Maxillary first molar extraction method decision method and system based on multi-parameter analysis

Abstract

The application discloses a maxillary first molar extraction method and system based on multi-parameter analysis. The method first acquires the patient's age and oral panoramic film; then, the panoramic film is segmented by the trained DeepLabV3+ semantic segmentation model to extract the mesial buccal root outer edge, mesial buccal root inner edge, distal buccal root outer edge, distal buccal root inner edge and root bifurcation point; based on the extracted contour, the minimum curvature radius R of the tooth root, the root bifurcation angle theta and the local standardized bone density index D are calculated; finally, according to the patient's age, the preset layered decision engine is entered, and the above quantitative parameters are combined to automatically output the recommended overall extraction or root extraction operation mode, and a decision basis and quantitative parameter report are generated. The application combines deep learning feature extraction with a clinical rule engine to realize objective quantitative decision and personalized recommendation of tooth extraction operation mode, and improves the accuracy and interpretability of the decision.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence-assisted diagnosis technology in oral medicine, and in particular to a method and system for decision-making regarding the extraction procedure of the maxillary first molar based on multi-parameter analysis. Background Technology

[0002] The maxillary first molar is one of the earliest erupting permanent teeth in the oral cavity, with the highest caries rate and the highest risk of loss due to periodontal disease. Its extraction is one of the most common procedures in oral and maxillofacial surgery. The maxillary first molar usually has three roots: the mesibular root, the distobuccal root, and the palatal root. The mesibular and distobuccal roots are often referred to as the "buccal double roots." The anatomical morphology of the buccal double roots is complex and variable. The degree of curvature, bifurcation angle, and the condition of the periostracal bone are key factors in determining the extraction procedure.

[0003] Currently, there are two main surgical procedures for extracting the palatal root and buccal roots of the maxillary first molar in clinical practice: one is complete extraction, which involves creating space and loosening the tooth before directly extracting it as a whole; the other is root-by-root extraction, which involves first using a drill to separate the palatal root and buccal roots in a "T-shape" cut, and then extracting them one by one. The choice of surgical procedure directly affects the extent of surgical trauma, operation time, risk of postoperative complications, and the patient's postoperative experience. An inappropriate choice may lead to serious complications such as root fracture, alveolar bone damage, postoperative infection, or even maxillary sinus perforation.

[0004] In current technology, doctors primarily rely on preoperative panoramic radiographs combined with their personal clinical experience to make decisions. Doctors subjectively assess the curvature of the tooth root, the bifurcation angle, and the bone density around the root by visually observing the panoramic radiographs. However, this traditional decision-making method has several limitations. First, visual assessment lacks standardized quantitative criteria; different doctors may interpret the same image differently, and even the same doctor may interpret it inconsistently at different times, leading to strong subjectivity and making it difficult to ensure standardized and homogenized treatment. Second, the human eye cannot accurately quantify the subtle curvature and bifurcation angle of the tooth root, and judgments of critical states are prone to bias; for example, the definition of "slight curvature" often varies from person to person. Furthermore, bone density assessment on two-dimensional panoramic radiographs is highly dependent on the doctor's personal perception, lacking objective quantitative indicators to eliminate image differences caused by different equipment and exposure parameters. Furthermore, patient age is an important biological factor. Younger patients have harder bones and more intact tooth roots, resulting in greater extraction resistance; while older patients have relatively porous bones, wider periodontal ligament spaces, and less extraction resistance, but also have reduced healing ability and are more sensitive to surgical trauma. Existing decision-making methods have failed to effectively integrate age factors with anatomical morphological parameters in their analysis.

[0005] A search of existing patent and non-patent literature revealed that while there are studies on measuring root anatomy and morphology, such as measuring root bifurcation angles and root trunk height using extracted teeth, these studies mostly remain at the level of basic anatomical description and have not been translated into clinically usable auxiliary decision-making tools. There are also some patent applications involving intelligent analysis of oral images, but most focus on caries detection, periodontal disease diagnosis, or implant planning. No methods or systems specifically addressing maxillary first molar extraction procedures that integrate age stratification and panoramic radiographs for intelligent decision-making have been found.

[0006] Therefore, there is an urgent need for a method that can provide objective, quantitative, and personalized surgical decision support for the extraction of the maxillary first molar, in order to overcome the shortcomings of existing technologies that rely on subjective experience, lack unified standards and quantitative evidence, reduce surgical risks, and improve treatment success rates. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention provides a method and system for decision-making regarding the extraction procedure of the maxillary first molar based on multi-parameter analysis.

[0008] The present invention provides a method for deciding on the extraction procedure of the maxillary first molar based on multi-parameter analysis, comprising the following steps: Step S1: Obtain the patient's age A and panoramic oral radiograph I; Step S2: Input the panoramic oral radiograph I into a trained semantic segmentation model to extract the mesial buccal root lateral border, mesial buccal root medial border, distal buccal root lateral border, distal buccal root medial border, and root bifurcation point; Step S3: Based on the extracted mesial buccal root lateral border and distal buccal root lateral border, calculate the minimum radius of curvature R of the tooth root; Based on the extracted mesial buccal root lateral border and distal buccal root lateral border, calculate the minimum radius of curvature R of the tooth root; Calculate the root bifurcation angle θ at the medial border of the mid-buccal root, the medial border of the distal-buccal root, and the root bifurcation point; calculate the local standardized bone mineral density index D based on the extracted lateral borders of the mid-buccal root and the distal-buccal root; Step S4: Compare the patient's age A with a preset age threshold, and select the corresponding decision submodule based on the comparison result; In the decision submodule, call the radius of curvature R, the angle θ, and the bone mineral density index D, and output the primary decision D_primary through a preset decision tree logic or weighted scoring model.

[0009] Furthermore, the semantic segmentation model in step S2 is a DeepLabV3+ model, which is trained to perform a six-class semantic segmentation task. The six classes are background, mesial buccal root lateral edge, mesial buccal root medial edge, distal buccal root lateral edge, distal buccal root medial edge, and root bifurcation point.

[0010] Furthermore, the calculation of the minimum radius of curvature R in step S3 specifically includes: smoothing the extracted lateral edges of the mesial buccal root and the lateral edges of the distal buccal root; calculating the instantaneous curvature k of each point on the contour line; taking the point with the largest absolute value of curvature among all points, whose radius of curvature R=1 / |k_max|, and converting it to physical units in combination with the pixel-millimeter scale of the panoramic image.

[0011] Furthermore, the calculation of the root bifurcation angle θ in step S3 specifically includes: locating four points 2mm and 4mm away from the coronal direction of the root bifurcation point on the medial edge of the mesial buccal root and the medial edge of the distal buccal root, respectively; connecting the points 2mm and 4mm coronally to obtain vectors V_MB and V_DB representing the central trends of the two roots; and calculating the angle between the two vectors as the root bifurcation angle θ.

[0012] Furthermore, the calculation of the standardized bone mineral density index D in step S3 specifically includes: using the lateral border of the mesial buccal root and the lateral border of the distal buccal root as baselines, expanding outwards to generate a region of interest; selecting a reference area on the panoramic image; and calculating the standardized Z-score of the pixel mean in the region of interest relative to the pixel mean in the reference area as the bone mineral density index D.

[0013] Furthermore, step S4 specifically includes: First-level diversion: If A≥60, then directly output "suggest root removal"; if 35≤A<60, then enter the comprehensive decision-making submodule for the middle-aged and young adults; if A<35, then enter the morphological decision-making submodule for the young adults; in the comprehensive decision-making submodule for the middle-aged and young adults, if R≤8mm, then output "suggest root removal"; if R>8mm and θ>20°, then start the weighted scoring model S=w1×(θ-20)+w2×D+w3×(A-35); if S≥threshold, then output "suggest root removal", otherwise output "tend to remove as a whole"; in the morphological decision-making submodule for the young adults, if R≤8mm or θ>20°, then output "suggest root removal", otherwise output "suggest removal as a whole".

[0014] Furthermore, in step S3, when calculating the standardized bone mineral density index D, the reference area is the dense cortical bone region at the lower edge of the mandible.

[0015] Furthermore, before calculating the minimum radius of curvature R in step S3, the extracted contour lines are further subjected to Savitzky-Golay filtering for smoothing.

[0016] Furthermore, the weights w1, w2, and w3 in the weighted scoring model are obtained by training historical clinical data using logistic regression.

[0017] This invention also provides a decision-making system for maxillary first molar extraction based on multi-parameter analysis, comprising: an input module for acquiring the patient's age A and a panoramic oral radiograph I; a feature extraction module for extracting the root contour and bifurcation point from the panoramic oral radiograph I based on a trained semantic segmentation model, and calculating the radius of curvature R, the included angle θ, and the bone mineral density index D; a decision engine module for selecting the corresponding decision path according to the patient's age A, and outputting the main decision D_primary according to a preset decision logic based on the R, θ, and D; and an output module for outputting the main decision D_primary, the decision basis, and a report of quantitative parameters.

[0018] Compared with the prior art, the present invention has the following beneficial effects: This invention is the first to combine an improved DeepLabV3+ semantic segmentation model, precise geometric parameter calculation, and a hierarchical rule engine that aligns with clinical thinking, achieving full automation from image input to surgical procedure recommendation. By automatically extracting root margins and key points through a deep learning model, it avoids the errors and subjectivity introduced by manual measurement in traditional methods, providing objective and accurate quantitative data for subsequent decision-making. This process guides the decision-making path through age stratification and then combines morphological parameters for refined judgment, ensuring the rigor of the decision-making logic and its clinical applicability.

[0019] This invention achieves a precise balance between a function-preservation orientation for young patients and a safety-first orientation for elderly patients through age stratification. For young patients (<35 years old), the decision-making logic is entirely based on root morphology (radius of curvature R and bifurcation angle θ), aiming to preserve root integrity while attempting minimally invasive extraction. For middle-aged and young patients (35-60 years old), a bone mineral density index (D) is introduced for comprehensive weighted scoring, taking into account both morphology and bone quality. For elderly patients (≥60 years old), root extraction is directly recommended to shorten surgical time and reduce trauma. This age-stratified design makes the decision-making results more in line with the physiological characteristics and clinical needs of patients in different age groups, reflecting a highly personalized treatment philosophy.

[0020] The decision-making method provided by this invention is highly transparent and interpretable. Each judgment step has clear quantitative parameters and threshold standards, and the output results are accompanied by detailed evidence. When the system outputs "suggests root removal," it will also explain whether it is due to "bent root (radius of curvature R = a certain value)," "excessive bifurcation angle (angle θ = a certain value)," or "priority given to elderly patients," etc. When it outputs "tends to remove the entire root," it will also provide a comprehensive score or low-risk indication. This interpretability allows clinicians to understand and verify the system's reasoning process, enhancing human-computer trust and meeting the reliability requirements of medical AI. At the same time, the quantitative parameters (R, θ, D) in the output report can be retained as medical record data, providing data support for subsequent treatment or research.

[0021] Furthermore, this invention possesses strong engineering practicality. The system employs a modular design, with the feature extraction module and decision logic module completely decoupled. This means that, with the accumulation of clinical data in the future, the deep learning model can be independently updated to improve the accuracy of feature extraction, or the thresholds and weights of the weighted scoring model in the decision tree can be optimized, without requiring significant changes to the entire system architecture, facilitating continuous system upgrades and maintenance. This system can be seamlessly integrated into existing oral imaging archiving and communication systems; doctors can obtain decision suggestions with a single click, without altering their existing workflows, demonstrating significant clinical application value. Attached Figure Description

[0022] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0023] Figure 1 This is a system overall workflow diagram provided in the embodiments of the present invention; Figure 2 This is a schematic diagram of the network structure of the DeepLabV3+ semantic segmentation model provided in this embodiment of the invention; Figure 3 This is an example diagram of panoramic image semantic segmentation and key structure annotation provided in an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the calculation of root curvature (radius of curvature R) provided in an embodiment of the present invention; Figure 5 This is a schematic diagram illustrating the calculation of the root bifurcation angle (θ) provided in an embodiment of the present invention; Figure 6 This is a region of interest definition diagram for calculating the localized normalized bone mineral density index (D) provided in an embodiment of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are part of this invention.

[0025] like Figure 1 As shown in the figure, this embodiment provides an overall workflow diagram of a decision-making system for maxillary first molar extraction procedures based on multi-parameter analysis.

[0026] like Figure 1 As shown, the patient's age A and panoramic dental radiograph I are first acquired. Patient age A is an integer scalar, such as 35 or 58. Panoramic dental radiograph I is a standard digital two-dimensional X-ray image, formatted as a grayscale matrix [H, W], where H is the image height, W is the image width, and the grayscale value of each pixel represents the X-ray absorption rate at that point. This image is typically stored in DICOM format and can be directly imported from the hospital's imaging system.

[0027] Then, feature extraction is performed using an internally integrated, pre-trained deep learning model to automatically and accurately extract three quantification parameters required for subsequent decision-making from the panoramic image I: the minimum radius of curvature of the tooth root R, the bifurcation angle θ, and the local standardized bone mineral density index D.

[0028] Finally, the final decision is made based on the patient's age, quantitative parameters R, θ, and D.

[0029] Specifically, the deep learning model is a semantic segmentation model based on DeepLabV3+. For example... Figure 2 As shown, the model employs an encoder-decoder architecture. In the encoder part, the input image is first processed by a deep convolutional network (DCNN) with dilated convolutions for feature extraction, such as using ResNet101 as the backbone. The extracted features are then fed into a dilated spatial pyramid pooling module for multi-scale contextual information capture. This module contains five parallel branches: branch 1 is a 1x1 convolution; branches 2-4 are three 3x3 dilated convolutions with different dilation rates of 6, 12, and 18, respectively; and branch 5 is a global image pooling layer. The output feature maps from these five branches are concatenated along the channel dimension and then fused using a 1x1 convolution. The resulting feature map is then upsampled by a factor of 4 and input into the decoder. In the decoder, a set of corresponding low-level features is extracted from the encoder backbone and adjusted using a 1x1 convolution. This adjusted low-level feature is then concatenated with the upsampled high-level semantic features from the encoder output. Then, a 3x3 convolution is applied to the stitched feature maps for further fusion, and finally, the resolution of the input image is restored by 4x upsampling to output the class prediction for each pixel.

[0030] This embodiment defines the problem as a six-class semantic segmentation task. The model has six output channels, corresponding to: Channel 0: background; Channel 1: lateral edge of the mesial buccal root; Channel 2: medial edge of the mesial buccal root; Channel 3: lateral edge of the distal buccal root; Channel 4: medial edge of the distal buccal root; Channel 5: root bifurcation point. During the training phase, at least 500 panoramic images precisely annotated by oral surgeons using professional software are required. Annotation uses thin lines of single-pixel width to precisely delineate the four edges, and root bifurcation points are marked with single pixels. The loss function used is cross-entropy loss. After sufficient training, the model can accurately perform segmentation inference on the input image I, obtaining the following results: Figure 3 The annotation results shown include the medial border of the mesial buccal root C_MB_I, the lateral border of the mesial buccal root C_MB_O, the medial border of the distal buccal root C_DB_I, the lateral border of the distal buccal root C_DB_O, and the root bifurcation point P_f.

[0031] In the inference and post-processing stages, after the model infers from the input image I, the predicted label map L_pred is obtained through the argmax operation. Connectivity analysis is performed on the output of channel 5, and its centroid coordinates are taken as the root bifurcation point P_f. Morphological thinning is performed on the outputs of channels 1-4, resulting in four single-pixel-wide edge contour lines: the lateral edge of the mesial buccal root C_MB_O, the medial edge of the mesial buccal root C_MB_I, the lateral edge of the distal buccal root C_DB_O, and the medial edge of the distal buccal root C_DB_I.

[0032] The minimum radius of curvature R is a quantitative indicator of the degree of root curvature. It is defined as the radius of the circle of curvature corresponding to the point of greatest curvature on the outer edge of the root contour. Curvature describes the degree of curvature of a curve; a larger curvature indicates greater curvature, and a smaller radius of curvature. Clinically, a smaller R indicates a more curved root, making root fracture more likely during extraction.

[0033] The specific calculation steps are as follows: 1. For example Figure 4 As shown, the Savitzky-Golay filter is used to smooth the two outer edge contour lines C_MB_O and C_DB_O to eliminate local fluctuations caused by image noise. The Savitzky-Golay filter is a smoothing method based on local polynomial fitting, which can preserve the shape characteristics of the curves well while removing noise.

[0034] 2. The instantaneous curvature k at each point on the contour line is calculated using a geometric method based on moving points. For a curve composed of discrete points, a circle is usually defined by three adjacent points, and the reciprocal of the radius of this circle is an approximation of the curvature at that point. The specific algorithm is as follows: For a point Pi on the contour line, take its two adjacent points Pi-1 and Pi+1, calculate the radius of the circle defined by these three points, and the reciprocal of this radius is the curvature ki. A curvature sequence is obtained by traversing all points.

[0035] 3. Find the point with the largest absolute value of curvature; this point is where the curvature is most severe, and its curvature is denoted as k_max. Then the minimum radius of curvature of this tooth root is R = 1 / |k_max|.

[0036] 4. Due to the pixel-to-millimeter scale of panoramic films (usually about 10 pixels / mm, meaning each pixel represents 0.1mm), the pixel unit R is converted to physical millimeter units. For example, if R_pixel = 80 pixels, then R = 80 / 10 = 8.0mm.

[0037] The critical threshold for the radius of curvature R was set at 8 mm. This setting is based on the typical working length of the maxillary first molar root (average approximately 10 mm) and clinical risk considerations. Using a clinically risky sharp bend (approximately a 90° arc) as a model, its radius of curvature R≈L / θ≈10 / (π / 2)≈6.4 mm, which is the theoretically significant bending radius. Considering the diversity of actual bending morphology, imaging measurement errors, and the need for algorithm robustness, the threshold was moderately relaxed from the theoretical benchmark to 8 mm. This setting ensures that all truly sharp bends (R<6.4 mm) are identified, while also covering risky bends in critical states. Retrospective clinical analysis confirmed that roots with R<8 mm have a significantly increased risk of root fracture or requiring special techniques (such as root separation) during extraction. Therefore, when R≥8 mm, the root is considered straight or slightly curved; when R<8 mm, it is considered a curved root, indicating a higher extraction risk.

[0038] The root bifurcation angle θ is a quantitative indicator that measures the degree of separation between the mesobuccal and distal buccal roots. It is defined as the angle between the central vectors of the two roots at a certain coronal distance from the bifurcation point. A larger angle indicates a greater separation between the two roots, resulting in greater interference between them during extraction and increasing the risk of root fracture.

[0039] The specific calculation steps are as follows: 1. For example Figure 5 As shown, four points are located on the two inner edge contour lines C_MB_I and C_DB_I, respectively, at distances of 2mm and 4mm coronally from the root bifurcation point P_f: P_MB_1, P_MB_2, P_DB_1, and P_DB_2. The distances are calculated along the curved path. The selection of points at 2mm and 4mm is based on a balance between anatomical rationality and measurement robustness. The area within 0-1.5mm coronally of the bifurcation point is the curvature transition zone of the root from fusion to separation, and its direction is unstable. Taking 2mm as the starting point can ensure that the measurement point is located in the main body of the root neck where the direction has become stable. The 2mm and 4mm distances are moderate, and the vector formed by them can represent the typical direction of the root body after bifurcation, eliminating the interference of the root tip's own curvature on the macroscopic geometric relationship of "bifurcation angle".

[0040] The diameter of the buccal root neck of the maxillary first molar is about 1.5-2.0 mm. The 2 mm spacing is slightly larger than the root diameter, so that the vector can stably reflect the overall axis of the tooth root and reduce the influence of local contour fluctuations. This distance is easy to clearly identify and stably locate in the image, reducing the amplification of subsequent calculations caused by the identification error of the bifurcation point P_f, and improving the robustness of the algorithm.

[0041] 2. Connect P_MB_1-P_MB_2 and P_DB_1-P_DB_2 respectively to obtain vectors V_MB and V_DB representing the two central trends.

[0042] 3. Calculate the angle between the two vectors: θ = arccos((V_MB·V_DB) / (|V_MB|×|V_DB|))×(180 / π).

[0043] The critical threshold for the bifurcation angle θ is set at 20°. This threshold is determined based on the following two considerations: Statistical anatomical basis: Related studies on a large number of maxillary first molars show that the bifurcation angle of their buccal roots exhibits a certain distribution pattern, with 20° being a key dividing point that can effectively distinguish between relatively concentrated (usually <20°) and significantly separated (usually ≥20°) root morphologies.

[0044] Clinical decision-making criteria: When θ < 20°, the root bifurcation is less pronounced and the overall morphology is more uniform. When extraction force is applied, the roots are less likely to interfere with each other, and the overall resistance to detachment is relatively small. Therefore, it is determined that the tooth can be extracted completely. When θ ≥ 20°, the root bifurcation is obvious. Forced extraction of the whole tooth will cause significant stress concentration at the bifurcation point due to leverage, which greatly increases the risk of root fracture or bone damage. Therefore, it is determined that root extraction is recommended.

[0045] Localized Standardized Bone Mineral Density Index (D) is a quantitative indicator of alveolar bone density around the tooth root. It is defined as the standardized Z-score of the grayscale values ​​of bone tissue pixels within a certain range on the outer side of the tooth root, relative to a stable reference area in the same image (such as the cortical bone at the lower border of the mandible). The Z-score represents the standard deviation of the bone density in this region from the mean of the reference area; a positive value indicates a higher density than the reference area, and a negative value indicates a lower density. This index eliminates the influence of different equipment and exposure parameters, achieving cross-image comparability.

[0046] The specific calculation steps are as follows: 1. Define the Region of Interest (ROI): such as Figure 6As shown, using the two outer edge contour lines C_MB_O and C_DB_O as baselines, morphological dilation is used to generate two annular region masks M_A and M_B, extending outward by 0.5mm (Region A) and 1.0mm (Region B, minus Region A), respectively. The size of the dilation kernel is determined by a pixel-to-millimeter scale. For example, if the scale is 10 pixels / mm, then a 0.5mm extension corresponds to a 5-pixel dilation kernel. Region A represents the bone tissue immediately adjacent to the tooth root, and Region B represents the bone tissue slightly further away; the combination of the two reflects the overall condition of the periroot bone.

[0047] 2. Extract original pixel values: Extract the set of all pixel values ​​V_A and V_B within regions M_A and M_B from the original grayscale image I.

[0048] 3. Internal Standardization: On the same image I, an anatomically stable region with minimal influence from projection conditions is automatically selected as a reference region, such as the dense cortical bone of the lower border of the mandible. This region typically has relatively constant bone density and exhibits minimal variation among individuals. The resulting pixel value set V_ref is obtained. The mean μ_ref and standard deviation σ_ref of the reference region are calculated.

[0049] 4. Calculate the standardization index: D_A = (mean(V_A) - μ_ref) / σ_ref; D_B = (mean(V_B) - μ_ref) / σ_ref. The D value is a dimensionless Z-score, which physically represents the degree of deviation of the target region's bone mineral density from the standard cortical bone mineral density in the image. In this embodiment, (D_A + D_B) / 2 is ultimately selected as the bone mineral density feature D. If the reference area cannot be automatically identified (such as a defect in the mandibular region), the user can manually select it, and the system will record the area for subsequent automatic learning.

[0050] The role of bone mineral density index (D) in decision-making: The higher the D value, the denser the bone around the tooth root, the greater the extraction resistance, and the more inclined to extract the tooth root separately to reduce trauma; conversely, the lower the D value (even a negative value), the more porous the bone, the less extraction resistance, and the more inclined to extract the tooth whole.

[0051] After feature extraction is completed, the age A and the calculated R, θ, and D are used for the final decision.

[0052] like Figure 1 As shown, the initial triage is based on age A, and the judgment is made strictly according to the predetermined decision tree logic.

[0053] The first level is age-stratified triage. Patients are divided into three levels based on age. Rule L1: If A ≥ 60, then D_primary = "Recommended root extraction", R_explain = "Elderly patients (≥ 60 years old), minimizing trauma is prioritized", and the process terminates. Rule L2: If 35 ≤ A < 60, proceed to the "Middle-aged and Young Adult Comprehensive Decision Submodule". Rule L3: If A < 35, proceed to the "Youth Morphological Decision Submodule".

[0054] The second level involves internal decision-making within each layer. In the morphological decision-making submodule for the youth layer (A<35), the decision completely ignores bone mineral density feature D and is based solely on root morphology. This is because young patients typically have denser bone but strong healing ability, and the decision should focus more on root morphology to avoid root fracture. Specific rules: If R≤8mm, then D_primary="suggest removing the root", R_explain="bent root (radius of curvature R=X mm)".

[0055] If R > 8mm, then: If θ>20°, then D_primary="suggest removing the root", R_explain="straight root but the bifurcation angle is too large (angle θ=Y°)".

[0056] If θ≤20°, then D_primary="It is recommended to remove the entire root", R_explain="It is recommended to try minimally invasive removal if the root is straight and the bifurcation angle is small".

[0057] In the comprehensive decision-making submodule for young and middle-aged individuals (35≤A<60), morphology and bone quality must be considered comprehensively. Specific rules: First, perform a preliminary morphological screening: if R≤8mm, then directly output "Recommend root removal", R_explain="bent root (radius of curvature R=X mm)".

[0058] If R > 8mm, then proceed to the comprehensive assessment of straight roots: If θ > 20°, the weighted scoring model S = w1 × (θ - 20) + w2 × D + w3 × (A - 35) is activated. Here, w1, w2, and w3 are positive weights obtained through training with historical clinical data. In this embodiment, based on preliminary clinical data, logistic regression training was used to obtain w1 = 0.4, w2 = 0.5, and w3 = 0.1. These weights reflect the contribution of each factor to the difficulty of tooth extraction: 0.4 points for every 1° increase in bifurcation angle, 0.5 points for every 1-degree increase in bone mineral density, and 0.1 points for every year exceeding 35 years of age. The empirical value of S_threshold is set to 5.0. If S ≥ S_threshold, then D_primary = "Recommended for root extraction", based on factors including large bifurcation angle, high bone mineral density, and age bonus; otherwise, D_primary = "Preferring whole extraction".

[0059] If θ≤20°, the system defaults to whole-root extraction. Only when the D value is abnormally high (e.g., D>2.0) will this tendency be reversed, and root separation be recommended. In this case, R_explain="extremely high bone density (D=Z), whole-root extraction faces great resistance".

[0060] Finally, the primary decision (D_primary), the structured decision basis (R_explain), and the quantitative parameter report (P_report) containing R, θ, and D values ​​are output to the user. The output format can be structured text, JSON data, or a PDF report, making it convenient for doctors to view and archive.

[0061] This embodiment also provides a decision-making system for maxillary first molar extraction based on multi-parameter analysis, including: an input module for acquiring the patient's age A and a panoramic oral radiograph I; a feature extraction module for extracting the root contour and bifurcation point from the panoramic oral radiograph I based on a trained semantic segmentation model, and calculating the radius of curvature R, the included angle θ, and the bone mineral density index D; a decision engine module for selecting the corresponding decision path according to the patient's age A, and combining the R, θ, and D to output the main decision D_primary according to a preset decision logic; and an output module for outputting the main decision D_primary, the decision basis, and a report of quantitative parameters.

[0062] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A method for deciding on the surgical approach for maxillary first molar extraction based on multi-parameter analysis, characterized in that, Includes the following steps: Step S1: Obtain patient age A and panoramic oral radiograph I; Step S2: Input the panoramic oral radiograph I into the semantic segmentation model to extract the lateral border of the mesial buccal root, the medial border of the mesial buccal root, the lateral border of the distal buccal root, the medial border of the distal buccal root, and the root bifurcation point; Step S3: Based on the extracted lateral borders of the mes buccal root and the distal buccal root, calculate the minimum radius of curvature R of the tooth root; based on the extracted medial borders of the mes buccal root, the distal buccal root, and the root bifurcation point, calculate the root bifurcation angle θ; based on the extracted lateral borders of the mes buccal root and the distal buccal root, calculate the local standardized bone mineral density index D. Step S4: Compare the patient's age A with a preset age threshold, and select the corresponding decision submodule based on the comparison result; In the decision submodule, the radius of curvature R, the included angle θ, and the bone mineral density index D are called, and the main decision D_primary is output through a preset decision tree logic or weighted scoring model. The calculation of the root bifurcation angle θ in step S3 specifically includes: positioning four points 2mm and 4mm away from the coronal direction of the root bifurcation point on the medial edge of the mesial buccal root and the medial edge of the distal buccal root, respectively; connecting the points 2mm and 4mm coronally to obtain vectors V_MB and V_DB representing the central trends of the two roots; calculating the angle between the two vectors as the root bifurcation angle θ, where θ=arccos((V_MB•V_DB) / (|V_MB|×|V_DB|))×(180 / π); The calculation of the standardized bone mineral density index D in step S3 specifically includes: using the lateral border of the mesial buccal root and the lateral border of the distal buccal root as baselines, expanding outwards to generate a region of interest; selecting a reference area on the panoramic image; and calculating the standardized Z-score of the pixel mean in the region of interest relative to the pixel mean in the reference area as the bone mineral density index D. Step S4 specifically includes: First-level branching: If A≥60, then directly output "Suggest root removal"; if 35≤A<60, then enter the comprehensive decision-making submodule for middle-aged and young people; if A<35, then enter the morphological decision-making submodule for young people. In the comprehensive decision-making submodule for the middle-aged and young adults, if R≤8mm, the output is "Recommend root removal"; if R>8mm and θ>20°, the weighted scoring model S=w1×(θ-20)+w2×D+w3×(A-35) is activated; if S≥threshold, the output is "Recommend root removal"; otherwise, the output is "Prefer whole removal"; where w1, w2, and w3 are weights obtained through training with historical clinical data. In the youth layer morphology decision submodule, if R≤8mm or θ>20°, the output is "suggest removing individual roots", otherwise the output is "suggest removing the whole plant".

2. The method for deciding on the extraction procedure of the maxillary first molar based on multi-parameter analysis according to claim 1, characterized in that, The semantic segmentation model in step S2 is the DeepLabV3+ model, which is trained to perform a six-class semantic segmentation task. The six classes are background, mesial buccal root lateral edge, mesial buccal root medial edge, distal buccal root lateral edge, distal buccal root medial edge, and root bifurcation point.

3. The method for deciding on the extraction procedure of the maxillary first molar based on multi-parameter analysis according to claim 1, characterized in that, The calculation of the minimum radius of curvature R in step S3 specifically includes: smoothing the extracted lateral edges of the mesial buccal root and the lateral edges of the distal buccal root; calculating the instantaneous curvature k of each point on the contour line; taking the point with the largest absolute value of curvature among all points, the maximum instantaneous curvature of this point is k_max, and its radius of curvature R=1 / |k_max|, and converting it to physical units in combination with the pixel-millimeter scale of the panoramic image.

4. The method for deciding on the extraction procedure of the maxillary first molar based on multi-parameter analysis according to claim 1, characterized in that, In step S3, when calculating the standardized bone mineral density index D, the reference area is the dense cortical bone region at the lower edge of the mandible.

5. The method for deciding on the extraction procedure of the maxillary first molar based on multi-parameter analysis according to claim 1, characterized in that, Before calculating the minimum radius of curvature R in step S3, the extracted contour lines are further subjected to Savitzky-Golay filtering for smoothing.

6. The method for deciding on the extraction procedure of the maxillary first molar based on multi-parameter analysis according to claim 1, characterized in that, The weights w1, w2, and w3 in the weighted scoring model are obtained by logistic regression training on historical clinical data.

7. A decision-making system for maxillary first molar extraction based on multi-parameter analysis, used to perform the method as described in any one of claims 1-6, characterized in that, include: The input module is used to obtain the patient's age A and the panoramic oral radiograph I; The feature extraction module is used to extract the root contour and root bifurcation point from the panoramic oral radiograph I based on the semantic segmentation model, and to calculate the radius of curvature R, the included angle θ and the bone mineral density index D. The decision engine module is used to select the corresponding decision path based on the patient's age A, and in combination with R, θ and D, output the main decision D_primary according to the preset decision logic; The output module is used to output the main decision D_primary, decision basis, and quantitative parameter report.

Images