An internet hospital-based breast benign and malignant tumor ai-assisted diagnosis and treatment system

Abstract

The present application relates to the technical field of medical image processing, in particular to a breast benign and malignant tumor AI auxiliary diagnosis and treatment system based on an internet hospital, which comprises receiving breast image data flow, monitoring the total quantity of image quantity, judging whether it exceeds the upper limit of system processing capacity, generating image applicability score by using a lightweight AI model, analyzing artifact distribution characteristics to construct a dynamic model and calculating interference degree index, deciding whether to start an optimization process, generating image acquisition complex factors, calculating priority value and sorting output processing sequence. The present application can dynamically perceive system load, accurately evaluate image quality, specifically process artifacts, intelligently decide optimization process, scientifically divide image processing priority, improve the flexibility and adaptability of breast image diagnosis, and meet diversified medical needs.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of medical artificial intelligence, and particularly relates to an AI auxiliary diagnosis and treatment system for breast benign and malignant tumors based on an Internet hospital. BACKGROUND

[0002] In the diagnosis and treatment of breast diseases, the diagnostic accuracy of breast X-ray and ultrasound images plays a crucial role in the development of treatment plans for patients. In the traditional mode, the evaluation of image quality is mostly completed manually, and doctors need to check whether the images have problems such as artifacts, incomplete markers or non-standard body positions one by one. This process not only consumes a large amount of time and labor cost, but also is unstable in evaluation effect due to fatigue or experience difference when the number of images is large. With the popularization of the Internet hospital mode, the amount of breast image data uploaded by patients or primary medical institutions has increased significantly, which has improved the diagnosis and treatment efficiency to a certain extent.

[0003] The prior art still has obvious limitations in processing breast images. In terms of image quality, the uploaded image data often shows great differences due to factors such as shooting equipment and operation standardization. For example, some molybdenum target films may not completely contain the pectoral muscle, or there may be motion artifacts; the ultrasound images may have problems such as probe focus offset and improper gain setting. These factors will interfere with the diagnosis results of the subsequent AI model and reduce the accuracy of lesion identification.

[0004] In addition, the prior art lacks a systematic evaluation mechanism for image applicability, and cannot judge in advance whether the image meets the basic requirements for diagnosis. This makes low-quality images directly enter the subsequent benign and malignant tumor diagnosis process, increasing the possibility of misdiagnosis. At the same time, the reviewing doctors need to spend extra time on manual review and feedback of unqualified images, further increasing the work burden. The current processing flow also fails to effectively distinguish the quality levels of images, resulting in mixed processing of high-quality images and low-quality images, which affects the overall efficiency. In addition, the prior art lacks sufficient exploration of the specific reasons for image quality problems, making it difficult to provide clear improvement suggestions and restricting the standardization improvement of the image acquisition link. SUMMARY

[0005] The purpose of the present application is to provide an AI auxiliary diagnosis and treatment system for breast benign and malignant tumors based on an Internet hospital to solve the problems raised in the background.

[0006] To achieve the above object, the present application provides a breast benign and malignant tumor AI auxiliary diagnosis and treatment system based on an Internet hospital, which comprises: receiving breast image data streams uploaded by patients or primary medical institutions, monitoring the total quantity of uploaded images through an image acquisition module, and judging whether the total quantity exceeds the preset upper limit of system processing capacity; when the total quantity exceeds the upper limit of system processing capacity, a lightweight AI model is used to evaluate the quality of the images, generate an image applicability score, and determine whether the images meet the basic requirements for diagnosis according to the score; a multi-dimensional analysis algorithm is applied to analyze the artifact distribution characteristics in the images, a dynamic artifact characteristic model is constructed, and the interference degree index of artifacts on key anatomical regions is calculated based on the dynamic artifact characteristic model; whether to start an image optimization processing flow is decided according to the image applicability score and the interference degree index; when it is decided to start the image optimization processing flow, the attribute data of the acquisition parameters of each independent image is analyzed, and an image acquisition complexity factor is generated; based on the image acquisition complexity factor, the priority value of each image is calculated, and the image processing sequence is output according to the priority value.

[0007] Preferably, the breast image data streams uploaded by patients or primary medical institutions are received, the total quantity of uploaded images is monitored through an image acquisition module, and it is judged whether the total quantity exceeds the preset upper limit of system processing capacity, specifically including: continuously receiving image data packets from the uploading terminal, the image data packets containing resolution parameters, gray depth parameters and image size parameters; using a classification algorithm to identify the image types in the image data packets to determine an effective image category list; comparing the total number of images in the effective image category list with the system preset processing capacity upper limit value to generate an input load intensity index; based on the comparison result of the input load intensity index and the system processing capacity upper limit, it is judged whether there is an image diagnosis safety hazard signal.

[0008] Preferably, the lightweight AI model is used to evaluate the quality of the images, generate an image applicability score, and determine whether the images meet the basic requirements for diagnosis according to the score, specifically including: extracting the preprocessed image feature vector, which is obtained through an edge enhancement algorithm and an artifact separation technology; inputting the preprocessed image feature vector into a lightweight neural network model to train the model to generate a multi-dimensional image applicability score map; analyzing the dependency relationship matrix between feature elements in the multi-dimensional image applicability score map to calculate the dependency anomaly probability; quantifying the image applicability score according to the dependency anomaly probability.

[0009] Preferably, the multi-dimensional analysis algorithm is applied to analyze the artifact distribution characteristics in the image, construct an artifact dynamic characteristic model, and calculate an interference degree index of the artifact on the key anatomical region based on the artifact dynamic characteristic model, specifically including: performing a frequency domain conversion operation on the de-artifacted image to extract artifact spectrum distribution data; processing the artifact spectrum distribution data using a multi-scale decomposition algorithm to construct a spatio-temporal artifact dynamic characteristic model; analyzing the artifact intensity variation curve in the spatio-temporal artifact dynamic characteristic model to generate a key anatomical region interference evaluation value; and outputting the interference degree index based on the key anatomical region interference evaluation value.

[0010] Preferably, the image optimization processing flow is decided to be started or not started according to the image applicability score and the interference degree index, specifically including: when the image applicability score is a low score value and the interference degree index is a high interference level, generating a start image optimization processing flow instruction; when the image applicability score is a medium score value or a high score value, or the interference degree index is a medium interference level or a low interference level, generating a do not start image optimization processing flow instruction; and outputting the start image optimization processing flow instruction or the do not start image optimization processing flow instruction as a decision result.

[0011] Preferably, the image acquisition complexity factor is generated by analyzing the attribute data of the acquisition parameters of each independent image, specifically including: calculating the acquisition parameter difference value of each image and other images; identifying the conflict region distribution map in the acquisition parameter difference value and marking the low-quality region coordinates; integrating the region quantity and severity parameter in the conflict region distribution map to generate an acquisition parameter attribute data set; and calculating the image acquisition complexity factor based on the acquisition parameter attribute data set.

[0012] Preferably, the acquisition parameter attribute data set is generated by integrating the region quantity and severity parameter in the conflict region distribution map, specifically including: counting the total quantity value of all marked regions in the conflict region distribution map; measuring the average value of the fuzziness index of each marked region; and fusing the total quantity value and the average value of the fuzziness index to construct the acquisition parameter attribute data set.

[0013] Preferably, the priority value of each image is calculated based on the image acquisition complexity factor, and the image processing sequence is output in order according to the priority value, specifically including: assigning an acquisition weight coefficient to each image, the acquisition weight coefficient being determined based on the image category and the clinical importance; multiplying the acquisition weight coefficient and the image acquisition complexity factor to generate a priority score; and comparing the priority scores of all images and arranging them in order from high to low.

[0014] Preferably, the receiving patient or primary medical institutions upload the breast image data stream, the total amount of image quantity is monitored through the image acquisition module, and the application of optical character recognition technology is used to analyze the annotation information segment in the image data stream; the annotation information segment is stored in the temporary database, the integrity and consistency of the annotation information segment are verified, and a data quality report is generated.

[0015] Preferably, the input pre-processed image feature vector is input into a lightweight neural network model, and a multi-dimensional image applicability score map is generated by training the model, specifically including: a lightweight neural network model is constructed by using a convolutional neural network combined with an attention mechanism, the lightweight neural network model is trained to learn the dependency relationship between feature vectors, and a multi-dimensional image applicability score map is output to a scoring and evaluation module.

[0016] Compared with the prior art, the effect of the present application is that the system can master the current processing pressure of the system in real time by receiving the breast image data stream uploaded by the patient or the primary medical institutions and monitoring the total amount of uploaded images. When the total amount exceeds the preset upper limit of the capacity, instead of simply refusing to process or continuing forcibly, a lightweight AI model is introduced to evaluate the quality of the image, generate an image applicability score to determine whether the image meets the basic requirements for diagnosis.

[0017] This way allows the system to still effectively perceive image quality problems under high load conditions, avoiding the risk of misjudgment caused by blind processing. In terms of artifact processing, the system breaks out of the framework of traditional fixed filtering methods and applies a multi-dimensional analysis algorithm to deeply analyze the distribution characteristics of artifacts, and further constructs an artifact dynamic characteristic model.

[0018] By calculating the interference degree index of artifacts on key anatomical regions through the model, the unique attributes of artifacts in different images can be accurately captured, making artifact processing more targeted, reducing the interference of artifacts on key information recognition, and making the diagnosis result closer to the actual content of the image.

[0019] According to the image applicability score and the interference degree index, it is decided whether to start the image optimization processing flow, so that the start of the optimization flow is more in line with the actual demand.

[0020] When the score and the interference are at a low level, the regular flow can be maintained to avoid resource waste; when the score or the interference reaches a certain degree, the optimization flow is started, so that the processing work is more focused on images with prominent problems.

[0021] For each individual image, the attribute data of its acquisition parameters are analyzed to generate an image acquisition complexity factor, providing a specific basis for priority division. Different images have different information complexity and acquisition difficulty. Calculating priority values ​​based on the complexity factor and sorting and outputting the image processing sequence allows for a more rational allocation of processing resources. Images with high acquisition difficulty and many potential problems are processed first, while simple images are processed quickly, reducing unnecessary waiting and repetitive processing and improving the smoothness of the overall processing flow.

[0022] Overall, the system forms a tightly integrated processing mechanism from image input to final processing sequence output, capable of adapting to image processing scenarios of varying scales, artifact environments, and complexities. Through dynamic load sensing, precise quality assessment, targeted artifact handling, intelligent decision-making processes, and scientific prioritization, the system makes the diagnostic process for breast imaging more flexible and adaptable, reducing various problems caused by technological rigidity and making the entire processing workflow more aligned with the diverse needs of actual medical work. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the overall system architecture of the present invention.

[0024] Figure 2 A logic diagram of the image suitability score generation process.

[0025] Figure 3 The flowchart shows the construction of the dynamic characteristic model of artifacts and the calculation of the interference level index.

[0026] Figure 4 A decision logic diagram for the image optimization processing workflow.

[0027] Figure 5 A flowchart for managing breast lumps or breast pain;

[0028] Figure 6 A flowchart for the treatment of nipple discharge;

[0029] Figure 7 This is a flowchart of the follow-up visit process after examination. Detailed Implementation

[0030] This invention provides an AI-assisted diagnosis and treatment system for benign and malignant breast tumors based on an internet hospital. The specific implementation method is described in detail with reference to the accompanying drawings. Figure 1 As shown, the system mainly consists of a breast imaging data stream, an image acquisition module, a lightweight AI model, an artifact dynamic characteristic model, an image optimization processing flow, an acquisition parameter attribute dataset, and a priority value sorting module. The modules work together through data transmission and logical control.

[0031] As a preferred mode, in practical application, the patient or primary medical institution transmits the breast image data stream to the system through the uploading terminal. The image acquisition module is responsible for receiving these data streams and monitoring them, ensuring that the total amount of uploaded images can be mastered in real time. Specifically, the image acquisition module continuously receives image data packets from the uploading terminal, and each data packet contains resolution parameters, gray depth parameters and image size parameters.

[0032] As a preferred mode, the image acquisition module uses a classification algorithm to identify the image type in the image data packet, generates a list of valid image categories, and compares the total number of images in the list of valid image categories with the system's preset upper limit of processing capacity, thereby generating an input load intensity index. This process is completed through logical operations and data comparison. When the input load intensity index exceeds the system processing capacity upper limit, the system triggers the subsequent processing mechanism to cope with the high load state.

[0033] As a preferred mode, as shown in Figure 2 When the total amount of images exceeds the system processing capacity upper limit, the lightweight AI model is started to evaluate the quality of the images.

[0034] First, the lightweight AI model extracts the pre-processed image feature vectors, which are obtained through edge enhancement algorithms and artifact separation techniques.

[0035] Next, the lightweight AI model inputs these feature vectors into a model constructed by combining convolutional neural networks with attention mechanisms, and trains the model to learn the dependency relationships between feature vectors.

[0036] Finally, a multi-dimensional image suitability score map is output. The multi-dimensional image suitability score map calculates the dependency anomaly probability by analyzing the dependency relationship matrix between feature elements, and quantifies the image suitability score based on the dependency anomaly probability. The score provides an important basis for subsequent decision-making, which is used to determine whether the image meets the basic requirements for diagnosis.

[0037] As a preferred mode, in terms of artifact processing, as shown in Figure 3 The system uses a multi-dimensional analysis algorithm to analyze the artifact distribution characteristics in the image and constructs an artifact dynamic characteristic model. Specifically, the system performs a frequency domain conversion operation on the de-artifacted image and extracts artifact spectral distribution data.

[0038] Subsequently, a multi-scale decomposition algorithm is used to process the artifact spectral distribution data to construct a spatiotemporal artifact dynamic characteristic model. By analyzing the artifact intensity change curve in the spatiotemporal artifact dynamic characteristic model, the system generates a key anatomical region interference evaluation value and outputs an interference degree index based on the evaluation value.

[0039] Through frequency domain analysis and multi-scale decomposition algorithm, the unique properties of different image artifacts can be accurately captured, providing a scientific basis for subsequent optimization processing.

[0040] As a preferred mode, as shown in Figure 4 The system decides whether to start the image optimization processing flow according to the image applicability score and the interference level index.

[0041] When the image applicability score is a low score and the interference level index is a high interference level, the system generates a start image optimization processing flow instruction; when the image applicability score is a medium score or a high score, or the interference level index is a medium interference level or a low interference level, the system generates a do not start image optimization processing flow instruction. Through logical judgment conditions, it is ensured that the start of the optimization process is more in line with actual needs.

[0042] As a preferred mode, during the image optimization processing, for each independent image, the system analyzes the attribute data of its acquisition parameters to generate an image acquisition complexity factor.

[0043] The system first calculates the acquisition parameter difference value of each image with other images, identifies the conflict region distribution map in the acquisition parameter difference value, and marks the low-quality region coordinates.

[0044] Subsequently, the system counts the total number value of all marked regions in the conflict region distribution map, and measures the average value of the fuzziness index of each marked region. By fusing the total number value and the average value of the fuzziness index, the system constructs an acquisition parameter attribute data set, and calculates the image acquisition complexity factor based on the acquisition parameter attribute data set. Data statistics and fuzziness analysis are realized, which provides specific basis for subsequent priority division.

[0045] As a preferred mode, after generating the image acquisition complexity factor, the system calculates the priority value of each image based on the image acquisition complexity factor, and outputs the image processing sequence according to the priority value. Specifically, the system assigns an acquisition weight coefficient to each image, and the acquisition weight coefficient is determined based on the image category and clinical importance.

[0046] Then, the system multiplies the acquisition weight coefficient with the image acquisition complexity factor to generate a priority score. By comparing the priority scores of all images, the system arranges them in order from high to low, and finally outputs the image processing sequence. Through mathematical operation and logical sorting, it is ensured that images with high acquisition difficulty and potential problems are given priority, and simple images are quickly processed, reducing unnecessary waiting and repeated processing.

[0047] As a preferred mode, during the uploading process of the image data stream, the system also applies optical character recognition technology to preliminarily analyze the annotation information segments in the image data stream and stores the annotation information segments to a temporary database. The system verifies the integrity and consistency of the annotation information segments and generates a data quality report. This process is realized through optical character recognition technology and data verification algorithms, ensuring the accuracy and usability of the annotation information.

[0048] In summary, the present application realizes the whole process management of breast image from uploading to processing sequence output through the cooperative work of the above-mentioned modules. The system accurately evaluates the quality through dynamic load sensing, processes the artifacts intelligently, makes scientific priority division, makes the diagnosis process of breast image more flexible and adaptive, reduces various problems caused by technical rigidity, and makes the whole processing process more suitable for the diversified needs of actual medical work.

[0049] The specific implementation principles of the present application are supplemented by the following two embodiments.

[0050] Example 1: Image centralized processing scene in peak period of physical examination in primary medical institutions:

[0051] A county township health center carries out female breast health examination activities, and 1200 breast ultrasound images are uploaded to the Internet hospital system within one day, far exceeding the preset upper limit of 800 processed per hour, triggering the high-load processing mechanism of the system.

[0052] Load monitoring and identification: the image acquisition module receives image data containing parameters such as resolution and gray depth, identifies 1180 valid images through classification algorithms, generates an input load density index, and determines that there is a diagnostic safety hazard.

[0053] Lightweight AI quality evaluation: start a lightweight neural network model including convolutional neural network + attention mechanism, pre-process the image, including edge enhancement + artifact separation, extract feature vectors, and generate a multi-dimensional applicability score map. Among them, 820 images get high scores (≥85) due to body position standards and no obvious artifacts, 230 images get medium scores (60-84) due to partial area blur, and 130 images get low scores (<60) due to severe motion artifacts.

[0054] Artifact interference analysis: analyze the artifact distribution of 130 low-score images, extract frequency spectrum data through frequency domain conversion, construct a spatiotemporal artifact dynamic model, and find that the artifacts of 90 of them are concentrated in the key anatomical regions of the breast gland, with a high interference degree index.

[0055] Optimization decision and priority ranking: 90 low-score + high-interference images are started for optimization process, and their acquisition parameters, including probe frequency and gain value, are analyzed. The difference value from the standard parameters is calculated, and the low-quality area, such as the blur area coordinates, is marked. The acquisition complexity factor is generated, ranging from 1.2 to 3.5, with higher values indicating greater acquisition difficulty. Combined with clinical importance, such as a weight coefficient of 1.5 for women over 40 years old and 1.0 for women under 40 years old, the priority score is calculated as the product of the complexity factor and the weight coefficient. Finally, the scores are ranked from high to low, and 45 images with a complexity factor of ≥2.0, mostly middle-aged and elderly patients, are prioritized for processing.

[0056] The system completes the processing of 800 high-score images within 2 hours, and simultaneously optimizes 90 low-quality images. The optimized images are inserted into the subsequent processing sequence, avoiding the burden of manual screening by primary doctors and ensuring that high-priority images are fed back within 4 hours.

[0057] Example 2: Intelligent optimization scenario for motion artifact images

[0058] A patient had a breast X-ray image with obvious motion artifacts due to slight movement during the examination, showing blurred stripes on the edge of the gland. The primary doctor uploaded the image and marked a suspected mass, requesting review by an Internet hospital specialist.

[0059] Artifact analysis and interference assessment: After receiving the image, the system performs frequency domain conversion on the de-artifact pre-processed image to extract artifact spectral distribution data. Through a multi-scale decomposition algorithm, a spatio-temporal artifact dynamic model is constructed to analyze the artifact intensity variation curve and find that it overlaps with the upper outer quadrant of the breast, generating an interference assessment value of 89 (out of 100, ≥80 is high interference).

[0060] Suitability score and optimization decision: A lightweight AI model extracts image feature vectors and analyzes the feature dependency relationship matrix, calculating a dependency anomaly probability of 0.72 and a suitability score of 58 (low score). Due to the low score and high interference, the system generates an optimization process instruction.

[0061] Acquisition parameter optimization and processing: The image acquisition parameters are analyzed as tube voltage 28kV and exposure time 0.8s. Compared with the standard image parameters of the same type, the difference value is 1.3, which is 0.5 beyond the normal range. The blur area coordinates are marked in the upper outer quadrant at 3 locations. The number of conflict areas is 3, with an average blur index of 0.65. The acquisition parameter attribute dataset is fused to calculate the acquisition complexity factor of 2.8. Combined with the suspected mass clinical marker, the weight coefficient is 2.0, and the priority score is 5.6, ranking first in the optimization sequence.

[0062] Optimization results: The system corrects the artifact interference through a dynamic model and enhances the contrast of key areas, outputting the optimized image. The expert review can clearly observe the glandular structure, exclude the suspected mass, and avoid misdiagnosis.

[0063] As Figures 5-7 The present application also relates to the process of AI-assisted diagnosis and treatment of breast benign and malignant tumors.

[0064] 1. Handling of breast lumps or breast pain

[0065] For patients with chief complaint of breast lumps or breast pain. The decision is first based on the age of the patient.

[0066] Determine if the age is greater than 40 years old:

[0067] If the age is less than or equal to 40 years old (≤40 years old):

[0068] Check if the patient has a breast ultrasound report within the past six months (breast ultrasound is the preferred test because young women have dense breast tissue, and B-ultrasound is more sensitive).

[0069] If there is a breast ultrasound report within the past six months: directly enter the shared step: interpretation after examination and improvement.

[0070] If there is no breast ultrasound report within the past six months: suggest that the patient complete a breast ultrasound examination. After the patient completes the B-ultrasound examination and obtains the report, enter the shared step: interpretation after examination and improvement.

[0071] Reason: Young women have a lower risk of breast cancer, and B-ultrasound as an initial examination is sufficient; avoid unnecessary radiation exposure.

[0072] If the age is greater than 40 years old (>40 years old):

[0073] Check if the patient has a breast ultrasound report within the past six months AND a mammogram report within the past year (mammography, i.e., breast X-ray photography, is more effective for women over 40 years old because the breast tissue gradually becomes fatty, and mammography can better detect calcification and other abnormalities).

[0074] If both reports are complete: directly enter the shared step: interpretation after examination and improvement.

[0075] If the mammogram report is missing (or both are missing): the text specifies "otherwise, the mammogram report needs to be improved", so it is recommended that the patient complete the mammogram examination first (if the B-ultrasound report is missing, B-ultrasound also needs to be improved; in practice, it should be ensured that the B-ultrasound basic examination has been completed). After the patient completes the mammogram report (or both B-ultrasound and mammography are complete), enter the shared step: interpretation after examination and improvement.

[0076] Reason: Women over 40 years old have an increased risk of breast cancer, and guidelines recommend combining B-ultrasound and mammography to improve detection rates; mammography is the gold standard for screening.

[0077] 2. Handling of nipple discharge

[0078] For patients with chief complaint of nipple discharge (non-milk-like discharge, which may be bloody, serous, etc.). The decision is also based on age, but nipple discharge may suggest intraductal lesions, and the examination requirements are similar to breast lumps. Breast MRI examination has strong diagnostic value at this time.

[0079] Determine if the age is greater than 40 years old:

[0080] If the age is less than or equal to 40 years old (≤40 years old):

[0081] Check if the patient has a breast B-ultrasound report and a breast MRI report within the past six months (MRI can accurately assess breast duct dilation or intraductal space-occupying lesions).

[0082] If there is a breast B-ultrasound report and a breast MRI report within the past six months: directly enter the shared step: interpretation after examination.

[0083] If there is no breast B-ultrasound report within the past six months: it is recommended that the patient complete a breast B-ultrasound examination and / or a breast MRI examination, and then enter the shared step: interpretation after examination.

[0084] Reason: Nipple discharge in young women is mostly benign (such as intraductal papilloma), and B-ultrasound is a safe and effective first choice, and breast MRI examination has higher sensitivity.

[0085] If the age is greater than 40 years old (>40 years old):

[0086] Check if the patient has a breast B-ultrasound report within the past six months AND a breast MRI+mammography report within the past year (mammography helps to detect calcification, which may be related to malignant lesions).

[0087] If all three reports are complete: directly enter the shared step: interpretation after examination.

[0088] If any one report is missing (or all three are missing): the text specifies "otherwise, complete a mammography+breast MRI report", so it is recommended that the patient complete a mammography+breast MRI examination (also, the B-ultrasound report should be present). After the patient completes the mammography+breast MRI report, enter the shared step: interpretation after examination.

[0089] The above-described and the above embodiments are only used to illustrate the technical solutions of the present application, rather than limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An AI-assisted diagnosis and treatment system for benign and malignant breast tumors based on an internet hospital, characterized in that, The system includes: The system receives breast image data streams uploaded by patients or primary healthcare institutions, monitors the total number of uploaded images through the image acquisition module, and determines whether the total number exceeds the preset system processing capacity limit. When the total number exceeds the system's processing capacity limit, a lightweight AI model is used to assess the quality of the images, generate an image applicability score, and determine whether the images meet the basic requirements for diagnosis based on the score. Multidimensional analysis algorithms were applied to analyze the distribution characteristics of artifacts in images, a dynamic characteristic model of artifacts was constructed, and an index of the degree of interference of artifacts on key anatomical regions was calculated based on the dynamic characteristic model of artifacts. Based on the image suitability score and the interference level index, a decision is made on whether to initiate the image optimization processing procedure. When the decision is made to initiate the image optimization processing flow, for each individual image, the attribute data of its acquisition parameters are analyzed to generate an image acquisition complexity factor. Based on the complexity factors of image acquisition, the priority value of each image is calculated, and the image processing sequence is output according to the priority value.

2. The AI-assisted diagnosis and treatment system for benign and malignant breast tumors based on an internet hospital, as described in claim 1, is characterized in that... The process of receiving breast image data streams uploaded by patients or primary healthcare institutions, monitoring the total number of uploaded images through the image acquisition module, and determining whether the total number exceeds the preset system processing capacity limit, specifically includes: Continuously receive image data packets from the uploading terminal, the image data packets containing resolution parameters, grayscale depth parameters and image size parameters; The image types in the image data package are identified using a classification algorithm to determine a list of valid image categories; The total number of images in the list of valid image categories is compared with the system's preset processing capacity limit to generate an input load intensity index. Based on the comparison between the input load intensity index and the upper limit of the system processing capacity, it is determined whether there are any potential safety hazards in image diagnosis.

3. The AI-assisted diagnosis and treatment system for benign and malignant breast tumors based on an internet hospital, as described in claim 1, is characterized in that... The process involves using a lightweight AI model to assess image quality, generate an image suitability score, and determine whether the image meets the basic diagnostic requirements based on this score. Specifically, this includes: The preprocessed image feature vector is extracted, which is obtained through edge enhancement algorithm and artifact separation technology; Input the preprocessed image feature vectors into a lightweight neural network model, train the model to generate a multi-dimensional image applicability score map; Analyze the dependency matrix among feature elements in the multidimensional image applicability scoring map and calculate the probability of dependency anomalies. Based on the probability of dependency anomalies, the applicability score of images is quantified.

4. The AI-assisted diagnosis and treatment system for benign and malignant breast tumors based on an internet hospital, as described in claim 1, is characterized in that... The application of multidimensional analysis algorithms analyzes the artifact distribution characteristics in images, constructs a dynamic artifact characteristic model, and calculates the interference degree index of artifacts on key anatomical regions based on the dynamic artifact characteristic model, specifically including: Perform a frequency domain transformation on the artifact-removed image to extract the artifact spectral distribution data; A multi-scale decomposition algorithm is used to process artifact spectral distribution data to construct a spatiotemporal artifact dynamic characteristic model; Analyze the artifact intensity variation curves in the spatiotemporal artifact dynamic characteristic model to generate interference assessment values ​​for key anatomical regions. Based on the interference assessment values ​​of key anatomical regions, an interference degree index is output.

5. The AI-assisted diagnosis and treatment system for benign and malignant breast tumors based on an internet hospital, as described in claim 1, is characterized in that... The step of deciding whether to initiate the image optimization processing procedure based on the image suitability score and the interference level index specifically includes: When the image suitability score is low and the interference level is high, a command to start the image optimization processing flow is generated. When the image suitability score is medium or high, or the interference level index is medium or low, an instruction to not start the image optimization processing flow is generated. The command to initiate or not initiate the image optimization process will be output as the decision result.

6. The AI-assisted diagnosis and treatment system for benign and malignant breast tumors based on an internet hospital, as described in claim 1, is characterized in that... For each individual image, the attribute data of its acquisition parameters are analyzed to generate an image acquisition complexity factor, specifically including: Calculate the difference in acquisition parameters between each image and other images; Identify the distribution map of conflict areas among the differences in collected parameters and mark the coordinates of low-quality areas; Integrate the region number and severity parameters from the conflict area distribution map to generate a dataset of collected parameter attributes; Based on the dataset of acquired parameter attributes, the image acquisition complexity factor is calculated.

7. The AI-assisted diagnosis and treatment system for benign and malignant breast tumors based on an internet hospital, as described in claim 6, is characterized in that... The integrated conflict area distribution map incorporates the number of regions and severity parameters to generate a dataset of collected parameter attributes, specifically including: The total number of all marked areas in the statistical conflict area distribution map; Measure the average blur index of each marked region; By combining the total quantity value and the average ambiguity index, a dataset of collected parameter attributes is constructed.

8. The AI-assisted diagnosis and treatment system for benign and malignant breast tumors based on an internet hospital, as described in claim 1, is characterized in that... The process of calculating the priority value of each image based on the image acquisition complexity factor and sorting and outputting the image processing sequence according to the priority values ​​specifically includes: Each image is assigned an acquisition weight coefficient, which is determined based on the image category and clinical importance; The acquisition weight coefficient is multiplied by the image acquisition complexity factor to generate a priority score; Compare the priority scores of all images and sort them in descending order.

9. The AI-assisted diagnosis and treatment system for benign and malignant breast tumors based on an internet hospital, as described in claim 1, is characterized in that... The method of receiving breast image data streams uploaded by patients or primary healthcare institutions, and monitoring the total number of uploaded images through the image acquisition module, also includes: Preliminary analysis of annotation information fragments in image data streams using optical character recognition technology; Store annotation information fragments to a temporary database, verify the completeness and consistency of the annotation information fragments, and generate a data quality report.

10. The AI-assisted diagnosis and treatment system for benign and malignant breast tumors based on an internet hospital, as described in claim 3, is characterized in that... The preprocessed image feature vector is input into a lightweight neural network model, and the model is trained to generate a multi-dimensional image applicability scoring map, specifically including: A lightweight neural network model is constructed by combining convolutional neural networks with attention mechanisms. The lightweight neural network model is trained to learn the dependencies between feature vectors and outputs a multi-dimensional image applicability score map to the scoring and evaluation module.

Images