A method and system for automatic calculation of left atrial volume
By combining a deep learning model with morphological erosion and dynamic expansion strategies, the device dependency and boundary ambiguity issues in the automatic analysis of left atrial volume were resolved, achieving efficient and stable left atrial volume calculation applicable to multimodal imaging and improving clinical diagnostic efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- TUOWEI MIXIN DATA TECH (NANJING) CO LTD
- Filing Date
- 2025-04-28
- Publication Date
- 2026-06-16
AI Technical Summary
Existing technologies for automatic left atrial volume analysis suffer from problems such as strong equipment dependence, blurred boundaries leading to segmentation difficulties, time-consuming and labor-intensive manual segmentation, and limited generalization ability of deep learning methods, which restrict their universality and practicality in clinical practice.
By employing a deep learning model combined with morphological erosion and dynamic expansion strategies, the left atrium is automatically segmented and its volume calculated using multimodal medical imaging data. This includes coarse segmentation, morphological erosion, center point adjustment, expansion operation, and second-order volume change rate analysis, enabling stable identification and accurate measurement of the left atrial structure.
It achieves efficient, stable, and accurate automatic calculation of left atrial volume without the need for special equipment, is applicable to multimodal imaging, improves clinical applicability and diagnostic efficiency, and provides stronger data support.
Smart Images
Figure CN120635177B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical image processing technology, and in particular to a method and system for automatic calculation of left atrial volume. This method is an automatic volume measurement approach based on multimodal medical image data, integrating deep learning segmentation, morphological processing, and dynamic change analysis. It can achieve stable identification and automatic measurement of left atrial structures even when image boundaries are blurred. Background Technology
[0002] Atrial fibrillation (AF) is a common arrhythmia, and the volume of the left atrium is of significant value in treatment planning and prognostic assessment. With the development of multimodal medical imaging technologies such as CT and MRI, automated analysis of cardiac structures based on multimodal imaging data has gradually become a research focus. Currently, some methods have attempted to achieve automated analysis of left atrial volume, but the following limitations still exist:
[0003] 1. Most methods rely on techniques such as modeling, trajectory tracking, or synchronous imaging of cardiac cycles. These methods usually require specific equipment or simultaneous recording of multiple cardiac cycles, which are complex, dependent on specific equipment, and have poor adaptability.
[0004] 2. The blurred boundaries between the left atrium and adjacent tissues such as the pulmonary veins and left atrial appendage make it difficult to accurately segment using traditional methods based on static modeling or rule filtering;
[0005] 3. Manual segmentation is not only time-consuming and labor-intensive, but also highly subjective, making it difficult to promote in routine clinical procedures;
[0006] 4. Although some deep learning methods have improved segmentation accuracy, they usually require a large amount of high-quality labeled data for training, have limited generalization ability, and lack good transfer ability between different image modalities. They usually need to retrain the model or adjust the model parameters.
[0007] The aforementioned problems severely limit the universality and practicality of existing methods in clinical practice. Therefore, there is an urgent need for a method that requires no special equipment, possesses high automation capabilities, and can directly extract left atrial volume change features from multimodal images, especially an algorithmic mechanism that can stably identify left atrial regions in areas with blurred boundaries, thereby providing more stable and efficient data support for the diagnosis and treatment of diseases such as atrial fibrillation. Summary of the Invention
[0008] This invention provides an automatic left atrial volume calculation method based on multimodal medical imaging data. It integrates artificial intelligence segmentation technology and image processing algorithms, enabling efficient and accurate segmentation of the left atrial structure and automatic volume calculation even in cases of blurred image boundaries and complex structures. The automatic left atrial volume calculation method includes the following steps:
[0009] Step 1: Use a deep learning model to perform coarse segmentation of the left atrium in medical image data and extract the initial structure containing the left atrium region;
[0010] Step 2: Perform morphological erosion on the initial left atrial structure obtained in Step 1 to remove accessory structures and retain the main morphological features;
[0011] Step 3: Adjust the center point position of the etched structure to align it spatially with the original coarsely segmented structure to obtain the initial expanded structure;
[0012] Step 4: Perform an expansion operation on the initial expansion structure and record the relevant volume change indicators during the expansion process;
[0013] Step 5: Analyze the volume change trend during the expansion process, and determine the optimal expansion termination point based on the analysis;
[0014] Step 6: Stop expansion at the optimal expansion termination point to obtain the final left atrial structure and calculate the left atrial volume.
[0015] Further, in step 1, the medical image data is three-dimensional medical image data containing information about the left atrium of the human body. The image data may be derived from multimodal images, including CT images, MRI images, and ultrasound images. A deep learning model is used to perform coarse segmentation of the left atrium on the image data. The deep learning models used include the SAM general segmentation model based on two-dimensional images, a model based on three-dimensional convolutional neural networks, a pre-trained model based on the U-Net structure, a segmentation model based on the Transformer architecture, and a deep learning segmentation model based on multimodal fusion.
[0016] Furthermore, in step 2, the specific method for progressively eroding the initial left atrial structure includes:
[0017] Step 2.1: The morphological erosion algorithm is used to perform iterative erosion on the coarse segmentation structure of the left atrium obtained in Step 1, gradually removing the pulmonary veins and the left atrial appendage and other accessory structures, thereby extracting the main morphological features of the left atrium.
[0018] Step 2.2: To avoid excessive corrosion, a dynamic termination mechanism is introduced. When the volume of the corroded structure shrinks to the threshold α of the initial left atrial volume, the corrosion operation is automatically stopped.
[0019] Furthermore, in step 3, by calculating the spatial center point of the original left atrial structure and using this center point as a reference, the corroded structure obtained in step 2 is spatially adjusted and moved accordingly so that its center point is aligned with the original left atrial structure. The adjusted center point position is represented by the following formula:
[0020] Cnew =C eroded +ΔC
[0021] Among them, C new C represents the adjusted center point position. eroded ΔC represents the spatial center point of the corroded structure, and ΔC represents the offset that the corroded structure needs to be adjusted relative to the original structure.
[0022] Furthermore, the specific method for performing the expansion operation on the initial expansion structure and recording the relevant volume changes in step 4 includes:
[0023] Step 4.1: Based on the initial expanded structure obtained in Step 3, apply the morphological expansion algorithm to perform a step-by-step expansion operation, increasing the volume of the structure in each iteration. The expanded volume V after the (n+1)th iteration is... n+1 It can be represented as:
[0024] V n+1 =V n +ΔV
[0025] Among them, V n+1 V represents the expansion volume after the current iteration. n Let ΔV be the volume of the previous iteration, and ΔV be the volume increment of the current iteration.
[0026] Step 4.2: During the expansion process, continuously record and update two key metrics:
[0027] a)A voxels This indicates the number of voxels that the current inflated structure exceeds the initial left atrial structure; this indicator reflects the degree of over-inflation.
[0028] b)B voxels : This indicates the number of elements that intersect the current expanded structure with the initial left atrial structure. This index reflects the degree of matching between the expanded structure and the original left atrial structure.
[0029] Step 4.3: For each expansion iteration, calculate and record A. voxels and B voxels The second-order rate of change, and the resulting trend, will be used to determine the optimal expansion termination point.
[0030] Step 4.4: The expansion process continues until a preset termination condition is met or the maximum number of iterations is reached. The termination condition is based on A. voxels and B voxels The setting is based on the changing trend to ensure that the optimal left atrial morphology and structure are captured.
[0031] Furthermore, in step 5, the specific method for determining the optimal expansion termination point by analyzing the volume change during the expansion process includes:
[0032] Step 5.1: Based on A recorded in Step 4 voxels and B voxels The data is used to calculate the second-order growth rate of these two indicators, which reflects the changing trend of the rate of change of these indicators and can more sensitively capture the key inflection points in the expansion process.
[0033] Step 5.2: For each expansion iteration, calculate A. voxels and B voxels The sum of the second-order growth rates, denoted as S, serves as a comprehensive indicator of the rate of structural change, reflecting the overall situation of structural changes during expansion. Its mathematical expression is:
[0034]
[0035] Among them, A voxels and B voxels The two key voxel indices recorded in step 4 are S, which represents the sum of their second-order growth rates and is used to determine the optimal termination point of the expansion.
[0036] Step 5.3: By analyzing the trend of S value changing with expansion iteration, find its minimum point. This minimum point usually corresponds to A. voxels and B voxels The moment of most gradual change, that is, the optimal state of expansion;
[0037] Step 5.4: Determine the iteration point with the smallest S value as the optimal expansion termination point.
[0038] Further, in step 6, the expansion operation is stopped at the optimal expansion termination point, and the final left atrial structure is obtained, specifically including the following steps:
[0039] Step 6.1: Based on the optimal expansion termination point determined in Step 5, terminate the expansion process at the corresponding number of iterations, and extract the expansion structure corresponding to the termination time. This structure represents the optimal left atrial morphology identified by the current algorithm.
[0040] Step 6.2: Perform a logical AND operation between the extracted dilated structure and the initial coarse segmentation result of the left atrium obtained in Step 1 to obtain their intersection, thereby removing the small overflow parts that may be generated during the dilation process;
[0041] Step 6.3: Perform morphological optimization on the obtained intersection structure, including slight smoothing, to eliminate possible jagged edges, improve structural continuity and smoothness, and make the final structure closer to the actual anatomical morphology of the left atrium.
[0042] Step 6.4: Define the optimized structure as the final left atrial structure. This structure retains the main morphological features of the initial segmentation and corrects the boundary details through a dynamic expansion process, thereby obtaining a more accurate left atrial structure morphology.
[0043] Step 6.5: Based on the final left atrial structure, count the number of voxels it contains, and calculate the actual volume of the left atrium by combining the voxel resolution of the image data in three dimensions.
[0044] Furthermore, the present invention also provides an automatic left atrial volume calculation system for implementing the above method, the system comprising:
[0045] The image acquisition module is used to acquire three-dimensional medical image data containing the left atrial region, and the image data includes multiple modal images such as CT, MRI and ultrasound images;
[0046] The segmentation module is used to perform coarse segmentation of the left atrium in the medical image data using a deep learning model and extract the initial structural region.
[0047] The corrosion module is used to perform morphological corrosion operations on the initial structure to remove accessory structures including the pulmonary veins and the left atrial appendage, and to extract the main morphological features of the left atrium.
[0048] The center point adjustment module is used to correct the spatial center point of the corroded structure so that it is aligned with the center of the original structure.
[0049] The expansion module is used to perform morphological expansion operations on the initial expansion structure and record the volume increment, the number of overflow voxels, and the number of cross voxels in each iteration.
[0050] The termination judgment module is used to determine the optimal expansion termination point based on the recorded change trend and its second-order rate of change.
[0051] The volume calculation module is used to extract the left atrial structure at the termination point, optimize the boundary through logic and operations, and calculate the final left atrial volume by combining the number of voxels and spatial resolution.
[0052] Technical effect
[0053] This invention proposes an automatic left atrial volume calculation method that integrates deep segmentation networks, morphological erosion, and dynamic expansion strategies, combined with second-order volume change rate curve analysis, and has the following significant technical advantages:
[0054] 1. Avoiding the dependence and limitations of traditional methods: Unlike existing schemes that rely on complex modeling or multi-period synchronous images, this invention starts from the temporal change trend of the image itself and combines image processing methods to automatically identify the left atrial boundary without the need for additional hardware or external physiological parameters, making it more adaptable.
[0055] 2. Innovative morphological processing and dynamic strategy fusion mechanism: Although existing literature involves the combination of deep learning and morphological operations, this invention introduces for the first time a synergistic strategy of morphological erosion + dynamic expansion + second-order volume change rate analysis, which effectively enhances the structural recognition stability of the blurred boundary region and overcomes the problem that traditional models are prone to distortion in the left atrium and pulmonary vein connection region.
[0056] 3. Applicable to multimodal images, not dependent on specific modal features: This method has good modal versatility and can achieve stable segmentation and volume extraction in various medical imaging such as CT and MRI, with strong clinical adaptability and scalability;
[0057] 4. Combining high automation and personalized analysis: The entire process is completed automatically without human intervention, greatly improving clinical efficiency; at the same time, it retains the ability to analyze individual characteristics, making the results closer to the patient's actual situation and improving the pertinence of the diagnosis and treatment plan.
[0058] In summary, this invention achieves automatic, stable, and accurate measurement of left atrial volume through innovative algorithm combination and image change trend analysis. It effectively solves the problems of equipment dependence, unclear boundaries, model distortion, and high training complexity in traditional methods, and provides strong data support for the clinical diagnosis and follow-up of atrial fibrillation and other heart diseases. Attached Figure Description
[0059] The invention will be better understood by referring to the description of specific embodiments of the invention given in the following figures, and the purpose, details, and features of each step of the invention will become more apparent.
[0060] Figure 1 This is a flowchart illustrating a method for automatically calculating left atrial volume according to the present invention.
[0061] Figure 2 This is a schematic diagram of the left atrium and sternum obtained after coarse segmentation of a medical image using a deep learning model in one embodiment.
[0062] Figure 3 This is a schematic diagram of the main morphological features of the left atrium obtained after morphological etching in one embodiment.
[0063] Figure 4 This is a schematic diagram of the volume changes of the inflatable body and the left atrium recorded during the expansion process in one embodiment.
[0064] Figure 5 This is a schematic diagram of the fine segmentation structure of the left atrium obtained according to the optimal expansion termination point in one embodiment. Detailed Implementation
[0065] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that the embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention. Furthermore, it should be understood that after reading the disclosure of this invention, those skilled in the art can make various modifications or alterations to the invention, and these equivalent forms also fall within the scope of protection defined by this invention.
[0066] This invention proposes an automatic method for calculating left atrial volume, such as... Figure 1 As shown, it includes the following steps:
[0067] Step 1: Acquire 3D medical image data of the patient's heart and use a deep learning model to coarsely segment the left atrium. The result is as follows: Figure 2 As shown, the specific methods include:
[0068] Step 1.1: Acquire 3D medical image data of the patient's heart. This data can come from various types of medical imaging equipment, including but not limited to computed tomography (CT), magnetic resonance imaging (MRI), and ultrasound imaging (US). 3D medical images can comprehensively cover the left atrium and its surrounding anatomical structures, including the pulmonary veins, left atrial appendage, and sternum. This multimodal medical image data provides multi-angle and multi-level information support for subsequent left atrial segmentation, ensuring sufficient accuracy and completeness in the segmentation results. The acquired image data typically includes multiple 2D image slices, which are then reconstructed and processed to obtain 3D image data. This image data will be used as input into a deep learning model for coarse segmentation of the left atrial structure.
[0069] Step 1.2: Coarse segmentation of the left atrium using the SAM (Segment Anything Model) general segmentation model based on 2D images. Specifically, after acquiring 3D medical image data, it is segmented into a series of 2D images, and the open-source SAM model is used for coarse segmentation of the left atrium. As an open-source general segmentation tool, the SAM model has strong adaptability and transfer learning capabilities. Through fine-tuning of medical images, it can quickly adapt to the left atrial segmentation task, greatly reducing the cumbersome model training process in traditional methods. The coarse segmentation of the left atrium can be completed in milliseconds, and the quality of the segmentation results meets clinical needs. The segmentation results are as follows: Figure 2 As shown, the red area represents the coarse segmentation of the left atrium, including accessory structures such as the pulmonary veins and the left atrial appendage, while the blue area represents the sternum, used for auxiliary localization.
[0070] Step 2: The initial left atrial structure is progressively eroded using a morphological algorithm to gradually reduce its volume, thereby removing excess parts and preserving the core structure of the left atrium. The results are as follows: Figure 3 As shown, it specifically includes:
[0071] Step 2.1: A morphological erosion algorithm is used to progressively erode the coarsely segmented left atrial structure obtained in Step 1, gradually removing accessory structures such as the pulmonary veins and left atrial appendage. The erosion operation uses three-dimensional structuring elements, reducing the volume of the left atrial structure in each iteration, specifically as follows:
[0072]
[0073]
[0074] LA represents the initial left atrial structure, and B is the three-dimensional structural element used for erosion. This represents a morphological erosion operation. After each erosion operation, the number of voxels V in the current structure is recorded. n The corrosion process will continue until the set termination conditions are met.
[0075] Step 2.2: To avoid excessive corrosion and preserve the main morphological features of the left atrium, a threshold termination mechanism is introduced. The corrosion operation stops when the volume of the corroded structure decreases to a threshold proportion α of the initial left atrial volume. The threshold α can be determined based on clinical experience or large-scale data statistics, and is usually set to 0.5 to ensure the preservation of the main morphological features of the left atrium, i.e.:
[0076]
[0077] Step 2.3: During the corrosion process, continuously monitor the topological changes of the structure. If structural separation or voids are detected, immediately stop the corrosion and revert to the previous stable state to ensure that the left atrial structure is ultimately preserved as continuous and intact.
[0078] Step 3: Adjust the center point position of the etched structure to obtain the initial expanded structure. This specifically includes:
[0079] Step 3.1: Based on the coarse segmentation results of the left atrial structure obtained in Step 1, calculate the spatial center point C of the original left atrial structure. orig The centroid calculation method is used, and the formula is as follows:
[0080]
[0081] Where (x) i ,y i ,z i ) represents the coordinates of the i-th voxel in the left atrial structure, and N represents the total number of voxels.
[0082] Step 3.2: Using the same centroid calculation method, calculate the spatial center point C of the corroded structure. eroded .
[0083] Step 3.3: Calculate the center point offset ΔC, which describes the spatial difference between the corroded structure and the original structure. The formula is as follows:
[0084] ΔC=C orig -C eroded
[0085] Step 3.4: Translate the eroded structure as a whole by ΔC, aligning its center with the original left atrial structure. The translation operation satisfies the following:
[0086] (x i ′,y i ′,z i ′)=(x i ,y i ,z i )+ΔC
[0087]
[0088] Where (x) i ,y i ,z i (x) represents the coordinates of the i-th voxel before corrosion. i ′,t i ′,z i ′) are the new coordinates after translation. The centroid C of the translated structure is recalculated. new After translation, we have C new =C orig This achieves consistency in the spatial center of gravity.
[0089] Step 3.5: Define the translated eroded structure as the initial expansion structure to prepare for subsequent expansion operations.
[0090] Step 4: Perform an expansion operation on the initial expansion structure and record the relevant volume changes during the expansion process. The second-order rate of change of the expansion index is as follows: Figure 4 As shown, it specifically includes:
[0091] Step 4.1: Based on the morphological dilation algorithm, using the initial dilated structure obtained in Step 3 as a foundation, a spherical structuring element is used to progressively perform the dilation operation. The dilation step size can be set to 2 voxels to balance accuracy and computational efficiency. Specifically, each iteration expands the structure using a spherical structuring element with a radius of 2 voxels. The dilation operation extends the structure boundary outward, gradually increasing the volume. The dilation formula is as follows:
[0092]
[0093] Among them, V n+1 V represents the expansion volume after the current iteration. nLet B(2) be the volume of the previous iteration, where B(2) represents a spherical structure with a radius of 2 voxels, and ⊕ represents the morphological expansion operation used for the left atrial structure V. n Perform an expansion operation.
[0094] Step 4.2: In each iteration of the expansion, continuously record the following two key metrics:
[0095] A voxels This indicates the number of voxels in the expanded structure that exceed the initial coarse segmentation of the left atrium, reflecting the overflow during the expansion process. The calculation formula is:
[0096]
[0097] Among them, S i,j,k L represents the position of a voxel in the expanded structure. i,j,k This indicates the position of a voxel in the initial coarse segmentation of the left atrium. In the formula, (S... i,j,k =1) indicates that the voxel of the inflated structure is 1, while (L i,j,k =0) indicates that the position is empty (unsegmented voxel) in the initial left atrium.
[0098] B voxels This represents the number of intersection elements between the expanded structure and the initial coarse segmentation result of the left atrium, reflecting the degree of matching between the expanded structure and the original structure. The calculation formula is:
[0099]
[0100] Among them, (S) i,j,k =1) and (L i,j,k =1) indicates that the voxel belongs to both the expanded structure and the initial left atrial structure.
[0101] Step 4.3: Record the volume increment ΔV for each iteration during the expansion process. The calculation formula is:
[0102] ΔV=V n+1 -V n
[0103] By recording the volume increment, the impact of the expansion procedure on the overall morphological changes of the left atrium can be better monitored.
[0104] Step 4.4: The expansion operation continues until a preset termination condition is met or the maximum number of iterations is reached. The specific termination condition will be further determined in step 5 based on the changing trend to ensure that the expansion captures the true actual morphology of the left atrium.
[0105] Step 5: By analyzing the volume changes during the expansion process, such as... Figure 4 As shown, the optimal expansion termination point is determined. Specifically, this includes:
[0106] Step 5.1: Based on A recorded in Step 4 voxcls and B voxels The second-order growth rates of these two indicators are calculated separately to capture key trends during the expansion process. The formula for calculating the second-order growth rate is as follows:
[0107]
[0108]
[0109] They are A voxcls and B voxels The first-order growth rate, They are A voxcls and B voxels The second-order growth rate.
[0110] Step 5.2: For each expansion iteration, calculate A. voxels and B voxels The sum of the second-order growth rates is denoted as S. The value of S reflects the overall situation of structural changes during the expansion process, and the calculation formula is as follows:
[0111]
[0112] Step 5.3: By analyzing the changing trend of the S value, find its minimum point. This minimum point usually corresponds to A during the expansion process. voxels and B voxels The moment of least change, i.e., the optimal expansion state, is calculated using the following formula:
[0113] S min =min(S1,S2,…,S) n )
[0114] Step 6: Stop expansion at the optimal termination point to obtain the final left atrial structure, as shown below. Figure 5 As shown, it specifically includes:
[0115] Step 6.1: Based on the optimal expansion termination point determined in Step 5, stop the expansion process and extract the expansion structure corresponding to the termination point as the final left atrial morphology.
[0116] Step 6.2: After obtaining the optimal expansion termination point, perform a logical AND operation between the extracted optimal expansion structure and the left atrial structure obtained from the initial coarse segmentation to obtain their intersection, thus obtaining the final left atrial structure. Let the expansion structure corresponding to the optimal expansion termination point be V. best The initial coarse segmentation structure is V initial The final left atrial structure V final It can be calculated using the following formula:
[0117] V final =V best ∩V initial
[0118] Step 6.3: Morphological optimization of the obtained intersection structure is performed. First, a Gaussian smoothing filter is applied to eliminate the jagged edges. Then, isolated regions are removed through connected component analysis to ensure that the final left atrial structure is smooth and conforms to the actual clinical anatomical morphology.
[0119] Step 6.4: Finally, based on the optimized final left atrial structure, the number of voxels is counted, and combined with the voxel resolution of medical images in three dimensions, the actual volume of the left atrium is calculated, providing accurate quantitative data for clinical diagnosis and treatment. The specific calculation formula is as follows:
[0120] V = N × spacing x ×spacing y ×spacing z
[0121] N is the number of voxels in the final left atrial structure, spacing x , spacing y , spacing z These represent the spatial resolution of the image data in the x, y, and z directions, respectively, in millimeters (mm). The final calculated left atrial volume is V.
Claims
1. A method for automatically calculating left atrial volume, characterized in that, Includes the following steps: Step 1: Use a deep learning model to perform coarse segmentation of the left atrium in medical image data and extract the initial structure containing the left atrium region; Step 2: Perform morphological erosion on the initial left atrial structure obtained in Step 1 to remove accessory structures and retain the main morphological features; Step 3: Adjust the center point position of the etched structure to align it spatially with the original coarsely segmented structure to obtain the initial expanded structure; Step 4: Perform an expansion operation on the initial expansion structure and record the relevant volume change indicators during the expansion process; Step 5: Analyze the volume change trend during the expansion process, and determine the optimal expansion termination point based on the analysis; Step 6: Stop expansion at the optimal expansion termination point to obtain the final left atrial structure and calculate the left atrial volume.
2. The automatic left atrial volume calculation method according to claim 1, in step 1, the medical image data is three-dimensional medical image data containing information about the left atrium of the human body, and the image data comes from multimodal images, including CT images, MRI images and ultrasound images; a deep learning model is used to perform coarse segmentation of the left atrium on the image data, and the deep learning model used includes a SAM general segmentation model based on two-dimensional images, a model based on three-dimensional convolutional neural networks, a pre-trained model based on U-Net structure, a segmentation model based on Transformer architecture, and a deep learning segmentation model based on multimodal fusion.
3. The automatic left atrial volume calculation method according to claim 2, wherein the specific method for progressively eroding the initial left atrial structure in step 2 includes: Step 2.1: The morphological erosion algorithm is used to perform iterative erosion on the coarsely segmented left atrium structure obtained in Step 1, gradually removing the pulmonary veins and the accessory structures of the left atrial appendage, thereby extracting the main morphological features of the left atrium. Step 2.2: To avoid excessive corrosion, a dynamic termination mechanism is introduced. The mechanism terminates when the corroded structure volume shrinks to a threshold value equal to the initial left atrial volume. The corrosion operation will automatically stop when the time is right.
4. The automatic left atrial volume calculation method according to claim 3, in step 3, by calculating the spatial center point position of the original left atrial structure and using this center point as a reference, the corroded structure obtained in step 2 is spatially adjusted and moved accordingly so that its center position is aligned with the original left atrial structure. The adjusted center point position is expressed by the following formula: in, The adjusted center point position. This indicates the location of the spatial center point of the corroded structure. This represents the offset that the corroded structure needs to be adjusted relative to the original structure.
5. The method for automatically calculating the left atrial volume according to claim 4, wherein the specific method for performing an expansion operation on the initial expansion structure and recording the relevant volume changes in step 4 includes: Step 4.1: Based on the initial dilated structure obtained in Step 3, apply the morphological dilation algorithm to perform a step-by-step dilation operation, increasing the volume of the structure in each iteration. The expansion volume after the next iteration It can be represented as: in, This represents the expansion volume after the current iteration. The volume of the previous iteration. This represents the volume increment for this iteration; Step 4.2: During the expansion process, continuously record and update two key metrics: a) This indicates the number of voxels that the current inflated structure exceeds the initial left atrial structure; this indicator reflects the degree of over-inflation. b) : Indicates the number of elements that intersect the current expanded structure with the initial left atrial structure. This index reflects the degree of matching between the expanded structure and the original left atrial structure. Step 4.3: For each expansion iteration, calculate and record the results. and The second-order rate of change, and the resulting trend, will be used to determine the optimal expansion termination point. Step 4.4: The expansion process continues until a preset termination condition is met or the maximum number of iterations is reached. The termination condition is based on... and The setting is based on the changing trend to ensure that the optimal left atrial morphology and structure are captured.
6. The automatic left atrial volume calculation method according to claim 5, wherein in step 5, the specific method for determining the optimal expansion termination point by analyzing the volume change during the expansion process includes: Step 5.1: Based on the information recorded in Step 4 and The data is used to calculate the second-order growth rate of these two indicators, which reflects the changing trend of the rate of change of these indicators and can more sensitively capture the key inflection points in the expansion process. Step 5.2: For each expansion iteration, calculate and The sum of the second-order growth rates, denoted as S, serves as a comprehensive indicator of the rate of structural change, reflecting the overall situation of structural changes during expansion. Its mathematical expression is: in, and These are the two key voxel indicators recorded in step 4. This represents the sum of its second-order growth rates, used to determine the optimal termination point of inflation; Step 5.3: By analyzing the trend of S-value changes with expansion iteration, find its minimum point. This minimum point usually corresponds to... and The moment of most gradual change, that is, the optimal state of expansion; Step 5.4: Determine the iteration point with the smallest S value as the optimal expansion termination point.
7. The automatic left atrial volume calculation method according to claim 6, wherein in step 6, the expansion operation is stopped at the optimal expansion termination point, and the final left atrial structure is obtained, specifically including the following steps: Step 6.1: Based on the optimal expansion termination point determined in Step 5, terminate the expansion process at the corresponding number of iterations, and extract the expansion structure corresponding to the termination time. This structure represents the optimal left atrial morphology identified by the current algorithm. Step 6.2: Perform a logical AND operation between the extracted dilated structure and the initial coarse segmentation result of the left atrium obtained in Step 1 to obtain their intersection, thereby removing the small overflow parts that may be generated during the dilation process; Step 6.3: Perform morphological optimization on the obtained intersection structure, including slight smoothing, to eliminate possible jagged edges, improve structural continuity and smoothness, and make the final structure closer to the actual anatomical morphology of the left atrium. Step 6.4: Define the optimized structure as the final left atrial structure. This structure retains the main morphological features of the initial segmentation and corrects the boundary details through a dynamic expansion process, thereby obtaining a more accurate left atrial structure morphology. Step 6.5: Based on the final left atrial structure, count the number of voxels it contains, and calculate the actual volume of the left atrium by combining the voxel resolution of the image data in three dimensions.
8. An automatic left atrial volume calculation system for implementing the method as described in any one of claims 1 to 7, characterized in that, include: The image acquisition module is used to acquire three-dimensional medical image data containing the left atrial region, and the image data includes multiple modal images such as CT, MRI and ultrasound images; The segmentation module is used to perform coarse segmentation of the left atrium in the medical image data using a deep learning model and extract the initial structural region. The corrosion module is used to perform morphological corrosion operations on the initial structure to remove accessory structures including the pulmonary veins and the left atrial appendage, and to extract the main morphological features of the left atrium. The center point adjustment module is used to correct the spatial center point of the corroded structure so that it is aligned with the center of the original structure. The expansion module is used to perform morphological expansion operations on the initial expansion structure and record the volume increment, the number of overflow voxels, and the number of cross voxels in each iteration. The termination judgment module is used to determine the optimal expansion termination point based on the recorded change trend and its second-order rate of change. The volume calculation module is used to extract the left atrial structure at the termination point, optimize the boundary through logic and operations, and calculate the final left atrial volume by combining the number of voxels and spatial resolution.