A medical image artifact identification and elimination method based on big data technology

Abstract

The application discloses a medical image artifact identification and elimination method based on big data technology and relates to the technical field of medical image processing.The application comprises the following steps: pre-processing collected image data based on gray scale normalization, then performing semantic segmentation and ROI positioning on the image content; performing lesion segmentation on the positioned image content, selecting lesion feature quantization, fusion, and performing model verification, and distributing the verified lesion segmentation model. The application solves the problems of high micro-nodular missed detection rate and high risk of missed diagnosis of malignant lesions by using a lesion detection model, reduces the false positive rate while improving the malignant lesion recall rate, improves the segmentation adaptability to lesions of different sizes by using a lesion segmentation model, obtains clear segmentation boundaries, assists in surgical planning, and reduces boundary positioning errors.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical image processing, and particularly relates to a medical image artifact identification and elimination method based on big data technology. BACKGROUND

[0002] In image diagnostics, PACS is a computer system specially used for storing, acquiring, sending and displaying medical images. The PACS technology is proposed for the purpose of realizing medical image non-filmization storage and communication, improving the overall medical work efficiency, and realizing the network, modernization and remote of doctors' work. It is a general solution in the field of medical image storage and communication.

[0003] In the prior art, the patent discloses an AI-based PACS system, which realizes image labeling, data set classification and AI model training, but does not solve the deep mining and prediction analysis of unstructured data, and lacks inspiration for the construction of structured feature library and disease prediction closed loop.

[0004] Therefore, there is an urgent need for a new medical image artifact identification and elimination method based on big data technology. SUMMARY

[0005] The present application provides a medical image artifact identification and elimination method based on big data technology, which solves the problems of the prior art.

[0006] In a first aspect, the present application provides a medical image artifact identification and elimination method based on big data technology, comprising:

[0007] The collected image data is preprocessed based on gray scale normalization, and then semantic segmentation and ROI positioning are performed on the image content.

[0008] The segmented image content is segmented, the lesion feature is quantized and fused, and the segmented model is distributed after model verification.

[0009] Further, in the preprocessing and semantic segmentation of the image data based on gray scale normalization, the preprocessing and semantic segmentation steps specifically include:

[0010] The field strength inhomogeneity of the MRI is corrected by a semantic segmentation model based on a convolutional neural network. At the same time, the contrast optimization suitable for all modalities is dynamically set and adjusted according to the image modalities, and the structure of the image content is checked for integrity to remove the low-confidence areas marked by the segmentation mask.

[0011] Further, the step of outputting the standardized image after preprocessing specifically further includes:

[0012] Step d1, access DICOM gateway data to obtain original image data, and correct by N4 bias field, including:

[0013] Model the image as original image, true signal, bias field and noise;

[0014] Iteratively estimate the bias field and compensate by the maximum likelihood algorithm;

[0015] Step d2, adjust the window width and window level, adjust the window level by adjusting the target tissue gray center value, adjust the window width by adjusting the gray scale display range, automatically match the preset window value through the DICOM tag, and when the DICOM tag is not marked, use adaptive histogram equalization instead of fixed window value;

[0016] Step d3, load into the encoder-decoder to perform image content segmentation and generate image content binary mask;

[0017] Step d4, perform confidence detection on the binary mask, output the image content ROI through the verified ROI positioning, and mark the non-artifact data if the verification fails;

[0018] Step d5, perform rule rejection on the artifact data;

[0019] Rule 1: integrity verification, calculate the artifact volume, if the artifact volume is less than 30% of the reference corresponding organ volume, mark it as incomplete scanning and reject;

[0020] Rule 2: morphological verification, extract the artifact contour, calculate the circularity, and if the circularity is less than the threshold, mark it as non-motion artifact and reject.

[0021] Further, the image data collected is preprocessed based on gray scale normalization, and then semantic segmentation and ROI positioning are performed on the image content, and the steps of ROI positioning include:

[0022] The data after semantic segmentation is processed into a preset voxel standardized image, the ROI positioning model is loaded, and the coordinates, confidence and target type label of the target bounding box are output.

[0023] Further, the ROI positioning model specifically includes:

[0024] The loss function of the ROI positioning model:

[0025]

[0026] Wherein, λ coord is a weight coefficient for controlling the coordinate loss of submillimeter level positioning required in diagnosis, λ obj is a weight coefficient for controlling target existence loss and suppressing false negatives,L obj is a focal loss function, L coord is a complete IoU loss function, λ cls is a weight coefficient for controlling the classification loss, L cls is a cross-entropy loss.

[0027] Further, the loss function of the ROI positioning model specifically includes:

[0028] The focal loss function formula is:

[0029]

[0030] wherein, p t is a confidence of the model predicting the target existing, α t is a class weight, γ is a focal parameter;

[0031] The formula of the complete IoU loss function is:

[0032]

[0033] wherein, IoU is an intersection over union of the prediction box and the real box, ρ 2 ( b pred , b gt ) is a Euclidean distance square of the center points of the prediction box and the real box, b pred is a prediction box, b gt is a real box, c is a diagonal line length of the minimum bounding box, v is a length-width ratio consistency factor; is a dynamic weight coefficient, = v / ((1- IoU )+ v ); when IoU is small, it means that the overlap degree is very low, at this time, more attention is paid to the overlap problem rather than the length-width ratio, will be automatically reduced to weaken the influence of v ;

[0034] v The formula for constraining the shape of the bounding box to conform to the lesion shape is:

[0035]

[0036] w gt , w pred , h gt , h pred respectively represent: true width, predicted width, true height, predicted height;

[0037] Classification loss function L cls The formula is:

[0038]

[0039] wherein, M is the total number of categories in the classification task, c is the category index, w c , y c , p c are the category weight, true label and predicted probability respectively.

[0040] Further, the positioned image content is executed lesion segmentation, lesion feature quantization, fusion, and model verification, and the verified lesion segmentation model is distributedly deployed, specifically including:

[0041] Based on the nnUNet adaptive lesion segmentation model, the lesion ROI data is loaded, 3D convolution downsampling is performed, transposed convolution upsampling is performed, and Voxel level mask is output;

[0042] Radiomics features extracted from the PyRadiomics library include GLCM entropy and GLSZM area size variance, deep learning features extracted from the lesion segmentation model include high-dimensional semantic features, specifically extracted from the second-to-last layer global average pooling, and the radiomics features and deep learning features are fused, and the complementary verification based on the AUC of benign and malignant discrimination is performed.

[0043] Further, regarding the model verification method, specifically including:

[0044] Five-fold cross-validation is performed by stratified sampling, and the average Dice coefficient is greater than the threshold Dice to be qualified, and the Dice coefficient is the overall segmentation overlap, that is, the segmentation accuracy;

[0045] Through robustness testing, motion artifacts are simulated under Gaussian noise, and the performance decline ratio is less than 5% to be qualified;

[0046] The model is verified by generalization test, and the AUC of benign and malignant identification of the model is less than 0.03.

[0047] Further, data drift detection is further performed on the lesion segmentation model, and when the data drift value is greater than 0.1, retraining of the lesion segmentation model is triggered.

[0048] When the accuracy of the lesion segmentation model is less than 0.85 for thirty consecutive days, retraining of the lesion segmentation model is triggered.

[0049] In a second aspect, the present application provides a medical image artifact recognition and elimination system based on big data technology, which is used to execute the method of any one of the first aspect, comprising a gateway and a server, the gateway accesses the server, the semantic segmentation model in the server pre-processes the MRI image data, the ROI positioning model outputs the image content positioning, the lesion segmentation model loads the ROI data to output the mask, the server extracts the deep learning features of the lesion segmentation model, fuses the radiomics features, and performs complementary verification based on the AUC of benign and malignant identification, and outputs the voxel-level lesion segmentation result based on the fused features, including: the prediction confidence of each voxel, the lesion volume, the surface area, and the centroid coordinates.

[0050] The medical image artifact recognition and elimination method provided by the present application solves the problem of inconsistent gray scale caused by differences in medical imaging equipment through bias field correction + window width and window level self-adaptation, reduces the gray scale standard deviation, and eliminates the device dependence difference;

[0051] Through image content segmentation + morphological verification, the problem of motion artifact interference and blurred organ boundary is solved, the Dice coefficient of organ segmentation is improved, and the artifact misjudgment rate is reduced;

[0052] Through the lesion detection model, the problems of high micro-nodular miss rate and high risk of missed diagnosis of malignant lesions are solved, the false positive rate is reduced, and the recall rate of malignant lesions is improved;

[0053] Through the lesion segmentation model, the segmentation adaptability to different sizes of lesions is improved, clear segmentation boundaries are obtained, surgical planning is assisted, and boundary positioning error is reduced. BRIEF DESCRIPTION OF DRAWINGS

[0054] The drawings described herein are used to provide further understanding of the embodiments of the present application, constitute a part of the present application, and do not constitute a limitation on the embodiments of the present application. In the drawings:

[0055] Figure 1 A medical image artifact recognition and elimination method based on big data technology is provided for an exemplary embodiment of the present application.

[0056] Figure 2A flow chart of positioning output as a standardized image after data preprocessing in a medical image artifact identification and elimination method based on big data technology provided for an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0057] The exemplary embodiments will be described in detail herein with reference to the accompanying drawings. In the following description, the same numbers refer to the same or similar elements throughout the drawings. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application.

[0058] First, the terms related to the present application are explained:

[0059] terminology definitions PACS system Picture Archiving and Communication System

[0060] The specific application scenario of the present application is the medical image processing technology field.

[0061] The medical image artifact identification and elimination method based on big data technology provided by the present application aims to solve the above technical problems of the prior art.

[0062] The technical solutions of the present application and how the technical solutions of the present application solve the above technical problems will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of the present application will be described below with reference to the accompanying drawings.

[0063] Embodiment 1: The present embodiment provides a medical image artifact identification and elimination method based on big data technology, which is realized by the following steps, including steps D1, D2, as shown in Figure 1

[0064] Step D1: Obtain images through DICOM gateway, use N4 bias field correction + window width adjustment to realize gray scale normalization, and then perform semantic segmentation and ROI positioning;

[0065] Here, semantic segmentation relies on the normalization result of the previous step, because uncorrected images will cause organ boundary blurring, and unnormalized images will cause the segmentation model to fail. Specifically, it includes: uneven gray scale causes organ boundary blurring, which reduces the segmentation accuracy;

[0066] As shown in Figure 2 Steps d1-d5 are the process of positioning output as a standardized image after data preprocessing, in detail:

[0067] Among them, the specific process of data preprocessing before semantic segmentation includes steps d1-d2:

[0068] ​Step d1, model the image as original image, true signal, bias field and noise, iteratively estimate the bias field and compensate by maximum likelihood algorithm, parameter setting: iteration number = 50, convergence threshold = 0.001;

[0069] Step d2, adjust the window width and window level based on the cv2.normalize operation of OpenCV, adjust the window level by adjusting the gray center value of the target tissue, adjust the window width by adjusting the gray scale display range, the dynamic adjustment strategy is: automatically match the preset window value through the DICOM label, when the DICOM label is not marked, use adaptive histogram equalization instead of fixed window value;

[0070] After preprocessing, perform semantic segmentation, i.e. image content segmentation, the semantic segmentation model based on convolutional neural network used in this embodiment is U-NetDice, if the image is not normalized, it will cause the segmentation model to fail, for example, when the coefficient of U-NetDice is ≤0.7, due to uneven gray scale, the organ boundary is blurred, the U-Net segmentation accuracy is reduced, at the same time, non-normalization also causes artifacts to remain, which affects the recognition result, for example, motion artifacts are judged as lung nodules;

[0071] The technical problem solved by normalization: the normalization process solves the inconsistency of image gray scale range output by different types of CT / MRI devices, for example, GE, Siemens, and avoids interference from scanning parameters, such as kVp, layer thickness and other parameter changes, which will cause the gray scale value of the same tissue to drift, it is found in practice that the bone CT value fluctuates between 2000~3000 HU; the bias field effect specific to MRI will cause the edge gray scale of the image to be distorted, with a maximum signal deviation of 40%, the de-artifact operation, i.e. the integrity check of the structure of the image content, has good effect on solving noise and artifacts. For subsequent operations, the influence of device noise is removed, more attention is paid to biological characteristics, and the data fidelity is improved.

[0072] Therefore, gray scale normalization and semantic segmentation of organs form a cascading preprocessing relationship, this embodiment segments the organs in the image content and removes artifacts, including the following steps d3-d5:

[0073] d3, load into the encoder-decoder to perform image content segmentation, and generate an image content binary mask;

[0074] Input specification: 128x128x64 voxels, (spatial resolution 1mm³), output result: generate organ binary mask (e.g. liver Mask=1, background=0);

[0075] d4, perform confidence detection on the binary mask, output the image content ROI through the verified ROI positioning, and mark the non-artifact data that does not pass the verification;

[0076] Step d5, performing rule rejection on artifact data;

[0077] Rule 1: integrity verification, calculate artifact volume, if artifact volume is less than 30% of reference corresponding organ volume, mark as incomplete scan and reject; here, the organ structure integrity check is the structure of the image content, in this embodiment, after normalization, the Dice coefficient of the segmentation model is improved by 0.15;

[0078] Rule 2: morphological verification, extract artifact contour, calculate circularity, if circularity is less than 0.7, mark as non-motion artifact and reject, normal organ circularity > 0.8.

[0079] d1-d5 Cascade verification data based on 1000 CT test results of LIDC-IDRI dataset:

[0080]

[0081] The ROI positioning model, i.e. the lesion detection model, has the following training process and use:

[0082] The training process of the lesion detection model is as follows:

[0083] The loss function of the lesion detection model is constructed as:

[0084]

[0085] Where λ coord is a weight coefficient (default value = 5) for controlling the coordinate loss of sub-millimeter level positioning required in diagnosis;

[0086] λ obj is a weight coefficient (default value = 1) representing the control target existence loss, which suppresses false negatives;

[0087] L obj is a focus loss function, which is used to dynamically adjust the difficulty of sample weight;

[0088] L coord is a complete intersection ratio loss function, which introduces a center point distance and aspect ratio penalty term, representing the complete intersection ratio loss;

[0089] λ cls is a weight coefficient (default value = 1) representing the control classification loss, which is used to improve the lesion type identification, for example, to distinguish between benign and malignant, such as ground glass nodule vs. solid nodule;

[0090] L clsFor cross-entropy loss, it represents the difference between the predicted label and the real label. In this embodiment, the multi-lesion type recognition is added, and the nodules, calcifications, masses, etc. can be detected at the same time.

[0091] The formula of the complete intersection-over-union loss function is:

[0092]

[0093] wherein, IoU The intersection-over-union of the predicted box and the real box is used to ensure the basic positioning accuracy.

[0094] ρ 2 b pred , b gt The Euclidean distance square of the center points of the predicted box and the real box is b pred The predicted box is b gt The real box is used to ensure the center positioning accuracy of the lesion;

[0095] c The diagonal length of the minimum bounding box is

[0096] v The aspect ratio consistency factor is used to constrain the shape of the bounding box to conform to the shape of the lesion, and the formula is:

[0097]

[0098] w gt , w pred , h gt , h pred The real width, the predicted width, the real height, and the predicted height are represented by

[0099] α The weight coefficient is used to weaken the influence of IoU when it is low; v

[0100] The formula of the focal loss function is:

[0101]

[0102] wherein, p t The confidence of the model predicting the existence of the target is

[0103] α ​​t As a class weight, the lesion class is set in this embodiment α t As 0.8, the background class is set α t As 0.2, when the lesion area is very small, the class weight is adaptively increased;

[0104] γ As a focus parameter, the default is 2; when the sample is a medium difficulty sample, it is optimized to 1.5 to avoid excessive suppression of sample data;

[0105] Classification loss function L cls The formula is:

[0106]

[0107] M As the total number of categories in the classification task, such as lung nodule diagnosis M As 3, indicating malignant / benign / uncertain, c As a class index, w c , y c , p c Class weight, true label and prediction probability, respectively, wherein the true label is encoded by one-hot;

[0108] The lesion detection model achieves high positioning accuracy and ensures accurate positioning of the lesion center.

[0109] Wherein, the training parameter optimizer is AdamW, and the batch size is 32 cases / GPU;

[0110] The use process of the lesion detection model is as follows:

[0111] This embodiment is implemented by using YOLOv5, the input data is processed into 128x128x64 voxel standardized images, the ROI positioning model is loaded, and the coordinates of the target bounding box, the confidence and the target type label are output. The target bounding box here is the lesion bounding box, which solves the problem of long time and low efficiency of manual marking of single CT by traditional radiologists; the lung nodule detection rate reaches 98.7% based on LIDC-IDRI benchmark, and the false positive rate is <0.5 cases / scan;

[0112] Step D2: After lesion detection, the lesion is segmented, the lesion segmentation model is loaded, the lesion feature is selected and quantified, fused, and the model verification is performed, and the verified lesion segmentation model is distributed.

[0113] The lesion segmentation model is based on nnUNet self-adaptive building, the input is the lesion ROI region output by YOLOv5, that is, 64x64x64 voxels, and the output is a binary mask of the lesion region with Voxel level precision, which solves the problem of fuzzy segmentation of irregular lesion boundaries.

[0114] The loss function of the lesion segmentation model is a hybrid loss composed of Dice loss and Hausdorff loss, and the loss function of the lesion segmentation model is L total The formula is:

[0115] L total = α ⋅ L Dice + β ⋅ L Hausdorff

[0116] The sum of the weight coefficients α,β is 1, usually set to α =0.6, β =0.4, and adjusted according to the training stage;

[0117] L Dice The Dice loss value is closer to 0, indicating that the predicted segmentation result has a higher degree of coincidence with the true result (gold standard), and the segmentation effect is better, and the formula is:

[0118]

[0119] Wherein, N is the total number of three-dimensional voxels, that is, the number of voxels of a single input image, p i is the model prediction probability, that is, the possibility of a voxel being judged as a lesion, g i is the gold standard mask, which is 0 or 1, and is the real lesion area marked by a doctor;

[0120] L Dice ∈[0,1] (0=perfect match, 1=complete mismatch), when the prediction is completely coincident with the gold standard L Dice is 0, when the prediction covers 50% of the gold standard, L Dice The calculation of 1-2*0.5 / 1+0.5=0.33;

[0121] L HausdorffThis is the loss value based on Hausdorff distance. A smaller value indicates a smaller maximum distance between the two contours, meaning a better boundary match. The formula is:

[0122]

[0123] Among them, ∂ G For the surface contour points of the gold standard lesion extracted by binary mask, ∂ P For the surface profile points of the predicted results; from ∂ G Find any point as x For point x, in predicting the contour ∂P Find it above x nearest point y Then iterate through all the points on the actual contour. x ,point x and y Not predefined, but Hausdorff During distance calculation, keypoint pairs are dynamically determined to find the maximum mismatch between two contours. ， 95perc represents the 95th percentile of the distance, and ||xy|| represents the Euclidean distance.

[0124] In terms of loss functions, the traditional Dice loss may give a high score but with coarse boundaries, while the Hausdorff loss is specifically optimized for sub-millimeter edge accuracy.

[0125] In feature fusion, the feature fusion matrix uses dimensions (128, 1536), where 1536 is a concatenation of 512 dimensions from radiomics and 1024 dimensions from deep learning. The penultimate layer of the lesion segmentation model uses global average pooling to extract dimensions of 1024. Structurally, the last layer of the segmentation network is typically a 1x1 convolution, outputting the same number of channels as the number of classes. For example, lesion / background binary classification uses 2 channels. This layer outputs the classification probability of each pixel, representing highly task-specific information. The penultimate layer is used here because it retains richer abstract features and is not yet constrained by the final task, thus exhibiting stronger generalization ability. From a medical perspective, the penultimate layer... Variational features need to balance generality and specificity. A 1024-dimensional embedding vector can express the general pattern of "malignant tumor" while retaining the unique signs of specific cases (such as calcification distribution). This is the value of high-dimensional semantic embedding. For example, in the medical interpretation of high-dimensional semantics, the activation mode is marginal high activation and the associated pathological feature is lobulation of lesions. In the embedding dimension of 384, the activation mode is center-edge gradient activation and the associated pathological feature is ground-glass nodule sign. In the embedding dimension of 768, the activation mode is punctate and clustered high activation and the associated pathological feature is microcalcification clusters.

[0126] Regarding model verification, five-fold cross-validation is performed by stratified sampling, and the average Dice coefficient greater than 0.90 is qualified, and the Dice coefficient is the overall segmentation coincidence degree, that is, the segmentation accuracy;

[0127] Through robustness testing, under Gaussian noise, motion artifact simulation, and performance degradation less than 5% is qualified;

[0128] Through generalization testing, the AUC of benign and malignant discrimination of the model is verified, and the decrease is less than 0.03.

[0129] Finally, the lesion segmentation model is distributed and deployed on multiple servers to expand the data collection range and enhance the model effect.

[0130] In the embodiments provided by the present application, it should be understood that the disclosed system and method can be implemented in other ways. For example, the system embodiments described above are only illustrative,

[0131] The modules described as separate components can or can not be physically separated, and the components shown as modules can or can not be physical modules, i.e., they can be located in one place or distributed to multiple network modules. Part or all of the modules can be selected according to actual needs to achieve the purpose of the present embodiment.

[0132] In addition, each functional module in the embodiments of the present application can be integrated into one processing module, or each module can exist physically, or two or more modules can be integrated into one module. The above integrated module can be realized in the form of hardware or in the form of hardware plus software functional module.

[0133] Those skilled in the art should understand that the embodiments of the present application can be provided as a method or a system. Therefore, the present application can adopt a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects.

[0134] It should be further noted that the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that the processes, methods, articles or devices including a series of elements not only include those elements, but also include other elements not explicitly listed or inherent to such processes, methods, articles or devices. Without more limitations, the element defined by the statement "comprises a" does not exclude the presence of additional identical elements in the process, method, article or device including the element.

[0135] The above merely provides the embodiment of the present application, but does not serve to limit the present application. For those skilled in the field, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in the scope of claims of the present application.

[0136] Other embodiments of the application will be apparent to those skilled in the art from consideration of the specification and practice of the application disclosed here. It is intended that the specification and examples be considered as exemplary only, with the true scope and spirit of the application being indicated by the following claims.

[0137] It is to be understood that the application is not limited to the precise details of construction and the above-described embodiments and that various modifications and changes can be effected therein by one skilled in the art without departing from the scope of the application. The scope of the application should be determined by the terms of the following claims.

Claims

1.A medical image artifact identification and elimination method based on big data technology, characterized in that, Comprise: Based on gray scale normalization to the collected image data preprocessing, and then to the image content semantic segmentation, ROI positioning; The step of outputting the standardized image after preprocessing specifically further comprises: step d1, accessing the DICOM gateway data to obtain the original image data, correcting the N4 bias field, including: modeling the image as the original image, the true signal, the bias field and the noise; iteratively estimating the bias field and compensating by the maximum likelihood algorithm; step d2, adjusting the window width and window level, adjusting the window level by adjusting the target tissue gray scale center value, adjusting the window width by adjusting the gray scale display range, automatically matching the C preset window value through the DICOM tag, and when the DICOM tag is not marked, using adaptive histogram equalization instead of fixed window value; Step d3, load into the encoder-decoder to perform image content segmentation and generate image content binary mask; step d4, perform confidence detection on the binary mask, output the ROI of the image content through the verified ROI positioning, and mark the non-artifact data if the verification fails; step d5, perform rule elimination on the artifact data; rule 1: integrity verification, calculate the artifact volume, if the artifact volume is less than 30% of the reference corresponding organ volume, mark it as incomplete scanning and eliminate; Rule 2: morphological verification, extract the artifact contour, calculate the circularity, and if the circularity is less than the threshold, mark it as non-motion artifact and eliminate; The steps of ROI positioning include: processing the data after semantic segmentation into a preset voxel standardized image, loading the ROI positioning model, and outputting the coordinates, confidence and target type label of the target bounding box; the ROI positioning model specifically comprises: the loss function of the ROI positioning model: wherein λ coord is a weight coefficient for controlling the coordinate loss in diagnosis requiring sub-millimeter level positioning, obj is a weight coefficient for controlling the target existence loss and suppressing false negatives, L obj is a focal loss function, L coord represents a coordinate positioning loss, which is used to measure the coincidence degree between the predicted lesion bounding box and the real bounding box and the consistency of the center point and the aspect ratio, wherein L coord is replaced by a complete intersection-over-union loss function, λ cls is a weight coefficient for controlling the classification loss, L cls is a cross-entropy loss; The loss function of the ROI positioning model specifically comprises: the focusing loss function formula is: wherein, p t a confidence that the model predicts the target exists, α t a class weight, The formula of the complete intersection-over-union loss function is: a focus parameter; The formula of the complete intersection-over-union loss function is: wherein, The formula of the complete intersection-over-union loss function is: is the intersection over union of the predicted box and the ground truth box, Perform lesion segmentation on the positioned image content, select lesion feature quantization, fusion, and perform model verification, and distribute the verified lesion segmentation model. 2 ( b pred , b gt is the squared Euclidean distance between the center points of the predicted box and the ground truth box, b pred is the predicted box, b gt is the ground truth box, c is the length of the diagonal of the minimum bounding box, v is the aspect ratio consistency factor; is the dynamic weight coefficient, = v / ((1- In the preprocessing and semantic segmentation steps of the image data preprocessing based on gray scale normalization and then performing semantic segmentation and ROI positioning on the image content, the preprocessing and semantic segmentation steps specifically comprise: ) + v ); v To constrain the bounding box shape to conform to the lesion shape, the formula is: w gt , w pred , h gt , h pred respectively: true width, predicted width, true height, predicted height; Classification loss function L cls The formula is: wherein, M is the total number of classes in the classification task, c is the class index, w c , y c , p c are the class weight, the true label, and the predicted probability, respectively; The semantic segmentation model based on convolutional neural network corrects the field inhomogeneity of MRI, at the same time, dynamically sets and adjusts the contrast optimization suitable for all modalities according to the image modalities, checks the integrity of the structure of the image content, and removes the low confidence area marked by the segmentation mask. 2.The medical image artifact identification and elimination method based on big data technology according to claim 1, characterized in that, The lesion segmentation model based on nnUNet adaptive construction loads the lesion ROI data, performs 3D convolution downsampling, transposes the convolution upsampling, and outputs the Voxel-level mask; ​ 3.The medical image artifact identification and elimination method based on big data technology according to claim 2, characterized in that, ​ ​ The radiomics features extracted from the PyRadiomics library include GLCM entropy and GLSZM area size variance, the deep learning features extracted from the lesion segmentation model include high-dimensional semantic features, specifically extracted from the second-to-last layer global average pooling, and the radiomics features and deep learning features are fused, and the complementarity verification based on the AUC of benign and malignant identification is performed. 4.The medical image artifact identification and elimination method based on big data technology according to claim 3, characterized in that, Regarding the model verification method, specifically includes: Through stratified sampling, five-fold cross-validation is performed, and the average Dice coefficient is greater than the threshold Dice to be qualified, and the Dice coefficient is the overall segmentation overlap, that is, the segmentation accuracy; Through robustness testing, motion artifacts are simulated under Gaussian noise, and the performance decline ratio is less than 5% to be qualified; Through generalization testing, the AUC of benign and malignant identification of the model is verified, and the decrease is less than 0.03 to be qualified. 5.The medical image artifact identification and elimination method based on big data technology according to claim 4, characterized in that, It also includes data drift detection for the lesion segmentation model, and when the data drift value is greater than 0.1, the retraining of the lesion segmentation model is triggered; When the accuracy of the lesion segmentation model is less than 0.85 for thirty consecutive days, the retraining of the lesion segmentation model is triggered.

Images