A cross-data heterogeneity intelligent optimization method, system and device based on data quality dynamic decision

Abstract

The present application relates to a kind of cross data heterogeneity intelligent optimization method, system and device based on data quality dynamic decision-making. Mainly include: S1, data collection and pretreatment, S2, data analysis and imaging quality division are carried out to the data after pretreatment, form training data set, S3, model construction and training, model includes a main network and multiple decoders, main network is used to extract general feature, multiple decoders are used to process the specific task of different quality data, the training of main network is unified using multiple training data sets of different quality joint training, the training of decoder is one-to-one training to single quality training data set, S4, the data set to be optimized is input into the model after training, by real-time analysis the pixel value distribution and eigenvalue distribution of input data, the most suitable decoder is automatically matched to output optimization. The method is conducive to subsequent model to carry out stable and accurate segmentation or classification to image.

Description

Technical Field

[0001] This invention relates to the field of image processing, and in particular to intelligent optimization methods, systems and apparatus for cross-data heterogeneity based on dynamic decision-making for data quality. Background Technology

[0002] In recent years, the development of artificial intelligence (AI) technology has brought revolutionary changes to the field of image analysis, especially the widespread application of deep learning in image classification and segmentation. However, with the increasing complexity of real-world applications, particularly in joint modeling of multiple datasets and processing heterogeneous data, existing technologies face significant challenges in terms of adaptability, generalization ability, and fine-grained processing. Modern image processing systems need to perform various tasks on image data from different domains. These tasks not only include common classification and segmentation but also require the ability to perform personalized optimization and intelligent task allocation when dealing with heterogeneous data.

[0003] Traditional single-task image processing models often face challenges such as large differences in cross-domain data features and insufficient model generalization ability when jointly modeling on multiple datasets, resulting in poor performance when processing heterogeneous datasets. Furthermore, with the gradual application of deep learning in medical imaging, the heterogeneity and acquisition differences of medical data place higher demands on image analysis techniques. Medical images often exhibit significant data heterogeneity due to differences in acquisition equipment, regional variations, and doctors' operating habits, such as… Figure 3 and Figure 4 As shown, this makes it difficult for existing image segmentation and classification models to process these heterogeneous data stably and accurately, thus affecting the accuracy and efficiency of medical diagnosis. Summary of the Invention

[0004] Based on this, a cross-data heterogeneity intelligent optimization method based on dynamic decision-making for data quality is proposed. This method is beneficial for improving the performance of image classification and segmentation, and facilitates stable and accurate image processing by subsequent models.

[0005] A cross-data heterogeneity intelligent optimization method based on dynamic data quality decision-making includes:

[0006] S1. Data collection and preprocessing

[0007] S2. Perform data analysis and image quality classification on the preprocessed data to form a training dataset.

[0008] S3, Model Construction and Training

[0009] The model consists of a backbone network and multiple decoders. The backbone network is used to extract general features, while the multiple decoders are used to process task-specific data of different quality. The backbone network is trained jointly using multiple training datasets of different quality, while the decoders are trained one-to-one on a single training dataset of different quality.

[0010] S4. Input the dataset to be optimized into the trained model, and use dynamic feature selection technology to automatically match the most suitable decoder for output by analyzing the pixel value distribution and feature value distribution of the input data in real time.

[0011] In one embodiment, S5 is also included, which evaluates the performance of the model and displays the evaluation results.

[0012] In one embodiment, different post-processing models are applied to data of different qualities.

[0013] In one embodiment, step S4 further includes a user operation platform for processing abnormal data. When data in the dataset to be optimized cannot be matched with any decoder, the data is considered abnormal. When abnormal data occurs, the user operation platform generates an alarm signal and retains the abnormal data.

[0014] In one embodiment, in the model, each decoder automatically adjusts its weights and task assignments based on its performance in different tasks.

[0015] In one embodiment, step S1 involves preprocessing the data using a data augmentation strategy. The data includes fetal ultrasound datasets or cardiac ultrasound datasets.

[0016] In one embodiment, step S2, data analysis and imaging quality classification includes: performing in-depth analysis on the preprocessed data and classifying the data quality according to image clarity, contrast, and the completeness of diagnostic information.

[0017] In one embodiment, in step S1, the data is medical image data.

[0018] In one embodiment, step S4 further includes: if the user needs to select a decoder independently, the user can independently select a decoder of appropriate quality.

[0019] In one embodiment, a decoder of appropriate quality is also included, which can be selected by the user.

[0020] A computer storage medium storing at least one executable instruction that causes a processor to perform operations corresponding to the intelligent optimization method for cross-data heterogeneity based on dynamic decision-making for data quality.

[0021] A computer device includes a processor, a memory, a communication interface, and a communication bus. The processor, memory, and communication interface communicate with each other via the communication bus. The memory stores at least one executable instruction, which causes the processor to perform an operation corresponding to the intelligent optimization method for cross-data heterogeneity based on dynamic decision-making for data quality.

[0022] The beneficial effects of this application are as follows:

[0023] This application significantly improves the performance of image classification and segmentation through adaptive optimization and multi-task collaboration. Employing a multi-decoder architecture and a dynamic weight allocation mechanism, the model possesses higher generalization ability, can efficiently adapt to datasets from different domains, and exhibits excellent stability in handling diverse tasks. A cross-level feature computation strategy ensures consistency in the feature transfer process, improving the model's convergence and robustness, and reducing the risk of overfitting and underfitting. The dynamic weight allocation mechanism enables the model to adaptively select optimal parameters, achieving personalized optimization for specific tasks and significantly improving task processing accuracy. Joint evaluation and dynamic optimization among multi-decoders achieve collaborative learning among decoders, effectively improving the model's overall performance and continuous optimization capabilities in classification and segmentation tasks.

[0024] In this method, the user platform can accurately display which port the input image is output to, and an alarm will be issued if no suitable port is available. In the field of medical imaging, this application reduces data errors, improves the standardization of analysis, shortens diagnostic time, and supports remote diagnosis when processing highly heterogeneous data, significantly improving diagnostic efficiency and accuracy. This application demonstrates significant advantages and superior performance compared to existing technologies in terms of adaptability, accuracy, and performance in multi-tasking environments. Attached Figure Description

[0025] Figure 1 This is a flowchart of the intelligent optimization method for cross-data heterogeneity based on dynamic decision-making for data quality, as described in this application.

[0026] Figure 2 This is a schematic diagram of the model used in the cross-data heterogeneity intelligent optimization method based on dynamic decision-making for data quality in this application, wherein the model has a backbone network and multiple decoders.

[0027] Figure 3These are fetal ultrasound datasets, arranged from top to bottom as follows: European fetal ultrasound dataset, African fetal ultrasound dataset, and Macau, China fetal ultrasound dataset. Figure 3 It is evident that differences in ultrasound equipment and the skill level of ultrasound physicians lead to variations in ultrasound data quality.

[0028] Figure 4 Two publicly available four-chamber echocardiography datasets, by Figure 4 It is evident that differences in ultrasound equipment and the skill level of ultrasound physicians lead to variations in ultrasound data quality. Detailed Implementation

[0029] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0030] like Figure 1 As shown, the first embodiment of this application provides an intelligent optimization method for cross-data heterogeneity based on dynamic decision-making for data quality, the method comprising:

[0031] S1. Data Collection and Preprocessing: Data preprocessing employs data augmentation strategies.

[0032] S2. Perform data analysis and image quality classification on the preprocessed data to form a training dataset. This data analysis and image quality classification includes: performing in-depth analysis on the preprocessed data and classifying the data quality according to image sharpness, contrast, and the completeness of diagnostic information.

[0033] S3, Model Construction and Training

[0034] The model consists of a backbone network and multiple decoders. The backbone network is used to extract general features, while the multiple decoders are used to process task-specific data of different quality. The backbone network is trained jointly using multiple training datasets of different quality, while the decoders are trained one-to-one on a single training dataset of different quality.

[0035] S4. Input the dataset to be optimized into the trained model. Employing dynamic feature selection technology, the system automatically matches the most suitable decoder for output optimization by analyzing the pixel and feature value distributions of the input data in real time. Each decoder automatically adjusts its weights and task allocation based on its performance in different tasks. Furthermore, if the user needs to manually select a decoder, they can choose one of suitable quality. Additionally, the system includes a user operation platform for handling anomalous data. When data in the dataset to be optimized cannot be matched with any decoder, it is considered anomalous. Upon encountering anomalous data, the user operation platform generates an alarm signal and retains the anomalous data.

[0036] S5 evaluates the model's performance and displays the evaluation results.

[0037] It should be noted that the data mentioned in step S1 of this application can be medical image data. In step S4, the data output by different decoders are processed using different post-processing models. It also includes allowing the user to choose a decoder of appropriate quality; for example, if there are a total of 3 decoders, and the user only needs to use 2 of them, then the user can choose to use 2 of the decoders.

[0038] The following details the intelligent optimization method for cross-data heterogeneity based on dynamic decision-making for data quality described in this application.

[0039] 1. Data collection and preprocessing:

[0040] Multiple datasets were acquired, including fetal ultrasound datasets from Europe, Africa, and Macau, China, for downstream classification tasks. A four-chamber echocardiogram dataset constructed by two research institutions in the United States and France was used for downstream segmentation tasks.

[0041] For image datasets with highly heterogeneous centers, preprocessing operations are first performed, including image cropping and label unification. In addition, various data augmentation strategies such as random flipping, rotation, and brightness adjustment are applied to increase data diversity and complexity, thereby improving the model's generalization ability.

[0042] 2. Data Analysis and Image Quality Classification:

[0043] In-depth analysis of the preprocessed data, combined with image quality assessment, is performed to classify the dataset quality according to image sharpness, contrast, and the completeness of diagnostic information.

[0044] 3. Model Construction and Training:

[0045] The model is constructed using a multi-head network framework, comprising a backbone network and multiple decoders. The backbone network is used to extract general features, while the multiple decoders are used to handle specific tasks with varying data quality. The choice of the backbone network is relatively broad; for the fetal ultrasound classification task, the VIT model was chosen as the backbone network; for the cardiac ultrasound segmentation task, the UNet model was chosen as the backbone network. For example... Figure 2 As shown, a multi-decoder-based USHydraNet framework is designed to detect the quality of input images and select an appropriate decoder for corresponding processing based on the image quality. Figure 2 In this context, NM represents a common and general network model, and D... i,i∈{1，2，3} The decoders represent different data qualities; for example, D1 represents low-quality images, D2 represents medium-quality images, and D3 represents high-quality images. Specifically, USHydraNet dynamically selects the most suitable decoder for corresponding operations by calculating the pixel-level data distribution of the images before they enter the neural network model and the feature-level data distribution after they pass through the neural network, combining the pixel-level and feature-level distributions. For example, in a dataset with 1500 test images, the first port (first decoder) outputs 200 images, the second port (second decoder) outputs 300 images, and the third port (third decoder) outputs 1000 images. The first port represents low-quality images, the second port represents medium-quality images, and the third port represents high-quality images. From the output results, it can be seen that the dataset has a high proportion of high-quality images, and this dataset can be defined as a high-quality dataset, which can then be processed using a suitable post-processing model. Similarly, if a dataset is a low-quality dataset, a dedicated post-processing model can be matched to it for processing. Furthermore, as an improvement, in addition to outputting the specific number of images, different ports can also output the specific codes of the images, with each image code corresponding to a unique image. In this way, the images corresponding to the image codes output by different ports can be used to form different datasets, which can then be matched with different trained post-processing models for targeted processing.

[0046] 4. Dynamic feature selection:

[0047] During the model's inference process, a dynamic feature selection technique is employed. By analyzing the pixel value distribution and feature value distribution of the input data in real time, the most suitable decoder is automatically matched for output optimization. Furthermore, to avoid the problem of input data not belonging to the range of any decoder, a user operation platform is added to handle abnormal data.

[0048] 5. Evaluation and Results Presentation:

[0049] The overall performance of the model is comprehensively evaluated using multi-dimensional performance metrics, including accuracy, precision, recall, and F1 score. Visualization tools such as confusion matrices, PR curves, and ROC curves are employed to demonstrate the model's performance across various tasks. Combined with dynamic evaluation results, this helps identify the model's strengths and limitations. In the results presentation phase, the visualization of feature maps and attention mechanisms provides an in-depth analysis of the model's prediction process, further supporting model improvement and optimization, especially in demanding applications such as medical imaging, ensuring the interpretability and diagnostic value of the prediction results.

[0050] This invention, through a multi-decoder architecture and dynamic weight allocation mechanism, enables the model to possess high adaptability and generalization ability, effectively handling cross-domain datasets and diverse tasks. Simultaneously, it employs cross-level feature computation and dynamic feature selection strategies to ensure consistency during feature transfer, resulting in stronger robustness and convergence. Through joint evaluation and dynamic optimization of multiple decoders, this invention achieves knowledge sharing and adaptive task allocation among decoders, improving the accuracy and efficiency of task collaborative processing. It particularly demonstrates continuous optimization capabilities in multi-task environments and features a well-functioning human-computer interface, displaying which output port is being used. If there is no output, an alarm is triggered, indicating which images cannot be classified into high, medium, or low quality categories (e.g., images with too low or too high quality cannot be output, and an automatic alarm will sound).

[0051] Furthermore, this invention demonstrates significant advantages in the field of medical imaging. Through a dynamic decision-making mechanism, it reduces errors caused by differences in data acquisition, improves the standardization of analysis and diagnostic efficiency, supports remote cross-border diagnosis, and facilitates the efficient utilization of medical resources. This invention combines cutting-edge technologies such as multi-task learning, dynamic feature selection, and flexible weight optimization, breaking through the bottlenecks of existing technologies and achieving intelligent optimization and refined processing under complex data conditions. It possesses broad application potential and significant performance improvements.

[0052] Table 1 presents the evaluation results of applying the method of this application (Hydra-vit, an improvement on the Vit model), comparing it with existing Vit models and existing Mixup-CIFAR10 models, after processing image data from different regions. Here, European represents European image data, African represents African image data, and Macao China represents Macao image data. The evaluation results show that the method of this application delivers significant improvements. Compared to the original Vit model, the method of this application achieves improvements on multiple metrics across multiple datasets and outperforms existing Mixup-CIFAR10 methods.

[0053] Table 1

[0054]

[0055] 1. This application surpasses traditional parallel processing methods for multiple problems by innovatively constructing a parallel multi-decoder architecture based on a combination of neural network models related to cardiac ultrasound and prenatal ultrasound. This architecture not only efficiently performs classification and segmentation tasks on multi-domain image datasets but is also specifically optimized for adapting to images of different quality. By introducing a multi-decoder structure, the model can dynamically adjust its processing strategy according to the quality of the input image, thereby effectively improving its performance on both low-quality and high-quality images and significantly enhancing its adaptability and generalization ability across domain tasks.

[0056] 2. Compared to existing technologies, this application classifies images of different qualities into low-quality, medium-quality, and high-quality images by comparing their quality based on relevant image feature indicators such as sharpness and contrast. A flexible dynamic port allocation mechanism allows the model to automatically select the optimal port, enabling personalized processing of datasets of different quality after segmentation. The multi-decoder structure further enhances the fine-grained processing capabilities for specific tasks and enables self-learning and knowledge sharing among decoders, effectively improving the model's task specificity and performance stability.

[0057] 3. In the feature extraction stage, this application employs a cross-level and cross-dimensional statistical feature strategy, calculating the mean and variance stepwise from the input image to the model feature layer, and considering relevant indicators such as contrast and sharpness. Simultaneously, the model is evaluated on datasets of the same quality to ensure data distribution consistency during feature mapping. Building upon this, this application further introduces a quality scoring mechanism, combining the distribution distance of image features to comprehensively evaluate and select the most suitable decoder. This globally consistent feature transfer mechanism enables the model to automatically adapt to different decoders based on image quality and distribution characteristics, thereby improving feature adaptation capability and ensuring efficient execution of cross-domain tasks.

[0058] 4. To achieve refined task processing across multiple domains and datasets, this application proposes a joint evaluation and progressive optimization strategy for multiple decoders. The performance of each decoder is monitored in real time, and the dynamic output of classification or segmentation ports of different qualities is optimized. The classification and segmentation ports each include low, medium, and high quality ports. To further improve the accuracy of task processing, this application sets one port for each classification and segmentation task, and each port is further subdivided into three sub-ports corresponding to images of different qualities (low, medium, and high). The model can automatically adapt to decoders of different qualities, and users can also choose the appropriate quality decoder according to their needs. In terms of image quality evaluation, the model scores images by comprehensively considering indicators such as clarity, contrast, and signal-to-noise ratio for the heart and fetal organs, and divides the images into three quality levels. Based on these quality scores, this application trains decoders corresponding to data of different qualities, thereby ensuring that each decoder can achieve optimal performance within its adapted quality range. This strategy can progressively optimize the performance of each decoder, effectively improving the overall performance of the model in classification and segmentation tasks.

[0059] 5. Addressing the heterogeneity and complexity of data acquisition in the field of medical imaging, this application provides a groundbreaking solution for medical image analysis. By combining dynamic weight allocation with the refined processing capabilities of multiple decoders, this application can effectively reduce data errors caused by individual physician habits, equipment differences, and regional factors, significantly improving data standardization. Simultaneously, this application supports remote cross-border diagnosis, enabling intelligent task allocation and adaptive data processing, greatly reducing the diagnostic burden on physicians.

[0060] A second embodiment of this application also provides a computer storage medium storing at least one executable instruction that causes a processor to perform the operation corresponding to the intelligent optimization method for cross-data heterogeneity based on dynamic decision-making for data quality.

[0061] A third embodiment of this application also provides a computer device, including: a processor, a memory, a communication interface, and a communication bus. The processor, the memory, and the communication interface communicate with each other through the communication bus. The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the intelligent optimization method for cross-data heterogeneity based on dynamic decision-making for data quality.

[0062] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A cross-data heterogeneity intelligent optimization method based on dynamic decision-making for data quality, characterized in that, include: S1. Data collection and preprocessing S2. Perform data analysis and image quality classification on the preprocessed data to form a training dataset. S3, Model Construction and Training The model consists of a backbone network and multiple decoders. The backbone network is used to extract general features, while the multiple decoders are used to process task-specific data of different quality. The backbone network is trained jointly using multiple training datasets of different quality, while the decoders are trained one-to-one on a single training dataset of different quality. S4. Input the dataset to be optimized into the trained model and use dynamic feature selection technology to automatically match the most suitable decoder for output by analyzing the pixel value distribution and feature value distribution of the input data in real time. The feature values ​​are extracted by using a cross-level and cross-dimensional statistical feature strategy in the feature extraction stage, by calculating the mean and variance of each level from the input image to the model feature layer, and taking into account contrast and sharpness related indicators. It also includes S5, which evaluates the model's performance and displays the evaluation results. In step S1, data augmentation strategies are used to preprocess the data, which includes fetal ultrasound datasets or cardiac ultrasound datasets. Step S4 also includes a user operation platform, which is used to handle abnormal data. When data in the dataset to be optimized cannot match any decoder, the data is considered abnormal. When abnormal data occurs, the user operation platform generates an alarm signal and retains the abnormal data. In the model, each decoder automatically adjusts its weights and task allocation based on its performance in different tasks. The performance of each decoder is monitored in real time, and the dynamic output of classification or segmentation ports of different qualities is optimized. Classification and segmentation tasks are respectively equipped with classification and segmentation ports, each containing low, medium, and high sub-ports corresponding to low, medium, and high image qualities, respectively, to improve the accuracy of task processing. In step S2, data analysis and imaging quality classification includes: performing in-depth analysis on the preprocessed data, and classifying the data quality according to image sharpness, contrast, and the completeness of diagnostic information. In step S4, different decoders are used to process data of different qualities. Step S4 also includes: if the user needs to select a decoder independently, the user can choose a decoder of appropriate quality.

2. A computer storage medium, characterized in that, The computer storage medium stores at least one executable instruction that causes the processor to perform the operation corresponding to the cross-data heterogeneity intelligent optimization method based on dynamic decision-making for data quality as described in claim 1.

3. A computer device, characterized in that, include: The system includes a processor, a memory, a communication interface, and a communication bus. The processor, memory, and communication interface communicate with each other through the communication bus. The memory stores at least one executable instruction, which causes the processor to perform the operation corresponding to the cross-data heterogeneity intelligent optimization method based on dynamic decision-making for data quality as described in claim 1.

Images