Medical image information processing method, system, equipment and medium

Abstract

The invention provides a medical image information processing method, system and device and a medium, and belongs to the technical field of medical image processing. Analyzing the image under two constraints through a pre-training model, and outputting first trend information for maintaining a core feature and second trend information for trending to a target feature; performing time sequence alignment on the trend information, extracting difference parameters, and generating a transformation rule; performing iterative adjustment on the image based on the rule, and verifying whether the image meets the original diagnosis constraint after each adjustment until the target diagnosis requirement is met; and finally, carrying out quality verification on the generated image, and storing related data into a medical image database to form a complete data set. Through a double-constraint mechanism, it is ensured that original diagnosis key information is reserved in the image adjustment process, target diagnosis requirements are effectively matched, and image availability and consistency are improved; the controllability and stability of the processing process are enhanced through an iterative adjustment mode, and the diagnosis efficiency and accuracy can be improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of medical image processing, and particularly relates to a medical image information processing method, system, device and medium. BACKGROUND

[0002] In the field of medical imaging, with the progress of digital imaging technology, such as the development of CT (Computed Tomography), MRI (Magnetic Resonance Imaging), ultrasound, etc., doctors can obtain more detailed internal structure images of the human body than ever before. However, these image data often contain noise, artifacts or other interference factors, which may affect the accuracy of diagnosis. In order to improve the efficiency and accuracy of diagnosis, it is necessary to process the original image to enhance the key features and reduce the interference of non-key areas.

[0003] In related technologies, in order to make the lung CT nodule boundary clear, excessive enhancement leads to misjudgment of the nodule as a blood vessel, or blurs the original relationship between the nodule and the pleura, destroying the core features of the original image.

[0004] In the prior art, the difference parameter calculation lacks time sequence constraints and boundary restrictions, and time step misalignment often occurs, for example, when calculating the lung CT difference, the nodule diameter needs to be reduced by 1 cm due to time step misalignment, but in fact it should be increased by 0.2 cm, resulting in a completely wrong adjustment direction. In the execution of iterative adjustment without hierarchical verification, only the lung CT iteration focuses on the clarity of the nodule boundary, without verifying whether it leads to misjudgment of the pleura as a nodule, resulting in an image with clear lesions but structural errors. SUMMARY

[0005] The application provides a medical image information processing method, which ensures that the image adjustment process retains the original diagnostic key information and effectively matches the target diagnostic requirements through a double constraint mechanism, improving image usability and consistency; the iterative adjustment method enhances the controllability and stability of the processing process, which helps to improve the efficiency and accuracy of diagnosis; the constructed dataset can provide high-quality, structured data support for medical research and AI model training.

[0006] The method comprises: S101: obtaining an original medical image to be processed and original image diagnostic description information corresponding to the original medical image, and simultaneously obtaining target diagnostic result information; S102: analyzing the original medical image based on two different constraint conditions through a pre-trained image analysis model: The original image diagnostic description information is used as a constraint to make the image analysis model maintain the core features such as the type and key structure of the displayed object in the original medical image unchanged during the analysis process, and only adjust the non-key details, and output first image state trend information; The target diagnosis result information is used as a constraint to make the image analysis model adjust the display content of the original medical image to meet the expected object characteristics in the diagnosis result information, and output second image state trend information; S103: The first image state trend information and the second image state trend information are time-aligned, a difference parameter at the same time step or feature dimension is extracted, and a transformation rule from the original state to the target state is formed; S104: Based on the transformation rule, the original medical image is adjusted through iterative image processing: in each iteration, the original image is locally or globally adjusted according to the difference parameter corresponding to the current step, and it is verified whether the adjusted image meets the display requirements of the target diagnosis result information; the iteration is repeated until the adjusted image meets the display requirements of the target diagnosis result information; S105: The generated diagnosis result image is quality-verified, and after verification, the original medical image, the original image diagnosis description information, the target diagnosis result information, and the diagnosis result image are stored in the medical image database to form a diagnosis image dataset.

[0007] Preferably, the step of outputting the first image state trend information further comprises: When the image analysis model performs the first image analysis process on the original medical image based on the original image diagnosis description information, the output parameters of the neural network are extracted in real time by a parameter acquisition module inside the image analysis model as first output parameters; The first output parameters are subjected to multi-dimensional feature extraction and trend analysis to generate the first image state trend information, which describes the natural change rule of the original medical image under the constraint of maintaining the status quo in the form of quantitative parameters; When the image analysis model performs the second image analysis process on the original medical image based on the diagnosis result information, the output parameters of the neural network are synchronously acquired by a model parameter monitoring module; The second output parameters are subjected to reverse analysis and target-oriented trend derivation to generate the second image state trend information, which describes the change rule of the original medical image from the current state to the target state in the form of path parameters.

[0008] Preferably, the step S102 specifically comprises: The core feature set defined in the original image diagnosis description information is acquired and converted into a constraint condition recognizable by the model; The constraint condition includes: limiting the object type to be retained during model analysis through a pre-defined classification label; labeling the region in the original image that needs to maintain the morphological characteristics to generate a structure mask; extracting the statistical quantities such as the gray mean value and contrast baseline of each region in the original image as the baseline range during model adjustment.

[0009] The expected features in the target diagnosis result information are converted into analysis targets to be achieved by the model; A pre-trained multi-task image analysis model is called, and the original medical image, the original diagnosis constraint, and the target diagnosis constraint are inputted to perform the following analysis operations: Multi-scale features of the original image are extracted through a CNN network; The features of the original diagnosis constraint are matched with the extracted features to generate a maintenance score; The features of the target diagnosis constraint are calculated with the extracted features to generate an adjustment demand heat map.

[0010] The first image state trend information and the second image state trend information are outputted respectively.

[0011] Preferably, between step S103 and step S104, there are further steps of: The first image state trend information is represented as a multi-scale tensor structure containing space-time dimensions; An improved time warping algorithm is adopted to perform space-time alignment on the first image state trend information tensor and the second image state trend information tensor, and through construction of a space-time distance matrix and optimal path planning, a first parameter sequence and a second parameter sequence after time step alignment are generated; A difference calculation model under diagnosis semantic constraint is constructed, and for each aligned time step: the Mahalanobis distance of the second parameter sequence and the first parameter sequence in the feature space is calculated to obtain a difference vector containing direction information; An attention mechanism is applied to weight the difference vector to enhance the parameter expression of the region related to the diagnosis; Through regularization processing of medical prior knowledge constraint, non-rational difference parameters are filtered out; The extracted difference vector sequence is inputted into a space-time convolution network to learn the time sequence dependency of parameter changes through a bidirectional LSTM layer to generate a continuous diagnosis result trend information tensor.

[0012] Preferably, the step of adjusting the original image according to the difference parameter corresponding to the current step, and verifying whether the adjusted image meets the original diagnosis description constraint, specifically includes: The difference parameter is divided into a core diagnosis area, an associated area, and a background area, and the core diagnosis area is preferentially adjusted locally; According to the core diagnosis area in the original image diagnosis description, it is checked whether the adjusted core diagnosis area still meets the preset features, and if the verification is passed, the contrast of the associated area is adjusted according to the difference parameter based on the adjustment result of the core diagnosis area, and the background area is subjected to noise reduction processing; it is checked whether the adjustment of the associated area has destroyed the features in the original description, and whether the noise reduction of the background area has affected the display of the core diagnosis area; If the core diagnostic area and the associated area pass the verification twice in a row, and there is no interference information in the background area, lock the current adjustment parameters and stop the adjustment of this area; if it fails, expand the adjustment range of the core diagnostic area and repeat the above steps until the verification is successful.

[0013] Preferably, S103 specifically includes: The contrast dimension is defined according to the clinical functional zoning of the images. The trend information of the first image status and the trend information of the second image status are divided into the direct diagnosis area, the auxiliary reference area, and the area with no diagnostic significance according to clinical function. Based on the operational purpose of image analysis, the time steps of the first image state trend information and the second image state trend information are mapped to the same operational purpose. For the aligned direct diagnostic region, extract diagnostic-related difference information; for the auxiliary reference region, extract the feature differences of the associated structure; for regions without diagnostic significance, only record whether there is interfering difference information. The extracted differential information is sorted according to clinical decision-making logic, then the vascular association differences in the auxiliary reference area are processed, and finally the interference differences in the area with no diagnostic significance are processed. Each sorting information defines the adjustment object, adjustment direction, and adjustment threshold, and all sorting information is labeled with constraints.

[0014] Preferably, step S104 specifically includes: Based on the clinical importance classification of the difference parameters in the transformation rules, high-priority regions are prioritized for local adjustment; After each local adjustment, the first layer verifies that the type and location of the core objects in the original image have not changed; the second layer verifies whether the adjusted regional features match the description of the target diagnosis results. After both layers of verification pass, the next iteration begins. If a high-priority area still fails to meet the criteria after two adjustments, the adjustment scope for that area will be expanded; if it has already met the criteria, the adjustment scope for a low-priority area will be narrowed. The iteration stops when the adjustment results for all regions pass the stratified verification twice consecutively, and the key features in the target diagnostic results are clearly displayed.

[0015] This application also provides a medical image information processing system, the system comprising: The data acquisition module is used to acquire the original medical image to be processed, as well as the original image diagnostic description information corresponding to the original medical image, and to acquire the target diagnostic result information. The image analysis module is used to analyze the original medical images based on two different constraints using a pre-trained image analysis model: Using the original image diagnostic description information as a constraint, the image analysis model maintains the core features such as the types and key structures of the objects displayed in the original medical image unchanged during the analysis process, only adjusting non-critical details, and outputting the first image state trend information. Using the target diagnostic result information as a constraint, the image analysis model adjusts the display content of the original medical image to make it conform to the expected object characteristics in the diagnostic result information, and outputs the second image state trend information. The trend calculation module is used to perform temporal alignment of the trend information of the first image state and the trend information of the second image state, extract the difference parameters at the same time step or feature dimension, and form the transformation rules from the original state to the target state. The iterative adjustment module, based on transformation rules, adjusts the original medical image through iterative image processing: in each iteration, the original image is locally or globally adjusted according to the difference parameters corresponding to the current step, and the adjusted image is verified to meet the original diagnostic description constraints; the iteration is repeated until the adjusted image meets the display requirements of the target diagnostic result information. The verification and association module is used to verify the quality of the generated diagnostic result images. After the verification is passed, the original medical image, the original image diagnostic description information, the target diagnostic result information, and the diagnostic result image are associated and stored in the medical image database to form a diagnostic image dataset.

[0016] According to another embodiment of this application, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the medical image information processing method.

[0017] According to another embodiment of this application, a storage medium is also provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the medical image information processing method.

[0018] As can be seen from the above technical solutions, the present invention has the following advantages: This invention provides a medical image information processing method that acquires the original medical image, its diagnostic description, and the target diagnostic result, ensuring that subsequent processing has a clear benchmark and endpoint, avoiding invalid operations due to incomplete data or ambiguous targets. Through dual-constraint analysis, the method optimizes target features while maintaining the core features of the original image, balancing the image's realism with the diagnostic practicality. By extracting difference parameters through temporal alignment and forming transformation rules, abstract adjustment requirements are transformed into quantifiable operational paths, and these paths are constrained by the original diagnosis, ensuring that the adjustment direction does not deviate from the rationality of the original image. Through iterative adjustment and verification, image features are gradually optimized, ensuring that each adjustment conforms to the original diagnostic description and ultimately meets the target requirements, improving the accuracy of the generated image. Quality verification and associated storage ensure the clinical validity of the generated image and completely preserve the processing chain, solving the problem of unreliable quality caused by the lack of verification in the generated image. The method retains the core clinical features of the original image while specifically optimizing the key information required for diagnosis, and ensures the reliability and traceability of the processing process through quantitative rules and verification mechanisms. Attached Figure Description

[0019] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart of medical image information processing methods; Figure 2 This is a schematic diagram of a medical image information processing system. Figure 3 This is a schematic diagram of an electronic device. Detailed Implementation

[0021] The medical image information processing method involved in this application will be described in detail below. Specific details such as particular system structures and technologies are presented for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application can also be implemented in other embodiments without these specific details.

[0022] It should be understood that, when used in this specification, terms include indicating the presence of a described feature, integral, step, operation, element, and / or component, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or collections thereof. The terms include, encompass, have, and variations thereof mean including but not limited to, unless otherwise specifically emphasized.

[0023] The statements such as "one embodiment" or "some embodiments" described in this application mean that one or more embodiments of this application include the specific features, structures, or characteristics described in that embodiment. Therefore, the statements such as "in one embodiment," "in some embodiments," "in other embodiments," and "in still other embodiments" in this application do not necessarily refer to the same embodiment, but rather mean one or more, but not all, embodiments, unless otherwise specifically emphasized.

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Please see Figure 1 The diagram shows a flowchart of a medical image information processing method in a specific embodiment. The method includes: S101: Obtain the original medical image to be processed, as well as the original image diagnostic description information corresponding to the original medical image, and at the same time obtain the target diagnostic result information.

[0026] In some embodiments, the acquired raw medical images include specific image types such as X-ray films, CT scans, and MRI sequences. Each image contains pixel-level raw data and device parameters, such as slice thickness for CT and weighted sequences for MRI. The original image diagnostic description information is structured text, such as "right upper lobe nodule, roughly round, approximately 1.8 cm in diameter, with rough edges, connected to the pleura," clearly recording the core features of the object in the image, including its type (nodule), location (right upper lobe), and shape (roughly round). The target diagnostic result information comes from the physician's preliminary diagnosis or clinical guideline recommendations, such as clearly showing the degree of adhesion between the nodule and the pleura, measuring the precise diameter, and specifying the desired display features. Data sources cover hospital imaging, electronic medical records, and clinical decision support systems to ensure complete data correlation.

[0027] S102: Using a pre-trained image analysis model, the original medical images are analyzed based on two different constraints: the original image diagnostic description information is used as a constraint, requiring the model to maintain the core features such as the types and key structures of the displayed objects in the original medical images unchanged during the analysis process, only adjusting non-key details, and outputting the first image state trend information; the target diagnostic result information is used as a constraint, requiring the model to adjust the display content of the original medical images to conform to the expected object features in the diagnostic result information, and outputting the second image state trend information.

[0028] In some embodiments, during the first analysis, the core structures in the original image diagnostic description, such as the outline of the right upper lobe and the boundaries of the nodules, are marked first. Only the noise and low-contrast areas outside these structures are adjusted to ensure that the core features such as the type and location of the nodules remain unchanged.

[0029] In the second analysis of this embodiment, the focus is on the degree of adhesion and the precise diameter requirements in the target diagnostic results. The contrast at the junction of the nodule and pleura is enhanced to clarify the adhesion boundary, and the nodule edges are sharpened while preserving non-target structures such as the lung lobe outline. In both analyses, if the original description of rough edges conflicts with the clear target showing adhesion, the core feature of rough nodule edges is prioritized, and only the clarity of the adhesion area is adjusted.

[0030] This embodiment balances the realism of the images with the practicality of diagnosis, avoiding structural distortion caused by over-processing.

[0031] In some embodiments, the step of outputting the first image state trend information further includes: S201: When the image analysis model performs the first image analysis process on the original medical image based on the original image diagnostic description information, the parameter acquisition module inside the image analysis model extracts the output parameters of the neural network in real time, such as the feature map distribution of the convolutional layer, the activation value vector of the fully connected layer, and the weight matrix of the attention mechanism, as the first output parameters. The first output parameters implicitly contain the model's analysis logic for the original image while maintaining the constraints of the original diagnostic description, including the potential direction and magnitude of image detail adjustments.

[0032] S202: Perform multi-dimensional feature extraction and trend analysis on the first output parameter: Specifically, this includes: the time dimension: the temporal changes of statistical parameters during the analysis process, such as the decay rate of noise parameters with the number of iterations.

[0033] Spatial dimension: Analyze the distribution pattern of parameters in the image space, such as the concentration of parameters in edge regions.

[0034] Feature dimensions: Extract the evolution trend of contrast and sharpness, such as the gradual change from blurry to clear.

[0035] The final result is the generation of first image state trend information, which describes the natural changes of the original medical image under the constraint of maintaining its current state, in the form of quantitative parameters. Examples include slight degradation trends over time and noise accumulation patterns.

[0036] S203: During the second image analysis process of the image analysis model based on diagnostic results, the output parameters of the neural network are synchronously collected through the model parameter monitoring module. Here, the adjusted convolutional kernel weights, the attention focus intensity of the target region, and the parameter offset of the artifact suppression module can be used as the second output parameters. The second output parameters reflect the model's intention to adjust the original image to achieve the diagnostic result target, including key information such as the coordinates of the region to be corrected and the adjustment threshold. Examples include clearly displaying lesion boundaries and enhancing the contrast of specific tissues.

[0037] S204: Perform reverse analysis and target-oriented trend derivation on the second output parameter.

[0038] Specifically, this includes: parameter difference analysis: calculating the difference between the second output parameter and the first output parameter to locate the image region that needs to be adjusted; Target feature matching: Match the second output parameter with the lesion location, size, and contrast in the diagnostic results information to deduce the adjustment direction; Ensure that the adjusted parameters conform to the original diagnostic description constraints, such as not changing the object type or losing key structures, and correct parameters that exceed the constraints.

[0039] Finally, second image state trend information is generated. This second image state trend information describes the change pattern of the original medical image from the current state to the target state in the form of path parameters. Such parameters include the location of areas to be enhanced, the range of noise distribution to be smoothed, and the amount of structural deformation to be corrected.

[0040] As can be seen, this embodiment uses the low-level output parameters of the neural network, rather than the final processed image, as the source of the state trend, reflecting the model's analytical logic and avoiding the indirectness of relying on image post-processing in traditional methods. Trend information is extracted based on both the original diagnostic description constraints and the target diagnostic result constraints, clarifying the differences in change patterns under the two scenarios of maintaining the status quo and adjusting to the target. When determining the second image state trend information, it is simultaneously verified whether the adjusted parameters conform to the original diagnostic description constraints, ensuring the legality and clinical applicability of the trend information and avoiding the diagnostic information distortion caused by blind adjustments in traditional methods.

[0041] S103: Perform temporal alignment of the first image state trend information and the second image state trend information, extract the difference parameters at the same time step or feature dimension, and form a transformation rule from the original state to the target state.

[0042] In this embodiment, during time alignment, the processing steps of the two analyses correspond to the corresponding time steps to ensure that feature changes in the same processing stage can be compared. The extracted difference parameters include: the adjustment amount of nodule edge sharpness, diameter measurement error, and contrast changes in the adhesion area.

[0043] The transformation rules are as follows: first, increase the contrast of the adhesion area by 30%, then optimize the clarity of the nodule edge by 20%, and finally correct the diameter measurement error to within 0.1cm. All adjustments must not exceed the boundary range of the lung lobe in the original image.

[0044] As can be seen, by comparing the characteristic change trends of the two analyses, the necessary adjustment amount from the original state to the target state is quantified, forming an adjustment path that conforms to clinical logic, ensuring that subsequent iterations have clear operational standards and boundaries. This ensures that the direction of adjustment is consistent with diagnostic needs, and that the scope is constrained by the original images.

[0045] The process between step S103 and step S104 also includes: Step S301: Represent the state trend information of the first image as a multi-scale tensor structure containing spatiotemporal dimensions. Each time step corresponds to a multi-channel feature matrix, and the matrix elements contain pixel-level change parameters, feature gradient information, and structural deformation vectors; similarly, construct the state trend information of the second image as a tensor structure of the same dimension to ensure that the two are comparable in spatiotemporal dimensions.

[0046] Step S302: An improved time warping algorithm is used to align the state trend information tensor of the first image with that of the second image in time and space. By constructing a time-space distance matrix and optimal path planning, the problem of inconsistent time steps between the two analysis processes is solved, and the first parameter sequence and the second parameter sequence after time step alignment are generated.

[0047] Step S303: Construct a difference calculation model under diagnostic semantic constraints, for each aligned time step: Calculate the Mahalanobis distance between the second parameter sequence and the first parameter sequence in the feature space to obtain the difference vector containing directional information; An attention mechanism is applied to weight the difference vectors, enhancing the parameter representation of diagnostically relevant regions; By using regularization to constrain prior medical knowledge, irrational difference parameters are filtered out.

[0048] Step S304: Input the extracted differential vector sequence into the spatiotemporal convolutional network, and learn the temporal dependency of parameter changes through the bidirectional LSTM layer to generate a continuous diagnostic result trend information tensor; this tensor contains the complete transformation path from the original image state to the target diagnostic state, while satisfying clinical rationality constraints and image quality evaluation indicators.

[0049] As can be seen, the trend information of the first and second images is first transformed into a multi-scale tensor structure containing spatiotemporal dimensions, making them comparable in terms of spatiotemporal features. Then, the two tensors are spatiotemporally aligned using an improved time warping algorithm to solve the problem of inconsistent time steps and generate an aligned parameter sequence. Next, the difference vector is calculated under diagnostic semantic constraints, and the parameter representation of diagnostic-related regions is strengthened by combining an attention mechanism, while unreasonable parameters are filtered out by medical prior knowledge. Finally, the difference vector is input into a spatiotemporal convolutional network, and the temporal dependency of parameter changes is learned using a bidirectional LSTM to generate a diagnostic result trend information tensor containing a complete transformation path and meeting the requirements of clinical rationality and image quality, providing accurate and continuous path guidance for subsequent iterations to generate target images.

[0050] In this embodiment, diagnostic semantic constraints and medical knowledge regularization ensure that the difference vector conforms to clinical logic, avoiding interference from unreasonable parameters in subsequent processing. The combination of spatiotemporal convolutional networks and bidirectional LSTM generates continuous transformation paths, providing a clear and reliable basis for subsequent iterative adjustments. Compared with the problems of coarse difference calculation and lack of clinical constraints in existing technologies, this significantly improves the accuracy and clinical adaptability of image adjustment.

[0051] S104: Based on the transformation rules, the original medical image is adjusted through iterative image processing: In each iteration, the original image is locally or globally adjusted according to the difference parameters corresponding to the current step, and the adjusted image is verified to meet the original diagnostic description constraints; the iteration is repeated until the adjusted image meets the display requirements of the target diagnostic result information.

[0052] In this embodiment, the adhesion area is adjusted first during iteration. After each adjustment, the original image diagnostic description is used for verification: the lung lobe outline has not shifted, and the nodule is still marked as round or oval. Then, the target diagnosis is used for verification: the contrast of the adhesion area reaches 50. If the standard is not met, the adjustment continues. If the adhesion area is still unclear after 3 adjustments, the adjustment range is expanded, from 1 times the original area to 1.2 times. If the standard is met, the adjustment range of the nodule edge is reduced, from 0.8 times the original area to 0.6 times. When the nodule diameter measurement error is ≤0.1cm and the adhesion boundary is clearly visible in 2 consecutive verifications, the iteration is stopped, and the contrast improvement of 30 and the edge sharpening of 20% are recorded as reference parameters for processing similar nodules.

[0053] This embodiment employs an iterative mechanism of priority adjustment, hierarchical verification, and range optimization to progressively refine image features, ensuring that each adjustment moves closer to the target diagnosis while remaining true to the core structure of the original image. The resulting image meets diagnostic requirements while retaining key information from the original image, improving diagnostic accuracy and efficiency.

[0054] The steps involve making local or global adjustments to the original image based on the difference parameters corresponding to the current step, and verifying whether the adjusted image meets the constraints of the original diagnostic description. Specifically, this includes: Step S4021: Divide the difference parameters into core diagnostic area, related area and background area, and prioritize local adjustment of the core diagnostic area.

[0055] For example, if the difference parameter is that the clarity of the lesion boundary needs to be improved by 40%, then the pixel contrast of the lesion edge should be adjusted first, while the related area and background area should not be adjusted for the time being, so as to ensure that key diagnostic information is optimized first.

[0056] Step S4022: Based on the core diagnostic area in the original image diagnostic description, check whether the adjusted area still meets the preset verification conditions. The preset verification conditions can be that the nodule is still located in the lower lobe of the left lung, the diameter measurement deviates from the original description by no more than 0.2 cm, and no new artifacts appear that cause the nodule to be misdiagnosed as other structures. If the verification is successful, proceed to the next step.

[0057] Step S4023: Based on the adjustment results of the core diagnostic area, adjust the contrast of the associated area according to the difference parameters, and perform noise reduction on the background area; check whether the adjustment of the associated area has destroyed the features in the original description, and whether the noise reduction of the background area has affected the display of the core diagnostic area, to ensure that the overall image conforms to the global features of the original diagnostic description.

[0058] Step S4024: When the core diagnostic area and the associated area are adjusted twice consecutively and both passes the verification, and there is no interference information in the background area, lock the current adjustment parameters and stop adjusting the area; if it fails, expand the adjustment range of the core diagnostic area and repeat the above steps until the verification is successful.

[0059] This involves continuously verifying and locking the adjustment parameters to ensure the stability of the adjustment results, while avoiding over-adjustment within the adjustment range, thus balancing adjustment accuracy and efficiency.

[0060] This embodiment achieves a precise and controllable transformation from the original image to the target image by prioritizing adjustments based on clinical impact grading, performing tiered verification to ensure dual compliance of the original image's core features and diagnostic needs, and optimizing the adjustment range. In terms of implementation, adjustments are first made sequentially according to clinical importance to ensure priority optimization of key diagnostic information; each adjustment is followed by two levels of verification to prevent damage to the core structure; the adjustment range is expanded or narrowed based on the verification results to reduce ineffective operations; finally, continuous verification locks in the adjustment parameters to ensure that the generated image both conforms to the target diagnostic result and retains the clinical baseline features of the original image. This approach ensures that the iterative process closely aligns with clinical diagnostic logic, improving the accuracy and reliability of the adjustments.

[0061] S105: Perform quality verification on the generated diagnostic result images. After verification, store the original medical images, original image diagnostic description information, target diagnostic result information, and diagnostic result images together in the medical image database to form a diagnostic image dataset.

[0062] In some embodiments, the quality verification process includes semantic consistency checks to determine whether the location of nodules in the comparative images matches the diagnostic description, and whether the measured values ​​match the target requirements. Clinical effectiveness verification involves evaluation by three attending physicians to assess whether the adjusted images make it easier to determine the degree of adhesion and whether the diagnostic time is reduced. After successful verification, the original images, original descriptions, target diagnoses, generated images, and processing parameters are stored together, labeled with patient ID, examination date, and processing personnel, forming a traceable and complete dataset.

[0063] This embodiment ensures the clinical usability of the output images while providing data support for research and process improvement, forming a closed loop of continuous improvement.

[0064] In one embodiment of the present invention, based on step S102, the following will provide a possible embodiment and its specific implementation will be described in a non-limiting manner. Step S102 specifically includes: Step S1021: Obtain the set of core features defined in the original image diagnostic description information and transform them into constraints that the model can recognize.

[0065] The constraints include: limiting the types of objects to be retained during model analysis by using predefined classification labels; labeling the regions in the original image that need to maintain morphological features and generating structural masks; and extracting statistical quantities such as the mean grayscale value and contrast baseline of each region in the original image as the benchmark range for model adjustment.

[0066] The core feature set can be organ type, key structural location, and normal morphological parameters.

[0067] Step S1022: Transform the desired features in the target diagnostic results information into the analysis objectives that the model needs to achieve. The analysis objectives can be defined as: marking the areas that need to be enhanced and setting the contrast threshold after enhancement; marking the areas that need to be smoothed and setting the noise reduction ratio; identifying the abnormal structures that need to be adjusted and providing the corrected reference positions.

[0068] Step S1023: Call the pre-trained multi-task image analysis model, and simultaneously input the original medical image, the original diagnostic constraints, and the target diagnostic constraints, and perform the following analysis operations: Multi-scale features of the original image are extracted using a CNN network; The features of the original diagnostic constraints are matched with the extracted features to generate a maintenance score; The differences between the features of the target diagnostic constraints and the extracted features are calculated to generate a heatmap of adjustment requirements.

[0069] Step S1024: Output the first image status trend information and the second image status trend information respectively, and ensure accuracy through double verification.

[0070] It should be noted that the first image state trend information can be obtained by calculating the rate of change of the matching degree between the original image features and the original constraint features over time steps, such as the change in SSIM every 10 seconds, which reflects the natural degradation or stabilization pattern of the original image under the constraint of maintaining the status quo.

[0071] The second image state trend information is obtained by calculating the rate of change of the difference between the target constraint features and the current image features over time steps, such as the decrease in MSE every 10 seconds, which reflects the rate and direction of the original image adjusting to the target state.

[0072] It should be noted that the first trend verification is to check whether the output trend only includes non-critical detail adjustments, such as noise attenuation, slight sharpening of blurred areas, and whether the core features have changed. The second trend validation is to check whether the output trend covers all target diagnostic needs and whether the adjustment range is within the clinically acceptable range.

[0073] Furthermore, in this embodiment, the original diagnostic description information and the target diagnostic result information are encoded into feature masks and parameter thresholds, respectively, and then input into the image analysis model. The image analysis model extracts local detail features and global semantic features of the original image through convolutional kernels at different levels, corresponding to the analysis requirements of the original constraints and the target constraints, respectively.

[0074] During the model training phase, a loss function is designed to balance the importance of the two constraints, ensuring that the image analysis model does not lose core information while effectively approximating the target features during adjustment. First trend information is generated by calculating the matching degree between the original image features and the constraint features; second trend information is generated by calculating the difference between the target features and the current features.

[0075] It can be seen that by limiting the model to adjust only non-critical details during analysis, the core diagnostic information of the original image is not destroyed, thus avoiding misdiagnosis caused by blind adjustment; by clarifying the characteristic requirements of the target diagnostic results, the model is provided with adjustment direction.

[0076] In one embodiment of the present invention, based on step S103, the following will provide a possible embodiment and its specific implementation will be described in a non-limiting manner. Step S103 specifically includes: Step S1031: Define the contrast dimensions according to the clinical functional zoning of the images. Divide the first image status trend information and the second image status trend information into the direct diagnosis area, the auxiliary reference area, and the area with no diagnostic significance according to clinical function.

[0077] The direct diagnostic area includes key structures in imaging used for core diagnosis, such as the "brain region where the lesion is located" in brain MRI.

[0078] The auxiliary reference area contains diagnostically relevant structures, such as the vascular distribution area around the lesion.

[0079] Areas of no diagnostic significance are background areas that do not affect diagnosis, such as blank areas at the edges of images, ensuring that differential parameter extraction focuses on areas of clinical value.

[0080] Step S1032: Based on the operational purpose of the image analysis, such as "removing artifacts," "strengthening boundaries," or "calibrating grayscale," map the time steps of the first image state trend information and the second image state trend information to the same operational purpose. For example, if the second time step of the first trend is used for "strengthening boundaries," then match the time step of the second trend with the same operational purpose to avoid incorrect comparison of parameters for different operational purposes.

[0081] Step S1033: For the aligned direct diagnostic area, extract diagnostically relevant difference information, such as differences in the continuity of lesion boundaries and differences in density uniformity within the lesion. For the auxiliary reference area, extract characteristic differences of associated structures, such as differences in the distance between blood vessels and lesions and differences in the degree of edema in surrounding tissues. For areas without diagnostic significance, only record whether there is interfering difference information, such as whether new artifacts are generated due to adjustments.

[0082] It should be noted that the difference information related to diagnosis can be described as follows: for the direct diagnostic area, it can be described as "the difference in the continuity of the lesion boundary is 25%, which needs to be adjusted to ≥80%"; for the auxiliary reference area, it can be described as "the degree of correlation influence", such as "the difference in the distance between the blood vessel and the lesion is 5mm, which has a moderate influence"; and for the area with no diagnostic significance, it can be judged as "whether there is interference". This ensures the accuracy of the core parameters and simplifies the expression of the secondary parameters, which meets the differentiated clinical needs for information accuracy.

[0083] Step S1034: Sort the extracted differential information according to clinical decision-making logic. For example, first process the lesion boundary differences in the direct diagnostic area, then process the vascular association differences in the auxiliary reference area, and finally process the interference differences in the area with no diagnostic significance. Each sorting information defines the adjustment object, such as "lesion boundary in the direct diagnostic area"; the adjustment direction, such as "from blurry to clear"; and the adjustment threshold. All sorting information is labeled with constraints, such as "the coordinate deviation between the adjusted lesion location and the original image diagnosis description does not exceed 1mm", to ensure that the sorting information directly guides subsequent image adjustments and does not deviate from the original diagnostic benchmark.

[0084] This embodiment combines the constraints of the original image diagnosis description, such as the unchangeable features of lesion location and size, so that the transformation rules not only clarify the operation path, but also limit the operation boundary. For example, the deviation of the adjusted lesion size from the original description is ≤0.3cm, which solves the problem of over-adjustment caused by the lack of clinical constraints in the existing technology.

[0085] This embodiment aligns clinical functional zoning with operational objectives, transforming abstract trend information into differential parameters closely related to diagnostic logic. These parameters are then integrated into constrained transformation rules according to the clinical decision-making sequence, ensuring that the differential extraction and ranking rule generation always serve clinical diagnostic needs. In terms of implementation, functional zoning first filters out regions with no diagnostic value, reducing interference from invalid parameters. Next, operational objectives align time steps, ensuring that compared parameters belong to the same processing stage. When extracting differential parameters, core differences in directly diagnostic areas are prioritized, while secondary differences in auxiliary and meaningless areas are downplayed. Finally, parameters are integrated according to the clinical decision-making sequence, resulting in ranking rules that clearly define the adjustment objects and goals while being constrained by the original image diagnostic description, making the rules more aligned with the actual diagnostic and treatment process.

[0086] In one embodiment of the present invention, based on step S104, the following will provide a possible embodiment and its specific implementation will be described in a non-limiting manner. Step S104 specifically includes: Step S1011: Based on the clinical importance classification of the difference parameters in the transformation rules, prioritize local adjustments to high-priority regions.

[0087] Optionally, the clinical importance grading method can be lesion boundary adjustment > normal tissue contrast adjustment > background noise smoothing.

[0088] For example, the edge clarity of the lesion area in the original medical image is first enhanced, then the grayscale of the surrounding normal tissue is adjusted, and finally the artifacts at the image edge are processed to ensure that the key diagnostic areas are optimized first.

[0089] Step S1012: Perform layered verification after each local adjustment.

[0090] The first layer verifies that the type and location of the core objects in the original image have not changed; the second layer verifies whether the adjusted regional features match the description of the target diagnosis results. After both layers of verification are passed, the next iteration begins.

[0091] Step S1013: If the high-priority area still fails to meet the standard after two adjustments, expand the adjustment range of the area; if it has met the standard, reduce the adjustment range of the low-priority area to reduce invalid adjustments.

[0092] Step S1014: Stop the iteration when the adjustment results of all regions pass the stratified verification twice consecutively, and the key features in the target diagnosis results are clearly displayed. Simultaneously, save the adjustment parameters of the high-priority regions in this iteration as a reference template for similar image processing.

[0093] This embodiment ensures dual compliance of core structure and diagnostic requirements through clinical priority-oriented local adjustments and stratified verification, and improves efficiency by adjusting the scope of iterations. Ultimately, it generates a stable image that retains the key features of the original image and meets the target diagnostic results. Adjustments are made in order of clinical importance to ensure that key areas are optimized first. Each adjustment is verified in two layers to avoid structural damage or diagnostic deviation. The adjustment scope is expanded or reduced based on the verification results to balance efficiency and accuracy. After the target is met, the parameter template is saved for subsequent reuse.

[0094] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0095] The following are embodiments of the medical image information processing system provided in this disclosure. This system and the medical image information processing methods in the above embodiments belong to the same inventive concept. For details not described in detail in the embodiments of the medical image information processing system, please refer to the embodiments of the above medical image information processing methods.

[0096] like Figure 2 As shown, the system includes: The data acquisition module 201 is used to acquire the original medical image to be processed, as well as the original image diagnostic description information corresponding to the original medical image, and to acquire the target diagnostic result information. Image analysis module 202 is used to analyze the original medical images based on two different constraints using a pre-trained image analysis model: Using the original image diagnostic description information as a constraint, the image analysis model maintains the core features of the displayed objects in the original medical images during the analysis process, only adjusting non-critical details, and outputs the first image state trend information. Using the target diagnostic result information as a constraint, the image analysis model adjusts the display content of the original medical image to make it conform to the expected object characteristics in the diagnostic result information, and outputs the second image state trend information. The trend calculation module 203 is used to perform temporal alignment of the trend information of the first image state and the trend information of the second image state, extract the difference parameters at the same time step or feature dimension, and form a transformation rule from the original state to the target state. The iterative adjustment module 204 adjusts the original medical image through iterative image processing based on transformation rules: in each iteration, the original image is locally or globally adjusted according to the difference parameters corresponding to the current step, and the adjusted image is verified to meet the original diagnostic description constraints; the iteration is repeated until the adjusted image meets the display requirements of the target diagnostic result information. The verification and association module 205 is used to verify the quality of the generated diagnostic result images. After the verification is passed, the original medical image, the original image diagnostic description information, the target diagnostic result information, and the diagnostic result image are associated and stored in the medical image database to form a diagnostic image dataset.

[0097] like Figure 3 As shown, this application also provides an electronic device, including a display module 103, a memory 102, a processor 101, and a computer program stored in the memory and executable on the processor 101, wherein the processor 101 executes the program to implement the steps of a medical image information processing method.

[0098] In embodiments of the present invention, electronic devices include, but are not limited to, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the embodiments described and / or claimed herein.

[0099] In this embodiment, processor 101 may be implemented using at least one of an application-specific integrated circuit, a programmable logic device, a field-programmable gate array, a processor, a controller, a microcontroller, a microprocessor, or an electronic unit designed to perform the functions described herein. In some cases, such an implementation may be implemented within a controller. For software implementation, implementations such as processes or functions may be implemented with separate software modules that allow the performance of at least one function or operation. Software code may be implemented by a software application (or program) written in any suitable programming language, and the software code may be stored in memory and executed by the controller.

[0100] The display module 103 is used to display information input by the user or information provided to the user. The display module 103 may include a display panel, which may be configured in the form of a liquid crystal display, an organic light-emitting diode, or the like.

[0101] The memory 102 can be used to store software programs and various data. The memory 102 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0102] This application also provides a storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the medical image information processing method.

[0103] The storage medium may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example,, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0104] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A medical image information processing method, characterized by comprising: Comprise: S101: obtain the original medical image to be processed, and the original image diagnostic description information corresponding to the original medical image, and simultaneously obtain the target diagnostic result information; S102: through the pre-trained image analysis model, the original medical image is analyzed based on two different constraint conditions respectively: The original image diagnostic description information is used as the constraint, so that the image analysis model maintains the core characteristics of the display object in the original medical image unchanged during the analysis process, and only adjusts the non-key details, and outputs the first image state trend information; The target diagnostic result information is used as the constraint, so that the image analysis model adjusts the display content of the original medical image to make it conform to the object characteristics expected in the diagnostic result information, and outputs the second image state trend information; S103: The first image state trend information and the second image state trend information are time-aligned, the difference parameters in the same time step or feature dimension are extracted, and the transformation rule from the original state to the target state is formed; S104: Based on the transformation rule, the original medical image is adjusted through iterative image processing: in each iteration, the original image is locally or globally adjusted according to the difference parameter corresponding to the current step, and it is verified whether the adjusted image meets the display requirements of the original diagnostic description constraint; repeat the iteration until the adjusted image meets the display requirements of the target diagnostic result information; S105: The generated diagnostic result image is verified, and after verification, the original medical image, the original image diagnostic description information, the target diagnostic result information and the diagnostic result image are stored in the medical image database to form a diagnostic image data set.

2. The medical image information processing method according to claim 1, wherein the step of outputting the first image state trend information further comprises: When the image analysis model performs the first image analysis process on the original medical image based on the original image diagnostic description information, the output parameters of the neural network are extracted in real time through the parameter acquisition module inside the image analysis model as the first output parameters; The first output parameters are subjected to multi-dimensional feature extraction and trend analysis to generate the first image state trend information, which describes the natural change rule of the original medical image under the maintenance status constraint in the form of quantitative parameters; When the image analysis model performs the second image analysis process on the original medical image based on the diagnostic result information, the output parameters of the neural network are synchronously collected through the model parameter monitoring module; The second output parameters are subjected to reverse analysis and target-oriented trend derivation to generate the second image state trend information, which describes the change rule of the original medical image from the current state to the target state in the form of path parameters.

3. The medical image information processing method according to claim 2, wherein the step S102 specifically comprises: Obtain the core feature set defined in the original image diagnostic description information, and convert it into a constraint condition for model recognition; ​ ​ The constraint conditions include: limiting the type of objects to be retained during model analysis by a pre-defined classification label; labeling the region in the original image that needs to maintain morphological features to generate a structure mask; extracting the gray mean value and contrast baseline of each region in the original image as the baseline range during model adjustment; Convert the expected features in the target diagnosis result information into the analysis objectives that the model needs to achieve; Call a pre-trained multi-task image analysis model, input the original medical image, the original diagnosis constraint, and the target diagnosis constraint, and perform the following analysis operations: Extract multi-scale features of the original image through a CNN network; Match the features of the original diagnosis constraint with the extracted features to generate a maintenance score; Calculate the difference between the features of the target diagnosis constraint and the extracted features to generate an adjustment requirement heat map; Output the first image state trend information and the second image state trend information respectively.

4. The medical image information processing method according to claim 1, wherein between step S103 and step S104, further comprising: representing the first image state trend information as a multi-scale tensor structure containing space-time dimensions; using an improved time warping algorithm to perform space-time alignment on the first image state trend information tensor and the second image state trend information tensor, generating the first parameter sequence and the second parameter sequence after time step alignment by constructing a space-time distance matrix and an optimal path planning; constructing a difference calculation model under the constraint of diagnosis semantics, for each aligned time step: calculating the Mahalanobis distance between the second parameter sequence and the first parameter sequence in the feature space to obtain a difference vector containing direction information; applying an attention mechanism to weight the difference vector to enhance the parameter expression of the region related to the diagnosis; filtering out irrational difference parameters through regularization processing constrained by medical prior knowledge; inputting the extracted difference vector sequence into a space-time convolutional network to learn the time sequence dependence of parameter changes through a bidirectional LSTM layer, generating a continuous diagnosis result trend information tensor.

5. The medical image information processing method according to claim 1, wherein the step of adjusting the original image according to the difference parameters corresponding to the current step, and verifying whether the adjusted image meets the original diagnosis description constraint, specifically comprises: dividing the difference parameters into a core diagnosis area, an associated area, and a background area, and preferentially adjusting the core diagnosis area locally; checking whether the adjusted core diagnosis area still meets the pre-set features according to the core diagnosis area in the original image diagnosis description, and if the verification is passed, adjusting the contrast of the associated area according to the difference parameters based on the adjustment result of the core diagnosis area, and performing noise reduction processing on the background area; checking whether the adjustment of the associated area has not destroyed the features in the original description, and whether the noise reduction of the background area has not affected the display of the core diagnosis area; when the core diagnosis area and the associated area pass the verification for two consecutive times, and the background area has no interference information, locking the current adjustment parameters and stopping the area adjustment; if not, expanding the adjustment range of the core diagnosis area and repeating the above steps until the verification is passed.

6. The medical image information processing method according to claim 1, wherein S103 specifically comprises: ​ ​ ​ According to the definition of the contrast dimension of the clinical function partition of the image, the first image state trend information and the second image state trend information are divided into a direct diagnosis area, an auxiliary reference area, and a non-diagnostic significance area according to the clinical function; According to the operation purpose of the image analysis, the time steps of the first image state trend information and the second image state trend information are corresponded to the same operation purpose; For the aligned direct diagnosis area, difference information related to diagnosis is extracted; for the auxiliary reference area, feature differences of associated structures are extracted; for the non-diagnostic significance area, only interference difference information is recorded; The extracted difference information is sorted according to the clinical decision logic, the vascular associated difference of the auxiliary reference area is processed, and finally the interference difference of the non-diagnostic significance area is processed; each sorting information defines an adjustment object, an adjustment direction, and an adjustment threshold, and all sorting information is marked with a constraint condition.

7. The medical image information processing method of claim 1, wherein Step S104 specifically comprises: According to the clinical importance classification of the difference parameters in the transformation rule, the high-priority area is preferentially subjected to local adjustment; After each local adjustment, the first layer verifies that the types and positions of the core objects in the original image have not changed; the second layer verifies whether the features of the adjusted area meet the description of the target diagnosis result; after both layers of verification pass, the next iteration is entered; If the high-priority area still does not meet the standard after 2 adjustments, the adjustment range of the high-priority area is expanded; if it has met the standard, the adjustment range of the low-priority area is reduced; When the adjustment results of all areas pass the layered verification for 2 consecutive times, and the key features in the target diagnosis result are clearly displayed, the iteration is stopped.

8. A medical image information processing system characterized by comprising: The system is used to implement the medical image information processing method of any one of claims 1 to 7; The system comprises: A data acquisition module is configured to acquire an original medical image to be processed and original image diagnosis description information corresponding to the original medical image, and to acquire target diagnosis result information; An image analysis module is configured to analyze the original medical image based on two different constraint conditions by using a pre-trained image analysis model: The original image diagnosis description information is used as a constraint to make the image analysis model maintain the core features of the displayed objects in the original medical image unchanged during the analysis process, and only adjust non-critical details, and output first image state trend information; The target diagnosis result information is used as a constraint to make the image analysis model adjust the display content of the original medical image to meet the expected object features in the diagnosis result information, and output second image state trend information; A trend calculation module is configured to perform time sequence alignment on the first image state trend information and the second image state trend information, extract difference parameters at the same time step or feature dimension, and form a transformation rule from the original state to the target state; An iterative adjustment module is configured to adjust the original medical image by iterative image processing based on the transformation rule: in each iteration, the original image is subjected to local or global adjustment according to the difference parameters corresponding to the current step, and it is verified whether the adjusted image meets the original diagnosis description constraint; the iteration is repeated until the adjusted image meets the display requirements of the target diagnosis result information; The verification association module is configured to verify the quality of the generated diagnostic result image, and after verification, the original medical image, original image diagnostic description information, target diagnostic result information and diagnostic result image are stored in the medical image database in association to form a diagnostic image data set.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor implements the steps of the medical image information processing method according to any one of claims 1 to 7 when executing the program.

10. A storage medium having stored thereon a computer program, characterized in that The computer program implements the steps of the medical image information processing method according to any one of claims 1 to 7 when executed by the processor.

Images