Method, system, and storage medium for generating clinical documentation based on breast mri images

Abstract

This invention relates to a method, system, and storage medium for generating clinical documents based on breast MRI images, belonging to the field of MRI image clinical processing technology. By constructing a unified, high-quality data transformation paradigm, it achieves precise semantic alignment between breast 3D MRI image features and clinical terminology, and utilizes an efficient feature compression mechanism to overcome the computational bottleneck of large-scale medical data processing. The core objective of the system is to endow artificial intelligence with cross-modal reasoning capabilities, enabling it not only to accurately identify breast tumor lesions and provide decision-making suggestions such as hierarchical management, but also to automatically and systematically generate four key clinical documents—imaging reports, outpatient medical records, pathology records, and initial progress notes—within a single technical architecture. Through the implementation of this invention, the aim is to significantly improve the objectivity and consistency of breast tumor diagnosis, greatly optimize clinical workflow efficiency, and provide comprehensive intelligent support for precision medicine in breast cancer.

Description

Technical Field

[0001] This invention relates to the field of MRI image clinical processing technology, and in particular to a method, system, and storage medium for generating clinical documents based on breast MRI images. Background Technology

[0002] Breast cancer is the most common malignant tumor among women worldwide, and its early detection and accurate classification are crucial for improving patient survival. Currently, multimodal magnetic resonance imaging (MRI), with its extremely high soft tissue resolution, has become a core tool for breast cancer screening, preoperative evaluation, and treatment monitoring. However, the diagnosis and treatment of breast cancer face significant challenges in clinical practice. First, the complex structure of breast tissue and the significant overlap in imaging manifestations between benign and malignant tumors (such as certain medullary carcinomas and fibroadenomas) make diagnosis highly dependent on the physician's subjective experience, easily leading to missed diagnoses or unnecessary biopsies. Second, with the development of precision medicine, the volume of clinical documentation (such as radiology reports, pathological diagnoses, and progress notes) has surged, requiring the integration of cross-modal information such as imaging, clinical history, and immunohistochemistry. Traditional manual writing methods are not only inefficient but also struggle to ensure structure and standardization.

[0003] Although existing deep learning models such as U-Net have made breakthroughs in lesion segmentation, most of them are small models with single functions. They lack a deep understanding of medical semantics and cannot perform logical reasoning and cross-modal narration like human doctors. Furthermore, when dealing with highly specialized breast 3D MRI data, general-purpose large models often have problems such as difficulty in visual feature alignment, huge consumption of computational resources, and "hallucinations" in specific medical tasks.

[0004] Therefore, existing technologies still need to be improved and enhanced. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the purpose of this invention is to provide a method, system and storage medium for generating clinical documents based on breast MRI images. The aim is to solve the problems of the prior art lacking a deep understanding of medical semantics, often resulting in difficulties in visual feature alignment, huge consumption of computing resources and "illusions" in specific medical tasks when processing highly specialized breast 3D MRI data.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, a method for generating clinical documentation based on breast MRI images, the method comprising: Collect historical MRI images and corresponding text data, including historical radiology reports, outpatient medical records, and pathology diagnosis records corresponding to the historical radiology reports; The historical MRI images and the text data are preprocessed; A multimodal dataset is constructed based on the preprocessed historical MRI images and the text data; Acquire the current MRI image; Based on a visual perception architecture, pathological feature recognition and analysis are performed on the current MRI image to obtain the image features of the current MRI image; Based on the image features of the current MRI image and the multimodal dataset, generate clinical documentation corresponding to the current MRI image; BI-RADS constraints are set within the visual perception architecture to generate an anti-illusion adjustment mechanism; The confidence level between the image features and the clinical document is detected based on the aforementioned anti-hallucination adjustment mechanism; Based on the confidence level, the clinical documentation is corrected to generate the final clinical documentation.

[0007] Furthermore, the preprocessing of the historical MRI images and the text data includes: A cross-modal semantic fusion method was used to extract the image features of the historical MRI images. The image features included the anatomical location, morphological features, blood supply characteristics, marginal infiltration, blood flow signal, enhancement morphology, internal heterogeneity, and marginal spiculation signal of the breast tumor. A shared semantic space is constructed based on the text data, the image features are mapped to the shared semantic space, and the image features and the text data are semantically aligned within the shared semantic space.

[0008] Furthermore, the step of generating a clinical document corresponding to the current MRI image based on the image features of the current MRI image and the multimodal dataset includes: The clinical documents include imaging reports, outpatient medical records, pathology records, and progress notes; The imaging report is generated based on the imaging features of the current MRI image; The outpatient medical record is generated by combining the medical history information corresponding to the image features of the current MRI image in the multimodal dataset; The pathological record is derived based on the pathological sections or puncture results corresponding to the imaging features of the current MRI image in the multimodal dataset. Based on the imaging report, the outpatient medical record, and the pathology record, the diagnosis and treatment approach is sorted out, and the medical record is generated.

[0009] Furthermore, the step of setting BI-RADS constraints within the visual perception architecture to generate an anti-hallucination adjustment mechanism includes: A low-level hard constraint is established with the BI-RADS standard knowledge matrix as the core, forming a strong constraint condition for the image features of the current MRI image initially extracted by the visual perception architecture. When the image features of the current MRI image match the BI-RADS standard knowledge matrix, the corresponding BI-RADS semantic association point is activated, and a BI-RADS classification is generated.

[0010] Further, the step of detecting the confidence level between the image features and the clinical document based on the anti-hallucination adjustment mechanism, and correcting the clinical document based on the confidence level to generate the final clinical document includes: Based on the pre-set confidence threshold of the anti-hallucination adjustment mechanism, the confidence level of the image features and the clinical document is detected in real time. The confidence level is compared with the confidence threshold. When the confidence level exceeds the confidence threshold, the clinical document conflicts with the image features, triggering a reverse modification route to forcibly modify the clinical document and correct it.

[0011] Furthermore, the step of performing pathological feature recognition and analysis on the current MRI image based on a visual perception architecture to obtain the image features of the current MRI image includes: The current MRI image is a multi-sequence MRI image, which includes T1WI sequence images and T2WI sequence images; Spatiotemporal feature decoupling and fine-grained lesion focus are embedded in the encoding end of the visual perception architecture. The encoding end includes a spatial path and a related path. The spatial path locks the T1WI sequence image and the T2WI sequence image and extracts the spatial features of the multi-sequence MRI image. The spatial features include extracting breast gland classification, mass anatomical morphology, and edge breathing. The relevant pathway dynamically enhances the multi-sequence MRI images, generating multiple continuous time windows. Based on these multiple continuous time windows, the dynamic evolution rate of the contrast agent inflow-termination cycle of the multi-sequence MRI images, as well as the rapid enhancement rate and reversal slope of the lesions in the early stage of the multi-sequence MRI images, are accurately extracted. The spatial features extracted by the spatial path and the dynamic evolution rate, rapid enhancement rate and reversal slope extracted by the related path are fused to generate the image features based on spatiotemporal feature coupling.

[0012] Secondly, a system for generating clinical documents based on breast MRI images, the system comprising: The acquisition module is used to acquire historical MRI images and text data corresponding to the historical MRI images. The text data includes historical radiology reports, outpatient medical records and pathology diagnosis records corresponding to the historical radiology reports. A preprocessing module is used to preprocess the historical MRI images and the text data; A dataset construction module is used to construct a multimodal dataset based on the preprocessed historical MRI images and the text data; The image processing module is used to acquire the current MRI image and perform pathological feature recognition and analysis on the current MRI image based on the visual perception architecture to obtain the image features of the current MRI image. The document generation module is used to generate clinical documents corresponding to the current MRI image based on the image features of the current MRI image and the multimodal dataset; The document correction module is used to set BI-RADS constraints within the visual perception architecture to generate an anti-illusion adjustment mechanism, detect the confidence level between the image features and the clinical document based on the anti-illusion adjustment mechanism, and correct the clinical document based on the confidence level to generate the final clinical document.

[0013] Thirdly, a system for generating clinical documents based on breast MRI images includes a memory and a processor, the memory storing programs or instructions executable on the processor, the programs or instructions, when executed by the processor, implementing the steps of the method for generating clinical documents based on breast MRI images as described above.

[0014] Fourthly, a readable storage medium having a program or instructions stored thereon, which, when executed by a processor, implement the steps of the method for generating clinical documentation based on breast MRI images as described above.

[0015] The technical solution adopted in this invention has the following beneficial effects: First, this study achieves high-quality alignment and transformation of cross-modal data. Addressing the challenge of mapping between breast tumor imaging features and complex clinical descriptions, the core work of this study lies in constructing a high-standard cross-modal multimodal dataset. Through a unified data transformation paradigm, the original breast tumor images are deeply correlated and precisely aligned with their corresponding three text modalities, significantly improving the accuracy and objectivity of breast cancer-assisted diagnosis: effectively reducing reliance on physicians' subjective experience during image interpretation and enhancing the reliability of differential diagnosis of complex lesions.

[0016] Second, during the dataset construction process, cross-modal semantic fusion technology is used to ensure that the image features in the images are highly consistent with the three types of text data in terms of semantic logic, which greatly optimizes the standardization and efficiency of clinical document generation: the system can automatically generate structured medical documents with complete information and rigorous logic, effectively alleviating the text processing burden of radiologists and breast surgeons.

[0017] Third, a text "anti-illusion" adjustment mechanism based on BI-RADS form report contract point constraints is implemented. To address the "medical illusion" and logical contradictions that easily arise when multimodal large models generate highly specialized, long clinical texts, a low-level hard constraint module is constructed with the BI-RADS (Breast Imaging and Data System) standard knowledge matrix at its core. This module corrects the clinical text by comparing the confidence levels of image features and clinical text in real time, thereby ensuring the absolute authenticity and reliability of the four types of generated clinical texts at the underlying algorithm architecture. This provides intelligent and personalized clinical decision support: the system can not only accurately locate lesions but also provide professional suggestions such as BI-RADS hierarchical management and immunohistochemical detection prompts based on multimodal information, assisting physicians in developing more scientific treatment plans.

[0018] Fourth, it achieves decoupling and fine-grained lesion focus on spatiotemporal features of multi-sequence (DCE-MRI). Addressing the technical shortcomings of multi-sequence breast MRI images where spatial anatomical structures and dynamic hemodynamic features easily interfere with each other, leading to the concealment of visual features of occult lesions, this model combines the spatial and temporal features of multi-sequence MRI images. This enables the large model to accurately identify small in situ carcinomas or special renal carcinomas with occult blood supply, significantly enhancing the system's accuracy in identifying and representing long-tailed and rare tumors, and significantly improving its ability to identify complex and rare tumors. It also improves the model's accuracy in representing rare breast pathological types, enabling it to keenly capture subtle lesion features and reducing the risk of misdiagnosis and missed diagnosis due to subjective experience differences. Attached Figure Description

[0019] Figure 1 This is one of the flowcharts for a method of generating clinical documentation based on breast MRI images according to an embodiment of the present invention; Figure 2 This is a second flowchart of a method for generating clinical documents based on breast MRI images according to an embodiment of the present invention; Figure 3 One of the block diagrams of a system for generating clinical documentation based on breast MRI images according to an embodiment of the present invention; Figure 4 This is a second block diagram of a system for generating clinical documents based on breast MRI images according to an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and effects of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0021] To address the current challenges in processing breast MRI images, this invention provides a method, system, and storage medium for generating clinical documentation based on breast MRI images.

[0022] In this embodiment, a method for generating clinical documents based on breast MRI images, combined with Figure 1 The method includes: Step 102: Collect historical MRI images and text data corresponding to the historical MRI images. The text data includes historical radiology reports, outpatient medical records and pathology diagnosis records corresponding to the historical radiology reports. Step 104: Preprocess the historical MRI images and the text data; Step 106: Construct a multimodal dataset based on the preprocessed historical MRI images and the text data; Step 108: Acquire the current MRI image; Step 110: Perform pathological feature recognition and analysis on the current MRI image based on the visual perception architecture to obtain the image features of the current MRI image; Step 112: Generate a clinical document corresponding to the current MRI image based on the image features of the current MRI image and the multimodal dataset; Step 114: Set BI-RADS constraints within the visual perception architecture to generate an anti-hallucination adjustment mechanism; Step 116: Detect the confidence level between the image features and the clinical document based on the anti-hallucination adjustment mechanism; Step 118: Based on the confidence level, the clinical document is corrected to generate the final clinical document.

[0023] In this embodiment, combined with Figure 2 Historical MRI images and corresponding text data are collected. The historical MRI images are deeply correlated and precisely aligned with three corresponding text modalities. The text data includes historical radiology reports, outpatient medical records and pathology diagnosis records corresponding to the historical radiology reports. The image features of historical MRI images are mapped to relevant text descriptions, providing a solid data foundation for the construction of the dataset and ensuring the stability of the mapping relationship between breast tumor image features and complex clinical descriptions.

[0024] In this embodiment, the preprocessing of the historical MRI images and the text data includes: A cross-modal semantic fusion method was used to extract the image features of the historical MRI images. The image features included the anatomical location, morphological features, blood supply characteristics, marginal infiltration, blood flow signal, enhancement morphology, internal heterogeneity, and marginal spiculation signal of the breast tumor. A shared semantic space is constructed based on the text data, the image features are mapped to the shared semantic space, and the image features and the text data are semantically aligned within the shared semantic space.

[0025] By utilizing the aforementioned cross-modal semantic fusion technology, we can ensure that the image features in the images (such as edge infiltration, blood flow signals, etc.) are highly consistent with the three types of text data in terms of semantic logic, thereby achieving high-quality data alignment. This provides a high-quality data foundation for the subsequent establishment of the dataset and a solid foundation for the model to achieve cross-dimensional reasoning from visual perception to the generation of professional medical temporary documents.

[0026] In this embodiment, combined with Figure 2 The step of generating a clinical document corresponding to the current MRI image based on the image features of the current MRI image and the multimodal dataset includes: The clinical documents include imaging reports, outpatient medical records, pathology records, and progress notes; The system is able to automatically generate a logically rigorous radiological report based on the input breast multi-sequence MRI images and the image's impact characteristics, by searching for relevant report information data in a multimodal dataset. Furthermore, in the multimodal dataset, we search for patient medical history records and other data corresponding to the current breast multi-sequence MRI images, and then generate outpatient medical records in relevant formats. Then, based on the imaging features of the current MRI images and the corresponding pathological sections or puncture results in the multimodal dataset, detailed pathological records are inferred. Based on the imaging report, the outpatient medical record, and the pathology record, the diagnosis and treatment approach is sorted out, and a standardized initial medical record for the admitted patient with the ability to automatically sort out the diagnosis and treatment approach is generated.

[0027] The above methods enable the automated and standardized generation of multidimensional clinical documents. Through a progressive fine-tuning strategy, the model is equipped with the ability to process four core medical documents simultaneously within a single architecture. Furthermore, this multi-task collaborative mode ensures the consistency and professionalism of information across various medical documents, significantly reducing the document writing burden on clinical medical staff.

[0028] In this embodiment, the visual perception architecture adopted in this application is a currently common model architecture, such as ResNet, which is a CNN-like architecture that extracts hierarchical features from edge texture to semantics by stacking convolutional layers and pooling layers, or introduces a self-attention mechanism to capture global dependencies after dividing the image into blocks, or the Swin Transformer architecture, which takes into account both local details and global semantics through hierarchical window design and has outstanding accuracy performance on large-scale image tasks. A related lightweight visual perception architecture can also be adopted for current MRI images to achieve feature recognition processing of MRI images.

[0029] In this embodiment, setting BI-RADS constraints within the visual perception architecture to generate an anti-hallucination adjustment mechanism includes: A low-level hard constraint is established with the BI-RADS standard knowledge matrix as the core, forming a strong constraint condition for the image features of the current MRI image initially extracted by the visual perception architecture. When the image features of the current MRI image match the BI-RADS standard knowledge matrix, the corresponding BI-RADS semantic association point is activated, and a BI-RADS classification is generated.

[0030] Among them, BI-RADS is a widely used breast imaging standardization system. Through a classification of 0-6 levels, it provides a clear and consistent framework for risk assessment and clinical management of breast lesions. This invention embeds this architecture into the visual perception architecture to form constraints and solve the problem of "medical illusion" and logical contradictions that are easily generated when multimodal large models generate highly professional and long clinical texts.

[0031] In one embodiment of the present invention, the step of detecting the confidence level between the image features and the clinical document based on the anti-hallucination adjustment mechanism, and correcting the clinical document based on the confidence level to generate the final clinical document includes: Based on the aforementioned anti-illusion adjustment mechanism, a pre-set confidence threshold is established. The confidence level of the image features and the clinical document is detected in real time. The confidence level is compared with the confidence threshold. When the confidence level exceeds the confidence threshold, a conflict occurs between the clinical document and the image features, triggering a reverse modification route to forcibly modify the clinical document and correct it. This ensures that the generated four types of clinical texts are absolutely authentic and reliable at the underlying algorithm architecture.

[0032] In this embodiment, the step of performing pathological feature recognition and analysis on the current MRI image based on a visual perception architecture to obtain the image features of the current MRI image includes: The current MRI image is a multi-sequence MRI image, which includes T1WI sequence images and T2WI sequence images; Spatiotemporal feature decoupling and fine-grained lesion focus are embedded in the encoding end of the visual perception architecture. The encoding end includes a spatial path and a related path. The spatial path locks the T1WI sequence image and the T2WI sequence image and extracts the spatial features of the multi-sequence MRI image. The spatial features include extracting breast gland classification, mass anatomical morphology, and edge breathing, etc. The relevant pathway dynamically enhances the multi-sequence MRI images, generating multiple continuous time windows. Based on these multiple continuous time windows, the dynamic evolution rate of the contrast agent inflow-termination cycle of the multi-sequence MRI images, as well as the rapid enhancement rate and reversal slope of the lesions in the early stage of the multi-sequence MRI images, are accurately extracted. The spatial features extracted by the spatial path and the dynamic evolution rate, rapid enhancement rate and reversal slope extracted by the correlation path are fused to generate the image features based on spatiotemporal feature coupling. This enables the large model to accurately identify small in situ carcinomas or special renal carcinomas with hidden blood supply, significantly enhancing the system's accuracy in identifying and characterizing long-tailed and rare tumors. It also solves the problem that spatial anatomical structures and dynamic hemodynamic features in current breast multi-sequence MRI images easily interfere with each other, leading to the concealment of visual features of occult lesions.

[0033] In one embodiment of the present invention, combined with Figure 3 A system 200 for generating clinical documents based on breast MRI images is provided. The system includes an acquisition module 210, a preprocessing module 220, a dataset construction module 230, an image processing module 240, a document generation module 250, and a document correction module 260. The acquisition module 210 is used to acquire historical MRI images and text data corresponding to the historical MRI images. The text data includes historical radiology reports, outpatient medical records and pathology diagnosis records corresponding to the historical radiology reports. Preprocessing module 220 is used to preprocess the historical MRI images and the text data; Dataset construction module 230 is used to construct a multimodal dataset based on the preprocessed historical MRI images and the text data; Image processing module 240 is used to acquire the current MRI image and perform pathological feature recognition and analysis on the current MRI image based on a visual perception architecture to obtain the image features of the current MRI image; Document generation module 250 is used to generate clinical documents corresponding to the current MRI image based on the image features of the current MRI image and the multimodal dataset; Document correction module 260 is used to set BI-RADS constraints within the visual perception architecture to generate an anti-illusion adjustment mechanism, detect the confidence level between the image features and the clinical document based on the anti-illusion adjustment mechanism, and correct the clinical document based on the confidence level to generate the final clinical document.

[0034] In this embodiment, the present invention addresses key pain points in the clinical diagnosis and treatment of breast tumors, such as complex imaging features, difficulty in integrating cross-modal information, and heavy burden of medical documentation. It achieves precise semantic alignment between breast 3D MRI imaging features and clinical terminology by constructing a unified, high-quality data transformation paradigm, and utilizes an efficient feature compression mechanism to overcome the computational bottleneck of large-scale medical data processing. The core objective of the system is to endow artificial intelligence with cross-modal reasoning capabilities, enabling it not only to accurately identify breast tumor lesions and provide decision-making suggestions such as hierarchical management, but also to automatically and systematically generate four key clinical documents—imaging reports, outpatient medical records, pathology records, and initial progress notes—within a single technical architecture. Through the implementation of this invention, the objectivity and consistency of breast tumor diagnosis are significantly improved, clinical workflow efficiency is greatly optimized, and comprehensive intelligent support is provided for precision medicine in breast cancer.

[0035] In one embodiment of the present invention, combined with Figure 4 A system 300 for generating clinical documentation based on breast MRI images is provided, including a memory 320 and a processor 310. The memory 320 stores programs or instructions that can run on the processor 310. When the program or instructions are executed by the processor 310, they implement the steps of the method for generating clinical documentation based on breast MRI images as described above. Therefore, this system for generating clinical documentation based on breast MRI images possesses all the beneficial effects of the method for generating clinical documentation based on breast MRI images of any of the above embodiments, which will not be elaborated further here.

[0036] In one embodiment of the present invention, a readable storage medium is provided, on which a program or instructions are stored, which, when executed by a processor, implement the steps of the method for generating clinical documentation based on breast MRI images as described above. Therefore, this readable storage medium possesses all the beneficial effects of the method for generating clinical documentation based on breast MRI images of any of the above embodiments, and will not be elaborated further here.

[0037] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the solutions disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of the invention are indicated by the claims.

Claims

1. A method for generating clinical documentation based on breast MRI images, characterized in that, The method includes: Collect historical MRI images and corresponding text data, including historical radiology reports, outpatient medical records, and pathology diagnosis records corresponding to the historical radiology reports; The historical MRI images and the text data are preprocessed; A multimodal dataset is constructed based on the preprocessed historical MRI images and the text data; Acquire the current MRI image; Based on a visual perception architecture, pathological feature recognition and analysis are performed on the current MRI image to obtain the image features of the current MRI image; Based on the image features of the current MRI image and the multimodal dataset, generate clinical documentation corresponding to the current MRI image; BI-RADS constraints are set within the visual perception architecture to generate an anti-illusion adjustment mechanism; The confidence level between the image features and the clinical document is detected based on the aforementioned anti-hallucination adjustment mechanism; Based on the confidence level, the clinical documentation is corrected to generate the final clinical documentation.

2. The method for generating clinical documents based on breast MRI images according to claim 1, characterized in that, The preprocessing of the historical MRI images and the text data includes: A cross-modal semantic fusion method was used to extract the image features of the historical MRI images. The image features included the anatomical location, morphological features, blood supply characteristics, marginal infiltration, blood flow signal, enhancement morphology, internal heterogeneity, and marginal spiculation signal of the breast tumor. A shared semantic space is constructed based on the text data, the image features are mapped to the shared semantic space, and the image features and the text data are semantically aligned within the shared semantic space.

3. The method for generating clinical documentation based on breast MRI images according to claim 1, characterized in that, The process of generating a clinical document corresponding to the current MRI image based on the image features of the current MRI image and the multimodal dataset includes: The clinical documents include imaging reports, outpatient medical records, pathology records, and progress notes; The imaging report is generated based on the imaging features of the current MRI image; The outpatient medical record is generated by combining the medical history information corresponding to the image features of the current MRI image in the multimodal dataset; The pathological record is derived based on the pathological sections or puncture results corresponding to the imaging features of the current MRI image in the multimodal dataset. Based on the imaging report, the outpatient medical record, and the pathology record, the diagnosis and treatment approach is sorted out, and the medical record is generated.

4. The method for generating clinical documentation based on breast MRI images according to claim 1, characterized in that, The step of setting BI-RADS constraints within the visual perception architecture to generate an anti-hallucination adjustment mechanism includes: A low-level hard constraint is established with the BI-RADS standard knowledge matrix as the core, forming a strong constraint condition for the image features of the current MRI image initially extracted by the visual perception architecture. When the image features of the current MRI image match the BI-RADS standard knowledge matrix, the corresponding BI-RADS semantic association point is activated, and a BI-RADS classification is generated.

5. The method for generating clinical documentation based on breast MRI images according to claim 1, characterized in that, The step of detecting the confidence level between the image features and the clinical document based on the anti-hallucination adjustment mechanism, and correcting the clinical document based on the confidence level to generate the final clinical document includes: Based on the pre-set confidence threshold of the anti-hallucination adjustment mechanism, the confidence level of the image features and the clinical document is detected in real time. The confidence level is compared with the confidence threshold. When the confidence level exceeds the confidence threshold, the clinical document conflicts with the image features, triggering a reverse modification route to forcibly modify the clinical document and correct it.

6. The method for generating clinical documents based on breast MRI images according to claim 1, characterized in that, The step of performing pathological feature recognition and analysis on the current MRI image based on a visual perception architecture to obtain the image features of the current MRI image includes: The current MRI image is a multi-sequence MRI image, which includes T1WI sequence images and T2WI sequence images; Spatiotemporal feature decoupling and fine-grained lesion focus are embedded in the encoding end of the visual perception architecture. The encoding end includes a spatial path and a related path. The spatial path locks the T1WI sequence image and the T2WI sequence image and extracts the spatial features of the multi-sequence MRI image. The spatial features include extracting breast gland classification, mass anatomical morphology, and edge breathing. The relevant pathway dynamically enhances the multi-sequence MRI images, generating multiple continuous time windows. Based on these multiple continuous time windows, the dynamic evolution rate of the contrast agent inflow-termination cycle of the multi-sequence MRI images, as well as the rapid enhancement rate and reversal slope of the lesions in the early stage of the multi-sequence MRI images, are accurately extracted. The spatial features extracted by the spatial path and the dynamic evolution rate, rapid enhancement rate and reversal slope extracted by the related path are fused to generate the image features based on spatiotemporal feature coupling.

7. A system for generating clinical documentation based on breast MRI images, characterized in that, The system includes: The acquisition module is used to acquire historical MRI images and text data corresponding to the historical MRI images. The text data includes historical radiology reports, outpatient medical records and pathology diagnosis records corresponding to the historical radiology reports. A preprocessing module is used to preprocess the historical MRI images and the text data; A dataset construction module is used to construct a multimodal dataset based on the preprocessed historical MRI images and the text data; The image processing module is used to acquire the current MRI image and perform pathological feature recognition and analysis on the current MRI image based on the visual perception architecture to obtain the image features of the current MRI image. The document generation module is used to generate clinical documents corresponding to the current MRI image based on the image features of the current MRI image and the multimodal dataset; The document correction module is used to set BI-RADS constraints within the visual perception architecture to generate an anti-illusion adjustment mechanism, detect the confidence level between the image features and the clinical document based on the anti-illusion adjustment mechanism, and correct the clinical document based on the confidence level to generate the final clinical document.

8. A system for generating clinical documentation based on breast MRI images, characterized in that, The method includes a memory and a processor, the memory storing a program or instructions executable on the processor, the program or instructions being executed by the processor to implement the steps of the method for generating clinical documentation based on breast MRI images as described in any one of claims 1 to 6.

9. A readable storage medium having a program or instructions stored thereon, characterized in that, When the program or the instructions are executed by the processor, they implement the steps of the method for generating clinical documentation based on breast MRI images as described in any one of claims 1 to 6.

Images