging to the lesion area, thereby identifying the lesion area.
[0021] Furthermore, the construction process of the S202 deep learning module is:
[0022] Input layer: extract features from the image through the convolution layer and pooling layer of CNN, and input the acquired scanned image of the target part into the CNN model;
[0023] Convolution layer: extract features by convolution operation between the convolution kernel and the input image, that is I(x,y) is the input image, K(s,t) is the convolution kernel, m and n are the half width of the convolution kernel, and O(x,y) is the feature map of the convolution output;
[0024] Pooling layer: Reduce the risk of overfitting, that is, P(x,y)=max (x′,y′)∈R I(x′,y′), I(x′,y′) is the element of the input feature map, R is the pooling region, and P(x,y) is the output after pooling;
[0025] Fully connected layer: Flatten the feature map and connect it to the fully connected layer for classification and regression z = W x + b, where z is the output of the fully connected layer, W is the weight matrix, x is the input feature vector, and b is the bias term;
[0026] Output layer: Output the final result and make classification probability prediction, that is z i is the i-th element of the fully connected layer output, softmax(z i ) is the predicted probability of the ith category.
[0027] Furthermore, the S203 generates a CNN model and performs training optimization to define a loss function to measure the difference between the model prediction and the true label, that is, N is the number of samples, y i is the real label, is the predicted probability value. The validation set is used to evaluate the model performance. The model hyperparameters, i.e., convolution kernel size, number of layers, and learning rate, are adjusted according to the evaluation results. The model is retrained and the performance of the final model is evaluated on the test set. The trained model is applied to new medical images for intelligent identification of tiny lesions.
[0028] Furthermore, the specific steps of S300 for tracking minor lesions are:
[0029] S301, feature extraction: extract multiple features based on the microlesion area identified by deep learning technology, namely shape features, texture features and motion information;
[0030] S302, feature matching: in a continuous image sequence, a feature matching algorithm is used to match the lesion features in the current frame with the features in the previous frame and the reference frame to determine the position change of the lesion;
[0031] S303, motion estimation and tracking: based on the result of feature matching, a motion estimation algorithm is used to predict and update the position and shape of the lesion area, and the motion of the lesion is estimated by calculating the motion vectors of pixels and feature points between consecutive frames;
[0032] S304, position and morphology update: according to the result of motion estimation, the position and morphology information of the lesion area are updated, and their changes are tracked in the continuous image sequence.
[0033] Furthermore, in the continuous image sequence, S302 selects the K feature points with the closest distance by calculating the distance between the feature to be matched and all known features, and estimates the position based on these points, that is, x i and i are the values of the two feature points in the i-th dimension, and n is the dimension of the feature.
[0034] Furthermore, the S303 obtains the motion information of the lesion area by calculating the optical flow between adjacent image frames based on the result of feature matching, that is, I x u+I y v+I t =0,I x and I y are the gradients of the image in the x and y directions, u and v are the components of the optical flow in the x and y directions, I t is the rate of change of the image over time.
[0035] Furthermore, the specific steps of S400 are:
[0036] S401, result annotation and visualization: annotating the identified and tracked microlesion areas on the original medical image, and highlighting the lesion areas by using borders, fills or marks of different colors;
[0037] S402, report generation: automatically generate a detailed diagnostic report, which includes the patient's basic information, scanning parameters, location, size, morphology and texture characteristics of the lesion, and provides quantitative analysis data of the lesion, i.e., volume, area, and gray value statistics of the lesion area, and compares it with historical data;
[0038] S403, result evaluation: carefully review the annotated lesion area and visualization results, and evaluate the generated diagnostic report;
[0039] S404, data storage and retrieval: the identification and tracking results, diagnosis report and doctor's evaluation opinions are stored in the database for subsequent retrieval and analysis.
[0040] Compared with the prior art, this medical image scanning image processing method has the following beneficial effects:
[0041] 1. The present invention can stably track the lesion area in a continuous image sequence and update its position and morphological information in real time by extracting the shape, texture and motion information of the lesion and adopting a tracking algorithm based on feature matching. This continuous tracking capability helps doctors to have a more comprehensive understanding of the development dynamics of the lesion, including key information such as its growth rate and morphological changes, so as to formulate a more reasonable and personalized treatment plan. At the same time, the continuous tracking data also provides valuable data support for medical research and helps to promote the development of the medical field.
[0042] 2. By introducing deep learning technology, the present invention can automatically learn discriminative features from medical images. These features are very effective in distinguishing lesions from normal tissues. At the same time, through morphological processing and other methods, the boundaries of lesions can be further accurately determined, thereby significantly improving the detection accuracy of tiny lesions. This improved accuracy helps to detect diseases at an early stage and provide patients with earlier treatment opportunities, thereby improving treatment outcomes and survival rates.
[0043] Other advantages, objectives and features of the present invention will be set forth in part in the following description and, in part, will be apparent to those skilled in the art based on an examination of the following or may be taught from the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0045] Figure 1 The present invention is an operation flow chart of a medical image scanning image processing method.
[0046] Figure 2 The present invention is a flowchart of the method steps of a medical image scanning image processing method. DETAILED DESCRIPTION
[0047] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0048] Embodiment 1
[0049] A medical image scanning image processing method, which uses deep learning technology to extract features and learn highly discriminative features from medical images, which is particularly important for the identification of tiny lesions. Compared with traditional image processing technology, this intelligent recognition method can more accurately detect tiny abnormal areas, thereby improving the accuracy of diagnosis, which is crucial for early detection of diseases and providing timely treatment opportunities.
[0050] First, enter the medical image data acquisition and preprocessing stage (S100), use a high-field strength MRI device to scan the patient's brain, determine the scanning area based on the location and size of the lesion, and set appropriate scanning parameters, such as scanning range, layer thickness, and layer spacing. After the scan is completed, the medical image data is transmitted to the image processing workstation, and the acquired image data is preprocessed, including image segmentation, to separate the target area from the background.
[0051] Then, the stage of intelligent identification of tiny lesions (S200) is entered. Medical imaging data sets containing tiny brain tumors are collected and annotated. A deep learning model is used for training to extract complex features in the image. Low-level to high-level features in the image are extracted through multiple convolutional layers. A feature map is obtained through weighted summation and activation function. The model finally outputs a probability map of each pixel belonging to the lesion area, thereby identifying the lesion area.
[0052] Next, the micro-lesion tracking stage (S300) is entered. Based on the micro-lesion area identified by deep learning technology, multiple features are extracted, including shape, texture and motion information. In a continuous image sequence, a feature matching algorithm is used to match the lesion features in the current frame with the features in the previous frame and the reference frame to determine the position change of the lesion. By calculating the optical flow between adjacent image frames, the motion information of the lesion area is obtained. According to the results of motion estimation, the position and morphological information of the lesion area are updated, and its changes are tracked in a continuous image sequence.
[0053] Finally, the result output and evaluation stage (S400) is entered, and the identified and tracked micro-lesion areas are annotated on the original medical image, highlighted with borders, fills or marks of different colors, and a detailed diagnostic report is automatically generated, including the patient's basic information, scanning parameters, and feature descriptions of the location, size, morphology and texture of the lesion, as well as quantitative analysis data of the lesion. The annotated lesion areas and visualization results are carefully reviewed, and the generated diagnostic report is evaluated. The identification and tracking results, the diagnostic report and the doctor's evaluation opinions are stored in a database for subsequent retrieval and analysis.
[0054] In summary, this embodiment provides specific steps and operation procedures for a medical image scanning image processing method. Through deep learning technology and feature extraction, the method can more accurately identify and locate tiny lesions, even low-contrast or difficult-to-find lesions. This high accuracy helps to detect diseases early and provides an opportunity for timely treatment, which may improve the patient's prognosis.
[0055] Embodiment 2
[0056] This embodiment provides an implementation process of a medical image scanning image processing method for identifying and tracking tiny brain tumors, and uses deep learning technology to identify the shape, texture and motion information of tiny lesion areas, and uses a tracking algorithm based on feature matching to track the lesion area in a continuous image sequence, and update its position and morphology information.
[0057] A high-field magnetic resonance imaging (MRI) device is used to scan the patient's brain. The scanning parameters, including the scanning range, layer thickness, layer spacing, and scanning time, are precisely set according to the possible location and size of the tumor to ensure that clear brain images are obtained. After the scan is completed, the generated brain MRI image data is transmitted to the image processing workstation and server for image segmentation to separate the brain from the background and reduce the interference of non-target areas on subsequent processing. A large number of medical imaging data sets containing tiny brain tumors are collected and accurately labeled to mark the location and category of the tumor area.
[0058] Based on the deep learning model, namely the convolutional neural network (CNN), the complex features in the image are automatically learned, the acquired scanned image of the target part is input into the CNN model, and the image features are extracted through the convolution layer and pooling layer of the CNN.
[0059] The features are extracted by convolution operation between the convolution kernel and the input image. The formula is: Reduce the risk of overfitting through the pooling layer, that is, P(x,y)=max (x′,y′)∈R I(x′,y′), flatten the feature map and connect it to the fully connected layer, perform classification and regression z=W·x+b, output the final result, and make classification probability prediction, that is, Define the loss function to measure the difference between the model prediction and the true label. The formula is Use the validation set to evaluate the model performance, adjust the model hyperparameters such as convolution kernel size, number of layers, learning rate, etc. according to the evaluation results, retrain, evaluate the performance of the final model on the test set, and apply the trained model to new medical images for intelligent identification of tiny lesions. Extract low-level to high-level features in the image through multiple convolutional layers, and obtain the feature map through weighted summation and activation function. The model finally outputs the probability map of each pixel belonging to the lesion area, thereby identifying the lesion area.
[0060] Based on the tiny lesion areas identified by deep learning technology, a variety of features are extracted, including shape features (such as area, perimeter, circularity, etc.), texture features (such as gray-level co-occurrence matrix, local binary pattern, etc.) and motion information. In the continuous image sequence, the feature matching algorithm is used to match the lesion features in the current frame with the features in the previous frame and the reference frame to determine the position change of the lesion. The K feature points with the closest distance are selected by calculating the distance between the feature to be matched and all known features, and the position is estimated based on these points. The formula is: Based on the result of feature matching, the motion estimation algorithm is used to predict and update the position and shape of the lesion area. The motion information of the lesion area is obtained by calculating the optical flow between adjacent image frames. The formula is I x u+I y v+I t =0, based on the results of motion estimation, the position and morphological information of the lesion area are updated, and its changes are tracked in continuous image sequences.
[0061] Finally, the identified and tracked tumor area is marked and highlighted with borders of different colors on the original MRI image, and a detailed diagnostic report is automatically generated, including the patient's basic information, scanning parameters, characteristic descriptions of the tumor's location, size, morphology and texture, as well as quantitative analysis data such as the volume, area, and grayscale value statistics of the lesion area, and compared with historical data. The doctor carefully reviews the marked lesion area and visualization results, evaluates the generated diagnostic report, and stores the identification and tracking results, diagnostic report, and doctor's evaluation opinions in the database for subsequent retrieval and analysis.
[0062] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.
Claims
1. A medical image scanning image processing method, characterized in that: The image processing method comprises the following steps: S100, medical image data acquisition and preprocessing: performing a medical image scan on the patient, acquiring a scanned image of a target part, and preprocessing the acquired image data; S200, Intelligent Identification of Minor Lesions: Based on deep learning technology, it is used to extract the features of minor lesions and learn discriminative features from the input medical images; S300, tracking of micro-lesions: for the identified micro-lesion area, extract its shape, texture, and motion information, and use a tracking algorithm based on feature matching to track the lesion area in a continuous image sequence, and update its position and morphology information; S400, result output and evaluation: The identification and tracking results are output to the doctor in a visual form, that is, an image and a report with the lesion area marked, and the doctor further evaluates and analyzes the lesion based on these results.
2. A medical image scanning image processing method according to claim 1, characterized in that: The specific steps of S100 are: S101, medical imaging scanning: use medical imaging equipment to scan the target area, and adjust the scanning parameters according to the type of lesion, i.e., scanning resolution, scanning time, and contrast agent usage; S102, data reception: after scanning is completed, the generated medical image data is transmitted from the device to the image processing workstation and server for subsequent processing; S103, image segmentation: separate the target area from the background to reduce the interference of non-target areas on feature extraction.
3. A medical image scanning image processing method according to claim 2, characterized in that: The S101 medical imaging scanning equipment includes: high-resolution magnetic resonance imaging MRI, multi-slice spiral computed tomography CT, and positron emission tomography PET. For the detection of tiny brain lesions, high-field strength MRI equipment is preferred, and for tiny lung nodules, multi-slice spiral CT is preferred. When performing image scanning, the scanning parameters are accurately set, including the scanning range, layer thickness, layer spacing, and scanning time. For brain scanning, the scanning coverage area is determined according to the location and size of the lesion. For lung scanning, the appropriate layer thickness and layer spacing are selected according to the size and distribution of the nodules to obtain sufficiently detailed image information.
4. The medical image scanning image processing method according to claim 1, characterized in that: The specific steps of S200 are: S201, data preparation: collect medical image datasets containing microlesions, annotate them, and mark the location and category of the lesion area; S202, Model selection and training: Automatically learn complex features in images based on deep learning models; S203, feature extraction: extract low-level to high-level features in the image through multiple convolutional layers, and obtain a feature map through weighted summation and activation function; S204, lesion area identification: The model finally outputs a probability map of each pixel belonging to the lesion area, thereby identifying the lesion area.
5. A medical image scanning image processing method according to claim 4, characterized in that: The construction process of the S202 deep learning module is: Input layer: extract features from the image through the convolution layer and pooling layer of CNN, and input the acquired scanned image of the target part into the CNN model; Convolution layer: extract features by convolution operation between the convolution kernel and the input image, that is I(x,y) is the input image, K(s,t) is the convolution kernel, m and n are the half width of the convolution kernel, and O(x,y) is the feature map of the convolution output; Pooling layer: Reduce the risk of overfitting, that is, P(x,y)=max (x′,y′)∈R I(x′,y′), I(x′,y′ is the element of the input feature map, R is the pooling region, and P(x,y) is the output after pooling; Fully connected layer: Flatten the feature map and connect it to the fully connected layer for classification and regression z = W x + b, where z is the output of the fully connected layer, W is the weight matrix, x is the input feature vector, and b is the bias term; Output layer: Output the final result and make classification probability prediction, that is z i is the i-th element of the fully connected layer output, softmax(z i ) is the predicted probability of the ith category.
6. A medical image scanning image processing method according to claim 5, characterized in that: The S203 generates a CNN model and performs training optimization, defines a loss function, and measures the difference between the model prediction and the true label, that is, N is the number of samples, y i is the real label, is the predicted probability value. The validation set is used to evaluate the model performance. The model hyperparameters, i.e., convolution kernel size, number of layers, and learning rate, are adjusted according to the evaluation results. The model is retrained and the performance of the final model is evaluated on the test set. The trained model is applied to new medical images for intelligent identification of tiny lesions.
7. A medical image scanning image processing method according to claim 1, characterized in that: The specific steps of S300 for tracking minor lesions are: S301, feature extraction: extract multiple features based on the microlesion area identified by deep learning technology, namely shape features, texture features and motion information; S302, feature matching: in a continuous image sequence, a feature matching algorithm is used to match the lesion features in the current frame with the features in the previous frame and the reference frame to determine the position change of the lesion; S303, motion estimation and tracking: based on the result of feature matching, a motion estimation algorithm is used to predict and update the position and shape of the lesion area, and the motion of the lesion is estimated by calculating the motion vectors of pixels and feature points between consecutive frames; S304, position and morphology update: according to the result of motion estimation, the position and morphology information of the lesion area are updated, and their changes are tracked in the continuous image sequence.
8. A medical image scanning image processing method according to claim 7, characterized in that: In S302, in a continuous image sequence, the K feature points with the closest distance are selected by calculating the distance between the feature to be matched and all known features, and the position is estimated based on these points, that is, x i and i are the values of the two feature points in the i-th dimension, and n is the dimension of the feature.
9. A medical image scanning image processing method according to claim 8, characterized in that: The step S303 obtains the motion information of the lesion area by calculating the optical flow between adjacent image frames based on the result of feature matching, i.e., x u+I y v+I t =0,I x and I y are the gradients of the image in the x and y directions, u and v are the components of the optical flow in the x and y directions, I t is the rate of change of the image over time.
10. A medical image scanning image processing method according to claim 1, characterized in that: The specific steps of S400 are: S401, result annotation and visualization: annotating the identified and tracked microlesion areas on the original medical image, and highlighting the lesion areas by using borders, fills or marks of different colors; S402, report generation: automatically generate a detailed diagnostic report, which includes the patient's basic information, scanning parameters, location, size, morphology and texture characteristics of the lesion, and provides quantitative analysis data of the lesion, i.e., volume, area, and gray value statistics of the lesion area, and compares it with historical data; S403, result evaluation: carefully review the annotated lesion area and visualization results, and evaluate the generated diagnostic report; S404, data storage and retrieval: the identification and tracking results, diagnosis report and doctor's evaluation opinions are stored in the database for subsequent retrieval and analysis.
Citation Information
Patent Citations
Focus tracking method under digestive endoscope based on sequential feature learning
CN111915573A
Medical image diagnosis, comparison and reading method
CN118262875A
