Focus analysis method based on medical image

Abstract

The invention provides a focus analysis method based on a medical image, and belongs to the field of image recognition. The method comprises the following steps: receiving medical image data to be analyzed, and calling a disease feature library corresponding to the medical image data; based on a disease feature library, carrying out preliminary judgment on a focus on the medical image data, identifying to obtain each first focus area, and determining a first identification confidence coefficient of each first focus area; for each first lesion area of which the first identification confidence is located in the target interval, carrying out area expansion by adopting a set step length to correspondingly obtain second lesion areas, and carrying out lesion re-judgment on each second lesion area to obtain second identification confidence of each second lesion area; and determining the first focus area with the first recognition confidence higher than the target interval and the second focus area with the second recognition confidence higher than the target interval as focus areas. According to the invention, the focus can be identified more accurately, and the probability of missed diagnosis and misdiagnosis is reduced.

Description

Technical Field

[0001] The present invention relates to the field of image recognition, and in particular to a lesion analysis method based on medical images. Background Art

[0002] In the field of medical diagnosis, medical imaging technologies such as X-rays, CT scans, and MRIs have become important tools for disease detection and diagnosis. By accurately identifying lesions in medical images, doctors can promptly detect diseases and develop appropriate treatment plans. However, because different types of diseases exhibit significant differences in image features in medical images, lesion identification is a challenging task.

[0003] Some disease lesions appear as isolated features in medical images, with weak correlations with surrounding tissue, relatively clear boundaries, and isolated features. For example, some benign tumors often appear as well-defined masses in images, easily distinguishable from surrounding normal tissue. For these lesions, existing image recognition algorithms, such as those based on threshold segmentation and edge detection, can achieve relatively accurate identification to a certain extent.

[0004] However, many diseases exhibit significant correlations between lesions and surrounding tissue. For example, interstitial lung lesions often intertwine with surrounding normal lung tissue, resulting in blurred boundaries. Furthermore, the texture and density of the lesion area exhibit continuity with changes in the surrounding tissue. In such cases, relying solely on traditional algorithms for identifying individual lesions is difficult to accurately delineate their extent, leading to missed or misdiagnosed conditions.

[0005] In summary, how to further improve the accuracy of lesion analysis in medical images is a technical problem that needs to be solved urgently. Summary of the Invention

[0006] In response to the above technical problems, the present invention provides a medical imaging-based lesion analysis method, electronic equipment, computer storage medium and computer program product.

[0007] The present invention discloses a lesion analysis method based on medical images, which includes the following steps: receiving medical image data to be analyzed, and retrieving a disease feature library corresponding to the medical image data; based on the disease feature library, performing a preliminary judgment on the lesions of the medical image data, identifying each first lesion area, and determining a first recognition confidence of each first lesion area; for each first lesion area whose first recognition confidence is within a target interval, using a set step size to expand the area to obtain a corresponding second lesion area, re-judging the lesion of each second lesion area, and obtaining a second recognition confidence of each second lesion area; and determining the first lesion area whose first recognition confidence is higher than the target interval and the second lesion area whose second recognition confidence is higher than the target interval as lesion areas.

[0008] Optionally, the receiving of medical imaging data to be analyzed and the retrieving of a disease feature library corresponding to the medical imaging data include: upon receiving the medical imaging data to be analyzed, determining the image type of the medical imaging data; judging whether the information on the disease to be diagnosed corresponding to the medical imaging data is received synchronously, and if so, retrieving a disease feature library corresponding to the image type and the information on the disease to be diagnosed; if not, sending a request message to a data interface that sends the medical imaging data, receiving the information on the disease to be diagnosed corresponding to the request message, and retrieving a disease feature library corresponding to the image type and the information on the disease to be diagnosed.

[0009] Optionally, based on the disease feature library, the medical image data is subjected to a preliminary judgment of lesions, and each first lesion area is identified, including: using an image segmentation algorithm to segment the medical image data according to multiple features of pixels to obtain multiple image sub-regions; wherein the multiple features include at least the grayscale value, color, and texture of the pixels; extracting image features of each image sub-region, and comparing the image features with typical features of the corresponding disease lesions in the disease feature library; wherein the image features include at least shape, size, edge features, internal texture complexity, and relative positional relationship with surrounding image sub-regions; if the similarity between multiple features of an image sub-region and the features of the corresponding disease lesions in the disease feature library exceeds a preset similarity threshold, the image sub-region is identified as a first lesion area.

[0010] Optionally, determining the first recognition confidence of each first lesion area includes: for each identified first lesion area, calculating the first recognition confidence based on the degree of matching between its features and the corresponding disease lesion features in the disease feature library, the number of matching features, and the importance weight of the features; wherein, the higher the degree of matching, the greater the number of matching features, and the greater the weight of important features, the higher the first recognition confidence.

[0011] Optionally, for each first lesion area whose first recognition confidence is located in the target interval, a set step size is used to perform area expansion to obtain a corresponding second lesion area, including: determining a boundary pixel set of each first lesion area whose first recognition confidence is located in the target interval; determining the set step size based on the resolution of the medical imaging data, the size of the first lesion area, and the preset expansion step size of the corresponding disease lesion in the disease feature library, and according to the set step size, starting from each pixel point in the boundary pixel set, expanding along the neighborhood direction of the pixel point; judging whether the expanded pixel point meets the preset conditions, and if so, adding the expanded pixel point to the first lesion area to form a new area, i.e., the second lesion area.

[0012] Optionally, the set step size is determined based on the resolution of the medical imaging data, the size of the first lesion area, and the preset expansion step size of the corresponding disease lesion in the disease feature library, including: determining the lateral resolution and longitudinal resolution of the medical imaging data, and the area of the first lesion area, and determining a first coefficient based on the lateral resolution, the longitudinal resolution, and the area; determining the disease level corresponding to the medical imaging data based on the information of the disease to be diagnosed, and predicting a corresponding second coefficient based on the disease level; wherein the second coefficient is positively correlated with the disease level; and using the first coefficient and the second coefficient to correct the preset expansion step size to obtain the set step size.

[0013] Optionally, after determining the lesion area, the method further includes: using an identification frame to mark each lesion area in the medical image data, and marking the first recognition confidence and the second recognition confidence around the corresponding identification frame respectively.

[0014] The present invention also discloses an electronic device comprising one or more processors, a memory, and one or more computer programs, wherein the one or more computer programs are stored in the memory, and the one or more computer programs include instructions, which, when executed by the processor, implement the medical imaging-based lesion analysis method as described in any one of the above items.

[0015] The present invention also discloses a computer storage medium, comprising computer instructions that can be executed on an electronic device, wherein the computer instructions are executed to implement any of the above-mentioned medical image-based lesion analysis methods.

[0016] The present invention also discloses a computer program product, which can be run on an electronic device to implement any of the medical image-based lesion analysis methods described above.

[0017] The beneficial effects of the present invention are at least that: the above-mentioned scheme of the present invention can expand the suspected lesion area and perform secondary identification of the lesion area based on the expanded image area, which greatly improves the accuracy of lesion identification, reduces the probability of missed diagnosis and misdiagnosis, and provides a reliable basis for doctors to accurately diagnose diseases and formulate effective treatment plans, and effectively promotes the development and clinical application of medical imaging diagnosis technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 This is a flow chart of a medical imaging-based lesion analysis method disclosed in an embodiment of the present invention.

[0020] Figure 2 It is a schematic diagram of the first lesion area and the second lesion area disclosed in an embodiment of the present invention.

[0021] Figure 3 It is a structural diagram of an electronic device disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0022] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0023] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0024] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0025] like Figure 1 As shown, in response to the above technical problems, an embodiment of the present invention discloses a lesion analysis method based on medical imaging, which includes the following steps: S10, receiving medical imaging data to be analyzed, and retrieving a disease feature library corresponding to the medical imaging data.

[0026] Medical imaging data to be analyzed can be received via a corresponding data interface. The data interface can be directly connected to an imaging device or to a doctor's diagnostic terminal. Thus, the medical imaging data can be obtained after the imaging device completes capturing the image or the doctor transmits the patient's medical image via the data interface. It is understood that the medical imaging data in the present invention can be derived from a variety of medical imaging technologies, such as image data generated by X-ray, CT, MRI, and other equipment mentioned in the background technology.

[0027] Different medical imaging technologies correspond to different disease characteristics, and different diseases also exhibit unique features in various types of images. To this end, the present invention pre-establishes and maintains multiple disease feature libraries. Each disease feature library stores typical feature information of various diseases in different medical images, such as the shape, size, density, and texture of the lesion. Upon receiving specific medical imaging data, the disease feature library corresponding to that medical imaging data can be retrieved.

[0028] S20, based on the disease feature library, perform preliminary lesion determination on the medical image data, identify each first lesion region, and determine a first recognition confidence of each first lesion region.

[0029] Using the disease feature information retrieved from the disease feature library, and based on the shape, density and other characteristics of the lesions in the disease feature library, each part of the received medical image data is analyzed and judged one by one, and the area that matches the lesion characteristics is identified, thereby identifying the area that may be the lesion. The identified suspected lesion areas are marked, and these areas are the first lesion areas (refer to Figure 2101 in the .

[0030] It should be noted that the identification of the first lesion area can be achieved based on conventional classification algorithms or models, such as support vector machine algorithms, naive Bayesian algorithms, etc. Such algorithms or models constructed based on them can achieve rapid identification of obvious lesion areas.

[0031] When each first lesion region is identified, a first identification confidence level is also determined for each first lesion region. For example, the confidence level that the region is a lesion, i.e., the first identification confidence level, is calculated based on factors such as the degree of match between the region and features in the disease feature library. The first identification confidence level reflects the reliability of the judgment that the region is a lesion. A higher confidence level indicates a higher likelihood that the region is a true lesion; a lower confidence level indicates a lower likelihood that the region is a lesion.

[0032] S30, for each first lesion area whose first recognition confidence is in the target interval, the area is expanded using a set step size to obtain a corresponding second lesion area, and the lesion of each second lesion area is re-judged to obtain a second recognition confidence of each second lesion area.

[0033] Since the first recognition confidence is only a preliminary judgment result, there is a certain degree of uncertainty. In order to more accurately determine the scope of the lesion, further processing is required for those first lesion areas whose first recognition confidence is in the target interval. These areas have a certain possibility of being lesions, but further analysis is needed to clarify. It can be understood that the target interval includes an upper confidence limit and a lower confidence limit; if the first recognition confidence exceeds the upper confidence limit, it can be directly identified as a lesion area. If the first recognition confidence is lower than the lower confidence limit, it can be directly identified as not a lesion area.

[0034] For the first lesion area with the first recognition confidence level in the target interval, these areas are expanded outward according to a preset step size (a certain number of pixels are expanded in the horizontal and vertical directions of the image each time), thereby obtaining a new, larger area, namely the second lesion area (refer to Figure 2 "102" in Figure 2 (The image is expanded outward twice. The middle black area is the first lesion area before expansion, and the outer gray and white areas together constitute the second lesion area.) It is understandable that the purpose of region expansion is to consider possible fuzzy lesion boundaries and expand the analysis scope to more comprehensively capture lesion information, such as the true lesion boundary.

[0035] For each secondary lesion region obtained after region expansion, the analysis algorithm or model based on the disease signature library is again applied to further determine the presence of a lesion. Similarly, a second recognition confidence level is calculated based on the judgment results, indicating the confidence level that the region is a lesion. This second recognition confidence level is a more accurate judgment result after considering the expanded region, and can more accurately reflect the likelihood of the presence of a lesion in the region.

[0036] S40: Determine the first lesion region whose first recognition confidence is higher than the target interval and the second lesion region whose second recognition confidence is higher than the target interval as lesion regions.

[0037] After the previous steps, two types of lesion regions and their corresponding confidence levels are obtained. For the first lesion region, if its first recognition confidence level is higher than the target range, it means that the system initially determines that the region has a high probability of being a true lesion and therefore identifies it as a lesion region.

[0038] If the second recognition confidence score for the second lesion region is also higher than the target range, this indicates that after expanding and re-evaluating the first lesion region (which had a first recognition confidence score within the target range), there is a high probability that this region contains a lesion, and it is also identified as a lesion region. These final lesion regions, identified after multiple rounds of analysis and judgment, can provide a reference for doctors' disease diagnosis and treatment planning.

[0039] The above-mentioned scheme of the present invention can expand the suspected lesion area and perform secondary identification of the lesion area based on the expanded image area, which greatly improves the accuracy of lesion identification, reduces the probability of missed diagnosis and misdiagnosis, and provides a reliable basis for doctors to accurately diagnose diseases and formulate effective treatment plans, and effectively promotes the development and clinical application of medical imaging diagnosis technology.

[0040] It should be noted that the solution of the present invention can be implemented on a dedicated lesion analysis server. This server is connected to the imaging equipment and / or doctor's diagnosis and treatment terminals of at least one medical institution, receives medical imaging data from these imaging equipment and / or diagnosis terminals through corresponding data interfaces, and identifies and annotates lesion areas therein. The disease feature libraries can also be pre-installed on this server or retrieved by the server when needed, without limitation.

[0041] Optionally, the receiving of medical imaging data to be analyzed and the retrieving of a disease feature library corresponding to the medical imaging data include: upon receiving the medical imaging data to be analyzed, determining the image type of the medical imaging data; judging whether the information on the disease to be diagnosed corresponding to the medical imaging data is received synchronously, and if so, retrieving a disease feature library corresponding to the image type and the information on the disease to be diagnosed; if not, sending a request message to a data interface that sends the medical imaging data, receiving the information on the disease to be diagnosed corresponding to the request message, and retrieving a disease feature library corresponding to the image type and the information on the disease to be diagnosed.

[0042] In this embodiment, when the lesion analysis server receives medical imaging data to be analyzed, it first identifies the data's source, specifically its image type. Medical imaging encompasses data generated by a variety of technologies, including X-rays, CT (computed tomography) scans, and MRI (magnetic resonance imaging). Different imaging types have unique imaging principles and characteristics, which determine how they depict human tissue structure and lesion characteristics. For example, X-rays primarily reflect differences in tissue density and are effective for imaging high-density tissues such as bone. CT images provide more detailed cross-sectional information and are advantageous for detecting lesions in areas such as the lungs and abdomen. MRI images excel at displaying soft tissue detail and are highly valuable for diagnosing soft tissue diseases such as those of the brain and spinal cord. The lesion analysis server accurately determines the image type of medical imaging data based on data file format, metadata, or specific identification fields.

[0043] After determining the image type, the next step is to check whether the corresponding disease information for the image data has been received. This disease information is typically entered by the doctor when the image data is collected, or retrieved from the hospital's information management system. For example, if a doctor suspects a patient has lung cancer, during a chest CT scan, the "suspected lung cancer" disease information will be recorded and transmitted along with the corresponding CT image data.

[0044] If the lesion analysis server confirms that it has synchronously received the information of the disease to be diagnosed, it will retrieve the corresponding part from the pre-built disease feature library based on the previously determined image type and the disease information. The disease feature library is not a single entity, but is classified and stored according to the image type and different diseases. For example, for lung cancer, the typical characteristics of lung cancer under X-ray images, the typical characteristics of lung cancer under CT images, and the typical characteristics of lung cancer under MRI images will be stored in the disease feature library. According to the received image type (such as CT) and the disease information to be diagnosed (such as lung cancer), the area in the disease feature library that stores the typical characteristics of lung cancer under CT images is accurately located to obtain relevant feature data.

[0045] If the lesion analysis server has not synchronously received the disease to be diagnosed, it sends a request message to the data interface that transmits the medical image data. This data interface can be a hospital's image storage system, a data transmission port of an image acquisition device, or a data port on a doctor's diagnosis and treatment terminal. The request message inquires about the possible disease information of the patient corresponding to the medical image data. Once this information is successfully obtained, the corresponding disease feature data is retrieved from the disease feature library according to the aforementioned method, based on the image type and the disease to be diagnosed.

[0046] Optionally, based on the disease feature library, the medical image data is subjected to a preliminary judgment of lesions, and each first lesion area is identified, including: using an image segmentation algorithm to segment the medical image data according to multiple features of pixels to obtain multiple image sub-regions; wherein the multiple features include at least the grayscale value, color, and texture of the pixels; extracting image features of each image sub-region, and comparing the image features with typical features of the corresponding disease lesions in the disease feature library; wherein the image features include at least shape, size, edge features, internal texture complexity, and relative positional relationship with surrounding image sub-regions; if the similarity between multiple features of an image sub-region and the features of the corresponding disease lesions in the disease feature library exceeds a preset similarity threshold, the image sub-region is identified as a first lesion area.

[0047] In this embodiment, the medical image data is a complex image set composed of a large number of pixels, and needs to be segmented according to the various features of the pixels using an image segmentation algorithm to obtain multiple image sub-regions.

[0048] Multiple features include at least the grayscale value, color, and texture of pixels. Among them, the grayscale value reflects the brightness information of the pixel. In some medical images (such as X-ray images and CT images), the grayscale values of different tissues and lesions vary. By analyzing the grayscale values, different areas can be preliminarily distinguished. Color features help identify specific tissues or lesions in some medical images (such as images after certain special staining). Texture features reflect the pixel distribution pattern of local areas in the image. Different tissues and lesions often have unique textures. For example, the texture of normal tissue is relatively regular, while the texture of some lesion areas may be more complex or show a specific disordered state. By comprehensively considering these pixel features, the image segmentation algorithm can divide medical imaging data into multiple relatively independent image sub-regions with similar characteristics, providing a more targeted basic unit for subsequent analysis.

[0049] After obtaining multiple image sub-regions, a series of image features of each image sub-region are extracted. These features include at least shape, size, edge features, internal texture complexity, and relative position relationship with surrounding image sub-regions.

[0050] Among them, shape features describe the geometric outline of an image subregion. Different types of disease lesions may exhibit specific shapes on an image, such as circular, elliptical, or irregular. Recognizing this shape allows for a preliminary assessment of whether a region is a lesion. Size features reflect the number of pixels or area occupied by an image subregion. Lesion size is often crucial for disease diagnosis. Early lesions of some diseases may be small but gradually increase in size as the disease progresses. Edge features characterize the boundaries of image subregions, such as edge clarity and smoothness. Some lesions have sharp edges, while others may have blurred edges. Internal texture complexity describes the distribution and complexity of pixels within a subregion. Internal texture complexity often differs between normal tissue and lesions. The relative position of a lesion to surrounding image subregions is also crucial. The positional relationship of the lesion to surrounding normal tissue on an image may indicate the type and stage of disease progression.

[0051] These extracted image features are compared with typical features of lesions in the corresponding disease categories in a pre-established disease feature library. This library contains a large number of clinically verified and researched typical manifestations of specific diseases on various types of medical images. This comparison can identify image subregions with features similar to those in the disease feature library, providing a basis for determining the presence of lesions.

[0052] After completing the comparison between the image sub-region features and the disease feature library features, if the similarity between multiple features of a certain image sub-region and the features of a certain type of disease lesions in the disease feature library exceeds a preset similarity threshold, the image sub-region is determined to be the first lesion region.

[0053] It should be noted that the preset similarity threshold is a key parameter determined based on extensive experimental data and clinical experience. Its setting requires comprehensive consideration of multiple factors, such as the degree of variability in disease characteristics, the quality of the medical imaging data, and the noise level. If the similarity threshold is set too high, some actual lesions may be missed because the feature similarity does not meet the threshold. Conversely, if it is set too low, some normal tissue areas may be misidentified as lesions, increasing the misdiagnosis rate. By properly setting this threshold, the true lesion area can be identified as accurately as possible while maintaining a certain level of accuracy, providing a reliable basis for subsequent diagnosis and treatment.

[0054] Optionally, determining the first recognition confidence of each first lesion area includes: for each identified first lesion area, calculating the first recognition confidence based on the degree of matching between its features and the corresponding disease lesion features in the disease feature library, the number of matching features, and the importance weight of the features; wherein, the higher the degree of matching, the greater the number of matching features, and the greater the weight of important features, the higher the first recognition confidence.

[0055] In this embodiment, for each identified first lesion area, its own characteristics (such as shape, size, edge characteristics, internal texture complexity, and relative position relationship with surrounding image sub-areas, etc.) need to be compared with the characteristics of the corresponding disease lesions in the disease feature library.

[0056] For example, using shape features as an example, algorithms such as contour keypoint matching and Fourier descriptor comparison are used to calculate the similarity score between the shape of the first lesion region and the corresponding lesion shape in the disease feature library. When the two shapes are completely identical, the similarity score is set to 1; as the shape difference gradually increases, the similarity score approaches 0.

[0057] The above processing method is also applicable to other types of features. For example, size features can be evaluated by calculating the proportional relationship of area or volume to obtain a size similarity score. Edge features can be extracted using an edge detection algorithm and then compared with edge sharpness, smoothness, and other attributes to obtain a similarity score. Internal texture complexity can be evaluated by using texture analysis algorithms (such as gray-level co-occurrence matrix and local binary pattern). Relative positional relationships are analyzed by comparing the relative position of the image with the surrounding image sub-regions and then comparing them with the corresponding positional relationships in the disease feature library to obtain a positional relationship similarity score. These similarity scores are all in the range of 0 to 1. The higher the score, the more closely the corresponding features of the first lesion region match the corresponding disease lesion features in the disease feature library.

[0058] After calculating the similarity of each feature, the number of features in the first lesion region that have a high degree of match (exceeding a certain threshold, such as a similarity score greater than 0.7) with the corresponding disease lesion features in the disease feature library is counted. A greater number of matching features indicates that the first lesion region more closely matches the characteristic pattern of the corresponding disease lesion in the disease feature library in multiple aspects, indicating a greater likelihood that the region is a true lesion.

[0059] Different features have different importance in determining whether it is a lesion and determining the type of lesion. For example, for some diseases, the shape characteristics of the lesion play a decisive role in diagnosis; while for other diseases, the internal texture complexity or relative position relationship may be more critical. In order to reflect this difference, the present invention sets an importance weight for each feature for different types of diseases through in-depth statistical analysis of a large amount of clinical case data and with the help of rich experience evaluation of medical experts. The importance weights of shape, size, edge characteristics, internal texture complexity, and relative position relationship are set as follows: , and the sum of these weights must satisfy This setting ensures that when comprehensively calculating the first recognition confidence, each feature can have a corresponding impact on the final result according to its importance.

[0060] Combining the above three aspects, the first recognition confidence The calculation formula is: ;in, Indicates the The similarity score of the features, Indicates the The importance weight of each feature. According to the above formula, the matching degree score of each feature is multiplied by its corresponding importance weight and then added together to obtain the first recognition confidence. , which can fully reflect the comprehensive matching of the first lesion area and the corresponding disease lesion features in the disease feature library. The higher the matching degree, the more matching features there are, and the greater the weight of important features, the higher the confidence level of the first recognition finally calculated. The higher the value, the greater the possibility that the first lesion area is the real lesion.

[0061] Optionally, for each first lesion area whose first recognition confidence is located in the target interval, a set step size is used to perform area expansion to obtain a corresponding second lesion area, including: determining a boundary pixel set of each first lesion area whose first recognition confidence is located in the target interval; determining the set step size based on the resolution of the medical imaging data, the size of the first lesion area, and the preset expansion step size of the corresponding disease lesion in the disease feature library, and according to the set step size, starting from each pixel point in the boundary pixel set, expanding along the neighborhood direction of the pixel point; judging whether the expanded pixel point meets the preset conditions, and if so, adding the expanded pixel point to the first lesion area to form a new area, i.e., the second lesion area.

[0062] In this embodiment, for those first lesion regions whose first recognition confidence levels fall within the target range, the boundary pixel set is first determined. This boundary pixel set represents the edge of the first lesion region. Edge detection algorithms (such as Canny edge detection and the Sobel operator) can be used to identify the boundary pixels of these first lesion regions from the medical image data.

[0063] Next, a step size is determined. This step size is the "length" of each outward expansion based on the boundary pixel set. The present invention considers at least three factors to determine the appropriate step size: The resolution of the medical imaging data: If the resolution of the medical imaging data is high, meaning the image contains more detail, a smaller step size can be set, allowing for more precise regional expansion and avoiding missing possible lesion pixels. Conversely, if the resolution is lower, the step size can be appropriately increased to improve expansion efficiency.

[0064] Size of the first lesion area: For a larger first lesion area, the step size can be relatively large, because a larger lesion usually has a larger associated range; while for a smaller first lesion area, the step size should be smaller to more accurately capture the boundary of the lesion.

[0065] The disease signature database also stores preset expansion steps for different disease lesions. This preset expansion step is based on a large number of clinical case summaries and reflects the typical possible expansion range of the disease lesion.

[0066] Starting from each pixel in the boundary pixel set, the algorithm expands along the neighborhood of that pixel, using the previously determined step size. Neighborhood directions typically include horizontal (left, right), vertical (up, down), and diagonal directions. For example, starting from a boundary pixel, the algorithm gradually expands to the right, above, and to the upper right, using the set step size.

[0067] Each time the image is expanded, it is necessary to determine whether the expanded pixel meets the preset conditions. The preset conditions are related to the pixel's characteristics, such as grayscale value, color, texture, etc. If the grayscale value, color, texture, etc. of the expanded pixel is similar to the pixel characteristics in the first lesion area within a certain threshold, the pixel is considered to meet the preset conditions.

[0068] When the expanded pixels meet the pre-set conditions, they are added to the first lesion region. As more pixels meeting the conditions are added, the first lesion region expands, eventually forming a new region, which becomes the second lesion region. This approach allows for a more comprehensive capture of the potential lesion range and reduces missed diagnoses due to blurred lesion boundaries.

[0069] It should be noted that the number of expansions should be pre-set, for example, a maximum of 3 expansions. If the number of expansions does not reach 3, and if it is determined that the pixel points obtained after the expansion do not meet the preset conditions, the expansion is terminated in advance, and the result of the previous expansion is used to determine the second lesion area.

[0070] Optionally, the set step size is determined based on the resolution of the medical imaging data, the size of the first lesion area, and the preset expansion step size of the corresponding disease lesion in the disease feature library, including: determining the lateral resolution and longitudinal resolution of the medical imaging data, and the area of the first lesion area, and determining a first coefficient based on the lateral resolution, the longitudinal resolution, and the area; determining the disease level corresponding to the medical imaging data based on the information of the disease to be diagnosed, and predicting a corresponding second coefficient based on the disease level; wherein the second coefficient is positively correlated with the disease level; and using the first coefficient and the second coefficient to correct the preset expansion step size to obtain the set step size.

[0071] In this embodiment, the medical image data has a horizontal resolution and a vertical resolution, measured in pixels per millimeter. These parameters represent the number of pixels per unit length in the horizontal and vertical directions, respectively. These two parameters reflect the image's level of detail. The horizontal and vertical resolutions of the medical image data can be determined by reading metadata or related parameter settings of the medical image data.

[0072] The size of the first lesion region is characterized by its area. For first lesion regions with regular shapes, such as approximately circular or rectangular, the area can be calculated using the corresponding geometric formula. For first lesion regions with irregular shapes, the pixel counting method can be used to count the number of pixels within the region and convert the number of pixels into the actual area based on the resolution of the medical imaging data.

[0073] The first coefficient mentioned above can be obtained by performing a comprehensive calculation on the horizontal resolution, vertical resolution, and area. The first coefficient is used to appropriately correct the preset expansion step size, that is, to increase or decrease it to adapt it to the actual situation of the medical image data. The correction method can be to multiply the first coefficient by the preset expansion step size. The calculation formula of the first coefficient is, for example: Where, is the first coefficient, is the area of each first lesion region where the first recognition confidence is located in the target interval, is the lateral resolution, is the vertical resolution. They are respectively normalization conversion functions, such as Min-Max normalization, Z-score normalization, etc., and are not specifically limited.

[0074] Based on the above calculation formula, when the area of the first lesion is larger and the resolution of the medical image is lower, The larger the value of , the larger the corresponding set step size will be corrected; conversely, when the area of the first lesion is smaller and the resolution of the medical image is higher, The smaller the value is, the smaller the corresponding setting step size will be corrected.

[0075] At the same time, the present invention also considers the actual condition of the patient to determine the second coefficient, and uses the second coefficient to correct the preset expansion step size. The correction method can also be a multiplication operation of the second coefficient and the preset expansion step size. The disease level can be analyzed from the aforementioned information on the disease to be diagnosed. The disease level can be obtained by the doctor after evaluating the patient's condition (for example, the stage of the disease, undiagnosed, early, middle, late, etc.). The higher the disease level, the more serious the patient's condition, and the corresponding prediction is that the lesion area is larger and the characteristics are more obvious. It should be noted that the information on the disease to be diagnosed should be cleared of privacy information to avoid the leakage of patient personal information.

[0076] Therefore, the present invention further sets the second coefficient to be positively correlated with the disease level, that is, the higher the disease level, the larger the second coefficient. The higher the disease level, the larger the potential extension range of the lesion (that is, the greater the probability that the tissue surrounding the significant lesion is also the lesion area), requiring a relatively large expansion step size to more comprehensively capture the possible lesion area. For example, if the disease level is divided into 1-5 levels, when the disease level is 1, the second coefficient is set to 1.0; when the disease level is 2, the second coefficient is set to 1.2; as the disease level increases, the second coefficient gradually increases. In this way, the corresponding second coefficient is predicted based on the determined disease level to reflect the impact of the disease characteristics on the expansion step size.

[0077] Finally, the first coefficient and the second coefficient are used in combination to correct the preset extension step length to the set step length.

[0078] It should be noted that, when the second lesion area is subjected to secondary identification, the present invention preferably uses a combination of multiple recognition algorithms, i.e., a plurality of recognition algorithms are used to identify the second lesion area, and then each recognition result is subjected to, for example, weighted fusion, and the result obtained after weighted fusion is used as the final result, i.e., the second recognition confidence. Of course, it is also possible to construct a recognition model based on each recognition algorithm, and then combine each recognition model to form a composite model, and the composite model is used to perform secondary identification on the second lesion area. Since the composite model technology is very mature, it will not be described in detail here. Compared with the aforementioned classification model based on the classification algorithm, the composite model of the present invention can achieve accurate identification of inconspicuous lesion areas and their true boundaries.

[0079] Optionally, after determining the lesion area, the method further includes: using an identification frame to mark each lesion area in the medical image data, and marking the first recognition confidence and the second recognition confidence around the corresponding identification frame respectively.

[0080] In this embodiment, each lesion region identified in the medical image data is clearly marked in the original medical image data using an identification frame. The identification frame can be rectangular, circular, or other suitable shapes. This marking method can help doctors quickly focus on the lesion region without having to search and judge it manually in complex medical images.

[0081] At the same time, the recognition confidence of each lesion area is also marked around the corresponding identification box. The marked position can be above, below, left or right of the identification box, etc., as long as it can clearly correspond to the lesion area corresponding to the identification box. For example, the value of the first recognition confidence is marked above the identification box, such as "Recognition confidence: 0.8". In this way, when the doctor views the annotated medical image, he can not only see the location and range of the lesion area, but also understand the reliability of the system's judgment that the area is a lesion, thereby more comprehensively assessing the condition and providing a more valuable reference for further diagnosis and treatment decisions. In addition, medical image data marked with recognition confidence can also be used to generate training data in the future, and then use these training data to train and generate a special lesion recognition model. The details will not be repeated here.

[0082] like Figure 3 As shown, an embodiment of the present invention also discloses an electronic device, including one or more processors, a memory, and one or more computer programs, wherein the one or more computer programs are stored in the memory, and the one or more computer programs include instructions, and when the instructions are executed by the processor, the medical imaging-based lesion analysis method as described in any of the preceding items is implemented.

[0083] An embodiment of the present invention further discloses a computer storage medium, comprising computer instructions that can be executed on an electronic device, wherein the computer instructions are executed to implement the medical imaging-based lesion analysis method as described in any of the preceding items.

[0084] An embodiment of the present invention further discloses a computer program product, which can be run on an electronic device to implement the medical image-based lesion analysis method as described in any of the preceding items.

[0085] The functional units in various embodiments of the present invention may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0086] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), a mobile hard drive, a magnetic disk, or an optical disk.

[0087] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A lesion analysis method based on medical imaging, characterized by: The method includes the following steps: receiving medical image data to be analyzed, and retrieving a disease feature library corresponding to the medical image data; based on the disease feature library, performing a preliminary judgment on the lesions of the medical image data, identifying each first lesion area, and determining a first recognition confidence of each first lesion area; for each first lesion area whose first recognition confidence is within a target interval, using a set step size to expand the area to obtain a corresponding second lesion area, re-judging the lesion of each second lesion area, and obtaining a second recognition confidence of each second lesion area; and determining the first lesion area whose first recognition confidence is higher than the target interval and the second lesion area whose second recognition confidence is higher than the target interval as lesion areas.

2. The method for analyzing lesions based on medical images according to claim 1, wherein: Receiving medical image data to be analyzed and retrieving a disease feature library corresponding to the medical image data, including: when receiving the medical image data to be analyzed, determining the image type of the medical image data; judging whether the disease information to be diagnosed corresponding to the medical image data is received synchronously, and if so, retrieving a disease feature library corresponding to the image type and the disease information to be diagnosed; if not, sending a request message to a data interface that sends the medical image data, receiving the disease information to be diagnosed corresponding to the request message, and retrieving a disease feature library corresponding to the image type and the disease information to be diagnosed.

3. The method for analyzing lesions based on medical images according to claim 1, wherein: Based on the disease feature library, a preliminary judgment of lesions is made on the medical image data, and each first lesion area is identified, including: using an image segmentation algorithm to segment the medical image data according to multiple features of pixels to obtain multiple image sub-regions; wherein the multiple features include at least the grayscale value, color, and texture of the pixels; extracting image features of each image sub-region, and comparing the image features with typical features of the corresponding disease lesions in the disease feature library; wherein the image features include at least shape, size, edge features, internal texture complexity, and relative positional relationship with surrounding image sub-regions; if the similarity between multiple features of an image sub-region and the features of the corresponding disease lesions in the disease feature library exceeds a preset similarity threshold, then the image sub-region is identified as a first lesion area.

4. The method for analyzing lesions based on medical images according to claim 1, wherein: Determining the first recognition confidence of each first lesion area includes: for each identified first lesion area, calculating the first recognition confidence based on the degree of matching between its features and the corresponding disease lesion features in the disease feature library, the number of matching features, and the importance weight of the features; wherein, the higher the degree of matching, the greater the number of matching features, and the greater the weight of important features, the higher the first recognition confidence.

5. The method for analyzing lesions based on medical images according to claim 1, wherein: For each first lesion area whose first recognition confidence is located in the target interval, a set step size is used to perform area expansion to obtain a corresponding second lesion area, including: determining a boundary pixel set of each first lesion area whose first recognition confidence is located in the target interval; determining the set step size based on the resolution of the medical imaging data, the size of the first lesion area, and the preset expansion step size of the corresponding disease lesion in the disease feature library; starting from each pixel point in the boundary pixel set, expanding along the neighborhood direction of the pixel point according to the set step size; judging whether the expanded pixel point meets the preset conditions, and if so, adding the expanded pixel point to the first lesion area to form a new area, namely the second lesion area.

6. The method for analyzing lesions based on medical images according to claim 2, wherein: The set step size is determined based on the resolution of the medical imaging data, the size of the first lesion area, and the preset expansion step size of the corresponding disease lesion in the disease feature library, including: determining the lateral resolution and longitudinal resolution of the medical imaging data, and the area of the first lesion area, and determining a first coefficient based on the lateral resolution, the longitudinal resolution, and the area; determining the disease level corresponding to the medical imaging data based on the information of the disease to be diagnosed, and predicting a corresponding second coefficient based on the disease level; wherein the second coefficient is positively correlated with the disease level; and using the first coefficient and the second coefficient to correct the preset expansion step size to obtain the set step size.

7. The method for analyzing lesions based on medical images according to claim 1, wherein: After determining the lesion area, the method further includes: marking each lesion area in the medical image data using an identification frame, and marking the first recognition confidence and the second recognition confidence around the corresponding identification frame respectively.

8. An electronic device, characterized in that: The system comprises one or more processors, a memory, and one or more computer programs, wherein the one or more computer programs are stored in the memory, and the one or more computer programs include instructions, which, when executed by the processor, implement the medical imaging-based lesion analysis method as described in any one of claims 1 to 7.

9. A computer storage medium, characterized in that: It includes computer instructions that can be executed on an electronic device, and the computer instructions, when executed, implement the medical image-based lesion analysis method according to any one of claims 1 to 7.

10. A computer program product, characterized in that: The computer program product can be run on an electronic device to implement the medical image-based lesion analysis method as described in any one of claims 1 to 7.

Images