Anatomical structure segmentation based medical age and gender prediction framework and explanation system

Abstract

The application provides an anatomic structure segmentation-based medical age and gender prediction framework and an explanation system, comprising: a data preprocessing and standardization module for cleaning, enhancing and normalizing input medical images; an age and gender feature extraction and prediction module for constructing and training a deep learning model to extract features from the medical images processed by the data preprocessing and standardization module and predict age or gender labels; a model explanation extraction module under structural constraints for taking medical prior segmentation results as structural constraints, generating model decision explanations aligned with medical structures based on intermediate layer feature maps and prediction outputs obtained by the age and gender feature extraction and prediction module through a dual-view technology path; and an explanation reliability verification module for quantitatively verifying the causal fidelity and rationality of the explanation results of the model explanation extraction module under structural constraints through a gradual intervention strategy, so as to improve the quality and readability of the explanation.

Description

Technical Field

[0001] This invention belongs to the field of medical imaging technology, and in particular relates to a medical age and sex prediction framework and interpretation system based on anatomical structure segmentation. Background Technology

[0002] With the widespread application of medical imaging in clinical diagnosis and treatment, deep learning-based image analysis techniques are increasingly being used for tasks such as age estimation, physiological development assessment, and gender recognition. However, existing models mainly rely on end-to-end convolutional neural networks or Transformer structures. While these models can achieve some improvement in predictive performance, they still face the following problems in medical applications: a lack of reliable medical interpretability. 1. General interpretation algorithms are not easy to interpret: Existing models mostly use the Grad-CAM general interpretation algorithm, but the interpretation diagrams are often scattered, noisy, and cannot correspond to specific medical anatomical structures, making it difficult for doctors to verify whether the "model's focus is reliable".

[0003] 2. Medical imaging noise and significant variability: Medical imaging presents challenges due to its complex structure and significant differences among patients. Under different imaging equipment, imaging angles, or anatomical variations, the model's region of interest can easily become confused, leading to unstable prediction results.

[0004] 3. Lack of structured anatomical knowledge integration: Most existing methods directly predict the overall image without establishing a salient region database based on medical structures, and cannot explain the basis of model judgment from the "anatomical principles". Summary of the Invention

[0005] To address the problems existing in the prior art, this invention provides a medical age and sex prediction framework and interpretation system based on anatomical structure segmentation.

[0006] This disclosure provides a medical age and sex prediction framework and interpretation system based on anatomical structure segmentation, including: The data preprocessing and standardization module is used to clean, enhance, and normalize the input medical images; The age and gender feature extraction and prediction module is used to build and train deep learning models to extract features from medical images processed by the data preprocessing and standardization module and predict age or gender labels. The model interpretation extraction module under structural constraints is used to generate a model decision interpretation aligned with the medical structure by taking the medical prior segmentation results as structural constraints, based on the intermediate layer feature map and prediction output obtained by the age and gender feature extraction and prediction module, and through a dual-view technical path. The reliability verification module is used to quantitatively verify the causal fidelity and rationality of the interpretation results of the model interpretation extraction module under structural constraints through a progressive intervention strategy.

[0007] Optionally, the data preprocessing and standardization module includes: The data cleaning and label alignment unit is used to remove samples with poor quality and verify the correspondence between images and labels; Histogram equalization unit is used to improve the contrast and visibility of anatomical structures in images; An image enhancement unit is used to improve the robustness of the model through at least one of the following strategies: random cropping, flipping, affine transformation, blurring, random erasure, and brightness / contrast adjustment. The image and label normalization unit is used to uniformly adjust the image size and normalize the pixel values, and to normalize the mean and variance of the age labels.

[0008] Optionally, the age and gender feature extraction and prediction module includes: The feature extraction unit uses a deep learning model to encode the features of the input image to obtain the extracted features. The prediction unit uses a fully connected layer to map the extracted features into predictions of age or gender. During deep learning model training, the gender prediction task uses the cross-entropy loss function, while the age prediction task uses the L1 loss function.

[0009] Optionally, the model interpretation and extraction module under structural constraints includes: The pixel interpretation extraction unit is used to generate pixel-level interpretations by calculating the gradient of the deep learning model output onto the feature map and weighting it with sensitivity to the feature saliency map. The structure-by-structure cumulative statistical unit is used to align pixel interpretation with physiological structural regions based on medical prior segmentation or unsupervised segmentation results, and to accumulate the contribution values ​​of pixels in each structural region that are above a set threshold. The structural perturbation interpretation extraction unit is used to generate a perturbation sample set by masking or replacing the target structural region, and extract the contribution weights of each structure in the local interpretable prediction model to the prediction results. The visualization and statistics unit is used to overlay pixel-level interpretations onto the original image in the form of a heatmap, and to summarize and analyze the structural interpretation results.

[0010] Optionally, the process of generating a perturbation sample set by masking or replacing the target structural region includes: At least one of the following: zero-pixel filling, mean replacement, noise filling, and blurring.

[0011] Optionally, the reliability verification module performs verification by defining a progressive intervention causal effect index, including: Define the Individual Intervention Causal Effect Index (ICE) to measure the difference in model predictions before and after applying structural intervention to individual samples; The ACE index of causal effect of group intervention is obtained based on the mean ICE of all samples in the group; The progressive intervention causal effect index PICE was obtained based on the average value of ACE under different intervention intensities. By comparing the PICE driven by explanation with that driven by randomization, the significance test was used to assess whether the difference between the two was significant, and to verify the causality and correlation between the explanation results and the model predictions at the pixel or structural level.

[0012] Optionally, the formula for calculating the individual intervention causal effect index ICE is as follows: , in, Indicates the first The original input image of each sample, Indicates that the first [condition] is applied to this sample. Images after intervention.

[0013] Optionally, the formula for calculating the Progressive Intervention Causality Index (PICE) is as follows: , in, Represents the total number of samples. Indicates the level of intervention.

[0014] Optionally, the medical prior segmentation results can be derived from any one of the following: manual annotation, regularized anatomical partitioning, automatic segmentation model, unsupervised segmentation algorithm, or weakly supervised structure extraction method.

[0015] Optionally, the system is applicable to at least one type of medical imaging, including X-ray images, CT images, cone-beam CT, and pathological slide images, and can be applied to at least one medical scenario, including age prediction, gender prediction, and related feature discovery.

[0016] The medical age and gender prediction framework and interpretation system based on anatomical structure segmentation provided by this invention achieves alignment between the interpretation and prior medical knowledge by imposing structural constraints on the model interpretation, thereby improving the quality and readability of the interpretation. Attached Figure Description

[0017] The above and other objects, features and advantages of this disclosure will become more apparent from the accompanying drawings, in which like reference numerals generally denote like parts.

[0018] Figure 1A principle block diagram of the medical age and sex prediction framework and interpretation system based on anatomical structure segmentation provided in the embodiments of this disclosure; Figure 2 This is a schematic diagram of the structural perturbation algorithm provided in an embodiment of this disclosure. Detailed Implementation

[0019] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0020] It should be understood that the following specific examples illustrate the implementation of this disclosure, and those skilled in the art can easily understand other advantages and effects of this disclosure from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. This disclosure can also be implemented or applied through other different specific implementation methods, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this disclosure. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0021] It should be noted that various aspects of embodiments within the scope of the appended claims are described below. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this disclosure, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using structures and / or functionalities other than one or more of the aspects set forth herein.

[0022] It should also be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this disclosure. The illustrations only show the components related to this disclosure and are not drawn according to the number, shape and size of the components in actual implementation. In actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0023] Furthermore, specific details are provided in the following description to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the described aspects can be practiced without these specific details.

[0024] The objective of this embodiment is to provide a medical age and sex prediction framework and interpretation system based on anatomical structure segmentation, addressing the issue that existing medical imaging AIs, while offering acceptable prediction performance, suffer from insufficient interpretability. This embodiment provides a complete prediction-interpretation-verification workflow. Based on pattern recognition, it imposes structural constraints on the model interpretation, aligning the interpretation with prior medical knowledge, thereby improving the quality and readability of the interpretation. It also provides perspectives and methods for discovering features of interest (such as developmental and aging features). This embodiment combines medical anatomical structure segmentation with interpretable model prediction, indicating both the direction of model optimization and providing medical interpretive evidence that can be understood and verified by doctors.

[0025] like Figure 1 As shown, this embodiment discloses a medical age and sex prediction framework and interpretation system based on anatomical structure segmentation, specifically including: The data preprocessing and standardization module provides a high-quality data foundation for model predictions and mainly includes: Data cleaning and label alignment unit: Removes samples with missing labels, abnormal labels, or severely degraded image quality; verifies the one-to-one correspondence between image data and label data to ensure that each input image has unique and correct label information; manages gender labels and age labels independently to ensure that the label format is consistent with the model output structure.

[0026] Histogram equalization unit: By redistributing the distribution of pixel gray values, the gray levels of the image are made more uniform globally, thereby improving the contrast and visibility of tooth contours, bone tissue and fine texture structures.

[0027] Image Augmentation Unit: To further improve the robustness of the model under different shooting conditions and individual differences, and to prevent overfitting of the deep model during training, various image augmentation strategies are introduced: Random cropping and scaling are used to simulate changes in shooting angle and imaging range; Horizontal or vertical flipping is used to enhance the model's ability to generalize to symmetrical structures; Affine transformation is used to simulate slight attitude changes; Blurring, random erasure, and brightness or contrast adjustments are used to improve the model's adaptability to noise and local occlusion.

[0028] Image and label normalization unit: The image size is uniformly adjusted to a preset size, preferably 224×224 or 512×512 pixels, and the pixel values ​​are normalized to the range of [-1,1] to adapt to the network structure and stabilize training; the age label is normalized by mean-variance and converted to a standard normal distribution to stabilize training.

[0029] Age and Gender Feature Extraction and Prediction Module: This module is used to extract patient features from images and predict label information, mainly including: Model Building and Training: Construct a deep learning model for age and gender prediction. The model consists of two parts: a feature extraction unit and a prediction unit. Feature extraction unit: Encodes features from the input dental X-ray image and automatically extracts image features related to age and gender. Feature extraction structures based on convolutional neural networks (such as ResNet and EfficientNet) or attention mechanisms (such as ViT) can be selected.

[0030] The feature extraction unit can be a convolutional neural network or a visual Transformer structure, or other deep learning models with automatic feature extraction capabilities, including but not limited to multi-scale convolutional networks, hybrid CNN-Transformer structures, and other image analysis models. Any implementation that can output intermediate features and support subsequent interpretation, extraction, and intervention analysis is considered an equivalent implementation of this embodiment.

[0031] Prediction Unit: A fully connected layer structure is used to connect to the output of the feature extraction unit, and is used to map the extracted features to the prediction result. During the model training phase, the cross-entropy loss function was used for training the gender prediction task, and the L1 loss function was used for training the age prediction task.

[0032] Training and Evaluation: For the gender classification task, accuracy, precision, recall, F1 score, and AUC were used for evaluation; for the age regression task, prediction error (E) and absolute error (AE) were extracted for statistical analysis, including indicators such as mean, variance, and quartiles.

[0033] The model interpretation extraction module under structural constraints uses medical prior segmentation as a structural constraint to align the interpretation and structural information of model decisions. This module mainly includes: Pixel Interpretation Extraction Unit: This unit generates pixel-level interpretations of the model using a feature saliency map sensitivity weighting method. Specifically, it calculates the gradient information of the model output on the intermediate layer feature maps and performs channel-level aggregation on the gradient results to obtain the contribution weights of each feature channel; subsequently, it uses these contribution weights to perform weighted superposition of the corresponding feature maps.

[0034] Pixel-level interpretation extraction methods are not limited to saliency weighting based on gradient sensitivity; they can also be other feature attribution algorithms or attention mapping methods, as long as they can generate spatial interpretation results related to model predictions and support subsequent structural aggregation and verification.

[0035] Structural cumulative statistical unit: Based on medical prior segmentation or unsupervised segmentation results, the attention map extracted from pixel interpretation is spatially aligned with each physiological structural region. The pixel contribution value of each structural region is cumulatively statistically analyzed, and low contribution pixels are filtered out by a significance threshold. Only pixels with a contribution value higher than the threshold are aggregated to eliminate noise and reduce the impact of structural area differences on the statistical results.

[0036] The structural perturbation interpretation and extraction unit: Based on the medical prior segmentation results of the input dental X-ray image, it acquires multiple physiological structural regions, designs structural masking strategies (such as mean, all-zero pixels, and noise), and randomly perturbs the structural regions to generate a set of structural perturbation samples. These neighborhood samples are input into a trained age and gender prediction model to obtain the corresponding prediction output, constructing a locally interpretable prediction model. The weight coefficients corresponding to each structure in the local model are extracted to obtain the contribution of different physiological structures to the model's age or gender prediction results, thereby achieving structural-level interpretation and extraction. In the process of structural perturbation interpretation and causality testing, the masking or perturbation methods used are not limited to all-zero, mean replacement, or noise filling; other equivalent structural destruction methods such as fuzzing can also be used, as long as they can progressively weaken the intensity of the target structural information. Figure 2 As shown.

[0037] Visualization and Statistical Unit: The pixel-level interpretation results are overlaid on the original dental radiographs in the form of heatmaps to provide an intuitive presentation of the areas of interest of the model; the cumulative statistical results of each structure and the interpretation results of structural perturbations are summarized and analyzed at the sample level.

[0038] Medical prior structure segmentation results can originate from manual annotation or regularized anatomical partitions, or from automatic segmentation models, unsupervised segmentation algorithms, or weakly supervised structure extraction methods. The specific number, morphology, and acquisition method of the structures do not constitute a limitation on this embodiment.

[0039] The reliability verification module is used to verify the rationality of the extracted structural interpretations, providing reliability for applications such as feature discovery. It mainly includes: Define the Progressive Intervention Causal Effect (PICE) metric: The aforementioned structural perturbation algorithm constitutes a progressive intervention process from weak to strong on physiological structural regions in images. Based on this intervention process, a Progressive Intervention Causal Effect (PICE) index is defined to quantify the degree of causal influence of the outcome on model decision-making. Its definition and parameter meanings are as follows.

[0040] set up This represents the original medical image sample input. This represents the completed prediction model; the model output is... Let be the predicted value for age or the predicted probability for gender. Denotes a set of intervention regions ordered by explanatory importance, where Intervention intensity varies Incremental. Through image analysis In the region By applying masking or perturbation operations, we obtain counterfactual samples after intervention: ,in, This indicates the intervention calculation applied to the specified structural region.

[0041] At the individual level, the Individual Causal Effect (ICE) is calculated to measure the effect of a single sample on the application of the first causal factor. The magnitude of change in model prediction output before and after intervention: , in, Indicates the first The original input image of each sample, Indicates that the first [condition] is applied to this sample. Images after intervention.

[0042] At the group level, the average causal effect (ACE) of the intervention is calculated to measure the effect at a given intervention intensity. The average impact of the intervention on the prediction results of the overall sample set is defined as follows: , in, This represents the total number of samples.

[0043] Furthermore, considering the intervention process as a continuous sequence from weak to strong, a progressive intervention causal effect index, PICE, is defined to cumulatively average the group causal effects under different intervention intensities. Its definition is: , Or equivalently expressed as: , in, The PICE value indicates the level of intervention. A higher PICE value indicates a more significant impact of the selected explanatory region on the model's predictions as it is progressively destroyed, thus demonstrating a stronger causal correlation and higher explanatory reliability between the explanatory result and the model's decision.

[0044] PICE-based analysis and verification: The reliability index of the pixel or structural region selected based on the interpretation results is compared with that of the randomly selected region under the same intervention conditions, and the difference between the two is evaluated by significance test; when the index corresponding to the interpretation region is significantly higher than that of the random region, it is determined that the interpretation results have strong causality and correlation with the model prediction at the pixel level or structural level.

[0045] This embodiment uses age and gender prediction in dental imaging in oral medicine as an example, but its technical framework is not limited to dental imaging applications. This method is also applicable to prediction and interpretation tasks in various medical imaging types, including but not limited to X-ray images, CT images, cone-beam CT (CBCT), and pathological slide images. It enables object recognition in different medical images. In all of the aforementioned image types, this embodiment can achieve structured interpretation and reliability verification of the model's decision-making basis through structural constraint interpretation and progressive intervention causal verification. It is applicable to various medical application scenarios such as age prediction, gender prediction, and related feature discovery, thereby facilitating the construction of smart healthcare.

[0046] This embodiment has the following effects: 1. Through a dual-perspective interpretation mechanism based on structural constraints, complementary extraction of pixel-level and structural-level interpretations aligned with medical priors is achieved: Traditional heatmap interpretation methods focus on the pixel level, resulting in fragmented and low-level interpretations that are difficult for medical professionals to understand. This approach aligns the interpretation with medical priors by accumulating structural saliency during the interpretation phase, improving readability and directionality. Simultaneously, it introduces two perspectives: pixel-level structural accumulation interpretation and structural perturbation interpretation. These perspectives characterize developmental features from two levels: "Where did the model see?" and "Which structures did the model's decisions depend on?" This ensures that the interpretation results possess both spatial accuracy and adhere to the logical understanding of medical structural hierarchies.

[0047] 2. Through gradual intervention and causal testing, quantitatively verify the decision fidelity and physiological rationality of the explained results: To address potential fidelity issues inherent in the explanation methods themselves, this embodiment further introduces an explanation reliability verification module. By constructing a sequence of structural or pixel interventions from weak to strong, gradually disrupting the explanation region and observing changes in model predictions, it is possible to directly verify whether the extracted region truly participates in the model's decision-making process. Based on this, the intervention causal effect (ICE) is defined at the individual level, and the average causal effect (ACE) is constructed at the group level. The PICE index is obtained through a progressive integral form. This index transforms the reliability of the explanation results into a comparable quantitative value, enabling the explanation to move beyond mere visual judgment and possess clear causal dependence on decision-making, thereby significantly improving the credibility of the explanation conclusions in research and application scenarios.

[0048] 3. The overall framework forms a closed loop of "prediction-interpretation-validation," improving the model's usability in medical scenarios: Through the modular design described above, this embodiment achieves a complete closed loop from model prediction and structured interpretation to causal reliability verification. The model not only provides predictions for age and gender but also clearly identifies the imaging regions and physiological structures upon which these predictions depend, and verifies, through quantitative indicators, whether these interpretations truly influence the decision-making process. This outcome-driven, structure-constrained, and causal verification-supported explanatory framework makes model behavior more transparent, reduces the risk of a black box approach, and provides a reliable foundation for medical feature discovery, clinical auxiliary analysis, and model safety assessment.

[0049] It should also be noted that in the systems and methods of this disclosure, the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered as equivalent solutions to this disclosure.

[0050] Various changes, substitutions, and modifications can be made to the technology described herein without departing from the teachings defined by the appended claims. Furthermore, the scope of the claims of this disclosure is not limited to the specific aspects of the processes, machines, manufactures, events, means, methods, and actions described above. Currently existing or later-developed processes, machines, manufactures, events, means, methods, or actions that perform substantially the same function or achieve substantially the same result as the corresponding aspects described herein can be utilized. Therefore, the appended claims include such processes, machines, manufactures, events, means, methods, or actions within their scope.

[0051] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of this disclosure. Therefore, this disclosure is not intended to be limited to the aspects shown herein, but rather to be carried out within the widest scope consistent with the principles and novel features disclosed herein.

[0052] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this disclosure to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations therein.

Claims

1. A medical age and sex prediction framework and interpretation system based on anatomical structure segmentation, characterized in that, include: The data preprocessing and standardization module is used to clean, enhance, and normalize the input medical images; The age and gender feature extraction and prediction module is used to build and train deep learning models to extract features from medical images processed by the data preprocessing and standardization module and predict age or gender labels. The model interpretation extraction module under structural constraints is used to generate a model decision interpretation aligned with the medical structure by taking the medical prior segmentation results as structural constraints, based on the intermediate layer feature map and prediction output obtained by the age and gender feature extraction and prediction module, and through a dual-view technical path. The reliability verification module is used to quantitatively verify the causal fidelity and rationality of the interpretation results of the model interpretation extraction module under structural constraints through a progressive intervention strategy.

2. The medical age and sex prediction framework and interpretation system based on anatomical structure segmentation according to claim 1, characterized in that, The data preprocessing and standardization module includes: The data cleaning and label alignment unit is used to remove samples with poor quality and verify the correspondence between images and labels; Histogram equalization unit is used to improve the contrast and visibility of anatomical structures in images; An image enhancement unit is used to improve the robustness of the model through at least one of the following strategies: random cropping, flipping, affine transformation, blurring, random erasure, and brightness / contrast adjustment. The image and label normalization unit is used to uniformly adjust the image size and normalize the pixel values, and to normalize the mean and variance of the age labels.

3. The medical age and sex prediction framework and interpretation system based on anatomical structure segmentation according to claim 1, characterized in that, The age and gender feature extraction and prediction module includes: The feature extraction unit uses a deep learning model to encode the features of the input image to obtain the extracted features. The prediction unit uses a fully connected layer to map the extracted features into predictions of age or gender. During deep learning model training, the gender prediction task uses the cross-entropy loss function, while the age prediction task uses the L1 loss function.

4. The medical age and sex prediction framework and interpretation system based on anatomical structure segmentation according to claim 1, characterized in that, The model interpretation and extraction module under structural constraints includes: The pixel interpretation extraction unit is used to generate pixel-level interpretations by calculating the gradient of the deep learning model output onto the feature map and weighting it with sensitivity to the feature saliency map. The structure-by-structure cumulative statistical unit is used to align pixel interpretation with physiological structural regions based on medical prior segmentation or unsupervised segmentation results, and to accumulate the contribution values ​​of pixels in each structural region that are above a set threshold. The structural perturbation interpretation extraction unit is used to generate a perturbation sample set by masking or replacing the target structural region, and extract the contribution weights of each structure in the local interpretable prediction model to the prediction results. The visualization and statistics unit is used to overlay pixel-level interpretations onto the original image in the form of a heatmap, and to summarize and analyze the structural interpretation results.

5. The medical age and sex prediction framework and interpretation system based on anatomical structure segmentation according to claim 4, characterized in that, The masking or replacement of the target structural region generates a perturbation sample set, including: At least one of the following: zero-pixel filling, mean replacement, noise filling, and blurring.

6. The medical age and sex prediction framework and interpretation system based on anatomical structure segmentation according to claim 1, characterized in that, The reliability verification module for the explanation is verified by defining a progressive intervention causal effect index, including: Define the Individual Intervention Causal Effect Index (ICE) to measure the difference in model predictions before and after applying structural intervention to individual samples; The ACE index of causal effect of group intervention is obtained based on the mean ICE of all samples in the group; The progressive intervention causal effect index PICE was obtained based on the average value of ACE under different intervention intensities. By comparing the PICE driven by explanation with that driven by randomization, the significance test was used to assess whether the difference between the two was significant, and to verify the causality and correlation between the explanation results and the model predictions at the pixel or structural level.

7. The medical age and sex prediction framework and interpretation system based on anatomical structure segmentation according to claim 6, characterized in that, The formula for calculating the Individual Intervention Causal Effect Index (ICE) is as follows: ， in, Indicates the first The original input image of each sample, Indicates that the first [condition] is applied to this sample. Images after intervention.

8. The medical age and sex prediction framework and interpretation system based on anatomical structure segmentation according to claim 7, characterized in that, The formula for calculating the Progressive Intervention Causality Index (PICE) is as follows: ， in, Represents the total number of samples. Indicates the level of intervention.

9. The medical age and sex prediction framework and interpretation system based on anatomical structure segmentation according to claim 1, characterized in that, The medical prior segmentation results are derived from any one of the following: manual annotation, regularized anatomical partitioning, automatic segmentation model, unsupervised segmentation algorithm, and weakly supervised structure extraction method.

10. The medical age and sex prediction framework and interpretation system based on anatomical structure segmentation according to claim 1, characterized in that, The system is applicable to at least one type of medical imaging, including X-ray images, CT images, cone-beam CT, and pathological slide images, and can be applied to at least one medical scenario, such as age prediction, gender prediction, and related feature discovery.

Images