Ultrasonic contrast time intensity curve quantitative analysis method, system and medium

Abstract

The invention discloses an ultrasonic contrast time intensity curve quantitative analysis method and system and a medium, and relates to the technical field of medical image ultrasonic contrast. Comprising the following steps: acquiring an ultrasound contrast dynamic image sequence of a tumor focus; delimiting a focus ROI (Region of Interest) of the dynamic image sequence; drawing an ultrasound contrast time intensity curve by taking time as an abscissa and the gray value of each frame of image in the focus ROI region as an ordinate; using an encoder to extract the features of the ultrasound contrast time intensity curve, and using a decoder to reconstruct the features of the ultrasound contrast time intensity curve into a smooth time intensity curve; identifying key time nodes of the smooth time intensity curve through an intelligent inflection point detection algorithm; acquiring an intensity parameter of the smooth time intensity curve through a self-adaptive segmentation integration technology; carrying out quantitative analysis on the ultrasound contrast time intensity curve based on the key time node and the intensity parameter of the corresponding tumor focus; according to the invention, the efficiency and accuracy of time intensity curve quantitative analysis are improved.

Description

Technical Field

[0001] This invention relates to the field of medical imaging ultrasound contrast imaging technology, and in particular to a method, system and medium for quantitative analysis of ultrasound contrast imaging time-intensity curves. Background Technology

[0002] Contrast-enhanced ultrasound (CEUS) is a safe, non-invasive, and real-time technology that has been widely used in the clinical diagnosis of tumors such as thyroid cancer and breast cancer. It involves intravenously injecting a contrast agent containing microbubbles, utilizing the nonlinear vibration of these microbubbles to enhance blood flow echo signals, clearly revealing the blood perfusion status of lesions, and providing crucial information for determining the nature of the lesion.

[0003] Traditional ultrasound contrast imaging analysis relies on physicians' subjective experience and qualitative judgment, resulting in poor diagnostic consistency, lack of standardized quantitative indicators, low efficiency, and poor reproducibility. While time-intensity curve (TIC) analysis, as a mainstream quantitative method, can extract objective parameters such as time to peak and peak intensity, traditional mathematical models are insufficiently accurate in fitting complex curves, are susceptible to noise and artifacts, and parameter calculations lack unified standards and often require manual operation, making it difficult to meet clinical needs.

[0004] Deep learning technology offers new insights into solving these challenges, but its research in intelligent TIC reconstruction, precise parameter extraction, and standardized analysis remains in the early stages, lacking mature and systematic solutions. Therefore, developing automated and accurate TIC reconstruction and quantitative analysis methods is of significant clinical value in improving the accuracy and standardization of tumor ultrasound contrast-enhanced diagnosis. Summary of the Invention

[0005] The purpose of this invention is to provide a method, system, and medium for quantitative analysis of ultrasound contrast imaging time-intensity curves, which improves the efficiency and accuracy of quantitative analysis of time-intensity curves.

[0006] To address the aforementioned technical problems, embodiments of the present invention provide a method for quantitative analysis of ultrasound contrast-enhanced imaging time-intensity curves, comprising the following steps: Acquire dynamic ultrasound contrast imaging sequences of different tumor lesions; delineate the region of interest (ROI) of the lesion in the dynamic image sequence; plot the ultrasound contrast imaging time-intensity curve with time as the x-axis and the grayscale value of each frame within the ROI region as the y-axis. The ultrasound contrast imaging time-intensity curve is reconstructed using a pre-trained reconstruction model to obtain a smooth time-intensity curve. The reconstruction model is a dual-channel structure of encoder and decoder. The encoder extracts the temporal features of the ultrasound contrast imaging time-intensity curve, and the decoder reconstructs the temporal features of the ultrasound contrast imaging time-intensity curve into a smooth time-intensity curve. The key time nodes of the smoothed time intensity curve are identified by an intelligent inflection point detection algorithm; the curve integrals of the rising segment, peak segment and falling segment of the smoothed time intensity curve are calculated by an adaptive piecewise integration technique to obtain the intensity parameters of the smoothed time intensity curve. Based on the tumor lesion threshold, the corresponding key time points, and the intensity parameters, a scoring model is constructed to quantitatively analyze the ultrasound contrast imaging time-intensity curve.

[0007] In some optional embodiments, the reconstruction model is an encoder-decoder dual-path structure. The encoder includes three convolutional layers with ReLU activation functions and two self-attention layers. Each convolutional layer integrates a residual block, which includes two convolutional layers and a batch normalization layer. The decoder includes convolutional layers and self-attention layers symmetrical to the encoder. The outputs of each layer of the encoder are added to the corresponding layer inputs of the decoder through residual connections, followed by a fully connected layer.

[0008] In some optional embodiments, the reconstruction model loss function includes a main loss term, a smoothness constraint term, and a physical constraint term. The main loss term is an improved mean squared error loss with a time weighting factor introduced. The smoothness constraint term is a constraint term based on the curve derivative. The physical constraint term is used to ensure that the reconstructed smooth time intensity curve conforms to hemodynamic characteristics.

[0009] In some optional embodiments, the key time points of the time-intensity curve include peak time, rise time, and enhancement duration; the intensity parameters include peak intensity, baseline intensity, and maximum enhancement intensity.

[0010] In some optional embodiments, the calculation of the key time points and intensity parameter scores corresponding to the tumor lesions specifically includes: The random forest algorithm learns key time points and intensity parameter data corresponding to different tumor lesions, and establishes a scoring model for key time points and intensity parameters corresponding to tumor lesion thresholds. Using the trained random forest algorithm, the scores of the key time points and intensity parameters of the corresponding tumor lesions are calculated, and the ultrasound contrast imaging time-intensity curve is quantitatively analyzed.

[0011] Embodiments of the present invention also provide a computer device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the above-described quantitative analysis method of ultrasound contrast imaging time-intensity curves.

[0012] Embodiments of the present invention also provide a computer-readable storage medium storing a computer program, which, when run by a processor, is capable of executing the above-described quantitative analysis method for ultrasound contrast imaging time-intensity curves.

[0013] The quantitative analysis method for ultrasound contrast imaging time-intensity curves provided by this invention has at least the following beneficial effects: This invention, through a reconstruction model using encoders and decoders, preserves details when extracting temporal features from ultrasound contrast-enhanced imaging time-intensity curves. The reconstructed smooth curves avoid noise and artifact interference. Combined with intelligent inflection point detection and adaptive piecewise integration technology, it accurately extracts key time nodes and intensity parameters, solving the problems of insufficient fitting accuracy and large parameter extraction errors in traditional methods. Simultaneously, it automates the entire process from dynamic ultrasound contrast-enhanced image acquisition and ROI delineation to curve reconstruction, parameter extraction, and quantitative analysis, reducing reliance on manual labor and subjective errors. Standardized procedures ensure consistency and repeatability of analysis results across different operators and institutions. The reconstruction model is highly adaptable to complex curve morphologies, meeting the analysis needs of various tumor lesions and providing objective evidence for clinical identification, evaluation, and follow-up. The automated process significantly shortens analysis time, supports large-scale case processing, alleviates diagnostic and treatment pressure, and is suitable for primary hospitals and multi-center studies, effectively improving accuracy, standardization, clinical applicability, and analytical efficiency. Attached Figure Description

[0014] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0015] Figure 1 This is a schematic flowchart of a quantitative analysis method for ultrasound contrast imaging time-intensity curves according to an embodiment of the present invention; Figure 2 This is a schematic diagram of an encoder-decoder dual-channel structure according to an embodiment of the present invention; Figure 3 This is a schematic diagram of a thyroid pathology type quantitative analysis report generation system according to an embodiment of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0017] One embodiment of the present invention relates to a method for quantitative analysis of ultrasound contrast imaging time-intensity curves. The implementation details of the method for quantitative analysis of ultrasound contrast imaging time-intensity curves in this embodiment are described in detail below. The following content is only for the convenience of understanding and is not necessary for implementing this solution.

[0018] The specific procedure for quantitative analysis of ultrasound contrast-enhanced time-intensity curves in this embodiment can be described as follows: Figure 1 As shown, it includes: Step 101: Obtain dynamic ultrasound contrast imaging sequences of different tumor lesions; delineate the region of interest (ROI) of the lesion in the dynamic image sequence; plot the ultrasound contrast imaging time-intensity curve with time as the x-axis and the grayscale value of each frame within the ROI region as the y-axis. Standardized ultrasound contrast imaging is used to standardize contrast agent injection, equipment parameter settings, and image acquisition operations. The ultrasound contrast imaging process and patient information are recorded to obtain high-quality dynamic image sequences of tumor lesions (such as thyroid cancer, breast cancer, etc.).

[0019] Regarding ultrasound imaging equipment, a high-end ultrasound diagnostic instrument with a resolution of ≥1024×768 pixels should be selected to ensure the imaging effect of capturing small blood vessels and providing sufficient image details, avoiding the impact on quantitative analysis due to unclear display of small blood vessels caused by low resolution; at the same time, an imaging software package should be equipped to achieve low mechanical index imaging and provide a dual-frame contrast display mode for convenient real-time observation and comparison; a dedicated imaging probe with a frequency range of 3-9 MHz should also be used. This frequency is suitable for imaging superficial thyroid tissues and can obtain good spatial resolution while ensuring penetration depth. Too high a frequency will lead to insufficient penetration, while too low a frequency will affect the resolution.

[0020] Contrast agent use must follow guidelines. The second-generation ultrasound contrast agent SonoVue is selected, which has better stability and longer duration of action compared to the first generation. Its microbubble size is uniform (average diameter 2.5μm), which is conducive to obtaining stable enhancement effects. The standard dose of 2.4ml is administered via bolus injection through the antecubital vein. This dose has been verified as optimal by numerous clinical studies, ensuring adequate contrast enhancement while avoiding image oversaturation due to excessive contrast agent. Immediately after injection, the area should be rapidly flushed with 5ml of normal saline to ensure that the contrast agent completely enters the circulatory system, reducing local retention and obtaining a more accurate time-intensity relationship.

[0021] There are also specific requirements for the preparation of the injection system. A 22G intravenous catheter must be selected, preferably the median cubital vein on the right side, and it should be fixed with a special transparent dressing. At the same time, the limb position should be adjusted so that the straight angle is maintained at 15-30°. The puncture site should be disinfected twice with povidone-iodine and then once with alcohol.

[0022] During the data acquisition process, patients should maintain an appropriate position and strictly control the mechanical index (MI < 0.1). After the injection of the standard dose of contrast agent, dynamic image sequences of no less than 180 seconds should be continuously acquired, with a sampling frequency maintained at 15-10 frames / second or higher, to fully record the entire process of the contrast agent inflow, peak, and decay phases. In addition, the time point of contrast agent injection should be recorded simultaneously, and the patient's age, gender, medical history, and other clinical data should be preserved to provide a basis for subsequent analysis.

[0023] Imaging data preprocessing is a crucial step in ensuring the accuracy of subsequent analysis, encompassing a multi-dimensional processing workflow. It involves optical flow compensation, filtering and noise reduction, and contrast correction for dynamic image sequences, followed by quality assessment.

[0024] First, raw bit depth detection is performed, covering 12 / 14 / 16-bit cases, and then the formula is used. Linear mapping to 8-bit grayscale is then performed. Next, histogram equalization is applied, setting the block size to 8×8 pixels and the contrast limit threshold to 0.02, using bilinear interpolation for optimization. Finally, brightness normalization is implemented to control the target average brightness within... The standard deviation range is maintained at 45-55, which standardizes the basic grayscale and brightness performance of the image.

[0025] Then conduct a noise assessment, Gaussian noise assessment with The local variance was calculated using a window, and the noise variance estimation threshold was set to 0.01. Shot noise was assessed by fitting a Rayleigh distribution and estimating parameters using the maximum likelihood method. Based on the assessment results, a hybrid noise model was constructed, employing a Gaussian-Rayleigh hybrid model, and parameter estimation was completed using the EM algorithm. This approach accurately suppressed interference such as speckle noise in ultrasound images, enhancing image quality.

[0026] Using a dense optical flow algorithm, set the window size. The system uses a pyramid layer count of 4, an iteration count of 3, and a precision threshold of 0.03. This optical flow algorithm can eliminate image drift caused by respiratory motion. Combined with the aforementioned grayscale conversion, denoising, and enhancement steps, the system completes the preprocessing of dynamic image sequences for thyroid ultrasound contrast imaging, laying a solid foundation for subsequent analysis. Quality assessment is required for each step, and images that fail to meet the standards are re-acquired or reprocessed.

[0027] The processed dynamic image sequence of tumor lesions was divided into high, medium and low grayscale layers by the improved Otsu multi-threshold segmentation algorithm. After optimization by morphological opening and closing operations, the initial ROI of the lesions was completed based on the quantization enhancement characteristics. Using a pre-trained YOLOv5 model to The system detects thyroid lesions in dynamic image sequences of pixels. Based on the initial location of the lesion ROI, it generates lesion ROI proposal boxes in areas with a confidence level greater than 0.85, while taking into account temporal continuity to prevent abrupt changes in the location of the lesion ROI. Based on the lesion ROI suggestion box, the optimal lesion ROI shape is determined by an adaptive ellipse fitting algorithm. The lesion ROI area is automatically adjusted to 40-100mm² with the boundary distance from the lesion 2-3mm. The shape factor and eccentricity are calculated during drawing to ensure measurement standardization.

[0028] A reference region of interest (ROI) was selected from the normal thyroid tissue on the same side as the thyroid lesion, ensuring that its area and depth were equal to or different from the lesion's ROI. Within, ensure the difference in signal strength during the baseline period. ; In the lesion ROI suggestion box, the optimal lesion ROI shape is used. Based on the reference ROI, the optical flow algorithm is used to track the lesion ROI in real time. The cumulative deviation exceeds 2mm and automatic correction is performed. The standard deviation of the signal intensity is monitored. If abnormal fluctuations are found, manual intervention and adjustment are prompted.

[0029] Calculate the average grayscale value of the lesion ROI per frame, plot the original time intensity curve (TIC), smooth the original TIC, record key time points, and perform quality control.

[0030] Step 102: The ultrasound contrast imaging time-intensity curve is reconstructed using a pre-trained reconstruction model to obtain a smooth time-intensity curve. The reconstruction model is a dual-channel structure of encoder and decoder. The encoder extracts the temporal features of the ultrasound contrast imaging time-intensity curve, and the decoder reconstructs the temporal features of the ultrasound contrast imaging time-intensity curve into a smooth time-intensity curve. The reconstruction model is a dual-path encoder-decoder structure. The encoder contains three convolutional layers with ReLU activation functions and two self-attention layers. Each convolutional layer integrates a residual block, which includes two convolutional layers and a batch normalization layer. The decoder contains convolutional layers and self-attention layers symmetrical to the encoder. The outputs of each layer of the encoder are added to the corresponding layer inputs of the decoder through residual connections, followed by a fully connected layer.

[0031] A deep sequence neural network model was pre-trained on large-scale simulated data to learn the basic TIC morphology, and then fine-tuned using real case data. During the training process, dynamic learning rate adjustment was applied, and the learning rate was gradually reduced in the later stage of training to optimize the TIC fitting accuracy, while data augmentation techniques were added. Ensemble learning is used to combine deep sequence neural network models from multiple training stages and fuse the prediction results through weighted averaging; cross-validation is used to evaluate the generalization ability of the deep sequence neural network model and training parameters are dynamically adjusted according to the learning curve; an early stopping mechanism is designed to terminate training when the validation set loss has not improved for several consecutive rounds; an incremental learning module and a contrastive learning strategy are designed on the deep sequence neural network model. The reconstruction model loss function includes a main loss term, a smoothness constraint term, and a physical constraint term. The main loss term is an improved mean squared error loss with a time weight factor introduced. The smoothness constraint term is a constraint term based on the curve derivative. The physical constraint term is used to ensure that the smoothed time intensity curve after reconstruction conforms to hemodynamic characteristics.

[0032] The original time-intensity curve (TIC) is input into a trained deep sequence neural network model, and the fitted continuous TIC is output. The fitting accuracy loss (measuring the difference between the predicted value and the actual value), smoothness loss (ensuring the continuity of the fitted curve), and physical constraint loss (ensuring that the curve conforms to hemodynamic characteristics) are comprehensively considered.

[0033] Step 103: Identify key time nodes of the smoothed time intensity curve using an intelligent inflection point detection algorithm; calculate the curve integrals of the rising segment, peak segment, and falling segment of the smoothed time intensity curve using an adaptive piecewise integration technique to obtain the intensity parameters of the smoothed time intensity curve. The calculation process for quantitative analysis parameters is as follows: A smart inflection point detection algorithm is used to identify key time nodes in continuous TIC (Transient Intracorporeal Cell) events, and an adaptive threshold method is used to determine the contrast agent arrival time; the peak time is determined by finding the maximum value point of continuous TIC; the slope change of the descending segment of continuous TIC is analyzed to determine the contrast agent clearance time; an intensity parameter calculation method based on interval integration is used, and the area under the curve of continuous TIC is calculated using adaptive piecewise integration technology; numerical integration is performed on the ascending segment, peak period, and descending segment of continuous TIC using the trapezoidal rule; simultaneously, the reference TIC is normalized to obtain a relative enhancement intensity value that can be used for comparison; finally, the maximum peak value, peak time, infiltration time, outfiltration time, and area under the curve are obtained.

[0034] A multi-validation mechanism is established to assess parameter reliability, and measurement errors are analyzed using Monte Carlo simulation. The system automatically detects outliers in the curves and identifies artifacts caused by respiratory motion or probe pressure changes. Measurement stability is evaluated by calculating the coefficient of variation of the parameters, and data with low reliability are marked and flagged.

[0035] In terms of quantitative parameter calculation, based on the fitted curve, multidimensional quantitative parameters are automatically calculated, mainly including key time nodes such as peak time, rise time and enhancement duration; and intensity parameters such as peak intensity, baseline intensity and maximum enhancement intensity.

[0036] Step 104: Calculate the key time points and intensity parameter scores for the corresponding tumor lesions, and perform quantitative analysis on the ultrasound contrast imaging time-intensity curve based on the threshold corresponding to the tumor lesion.

[0037] The random forest algorithm learns key time points and intensity parameter data corresponding to different tumor lesions, and establishes a scoring model for key time points and intensity parameters corresponding to tumor lesion thresholds. Using the trained random forest algorithm, the scores of the key time points and intensity parameters of the corresponding tumor lesions are calculated, and the ultrasound contrast imaging time-intensity curve is quantitatively analyzed.

[0038] Based on the quantitative analysis of the time-intensity curves, a standardized analysis report is generated, which includes the original image sequence, time-intensity curve graph, quantitative parameter table, statistical analysis results, and diagnostic suggestions.

[0039] The complete analysis system provided by this invention includes multiple functional modules: an image processing module responsible for preprocessing and ROI analysis; a deep learning module performing curve fitting and parameter extraction; a results analysis module conducting statistical analysis and generating reports; a data management module managing case information and analysis results; a user interface module providing interactive operations and results display; a system maintenance module undertaking system updates and optimizations; and a remote access module supporting remote diagnosis and data sharing.

[0040] This invention relates to a deep neural network-based ultrasound contrast-enhanced imaging time-intensity curve fitting and quantitative analysis system. As an integrated hardware and software platform, it mainly consists of several parts: an ultrasound contrast-enhanced imaging data acquisition system for acquiring high-quality dynamic image sequences of thyroid ultrasound contrast imaging, supporting DICOM format data import and standardized processing; a high-performance computing platform equipped with a professional GPU accelerator for real-time execution of deep neural network models and complex image processing tasks; an image processing module including motion compensation, noise suppression, and contrast enhancement preprocessing functions to ensure image quality; an ROI analysis system providing intelligent region of interest annotation and tracking tools, supporting the drawing and automatic tracking of various ROI shapes; a deep learning analysis platform deploying a dedicated neural network model to achieve intelligent time-intensity curve fitting and parameter extraction; a quantitative parameter calculation module automatically calculating multi-dimensional quantitative indicators; a results display system providing visualization methods such as curve graphs, parameter tables, and statistical charts; a report generation module automatically generating standardized analysis reports and supporting custom templates; and a data management system for secure storage and efficient management of patient information, examination records, and analysis results. The system adopts a modular design architecture, with each functional module connected through standard interfaces to ensure scalability and maintainability.

[0041] This invention also provides an intelligent tumor lesion diagnostic report generation system, the specific process of which is as follows: The tumor lesion diagnostic system adopts a modular design, integrating the examination information of patients with tumor lesions, including a basic information module for patients with tumor lesions, an image acquisition parameter module, and a quantitative parameter analysis result module; The image acquisition parameter module is based on a deep learning model to analyze the morphological features of tumor lesions and, combined with the dynamic image sequence of tumor lesions acquired by ultrasound contrast imaging equipment, generates descriptive text that conforms to radiological terminology standards, and uses a knowledge graph to ensure the logical integrity of the content.

[0042] The system features an intelligent module for generating quantitative parameter analysis results, which can convert continuous TIC and quantitative parameters into intuitive and visual charts. The system automatically labels key time points and quantitative parameter values, generating charts comparing normal parameter ranges and highlighting abnormal indicators. Gradient colors are used to indicate different risk levels, providing a reference for diagnosis.

[0043] The tumor lesion diagnostic system is an evidence-based diagnostic reasoning system that integrates clinical guidelines and expert experience. It uses decision tree algorithms to analyze quantitative parameters and dynamic image sequences of tumor lesions to generate diagnostic and follow-up recommendations.

[0044] The tumor lesion diagnosis system establishes a multi-level quality control system, conducts terminology standardization checks, logical verification and completeness assessments, monitors key information omissions and anomalies in real time, and sets up manual review and modification to ensure report quality.

[0045] The tumor lesion diagnostic system automatically selects a descriptive template based on different pathological types, generating a tumor lesion diagnostic report that includes dynamic image sequences of the tumor lesion, continuous TICs, quantitative parameter tables, statistical analysis results, and diagnostic suggestions. The analysis results are stored in DICOM format in the hospital system, supporting data integration with other clinical systems.

[0046] Time-intensity curve quantitative analysis report generation system, such as Figure 3 As shown in the figure, the process involves first acquiring contrast-enhanced ultrasound imaging data (S101), preprocessing and quality control of the images (S102), selecting the ROI region to obtain the TIC curve (S103), then visualizing and describing the TIC curve (S104), fitting the curve using a deep learning network (S105), and quantitatively analyzing the parameters of the fitted curve (S106). Finally, the system is optimized through clinical validation (S107) to achieve a complete analysis from data acquisition to clinical application validation.

[0047] This invention underwent systematic application evaluation and comparative analysis in clinical practice. Through multi-center clinical studies, the diagnostic efficacy and practical value of the system were verified. The implementation process was as follows: A prospective controlled study design was adopted, including 200 patients with benign thyroid nodules, 150 patients with malignant thyroid cancer, and 100 normal controls from the ultrasound outpatient departments of three tertiary hospitals (all cases were diagnosed by surgical pathology or fine-needle aspiration biopsy). Standardized examination procedures were established to ensure data collection consistency and comparability. In a double-blind controlled manner, three chief physicians with over 10 years of experience in ultrasound departments performed routine qualitative analysis, while the system performed quantitative analysis. ROC curve analysis was used to assess diagnostic efficacy, and sensitivity and other indicators were calculated. The statistical significance of differences in AUC values ​​was evaluated to complete the diagnostic performance assessment; examination and report generation times and physician satisfaction scores were recorded; physician feedback on user experience was collected through questionnaires; a follow-up database was established to track the clinical outcomes of diagnostic recommendations; and clinical application value analysis was conducted. Thirty typical cases were selected for repeated analysis to calculate the coefficient of variation of key quantitative parameters, analyze the measurement consistency among different operators and equipment, assess reliability using intragroup correlation coefficients, analyze the causes of cases with large deviations and optimize the system, and verify system stability; the improvement in examination efficiency and savings in labor costs were statistically analyzed to assess the contribution to diagnostic accuracy and optimization of medical resources; cost-benefit analysis demonstrated the promotional value, achieving economic benefit assessment.

[0048] This invention represents a groundbreaking innovation in the field of contrast-enhanced ultrasound time-intensity curve analysis. It is the first to apply deep neural network technology to contrast-enhanced ultrasound image analysis, developing a specialized hybrid neural network architecture capable of simultaneously processing spatial feature information and temporal data. This network model extracts image features through a multi-layer convolutional neural network and combines it with a bidirectional long short-term memory network to process temporal relationships, achieving accurate fitting of complex time-intensity curves. The innovatively designed loss function comprehensively considers fitting accuracy, curve smoothness, and physical constraints, ensuring the accuracy and reliability of the analysis results. In terms of performance, this invention offers significant advantages over traditional methods: fitting accuracy is improved by more than 20%, the average processing time is reduced to less than 1.3 seconds, and noise resistance is greatly enhanced. Particularly when processing complex time-intensity curves, this invention exhibits strong adaptability, accurately capturing various characteristic changes in the curve. Furthermore, this invention establishes a complete quantitative parameter system, including multiple dimensions such as time parameters, intensity parameters, dynamic parameters, and area parameters, providing comprehensive quantitative indicators for clinical diagnosis. The innovative adaptive algorithm significantly improves the repeatability of the analysis results, controlling the coefficient of variation to within 5%, providing reliable data support for clinical diagnosis.

[0049] This invention has outstanding practical value in clinical applications. The system adopts a modular design, constructing a complete analysis platform including functional modules such as data management, image processing, deep learning analysis, and result display. The intuitive graphical user interface significantly lowers the barrier to entry, allowing doctors to quickly master the operation without requiring specialized computer knowledge. The system supports DICOM format image import and is compatible with mainstream ultrasound equipment on the market, ensuring good versatility. In practical applications, the system can automatically complete complex operations such as image preprocessing, ROI tracking, and curve fitting, significantly reducing manual steps and improving work efficiency. The standardized analysis process effectively reduces the influence of subjective factors, ensuring good consistency in analysis results among different operators. The system also supports batch processing, allowing simultaneous analysis of multiple cases, greatly improving diagnostic efficiency. The generated analysis reports are standardized in format and comprehensive in content, including important information such as quantitative parameters, fitted curves, and statistical results, facilitating clinical decision-making and case management for doctors. Furthermore, the system supports custom report templates, allowing for flexible adjustment of report formats according to the needs of different hospitals.

[0050] This invention boasts excellent system scalability and technological advancement. In terms of system architecture, it adopts an advanced microservice design concept, with functional modules communicating through standard interfaces, facilitating functional expansion, upgrades, and maintenance. The deep learning model employs a modular design, allowing for easy updates to network structure and training parameters, supporting online learning and model optimization. The system provides abundant secondary development interfaces, enabling researchers to add new analytical functions and quantitative parameters. The data management module supports importing and exporting various data formats, facilitating data exchange and sharing with other medical information systems. The system also features remote upgrade capabilities, pushing model updates and functional optimizations via the network to ensure its continued advancement. Regarding security, the system employs multi-layered data encryption and access control mechanisms to ensure the security of patient information and medical data. Furthermore, the system supports multi-center collaborative research, facilitating data integration and statistical analysis, providing strong support for clinical research. The technological advancement and system scalability provide a solid foundation for the continued development of this technology.

[0051] This invention offers significant economic benefits and important social value. From an economic perspective, the system significantly improves the efficiency of ultrasound diagnostic equipment, reduces examination time, increases diagnostic efficiency, and directly lowers medical costs. The automated analysis process reduces the workload of doctors, allowing them to focus on more cases or more complex diagnostic problems. The widespread application of the system helps improve the diagnostic level of primary hospitals, reduces redundant examinations, and saves medical resources. From a social value perspective, this invention improves the accuracy of thyroid disease diagnosis, facilitates early detection and intervention of thyroid diseases, and improves patient prognosis. The standardized analysis methods provide reliable technical support for multi-center clinical research, promoting the development of medical research. The system's remote diagnostic function helps achieve a more balanced distribution of medical resources, allowing patients in remote areas to enjoy high-quality medical services.

[0052] The steps of the various methods described above are only for clarity. In practice, they can be combined into one step or some steps can be split into multiple steps. As long as they include the same logical relationship, they are all within the protection scope of this invention. Adding insignificant modifications or introducing insignificant designs to the algorithm or process, without changing the core design of the algorithm and process, are also within the protection scope of this invention.

[0053] Another embodiment of the present invention relates to a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the method embodiments described above.

[0054] That is, those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0055] Those skilled in the art will understand that the above embodiments are specific embodiments for implementing the present invention, and in practical applications, various changes can be made to them in form and detail without departing from the spirit and scope of the present invention.

Claims

1. A method for quantitative analysis of ultrasound contrast-enhanced imaging time-intensity curves, characterized in that, The method includes: Acquire dynamic ultrasound contrast imaging sequences of different tumor lesions; delineate the region of interest (ROI) of the lesion in the dynamic image sequence; plot the ultrasound contrast imaging time-intensity curve with time as the x-axis and the grayscale value of each frame within the ROI region as the y-axis. The ultrasound contrast imaging time-intensity curve is reconstructed using a pre-trained reconstruction model to obtain a smooth time-intensity curve. The reconstruction model is a dual-channel structure of encoder and decoder. The encoder extracts the temporal features of the ultrasound contrast imaging time-intensity curve, and the decoder reconstructs the temporal features of the ultrasound contrast imaging time-intensity curve into a smooth time-intensity curve. The key time nodes of the smoothed time intensity curve are identified by an intelligent inflection point detection algorithm; the curve integrals of the rising segment, peak segment and falling segment of the smoothed time intensity curve are calculated by an adaptive piecewise integration technique to obtain the intensity parameters of the smoothed time intensity curve. Based on the tumor lesion threshold, the corresponding key time points, and the intensity parameters, a scoring model is constructed to quantitatively analyze the ultrasound contrast imaging time-intensity curve.

2. The quantitative analysis method for ultrasound contrast-enhanced imaging time-intensity curves as described in claim 1, characterized in that, The reconstruction model is a dual-path encoder-decoder structure. The encoder contains three convolutional layers with ReLU activation functions and two self-attention layers. Each convolutional layer integrates a residual block, which includes two convolutional layers and a batch normalization layer. The decoder contains convolutional layers and self-attention layers symmetrical to the encoder. The outputs of each layer of the encoder are added to the corresponding layer inputs of the decoder through residual connections, followed by a fully connected layer.

3. The quantitative analysis method for ultrasound contrast-enhanced imaging time-intensity curves as described in claim 1, characterized in that, The reconstruction model loss function includes a main loss term, a smoothness constraint term, and a physical constraint term. The main loss term is an improved mean squared error loss with a time weighting factor introduced. The smoothness constraint term is a constraint term based on the curve derivative. The physical constraint term is used to ensure that the smoothed time intensity curve after reconstruction conforms to hemodynamic characteristics.

4. The quantitative analysis method for ultrasound contrast-enhanced imaging time-intensity curves as described in claim 1, characterized in that, The key time points of the time-intensity curve include peak time, rise time, and enhancement duration; the intensity parameters include peak intensity, baseline intensity, and maximum enhancement intensity.

5. The quantitative analysis method for ultrasound contrast-enhanced imaging time-intensity curves as described in claim 1, characterized in that, The calculation of the key time points and intensity parameter scores corresponding to the tumor lesions specifically includes: The random forest algorithm learns key time points and intensity parameter data corresponding to different tumor lesions, and establishes a scoring model for key time points and intensity parameters corresponding to tumor lesion thresholds. Using the trained random forest algorithm, the scores of the key time points and intensity parameters of the corresponding tumor lesions are calculated, and the ultrasound contrast imaging time-intensity curve is quantitatively analyzed.

6. A computer system, characterized in that, include: At least one processor; And a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the quantitative analysis method of ultrasound contrast imaging time-intensity curves as described in any one of claims 1 to 5.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, is capable of performing any of the quantitative analysis methods for ultrasound contrast imaging time-intensity curves as defined in claims 1 to 5.

Images