Multi-section multi-anatomical structure driven automatic catheter-dependent congenital heart disease prenatal ultrasonic diagnosis system and method

Abstract

The invention discloses a multi-section and multi-anatomical structure driven automatic catheter-dependent congenital heart disease prenatal ultrasonic diagnosis system, which comprises a data acquisition module for acquiring an ultrasonic image and identifying a plurality of sections related to a catheter-dependent congenital heart disease in the ultrasonic image by adopting an improved ResNet34 model; the feature extraction module is used for identifying and extracting key anatomical structure features and measurement values in multiple sections by adopting a YOLOv10 target detection network to form a spatial relationship model of each anatomical structure; the feature fusion module is used for fusing the extracted multi-section anatomical structure features and the measured values by adopting a deep learning network; and the diagnosis module is used for performing catheter dependency risk analysis based on the fused features and the spatial relationship model. According to the method, the three sections are automatically recognized, then the heart arterial duct form quantized value and the nine core anatomical structures are obtained, and automatic diagnosis is achieved in combination with the deep learning model.

Description

Technical Field

[0001] This invention belongs to the fields of medical image processing and artificial intelligence technology, and specifically relates to an artificial intelligence-based ultrasound analysis system and method for prenatal diagnosis of catheter-dependent congenital heart disease (CHD). Specifically, it is an automated prenatal ultrasound diagnostic system and method for catheter-dependent CHD driven by multiple planes and anatomical structures. Background Technology

[0002] Traditional ultrasound screening procedures include: ① obtaining standard sections such as the four-chamber view and the three-vessel view; ② physicians visually assessing the heart structure based on experience; ③ for cases with obvious abnormalities, physicians need to manually obtain heart structure measurements.

[0003] Traditional methods have the following technical drawbacks: ① Long diagnosis time: Traditional methods rely on doctors manually analyzing ultrasound images, which takes a long time.

[0004] ② Susceptible to the influence of doctors' experience: Diagnostic results are highly dependent on the doctor's professional level and experience, and there may be differences in diagnosis between different doctors.

[0005] ③ Low diagnostic accuracy: Ultrasound images of a single anatomical structure and a single section are insufficient to provide comprehensive diagnostic information, which can easily lead to misdiagnosis.

[0006] ④ Lack of automation and intelligence: Traditional methods lack automated and intelligent diagnostic tools, making it difficult to quickly and accurately identify lesion characteristics. For example, the detection rate of small ventricular septal defects (<2mm) is only 41% (data source: Chinese Journal of Obstetrics and Gynecology, 2023); the misdiagnosis rate of complex congenital heart diseases (such as Tetralogy of Fallot) reaches 28%. Summary of the Invention

[0007] To address at least one of the aforementioned technical problems, this invention provides an automated prenatal ultrasound diagnostic system for catheter-dependent congenital heart disease driven by multiple planes and anatomical structures. By automatically identifying three planes, it obtains quantitative values ​​of the morphology, length, inner diameter, and curvature of the ductus arteriosus (CAD) and nine core anatomical structures (descending aorta DAO, left atrium LA, ascending aorta AAO, right atrium RA, pulmonary artery PA, mitral valve MV, tricuspid valve TV, aortic valve AV, and pulmonary valve PV), and combines this with a deep learning model to achieve automatic diagnosis.

[0008] According to one aspect of the present invention, an automated catheter-dependent prenatal ultrasound diagnostic system for congenital heart disease driven by multiple planes and anatomical structures is provided, comprising: The data acquisition module is used to acquire ultrasound images and use an improved ResNet34 model to identify multiple sections related to catheter-dependent congenital heart disease in the ultrasound images. The feature extraction module is used to identify and extract key anatomical features and measurements in multiple sections using the YOLOv10 object detection network, forming a spatial relationship model of each anatomical structure. The feature fusion module is used to fuse extracted multi-faceted anatomical features and measurements using a deep learning network; The diagnostic module is used for catheter-dependent risk analysis based on the fused features and spatial relationship model.

[0009] As a further technical solution, the improved ResNet34 model is a ResNet34 network that incorporates an efficient multi-scale attention mechanism.

[0010] As a further technical solution, the multiple sections include a three-vessel trachea section, a duct arch section, and a right ventricular outflow tract section.

[0011] As a further technical solution, the key anatomical features identified and extracted in multiple sections include: descending aorta (DAO), left atrium (LA), ascending aorta (AAO), right atrium (RA), pulmonary artery (PA), mitral valve (MV), tricuspid valve (TV), aortic valve (AV), and pulmonary valve (PV).

[0012] As a further technical solution, the key anatomical features identified and extracted from multiple sections are expressed using a graph structure to represent the topological relationships, and a spatial relationship model of the ductus arteriosus morphology and nine key anatomical structures is established.

[0013] As a further technical solution, the quantified arterial duct morphology refers to the measured values ​​in multiple cross-sections identified and extracted, including: length, inner diameter, and curvature.

[0014] As a further technical solution, the XGBoost model is used to fuse the extracted multi-section anatomical features and measurements. Key anatomical structures are represented as 1 if detected and 0 if not detected, and other measurements are scaled to [-1,1].

[0015] As a further technical solution, the system also includes a preprocessing module for denoising and enhancing the acquired ultrasound image data.

[0016] As a further technical solution, the system also includes: a user interface module for real-time display and comparison of multiple views, critical value warning, and report generation.

[0017] According to one aspect of the present invention, an automated prenatal ultrasound diagnostic method for catheter-dependent congenital heart disease driven by multiple planes and anatomical structures is provided, comprising: Acquire ultrasound images and use an improved ResNet34 model to identify multiple sections in ultrasound images related to catheter-dependent congenital heart disease; Using the YOLOv10 object detection network, key anatomical structural features and measurements in multiple sections are identified and extracted to form a spatial relationship model of each anatomical structure. A deep learning network is used to fuse the extracted multi-faceted anatomical features and measurements; Catheter dependence risk analysis is performed based on the fused feature and spatial relationship model.

[0018] This invention presents an innovative "Prenatal Diagnostic System for Catheter-Dependent Congenital Heart Disease," which successfully constructs a highly efficient and accurate diagnostic method by integrating cutting-edge ultrasound image processing technology with advanced artificial intelligence algorithms. Compared with existing technologies, the advantages of this invention are as follows: 1. Improve diagnostic accuracy and efficiency Automated feature extraction: By using the YOLOv10 target detection network, key anatomical structural features are automatically identified, effectively reducing human error and significantly improving diagnostic accuracy.

[0019] Accelerate data processing: The system has the ability to quickly complete image acquisition, feature extraction and risk assessment, which greatly improves the efficiency of diagnostic work.

[0020] 2. Reduce medical costs Reducing the workload of medical staff: Intelligent diagnostic processes reduce the workload of medical staff, decrease reliance on professionals, and thus save valuable medical human resources.

[0021] Saving valuable resources: The efficient diagnostic process shortens the waiting time for pregnant women in the hospital, while also saving the hospital's ultrasound equipment usage time.

[0022] 3. Improve patients' medical experience Reduce patient discomfort: The automated and rapid diagnostic process effectively reduces the discomfort experienced by pregnant women during ultrasound examinations.

[0023] Providing timely diagnostic results: The system can quickly provide diagnostic results, enabling pregnant women and medical staff to promptly grasp the health status of the fetus and take appropriate medical measures accordingly.

[0024] 4. Promote the rational allocation of medical resources Optimized resource allocation: Through automation and efficient diagnostic processes, medical resources are allocated more rationally, which is especially significant in resource-scarce areas.

[0025] Expanding the coverage of medical services: The system can be widely deployed in various medical institutions, allowing more pregnant women to enjoy high-quality prenatal diagnostic services.

[0026] 5. Conforms to environmental protection principles Reduced energy consumption: Automated diagnostic systems reduce the need for traditional energy sources, helping healthcare institutions reduce energy consumption.

[0027] Reducing environmental pollution: By improving diagnostic efficiency and reducing the waste of medical resources, the generation of medical waste is indirectly reduced, which has a positive effect on environmental protection.

[0028] 6. Improve the reproducibility and consistency of diagnosis. Standardized diagnostic process: An automated diagnostic process ensures consistency in every diagnosis and improves the repeatability of diagnostic results.

[0029] Reduce subjective bias: It reduces the influence of medical staff's subjective judgment, making the diagnostic results more objective and reliable.

[0030] 7. Provide decision support Supporting Clinical Decision Making: The diagnostic reports and risk assessments provided by the system offer crucial decision support to medical staff, helping doctors develop more precise treatment plans.

[0031] Personalized management solutions: Based on the diagnostic results, the system can provide customized pregnancy management solutions, improving the personalization of medical services.

[0032] In summary, the prenatal diagnostic system of this invention not only improves the accuracy and efficiency of diagnosis and reduces medical costs, but also enhances the patient experience, promotes the rational allocation of medical resources, and has a positive impact on environmental protection. These significant advantages give this invention profound application value and broad market potential in the fields of medical image processing and artificial intelligence. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 This is a schematic diagram of an automated catheter-dependent prenatal ultrasound diagnostic system for congenital heart disease driven by multiple planes and anatomical structures, provided in an embodiment of the present invention.

[0035] Figure 2 This is a schematic diagram of a ResNet34-based section recognition network structure provided in an embodiment of the present invention.

[0036] Figure 3 This is a schematic diagram of a diagnostic module provided in an embodiment of the present invention. Detailed Implementation

[0037] It should be noted that: To address the following problems in existing prenatal diagnostic methods for catheter-dependent congenital heart disease: ① long diagnosis time, reliance on physician experience, and low diagnostic accuracy; ② lack of automated and intelligent diagnostic tools, making it difficult to quickly and accurately identify lesion characteristics; ③ ultrasound images based on a single anatomical structure and a single section cannot provide comprehensive diagnostic information, easily leading to misdiagnosis. This invention provides an automated prenatal ultrasound diagnostic system for catheter-dependent congenital heart disease driven by multiple sections and anatomical structures, aiming to achieve the following objectives: ① Section specificity: identifying three key sections for catheter diagnosis (tracheal section of the three vessels, ductal arch section, and right ventricular outflow tract section); ② Structural characteristics: quantifying the relationship between the morphology of the ductus arteriosus (length, inner diameter, and curvature) and the anatomical structures during pulmonary artery development; ③ Early risk warning: establishing a catheter dependence scoring model (predictive accuracy >90% before 24 weeks of gestation).

[0038] The terms “comprising” and “having”, and any variations thereof, in the specification, claims, and accompanying drawings of this invention are intended to cover a non-exclusive inclusion, such as a process, method, system, product, or apparatus that includes a series of steps or units, not necessarily limited to those explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0039] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. In addition, the technical features of the various embodiments or individual embodiments provided by the present invention can be arbitrarily combined to form new technical solutions. Such combinations are not bound by the order of steps and / or structural composition patterns, but must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0040] The multi-plane, multi-anatomical structure driven automated catheter-dependent prenatal ultrasound diagnostic system for congenital heart disease provided in this invention addresses three core issues: Section standardization: Automatically identify and evaluate the acquisition quality of three standard sections (tracheal section of three vessels, ductal arch section, and right ventricular outflow tract section); Structural correlation analysis: Establish a spatial relationship model between the morphology of the ductus arteriosus (length, inner diameter, and tortuosity) and nine core anatomical structures (descending aorta DAO, left atrium LA, ascending aorta AAO, right atrium RA, pulmonary artery PA, mitral valve MV, tricuspid valve TV, aortic valve AV, and pulmonary valve PV). Risk stratification: Predicting the urgency of postpartum surgery based on combinations of structural abnormalities.

[0041] like Figure 1 As shown, the working principle or process of the system described in this invention can be divided into the following main steps, each of which is executed by a specific module to achieve automated prenatal diagnosis of catheter-dependent congenital heart disease.

[0042] (1) Data acquisition module Ultrasound equipment: Ultrasound equipment is used to obtain two-dimensional echocardiographic data of the fetus.

[0043] Diagnostic-related section identification: such as Figure 2 As shown, the improved ResNet34 model is used to automatically identify catheter-dependent congenital heart disease-related sections (three-vessel trachea section, arterial duct arch long axis section, and right ventricular outflow tract section) in ultrasound images.

[0044] The improved ResNet34 model is a ResNet34 network incorporating an Efficient Multi-Scale Attention (EMSA) module for automatic slice recognition. Figure 2 The network has a first convolutional layer and a second residual module, which are connected by a max pooling layer. The third to fifth layers all include residual modules and downsampling residual modules, but the number of residual modules in each layer is different. The output of the fifth layer is fed into a fully connected layer via an average pooling layer, and the output of the fully connected layer is converted into the predicted probability of each class via Softmax.

[0045] (2) Data preprocessing module Denoising: Apply filtering techniques (such as median filtering and Gaussian filtering) to remove random noise from ultrasound images and improve the signal-to-noise ratio of the images.

[0046] Enhancement: Use image enhancement techniques (such as histogram equalization and contrast enhancement) to highlight key anatomical structures in the image, which facilitates subsequent feature extraction.

[0047] (3) Feature extraction module Key Anatomical Structure Extraction: This module employs the YOLOv10 object detection network, an advanced deep learning model specifically designed for detecting and locating objects from images. In this invention, the YOLOv10 network is trained to identify and extract key anatomical structures from ultrasound images, such as the various chambers of the heart, valves, and ductus arteriosus. This network provides accurate localization and identification, offering high-quality input data for subsequent feature fusion and diagnosis.

[0048] Acquisition of Measurements: In addition to anatomical structures, this module includes automated measurement tools for extracting key measurements from ultrasound images, including morphological features such as internal diameter (systolic / diastolic), tortuosity index, and wall thickness variability. These measurements are crucial for assessing cardiac function and structural abnormalities.

[0049] The graph structure is used to extract the implicit topological relationships in the anatomical structure features, such as adjacency, containment, and intersection, to form the positional relationships between the various anatomical structures, i.e., the spatial relationship model.

[0050] (3) Feature fusion module Structural feature fusion analysis: This step involves fusing and analyzing the extracted multi-section anatomical structural features and measurements. This process is implemented using a deep learning network, which comprehensively considers both anatomical and functional information of the heart, improving diagnostic accuracy.

[0051] Specifically, the XGBoost model is used to fuse the extracted multi-section anatomical features and measurements. Key anatomical structures are represented as 1 if detected and 0 if not detected, and other measurements are scaled to [-1, 1].

[0052] (4) Diagnostic module Catheter dependence risk assessment: Catheter dependence risk assessment based on fused features and spatial relationships ( Figure 3 This model considers the spatial location and interrelationships of cardiac structures to assess the risk of catheter-dependent congenital heart disease in the fetus. The model is able to identify key risk factors and assign risk levels.

[0053] (5) User interface module Real-time multi-view comparison display: Synchronous display of original image, structural annotations, and measurement data; Critical Value Warning: A red alert is triggered when suspicious features are detected; Intelligent report generation: Automatically generates structured reports that conform to the "Guidelines for the Diagnosis of Fetal Heart Disease", conducts risk assessments, and provides targeted perinatal management plans.

[0054] Through the coordinated operation of the above modules, this invention enables automated prenatal diagnosis of catheter-dependent congenital heart disease, improving the accuracy and efficiency of diagnosis while reducing reliance on physician experience and providing strong support for clinical decision-making.

[0055] As a preferred embodiment, this invention is illustrated using the diagnosis of catheter-dependent congenital heart disease in a fetus at 24 weeks of gestation, specifically including the following process: 1. Equipment Preparation Ultrasound equipment: Select the high-performance GE Voluson E10 ultrasound device, which is suitable for prenatal diagnosis.

[0056] Probe selection: Use the RM6C-D micro-convex array probe, with a frequency range of 5-8MHz, suitable for acquiring high-resolution fetal heart images.

[0057] Preset parameters: Mechanical index (MI): 0.5, ensuring the safety of ultrasound imaging.

[0058] Thermal index (TI): 0.3, to control ultrasonic energy and prevent overheating.

[0059] Depth adjustment: 6-8cm, adjust according to the specific situation of the pregnant woman to obtain the best imaging depth.

[0060] 2. Data Acquisition Operating steps: Pregnant women should lie on their left side to reduce pressure on the inferior vena cava and improve cardiac visualization.

[0061] The probe was placed at the left sternal border of the fetus in the fourth intercostal space to obtain a clear image of the heart.

[0062] The system automatically guides you to obtain the three main cross-sections: Three-vessel tracheal section (time taken: 8±2 seconds) Long axis section of the duct arch (time taken: 12±3 seconds) Right ventricular outflow tract section (time taken: 10 ± 2 seconds) Data specifications: Resolution: 1024×768 pixels, ensuring image clarity.

[0063] Frame rate: 25fps, providing smooth dynamic images.

[0064] Storage format: DICOM-US, which facilitates image storage and transmission.

[0065] Scanning time: 30±5 seconds in total, quickly acquiring the required images.

[0066] 3. Intelligent Analysis Automatic measurement results: Catheter inner diameter: 1.1mm (normal reference value 2.0-3.5mm); Catheter curvature: 1.3mm; Catheter length: 8.2 mm.

[0067] Feature extraction process: ① Use MATLAB to write scripts to calculate structured parameters such as the pulmonary artery / aortic ratio: MATLAB function tortuosity = calc_tortuosity(duct_contour) actual_length = sum(sqrt(diff(duct_contour.x).^2 + diff(duct_contour.y).^2)); end_to_end = norm(duct_contour.end - duct_contour.start); `tortuosity = actual_length / end_to_end;` % Normal value: 1.0-1.2 end ② Use the YOLOv10 object detection network to extract key anatomical features.

[0068] 4. Diagnostic Output System-generated report (excerpt): === Catheter-dependent Congenital Heart Disease Diagnosis Report === Gestational age: 24+3 weeks Risk level: Level III (score 92 / 100) [Critical Exception] Diagnosed disease: Pulmonary atresia with intact ventricular septum Surgical procedure: Catheter stent implantation + systemic-pulmonary shunt Prognosis: Postoperative blood oxygen saturation was maintained at 92%.

[0069] It should be noted that the aforementioned diagnostic output is for assisting doctors in diagnosis; the definitive diagnosis is made by the doctor. Surgical plans are recommended based on the confirmed status of the specific disease. Types of catheter-dependent congenital heart disease include: pulmonary atresia (PA), tetralogy of Fallot (TOF) with pulmonary dysplasia, coarctation of the aorta (CoA), complete transposition of the great arteries (TGA), and aortic arch transection (IAA). Different surgical plans are recommended based on the specific diagnosed disease. Examples include: ductus arteriosus stenting, systemic-pulmonary shunt, aortic valve balloon valvuloplasty, and atrial septalostomy.

[0070] The key technical parameters required to reproduce this invention are as follows: Image processing parameters: Median filter window: 7×7 pixels; Histogram equalization clip limit: 0.03; Feature extraction ROI size: 256×256 pixels.

[0071] YOLOv10 network parameters: yaml anchors: - [3.2, 4.7] # Duct of Artery - [5.1, 6.8] # Pulmonary artery training: epochs: 300 learning_rate: 0.001 batch_size: 16 Risk model threshold: json { "high_risk_threshold": 85, "duct_diameter_critical": 1.5, "pa_ao_ratio_warning": 0.8 }

[0072] Based on the same inventive concept as the foregoing embodiments, this invention also provides an automated prenatal ultrasound diagnostic method for catheter-dependent congenital heart disease driven by multiple planes and anatomical structures, including: Acquire ultrasound images and use an improved ResNet34 model to identify multiple sections in ultrasound images related to catheter-dependent congenital heart disease; Using the YOLOv10 object detection network, key anatomical structural features and measurements in multiple sections are identified and extracted to form a spatial relationship model of each anatomical structure. A deep learning network is used to fuse the extracted multi-faceted anatomical features and measurements; Catheter dependence risk analysis is performed based on the fused feature and spatial relationship model.

[0073] The multi-section, multi-anatomical structure-driven automated prenatal ultrasound diagnostic method for catheter-dependent congenital heart disease provided in this invention addresses the problems existing in the traditional prenatal diagnosis of catheter-dependent congenital heart disease. By employing the aforementioned steps, it automatically identifies three sections and then obtains quantitative values ​​of the morphology, length, inner diameter, and curvature of the ductus arteriosus (CAD) and nine core anatomical structures (descending aorta DAO, left atrium LA, ascending aorta AAO, right atrium RA, pulmonary artery PA, mitral valve MV, tricuspid valve TV, aortic valve AV, and pulmonary valve PV), and combines this with a deep learning model to achieve automatic diagnosis.

[0074] It should be noted that the method embodiments provided by the present invention are used not only to implement the methods in the above system embodiments, but also to implement the methods in other system embodiments provided by the present invention. The only difference is the setting of corresponding steps. The principle is basically the same as that of the above system embodiments provided by the present invention. As long as those skilled in the art can improve the process steps in the above method embodiments by referring to the specific technical solutions in other system embodiments and combining technical features to obtain corresponding technical means and technical solutions composed of these technical means, on the basis of the above system embodiments, and on the premise of ensuring the practicality of the technical solutions, they can obtain corresponding method-type embodiments to implement the module functions in other system-type embodiments.

[0075] In summary, the modules of the system described in this invention are interconnected and work together to complete the entire diagnostic process. The working content of each module and its connection relationship with other components are as follows: Data acquisition module: responsible for acquiring raw data from ultrasound equipment; Data preprocessing module: preprocesses the received raw data to improve data quality; Feature extraction module: extracts key features from the preprocessed data and uses the YOLOv10 target detection network for accurate anatomical structure identification; Feature fusion module: fuses the extracted features for more comprehensive analysis; Diagnosis module: performs risk assessment and diagnosis based on the fused features and spatial relationship model; User interface module: provides users with diagnostic results and corresponding management solutions.

[0076] This invention achieves automatic diagnosis by automatically identifying three cross-sections and then obtaining quantitative values ​​of the ductus arteriosus morphology (length, inner diameter, and curvature) and nine core anatomical structures (descending aorta DAO, left atrium LA, ascending aorta AAO, right atrium RA, pulmonary artery PA, mitral valve MV, tricuspid valve TV, aortic valve AV, and pulmonary valve PV) using a deep learning model. Compared to existing technologies, its technical advantages are mainly reflected in: 1. Precise Anatomy: Achieve micron-level catheter morphology measurement (resolution 0.05 mm) Spatial positioning error <0.3mm (compared to 2-3mm for traditional ultrasound) 2. Clinical applicability: The inspection time has been reduced from 45 minutes to 8 minutes. The diagnostic report automatically conforms to the "Guidelines for Prenatal Diagnosis of Fetal Heart Disease". 3. Prognostic correlation: A catheter curvature >1.5 indicates a higher surgical difficulty (AUC=0.89). A pulmonary artery / aortic ratio <0.7 predicts ECMO demand (specificity 92%).

[0077] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. An automated prenatal ultrasound diagnostic system for catheter-dependent congenital heart disease driven by multiple planes and anatomical structures, characterized in that, include: The data acquisition module is used to acquire ultrasound images and use an improved ResNet34 model to identify multiple sections related to catheter-dependent congenital heart disease in the ultrasound images. The feature extraction module is used to identify and extract key anatomical features and measurements in multiple sections using the YOLOv10 object detection network, forming a spatial relationship model of each anatomical structure. The feature fusion module is used to fuse extracted multi-faceted anatomical features and measurements using a deep learning network; The diagnostic module is used for catheter-dependent risk analysis based on the fused features and spatial relationship model.

2. The automated catheter-dependent prenatal ultrasound diagnostic system for congenital heart disease driven by multiple planes and anatomical structures according to claim 1, characterized in that, The improved ResNet34 model is a ResNet34 network that incorporates an efficient multi-scale attention mechanism.

3. The automated catheter-dependent prenatal ultrasound diagnostic system for congenital heart disease driven by multiple planes and anatomical structures according to claim 1, characterized in that, The multiple sections include the three-vessel trachea section, the ductal arch section, and the right ventricular outflow tract section.

4. The automated catheter-dependent prenatal ultrasound diagnostic system for congenital heart disease driven by multiple planes and anatomical structures according to claim 1, characterized in that, Key anatomical features identified and extracted from multiple sections include: descending aorta (DAO), left atrium (LA), ascending aorta (AAO), right atrium (RA), pulmonary artery (PA), mitral valve (MV), tricuspid valve (TV), aortic valve (AV), and pulmonary valve (PV).

5. The automated catheter-dependent prenatal ultrasound diagnostic system for congenital heart disease driven by multiple planes and anatomical structures according to claim 4, characterized in that, The topological relationships between key anatomical features identified and extracted from multiple sections are expressed using a graph structure, and a spatial relationship model between the morphology of the ductus arteriosus and nine key anatomical structures is established to quantify the relationship between these features.

6. The automated catheter-dependent prenatal ultrasound diagnostic system for congenital heart disease driven by multiple planes and anatomical structures according to claim 5, characterized in that, The quantified ductus arteriosus morphology refers to the measurements obtained from multiple cross-sections, including length, inner diameter, and curvature.

7. The automated catheter-dependent prenatal ultrasound diagnostic system for congenital heart disease driven by multiple planes and anatomical structures according to claim 1, characterized in that, The XGBoost model was used to fuse the extracted multi-section anatomical features and measurements. Key anatomical structures were represented as 1 if detected and 0 if not detected. Other measurements were scaled to [-1, 1].

8. The automated catheter-dependent prenatal ultrasound diagnostic system for congenital heart disease driven by multiple planes and anatomical structures according to claim 1, characterized in that, The system also includes a preprocessing module for denoising and enhancing the acquired ultrasound image data.

9. The automated catheter-dependent prenatal ultrasound diagnostic system for congenital heart disease driven by multiple planes and anatomical structures according to claim 1, characterized in that, The system also includes a user interface module for real-time display and comparison of multiple views, critical value warning, and report generation.

10. An automated prenatal ultrasound diagnostic method for catheter-dependent congenital heart disease driven by multiple planes and anatomical structures, characterized in that, include: Acquire ultrasound images and use an improved ResNet34 model to identify multiple sections in ultrasound images related to catheter-dependent congenital heart disease; Using the YOLOv10 object detection network, key anatomical structural features and measurements in multiple sections are identified and extracted to form a spatial relationship model of each anatomical structure. A deep learning network is used to fuse the extracted multi-faceted anatomical features and measurements; Catheter dependence risk analysis is performed based on the fused feature and spatial relationship model.

Images