A disease grading evaluation method based on image feature analysis

Abstract

The present application relates to the technical field of medical image analysis, and particularly relates to a disease grading evaluation method based on image feature analysis, comprising the following steps: generating a level difference borrowing map according to the dislocation relationship of the surface sign and the latent sign in spatial position; identifying whether there is a borrowing situation of low-grade appearance wrapping high-grade sign, local high-grade sign departing from the overall imaging order or the same grade sign appearing discontinuously across regions between the surface sign and the latent sign, and generating a level difference conflict evidence sequence; performing conflict resolution evaluation on a preset disease grade, and determining the target grade which still maintains a continuous support relationship after resolution as the disease grading result. Through the introduction of conflict resolution and weight mechanism, the present application can differentially process different types of borrowing, reduce the interference of noise, artifacts or incidental abnormalities on the grading result, and improve the stability of overall determination.

Description

Technical Field

[0001] This invention relates to the field of medical image analysis technology, and in particular to a disease grading assessment method based on image feature analysis. Background Technology

[0002] Medical imaging plays a crucial role in disease diagnosis and grading, especially in diseases such as tumors, inflammation, and degenerative diseases, where imaging grading results directly affect the selection of clinical treatment plans and prognosis. Existing disease grading methods are typically based on information such as the morphological characteristics, grayscale distribution, boundary clarity, or texture changes of lesion regions in images. After segmenting and extracting features from the target region, rules or models are then used to complete the grading determination.

[0003] However, in practical applications, the above methods generally suffer from the following shortcomings: First, most existing technologies assume overall imaging consistency, meaning they assume the appearance of the main lesion area in the image can represent the overall disease grade. However, in many cases, disease progression is not uniform but rather manifests as changes occurring in localized areas first. For example, within a large area of ​​low-grade tissue, small areas of high-grade lesions may already be present. Traditional methods, when performing overall feature statistics, often average or weaken these local abnormalities, leading to lower grading results and failing to reflect the true progression of the disease in a timely manner. Second, existing methods typically focus on analyzing single lesion areas, lacking a systematic characterization of the spatial relationships between different signs. In fact, the spatial distribution of signs of different grades in images has significant diagnostic value. For example, high-grade signs being enveloped by low-grade areas, different grade areas intersecting, or the same grade area being fragmented by other grade areas all reflect the instability and evolutionary trend of the disease structure. However, existing technologies often only focus on the characteristics of the area itself, ignoring the spatial hierarchy and structural relationships between areas, resulting in the inability to identify these crucial grade misalignments. Summary of the Invention

[0004] This invention provides a disease grading assessment method based on image feature analysis, which effectively identifies the spatial relationships between different grades of signs in medical images, especially identifying masked high-grade signs, local abnormalities detached from the overall structure, and discontinuously distributed areas of the same grade. Based on this, a unified conflict resolution mechanism and structural constraint mechanism are constructed, which improves the accuracy and stability of disease grading assessment.

[0005] A disease grading assessment method based on image feature analysis includes the following steps: S1. Perform graded sign decomposition on the target disease image, distinguish the surface signs that dominate the current overall imaging judgment and the latent signs hidden in the local structure, and generate a graded borrowing map based on the spatial misalignment relationship between the surface signs and the latent signs. S2. Based on the graded borrowing diagram, identify whether there are borrowing situations between surface signs and latent signs, such as low-level appearances enveloping high-level signs, local high-level signs breaking away from the overall imaging order, or signs of the same level appearing intermittently across regions, and generate a graded conflict evidence sequence. S3. Based on the aforementioned differential conflict evidence sequence, perform conflict resolution assessment on the preset disease level, and determine the target level that still maintains a continuous support relationship after resolution as the disease classification result.

[0006] Optionally, S1 further includes a preset sign level label library, which includes multiple preset disease level labels. After acquiring the target disease image, the target disease image is segmented at the pixel level or region level to extract multiple sets of image feature regions corresponding to different disease levels. The multiple sets of image feature regions are used to match the disease level labels with the sign level label library.

[0007] Optionally, the distribution range, visual prominence, and coverage level of the image feature regions in the image space are obtained. Image feature regions that occupy the main field of view of the image and constitute the overall visual continuity are marked as surface features, while image feature regions that are distributed in the local deep structure of the image, are spatially discrete, and visually constitute a front-to-back layer distinction from surface features are marked as latent features.

[0008] Optionally, the spatial location information of the surface signs and the latent signs is extracted, the relative positional relationship of their spatial location information in the image space is calculated, and the misaligned regions of spatial inclusion, local overlap or spatial separation between the surface signs and the latent signs are identified. The misaligned regions are mapped to the same spatial reference system to generate a graded borrowing map. The graded borrowing map is used to characterize the spatial misalignment distribution of different levels of signs.

[0009] Optionally, the sign levels and spatial distribution boundaries corresponding to the surface signs and latent signs in the grade difference borrowing map are extracted; and borrowing situations are identified and marked as multiple borrowing units; the multiple borrowing units are structurally combined according to their respective spatial locations, borrowing types and the sign levels involved to generate a grade difference conflict evidence sequence.

[0010] Optionally, the borrowing situation includes: Identify instances where a lower-level appearance encloses a higher-level feature through misplacement; Identify cases where localized high-level features deviate from the overall imaging order; Identify instances of borrowing where similar signs appear intermittently across different regions.

[0011] Optionally, the identification of the borrowing situation where a low-level appearance covers a high-level feature specifically includes: when the feature level of the surface feature is detected in the differential borrowing map as being lower than the feature level of the latent feature covered or covered by the surface feature, the corresponding spatial area is marked as the first borrowing unit. The identification of local high-level features deviating from the overall imaging order specifically includes: when the feature level of the latent feature detected in the differential borrowing map is higher than the main feature level corresponding to the overall imaging, and the latent feature forms a non-continuous level transition with the surrounding features in spatial distribution, the corresponding spatial area is marked as the second borrowing unit. The identification of the intermittent occurrence of the same level of sign across regions specifically includes: when multiple image feature regions of the same sign level are detected in the differential borrowing map and are spatially separated by regions of different levels, forming an intermittent distribution pattern, the corresponding intermittent distribution region is marked as the third borrowing unit.

[0012] Optionally, S3 acquires the differential conflict evidence sequence and a preset disease level, the preset disease level including multiple candidate levels in ascending order of severity; for each borrowing unit in the differential conflict evidence sequence, the sign level involved in the borrowing unit is compared with the preset disease level to identify the support or conflict relationship between the borrowing unit and each candidate level, and a corresponding resolution weight is assigned to the conflict relationship according to the borrowing type.

[0013] Optionally, the candidate levels are traversed, and the confidence level of candidate levels with conflicting relationships is reduced according to the resolution weight. At the same time, the confidence level of candidate levels with supporting relationships is accumulated to form the conflict resolution result of each candidate level.

[0014] Optionally, S3 further includes determining the candidate level with the highest confidence level after conflict resolution and supported by a continuous spatial distribution of signs as the disease grading result.

[0015] The beneficial effects of this invention are: This invention constructs a graded borrowing map and further identifies the first borrowing unit, enabling the explicit extraction and grading of high-grade latent signs that are obscured by surface low-grade signs in the image. Compared with traditional methods that grade based solely on overall imaging or lesion features, this invention effectively avoids the underestimation problem caused by small-area high-grade signs being masked by large-area low-grade signs. Thus, it can identify potential progression signals in the early stages of disease progression or in mixed manifestations, improving the sensitivity and foresight of grading results.

[0016] This invention, through the identification mechanism of the second borrowing unit, compares the level of latent signs with that of the main signs, and combines spatial continuity analysis to identify the discontinuous transition relationship between local high-level signs and the overall imaging. This mechanism breaks through the limitation of the overall consistency assumption in traditional grading methods, so that local anomalies are no longer averaged or weakened, thus accurately capturing the key features of local pre-deterioration. Simultaneously, by introducing conflict resolution and weighting mechanisms, different types of borrowing can be treated differently, reducing the interference of noise, artifacts, or sporadic anomalies on the grading results and improving the stability of the overall judgment.

[0017] This invention structurally combines three types of borrowing units to form a graded conflict evidence sequence. Based on this sequence, a conflict resolution process involving weighted accumulation and subtraction is performed on each candidate grade to obtain the confidence level results for each grade. Furthermore, a spatial continuity support factor is introduced to screen candidate grades for structural consistency, retaining only grades that form a continuous spatial distribution of signs as valid results. This mechanism not only avoids misjudgments caused by splicing together scattered abnormalities but also provides clear spatial structural support for the final grading results, achieving a unity of numerical and structural judgments, and improving the interpretability and clinical credibility of the results. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of a method according to an embodiment of the present invention; Figure 2 This is a logical framework diagram of an embodiment of the present invention. Detailed Implementation

[0020] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. For some well-known technologies, those skilled in the art may also use other alternative methods to implement the invention. Moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0021] like Figures 1-2 As shown, a disease grading assessment method based on image feature analysis includes the following steps: S1. Perform graded sign decomposition on the target disease image, distinguish between the surface signs that dominate the current overall imaging judgment and the latent signs hidden in the local structure, and generate a graded borrowing map based on the spatial misalignment relationship between the surface signs and the latent signs.

[0022] S11, Grade sign breakdown of target disease imaging: Pre-set symptom level label library Acquire images of the target disease Pixel-level or region-level segmentation is performed on the target disease image to obtain multiple sets of image feature regions corresponding to different disease levels, forming an image feature region set:

[0023] Each image feature region Corresponding to a disease severity label: ; in, Indicates the image of the target disease. This indicates a symptom level label library, including multiple preset disease level labels. Represents a set of image feature regions. Indicates the first Each image feature region Represents image feature regions The corresponding disease level label, This represents the total number of image feature regions.

[0024] The segmentation process for the target disease image is as follows: 1. Perform standardized processing on images of the target disease to make images from different devices and under different acquisition conditions comparable, forming standardized image input, including grayscale normalization processing to ensure that the grayscale distribution of different images is within a uniform range; spatial resolution alignment to ensure that the pixel spacing or voxel size is consistent, and output standardized images.

[0025] 2. Based on the structural continuity and display change trends of the image, initial region division is performed on the standardized image to obtain a set of candidate regions. Continuous change bands of grayscale or signal are identified to avoid cutting off gradient boundaries, maintain spatial connectivity, prevent the same structure from being split, separate obviously discontinuous regions, and output a set of candidate image regions.

[0026] 3: Perform consistency correction on the candidate image region set to avoid oversegmentation or undersegmentation. Merge regions with consistent internal features and unclear boundaries, and further subdivide regions with significant internal differences to ensure that the internal features of each region are relatively stable and the differences between regions are obvious. Output an image feature region set, remove regions that are too small and lack continuity, and remove regions formed only by noise or artifacts. Finally, output multiple sets of image feature regions.

[0027] The sign level label library is not a simple set of labels, but a ternary mapping library of sign-level-spatial relationships. It is used to solve the problem that the same image area may appear to be of a low level, but locally contains high-level signs. The specific construction steps are as follows: 1. Standard sign definition: Based on clinical knowledge and historical imaging data, define typical sign patterns for different disease grades, including the clarity of boundaries, homogeneity of internal structure, reaction mode of surrounding tissues, and regularity or destructiveness of morphology; output a set of sign descriptions.

[0028] 2. Symptom Stratification: Symptoms are organized into hierarchical levels, forming a progressive hierarchy, including: Low-level characteristics (stable, regular, limited); Moderate signs (boundary changes, mild invasion); High-level signs (structural damage, cross-regional impact); Output: A hierarchical set of characteristics.

[0029] 3. Spatial Relationship Encoding: Supplement each type of symptom with a spatial relationship description to enable it to recognize borrowed positions, including whether it is allowed to be wrapped (lower level wrapping higher level), whether it has a diffusion trend, whether there is a cross-regional break, and whether it depends on the response of neighboring tissues; output the spatial behavior rules of the symptom.

[0030] 4. Tag Mapping Construction: Bind the symptom description, level, and spatial relationship to form symptom level tag library entries. Each entry = [symptom identifier + corresponding level + spatial relationship rule + judgment condition].

[0031] Table 1. Examples of the Symptom Level Tag Library Symptom ID Symptom Name Corresponding level Spatial distribution characteristics Boundary characteristics Internal structural characteristics Spatial Relationship Rules Priority weight E1 Uniform dense region low level concentrated Clear uniform Do not spread Low E2 Blurred boundary area Intermediate level Local extension Vague Mild unevenness Can be extended outwards middle E3 Invasion and destruction zone High level Discrete diffusion irregular Inequality Penetrable package high E4 Local anomaly islands High level Discrete Clear or blurry High heterogeneity Can be embedded in low-level areas high In the table: E1 is a low level, representing a stable organizational state: the area is concentrated, the boundaries are clear and there is no diffusion, which usually constitutes a surface phenomenon. E2 (Medium) indicates the beginning of abnormal changes: the boundaries become blurred and show a tendency to expand, which is a transitional sign. E3 is a high-grade lesion, representing obvious lesions: structural damage, the ability to break through the original boundaries, and strong diagnostic capabilities. E4: A small area of ​​high-level features is embedded in a low-level area, which is a latent phenomenon. This is the key point of the present invention: the high-level features are wrapped by the low-level appearance.

[0032] In summary, this invention first constructs structurally stable image feature regions, then identifies the hierarchical semantics carried by the regions through a sign level label library, and further supports the construction of subsequent hierarchical borrow maps.

[0033] S12, Determination of surface signs and latent signs: S121, for each image feature region Calculate their spatial distribution range respectively. Visual prominence and coverage levels ; Spatial distribution range This is used to characterize the occupancy of a feature region within the overall image, reflecting whether it constitutes a dominant visual region. The calculation is as follows: Extract all pixels or voxels from the feature regions of an image to form a region point set; count the number of elements in the region point set to obtain the two-dimensional area or three-dimensional volume of the region; perform the same statistical analysis on the entire target disease image to obtain the total area or volume of the image; normalize the region area or volume with the overall image range to obtain the spatial distribution range. A larger value indicates a higher region proportion, which is more likely to be a superficial sign; a smaller value indicates a localized region, which is more likely to be a latent sign.

[0034] Visual prominence This is used to characterize the visual prominence of a region in an image, that is, whether the region is significant relative to its surrounding environment. The calculation is as follows: Calculate the average grayscale value or signal intensity within the image feature region as the region's internal characterization value. Construct a neighboring region, i.e., a ring-shaped or extended region, around the image feature region as the background reference region. Calculate the average grayscale value or signal intensity of the neighboring region as the background characterization value. Compare the degree of difference between the region and the background. If the difference is significant, it indicates high visual prominence of the region; if the difference is small, it indicates high integration between the region and the background. Higher values ​​indicate a more prominent region, more likely to be a surface feature; lower values ​​indicate a less prominent region, more likely to be a latent feature.

[0035] Coverage level This is used to characterize the dominance of an image feature region within the visual hierarchy, i.e., whether the region constitutes a higher layer of the overall visual structure. The calculation is as follows: Analyze the spatial relationships between image feature regions to identify whether there are overlapping, nesting, or occlusion relationships between regions. Prioritize regions according to the following principles: regions with larger coverage areas have higher priority; regions with stronger continuity have higher priority; regions with a greater impact on the overall structure have higher priority. Number all regions in hierarchical order; the outermost or dominant region has a higher priority value, and embedded or local regions have a lower priority value. Normalize the hierarchical values ​​to ensure they are within a uniform scale. Larger values ​​indicate a region that covers the entire area, representing a surface feature; smaller values ​​indicate a region that is more partially enveloped, representing a latent feature.

[0036] S122, Constructing comprehensive judgment indicators: ; in, Represents image feature regions Spatial distribution range, Indicates the overall spatial extent of the target disease image. Represents image feature regions Visual salience is used to characterize its salience relative to surrounding tissues. Represents image feature regions The coverage level is used to characterize the degree of hierarchical dominance in the image. This indicates a comprehensive judgment index. The weighting coefficients, denoted as 0.4, 0.35, and 0.25, are used to adjust the influence of different judgment factors. Surface features are primarily reflected in occupying the main field of vision and are the most stable and least affected by noise, thus having the highest weight. Visual salience directly influences the first judgment of the human eye or model, but is easily affected by imaging conditions and contrast, so it has the second highest weight. Coverage level is used to identify enclosed or latent relationships and calculate dependency inferences; its stability is slightly lower, so its weight is relatively lower. Essentially, this formula is a weighted fusion of three dimensions: spatial distribution range reflects whether it occupies the main field of vision, visual salience reflects whether it is visually significant, and coverage level reflects whether it is structurally dominant.

[0037] S123, based on comprehensive judgment indicators The size of the region is used to classify the image feature regions: ; in, The classification threshold is used to distinguish between surface signs and latent signs. Based on statistics from a large number of samples, surface signs usually meet the criteria of higher comprehensive indicators, while latent signs are usually below a certain threshold. The median or the inflection point of the distribution can be used as the classification threshold. Surface signs and latent signs are essentially in a continuous transitional relationship, so a single threshold is used to achieve a clear distinction.

[0038] S124 further forms two sets: ; ; in, Represents a set of surface signs. This indicates a collection of latent signs.

[0039] S13, Construction of the differential borrowing chart: The purpose is to determine the spatial relationship between surface signs and latent signs, including whether they are contained within each other, pressed together, or completely separated. The core of the differential borrowing chart is to determine whether signs of different levels have been misaligned.

[0040] S131, Extract the surface symptom set Collection of latent signs The spatial location information of each image feature region is defined as follows: ; For any set of surface sign regions Areas with latent signs Calculate the relative positional relationship index between the two: S1311, Spatial coverage: ;in, Indicates the area of ​​latent signs Surface signs area The proportion included This represents the area or volume of the intersection of two regions. This indicates the area or volume of the latent sign region. It determines whether a latent sign is enclosed by a surface sign. If a small area (latent sign) is completely contained within a large area (surface sign), it is considered enclosed.

[0041] S1312, Spatial overlap: ;in, This indicates the degree of overlap between surface and latent symptom areas. It represents the area or volume of the union of two regions. It determines whether two features overlap but do not completely contain each other, meaning that the two regions partially overlap but also have independent parts.

[0042] S1313, Spatial Separation: ;in, This represents the minimum distance between the surface sign area and the latent sign area. These are spatial points within the region. This represents the Euclidean distance between two points. It determines whether two features are completely separated, essentially finding the shortest distance between two regions.

[0043] S132, Based on the above spatial relationship indicators, identify the following misaligned areas: ; in, This is a misalignment region type between surface signs and latent signs. Thresholds corresponding to different misalignment determinations; A value of 0.6 indicates that when more than 60% of the latent area falls within the surface area, a clear encapsulation relationship has been formed. The value is 0.3, when There is already a clear spatial intersection. , The characteristic scale of the target image is the maximum diameter of the lesion or the regional scale, indicating whether two regions are separated from the same structural system. When the distance exceeds 15% of the overall structural scale, they are considered to no longer belong to the same continuous structure.

[0044] S133, map all misaligned regions to the same spatial reference frame. This forms a graded borrowing chart: ;in, This represents a borrowing chart for grade differences. To unify spatial reference frames, This is an aggregation operation for all combinations of surface and latent symptoms. This represents a spatial mapping function, used to uniformly transform different misaligned regions to the same spatial reference frame, making various misalignment relationships comparable, superimposed, and consistently expressive, thus forming a graded borrowing map. The specific steps are as follows: The spatial coordinate system of the target disease image itself is selected as the initial reference and standardized to establish a unified spatial reference system. This includes scaling the original coordinate system of the image to ensure that different image sizes are in a uniform proportion; setting the image center or lesion center as the unified coordinate origin; and standardizing the coordinate axis directions to ensure that different samples are consistent in direction.

[0045] For each misaligned region, its spatial location is standardized and transformed to map it into a unified spatial reference system. This includes extracting the original spatial location range of the misaligned region, i.e., the bounding box or point set; translating and adjusting the position of the region relative to the lesion center to align it with the unified reference origin; scaling the region according to a unified scale to make the regions in different images comparable in size; performing rotation correction on regions with directional differences to make their spatial orientation consistent with the unified reference system; and outputting the standardized misaligned region.

[0046] After unifying the position, scale, and orientation, a morphological mapping is performed on the misaligned region to ensure that the structural features of the region are not destroyed. This includes maintaining the relative positional relationships of the internal structures of the region; maintaining the continuity of the region boundary morphology; constraining local deformations to avoid structural distortion caused by mapping; and outputting the morphologically preserved mapped region.

[0047] During the spatial mapping process, the additional attributes of the misaligned regions are synchronously mapped to the unified reference system. This includes marking the misalignment type, which may include inclusion, overlap, or separation; attaching the corresponding symptom level information to the mapped region; preserving the correlation between regions; and outputting standardized misaligned regions with attributes.

[0048] Each misaligned region is projected onto a unified reference system according to its spatial location; overlapping regions are fused to retain the distinguishing information of different misalignment types; conflicting locations of multiple regions are recorded in layers or marked with priorities. In general, all misaligned regions processed by the spatial mapping function are superimposed onto a unified spatial reference system to form a unified expression result and output a differential borrowing map.

[0049] S2. Based on the grade difference borrowing diagram, identify whether there are borrowing situations between surface signs and latent signs, such as low-level appearances enveloping high-level signs, local high-level signs breaking away from the overall imaging order, or signs of the same level appearing intermittently across regions, and generate a grade difference conflict evidence sequence.

[0050] S21, Extract the sign information from the grade borrowing chart: Obtaining the borrowed position chart The sign levels and spatial distribution boundaries of each image feature region in the surface sign set and the latent sign set are extracted and represented as follows: ; ; in, This represents a borrowing chart for grade differences. This indicates the image feature region corresponding to surface signs. This indicates the image feature area corresponding to latent signs. This indicates the severity level corresponding to the surface symptom area. This indicates the severity level corresponding to the area of ​​latent symptoms. Indicates the spatial distribution boundary of the surface phenomenon area. This represents the spatial distribution boundary of the latent sign area. When the image feature area is obtained through S1 segmentation, each area already has its own boundary, namely the outer contour of the pixel / voxel set. The calculation of constructing the hierarchical borrow map is essentially based on the region boundary.

[0051] S22, the first borrowing unit identification, that is, a low-level appearance enveloping a high-level feature: For any set of surface signs areas Areas with latent signs Under the premise of satisfying spatial coverage or inclusion relationships, determine the relationship of their symptom levels: ;in, For the first borrow unit set, This indicates that the surface sign level is lower than the latent sign level. For region pairs that meet the above conditions, their corresponding spatial regions are marked as the first borrowing units: ;in, For the first borrowed unit space set, The spatial intersection area of ​​surface signs and latent signs.

[0052] The S22 first borrowing unit identification section primarily analyzes areas that appear to be of low level but actually conceal higher level areas. Specifically, it first determines whether a surface sign area and a latent sign area have a spatial enclosing relationship, meaning whether the latent sign is mostly contained within the surface sign. If the enclosing ratio reaches a set threshold, it indicates a true enclosing. After confirming a true enclosing, it then examines the level. If the level of the surface sign is lower than the latent sign inside, it indicates that a low-level appearance is obscuring a high-level sign. Once both spatial enclosing and level conditions (lower outside, higher inside) are met, this pair of areas is identified as the first borrowing unit. The overlapping portions of all areas satisfying this relationship are extracted as a dedicated set of areas—that is, the spatial locations of all high-level areas enclosed by low-level ones.

[0053] S23, the second borrowing unit identification, that is, high-level signs are out of the overall imaging order: First, define the main feature levels of the overall imaging. It can be determined by the dominant symptom level in the surface symptom set: This formula indicates that, among surface features, the feature level that occupies the largest spatial area and contributes the most to the overall visual continuity is taken as the dominant tone level of the current image. in, Indicates the level of the main symptoms, The area or volume of the surface manifestation region. This indicates taking the highest possible grade value from the expression.

[0054] Subsequently, each area of ​​latent signs was analyzed. Determine its level and spatial continuity: ; The degree of spatial discontinuity level transition is defined as follows: ; in, For the second borrow unit set, This refers to the degree of hierarchical difference between the area of ​​latent symptoms and its neighboring areas. Representation and region Adjacent region sets Indicates the severity level of the phenomenon in the surrounding area. The threshold for determining discontinuous transitions is 1-2, calculated based on the level difference. If the average level difference between the latent sign and the surrounding area is close to 0, it indicates a continuous change and is not abnormal. When the average difference reaches 1, it indicates that at least one level has been crossed, which is a significant transition. When it reaches 2 or above, it indicates a skipped level change and is determined to be an abnormal departure.

[0055] Regions that meet the conditions are marked as second borrowing units: ;in, This is the second borrowed unit space set.

[0056] The core of the S23 second borrowing unit identification section is to identify more concealed but dangerous situations, namely, the overall image may be at a certain level, but localized areas have already shown signs of a higher level, and these higher-level signs are detached from the overall structural order. Specifically, the first step is to determine the overall level, also known as the dominant sign level. This is determined by selecting the dominant level from the surface signs; that is, the level with the largest and widest surface area is considered the current main manifestation of the entire image, which is the benchmark for overall imaging. After the benchmark is determined, each latent sign area is analyzed. If a latent sign is found to be at a higher level than the dominant level, it indicates that it has exceeded the overall manifestation. Furthermore, it must be determined whether it is integrated into the overall structure or isolated. This can be done by analyzing its surrounding neighborhood. If the level of most surrounding areas is significantly lower than its own, and the difference is not a smooth transition but a sudden jump, it indicates that it is not on the same evolutionary level as its surroundings, representing a discontinuous level transition. When both conditions are met—a higher level and a clear discontinuity in the surrounding level relationships—this type of area is identified as a second borrowing unit.

[0057] S24, Third Borrowing Unit Identification, i.e., Discontinuous Distribution of Features of the Same Level: Features of the same level, which should be continuously distributed, are fragmented by other levels. In this case, it is necessary to first identify all image feature regions of the same level and observe whether they are spatially connected. If these regions are close to each other, they are considered to belong to the same continuous structure; if they are separated by a distance, and there are regions of other levels in between, it indicates that the feature of that level is fragmented and no longer a whole, but appears intermittently. The key judgment is whether they are connected. Take two regions of the same level and calculate the nearest distance between them. If the distance is very small, they are considered connected; if the distance exceeds a set threshold, they are considered separated. When many similar regions of the same level are divided into multiple unconnected small blocks, and these small blocks are indeed separated by regions of other levels, it is identified as a discontinuous distribution, which is the third borrowing unit.

[0058] The specific plan is as follows: For the same symptom level Extract the corresponding image feature region set: ; To determine the spatial connectivity of these regions, define a connectivity function: ; in, For connectivity determination function, Represents the minimum distance between two regions, from region Take one point from all boundary points, from the region We select one point from all boundary points, calculate the spatial distance between the two points, and take the minimum value among all point pairs. The Euclidean distance is used. The threshold for connectivity determination. , The spatial size of a single pixel or voxel is such that if the distance is less than 1 to 2 pixels, it is usually just a small break at the boundary and should be considered as connected. If it exceeds this range, it means that there is already a real gap.

[0059] If a set of regions of the same level is divided into multiple disconnected subsets, and different level regions exist as intervals between the subsets, then it is considered a discontinuous distribution, represented as: ; The corresponding spatial region is marked as the third borrowing unit: ;in, Represents the third borrow unit space set. This represents the third borrow unit set.

[0060] S25, the first, second, and third borrowing units are structurally combined according to their spatial location, borrowing type, and symptom level to form a graded conflict evidence sequence: ; in, Indicates a sequence of conflicting evidence. Represents spatial location coordinates, This indicates the borrow type identifier, i.e., the first, second, or third borrow unit. To correspond to the symptom level information, Represents a spatial point or region index.

[0061] The grade conflict evidence sequence essentially describes: in the image, where the anomaly begins, what type of anomaly it is, what its corresponding grade is, and how the anomaly is distributed and combined.

[0062] S3. Based on the differential conflict evidence sequence, perform conflict resolution assessment on the preset disease level, and determine the target level that still maintains a continuous support relationship after resolution as the disease classification result.

[0063] S31, Obtain the sequence of evidence for graded conflicts And a preset set of disease levels: ;in, For a pre-defined set of disease levels, For the first One candidate level, Number of candidate grades; The levels are arranged in ascending order of severity.

[0064] The sequence of conflicting evidence is represented as follows: ;in, For the first One unit of borrowed evidence Hey, spatial location, Borrow type identifier, This represents the symptom level corresponding to the borrowed unit.

[0065] A typical disease severity level set can be set to four levels: L1: Low risk or benign performance, stable structure, no invasive features. L2: Mild abnormality, local structures are beginning to change, but the overall structure remains stable; L3: Moderate anomaly, showing obvious structural damage or expansion trend; L4: Highly abnormal or malignant manifestations, with invasive, spreading or systemic effects; It can also be expanded to level five or more depending on the specific disease.

[0066] S32, Identification and weighting of supporting and conflicting relationships: For each borrow unit Define its relationship to candidate levels Relational functions: ; in, This indicates the relationship determination result of the borrow unit on the candidate level, where +1 indicates a supporting relationship, -1 indicates a conflicting relationship, and 0 indicates no significant relationship. This is a borrowing type. The third borrowing is a discontinuity of the same level, and in essence, it does not directly change the level relationship.

[0067] Assigning weights to resolve conflicting relationships: ; in, This indicates the resolution weight corresponding to the borrow unit. These correspond to the weight coefficients of the three types of borrowed units.

[0068] The first borrowing position in a lower-level package within a higher-level package, weighting coefficient. A value of 0.6 indicates that a higher-level conflict is masked, representing a hidden conflict rather than a direct negation of the current level; the weight is moderately high. The second borrowing action separates a higher-level conflict from the overall structure, with a weight coefficient... A value of 1.0 indicates a higher level occurring locally, discontinuous with the overall trend, constituting a direct conflict; this is the strongest upgrade signal and has the highest weight. In the third borrowing phase, which involves discontinuity at the same level, the weight coefficient... A value of 0.4 indicates that the structure is interrupted, but the grade is not directly changed. This is considered structural supporting evidence and has a low weight.

[0069] S32 essentially involves each piece of anomalous evidence voting for each candidate level, then determining which level is most reasonable. After obtaining the sequence of conflicting evidence, each piece of evidence represents an anomaly in a certain location within the image and carries two key pieces of information: The level of the phenomenon itself: whether it is high or low; Which type of borrowing is it: enclosed, detached, or discontinuous? Each candidate level L1, L2, L3, and L4 is evaluated individually. For a given level... Go and see all the evidence: 1. If the level of evidence is exactly equal to This indicates that it supports this level, i.e., it adds points; 2. If a piece of evidence is more... Higher, divided into two types: If it's a wrap-up type, it means higher levels are hidden, but upgrades and bonuses are still supported. If it's a departure type, it means a significantly higher level has emerged, negating the current level and deducting points. 3. If a piece of evidence is more... A lower score indicates a higher current level, resulting in a deduction of points.

[0070] Each point added or subtracted is not weighted equally; instead, it is multiplied by a weighting coefficient based on the borrowing type.

[0071] Finally, by summing up the impact of all the evidence, we get a total score for each candidate level, which is the confidence level. The one with the highest score is the final level.

[0072] S33, Rank Confidence Update and Conflict Resolution: Initialize the confidence scores for each candidate level: ;in, Candidate level The initial confidence level, To ensure a unified initial value, either 0 or 1 can be used, as both represent a unified basis.

[0073] For each candidate level Iterate through all borrowed units and update the confidence level: ;in, Indicates candidate level Confidence level after conflict resolution This indicates that the summation is applied to all borrowed units. This represents the weighted impact of the borrowed unit on this level. when This indicates that the confidence level is accumulated for that level; when This indicates that the confidence level is reduced for that level.

[0074] S33 sums up all the supporting or opposing evidence for borrowing and calculates a final confidence score for each candidate level.

[0075] Extract each of the previously obtained conflicting evidence sequences and use each piece of evidence to evaluate each candidate level. For a given candidate level, make a judgment for each piece of evidence encountered: does this evidence support or oppose the level? If it supports it, add points to the level; if it opposes it, subtract points. The amount added or subtracted is not fixed but determined by the borrowing type of the evidence, reflected through weighting. Accumulating all the evidence, each candidate level receives a final total confidence score, representing the degree to which the current level is supported by the combined effect of all anomalous evidence. The level with the highest score indicates that the overall evidence most strongly supports it.

[0076] S34, for each candidate level, there is already a total confidence score. The score indicates the overall support for this grade based on all borrowed evidence. However, in actual imaging, even if a grade has a high score, if it is spatially fragmented and pieced together, it should not be considered a final conclusion. Therefore, a spatial continuity support factor is introduced to determine whether the current grade forms a continuous and systematic distribution of signs in the image. If a grade has a high score but is only pieced together from several scattered small areas without forming a continuous structure, then this grade is directly excluded, i.e., multiplied by 0. Only those grades that have both high scores and form a continuous spatial distribution participate in the final comparison. Finally, among all grades that meet the continuity condition, the one with the highest score is selected as the final disease grading result.

[0077] Specifically, as follows: Select the target level from all candidate levels that meets the following criteria: ;in, The final disease classification result is the disease level that best supports the assessment of borrowed lobe evidence and forms a continuous pattern distribution in the imaging space among all candidate levels. It serves as a supporting factor for spatial continuity.

[0078] The spatial continuity support factor is defined as: ; It is an indicator variable used to screen whether spatial continuity support is available.

[0079] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0080] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A disease grading assessment method based on image feature analysis, characterized in that, Includes the following steps: S1. Perform graded sign decomposition on the target disease image, distinguish the surface signs that dominate the current overall imaging judgment and the latent signs hidden in the local structure, and generate a graded borrowing map based on the spatial misalignment relationship between the surface signs and the latent signs. S2. Based on the graded borrowing diagram, identify whether there are borrowing situations between surface signs and latent signs, such as low-level appearances enveloping high-level signs, local high-level signs breaking away from the overall imaging order, or signs of the same level appearing intermittently across regions, and generate a graded conflict evidence sequence. S3. Based on the aforementioned differential conflict evidence sequence, perform conflict resolution assessment on the preset disease level, and determine the target level that still maintains a continuous support relationship after resolution as the disease classification result.

2. The disease grading assessment method based on image feature analysis according to claim 1, characterized in that, S1 further includes a preset sign level label library, which includes multiple preset disease level labels. After acquiring the target disease image, the target disease image is segmented at the pixel level or region level to extract multiple sets of image feature regions corresponding to different disease levels. The multiple sets of image feature regions are used to match the disease level labels with the sign level label library.

3. The disease grading assessment method based on image feature analysis according to claim 2, characterized in that, The distribution range, visual prominence, and coverage level of the image feature regions in the image space are obtained. Image feature regions that occupy the main field of view of the image and constitute the overall visual continuity are marked as surface features, while image feature regions that are distributed in the local deep structure of the image, are spatially discrete, and visually constitute a front-to-back layer distinction from surface features are marked as latent features.

4. The disease grading assessment method based on image feature analysis according to claim 3, characterized in that, Extract the spatial location information of the surface signs and the latent signs respectively, calculate the relative positional relationship of their spatial location information in the image space, identify the spatial inclusion, local overlap or spatial separation misalignment areas between the surface signs and the latent signs, map the misalignment areas to the same spatial reference system, and generate a graded borrowing map. The graded borrowing map is used to characterize the spatial misalignment distribution of different levels of signs.

5. The disease grading assessment method based on image feature analysis according to claim 1, characterized in that, Extract the sign level and spatial distribution boundary of the surface signs and latent signs in the grade difference borrowing map; identify borrowing situations and mark them as multiple borrowing units; and combine the multiple borrowing units in a structured way according to their spatial location, borrowing type and the sign level involved to generate a grade difference conflict evidence sequence.

6. The disease grading assessment method based on image feature analysis according to claim 1, characterized in that, The borrowing situations include: Identify instances where a lower-level appearance encloses a higher-level feature through misplacement; Identify cases where localized high-level features deviate from the overall imaging order; Identify instances of borrowing where similar signs appear intermittently across different regions.

7. The disease grading assessment method based on image feature analysis according to claim 1, characterized in that, The identification of the borrowing situation where a low-level appearance covers a high-level feature specifically includes: when the feature level of the surface feature is detected in the differential borrowing map as being lower than the feature level of the latent feature covered or covered by the surface feature, the corresponding spatial area is marked as the first borrowing unit. The identification of local high-level features deviating from the overall imaging order specifically includes: when the feature level of the latent feature detected in the differential borrowing map is higher than the main feature level corresponding to the overall imaging, and the latent feature forms a non-continuous level transition with the surrounding features in spatial distribution, the corresponding spatial area is marked as the second borrowing unit. The identification of the intermittent occurrence of the same level of sign across regions specifically includes: when multiple image feature regions of the same sign level are detected in the differential borrowing map and are spatially separated by regions of different levels, forming an intermittent distribution pattern, the corresponding intermittent distribution region is marked as the third borrowing unit.

8. The disease grading assessment method based on image feature analysis according to claim 1, characterized in that, S3 acquires the grade conflict evidence sequence and a preset disease level, the preset disease level including multiple candidate levels in ascending order of severity; for each borrowing unit in the grade conflict evidence sequence, the symptom level involved in the borrowing unit is compared with the preset disease level to identify the support or conflict relationship between the borrowing unit and each candidate level, and a corresponding resolution weight is assigned to the conflict relationship according to the borrowing type.

9. A disease grading assessment method based on image feature analysis according to claim 8, characterized in that, The process iterates through all candidate levels, deducts the confidence level of candidate levels with conflicting relationships based on the resolution weights, and accumulates the confidence level of candidate levels with supporting relationships to form the conflict resolution results for each candidate level.

10. A disease grading assessment method based on image feature analysis according to claim 9, characterized in that, The S3 further includes identifying the candidate level with the highest confidence level after conflict resolution and supported by a continuous spatial distribution of signs as the disease grading result.

Images