Computer program product, device and apparatus for deviated nasal septum recognition

Abstract

The present disclosure provides a computer program product, a nasal septum deviation identification device and equipment, and relates to the technical field of artificial intelligence. In some embodiments of the present disclosure, a nasal CBCT image photographed by a cone beam CT device is obtained; the nasal CBCT image is input into a nasal septum detection model, target detection is performed on the nasal CBCT image, and a nasal septum region image corresponding to the nasal CBCT image is obtained; the nasal septum region image is input into a nasal septum deviation classification model, nasal septum deviation classification is performed on the nasal septum region image, and a nasal septum deviation result is obtained; the nasal septum region image and the nasal septum deviation result are sent to a film reading terminal, so that the film reading terminal displays the nasal septum region image and the nasal septum deviation result; by means of the nasal septum detection model and the nasal septum deviation classification model, the present disclosure can automatically read the three-dimensional CBCT image, and compared with the existing artificial reading method, the reading efficiency of the nasal CBCT image is improved, and the reading accuracy is improved.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of intelligent medical treatment, and in particular to a computer program product, a nasal septum deviation identification device and equipment. BACKGROUND

[0002] Nasal septum deviation is a global common nasal disease, which can cause nasal obstruction, sinusitis, nosebleed, headache and other symptoms in patients, and seriously affects the physical health and quality of life of patients. According to statistics, the incidence of nasal septum deviation is 86.6% according to the analysis of nasal CT scan data of adults. Studies have found that the incidence of nasal septum deviation increases with age, so it is crucial to conduct early screening of the nasal septum deviation condition of patients.

[0003] Imaging examination is an important means of nasal septum deviation screening. Through imaging examinations such as X-ray examination, computed tomography (CT) and magnetic resonance imaging (MRI), the structure and deviation of the nasal septum can be more accurately observed. With the popularization of three-dimensional imaging technologies such as CBCT (Cone Beam Computed Tomography) and CT, the traditional manual film reading mode faces the dual challenges of efficiency bottleneck and low diagnostic consistency, and clinical doctors urgently need automated tools to assist doctors in quickly screening lesions.

[0004] At present, manual film reading is adopted, and the film reading efficiency is low; the results are affected by human interference factors, and the film reading accuracy is low. SUMMARY

[0005] The present disclosure provides a computer program product, a nasal septum deviation identification device and equipment to at least solve the problems of low film reading efficiency and low film reading accuracy.

[0006] The technical solutions of the present disclosure are as follows:

[0007] The present disclosure provides a computer program product, which includes computer programs / instructions that are executed by a processor to implement the following steps:

[0008] Obtain a nasal CBCT image taken by a cone beam CT device;

[0009] Input the nasal CBCT image into a nasal septum detection model, perform target detection on the nasal CBCT image, and obtain a nasal septum region image corresponding to the nasal CBCT image;

[0010] Input the nasal septum region image into a nasal septum deviation classification model, perform nasal septum deviation classification on the nasal septum region image, and obtain a nasal septum deviation result;

[0011] The nasal septum area image and the nasal septum deviation result are sent to a film reviewing terminal, so that the film reviewing terminal displays the nasal septum area image and the nasal septum deviation result.

[0012] Optionally, the nasal CBCT image taken by the cone beam CT device comprises:

[0013] The nasal CBCT image uploaded by the film reviewing terminal is received, wherein the nasal CBCT image is an image uploaded by the film reviewing terminal in response to an image uploading operation after receiving the nasal CBCT image sent by the cone beam CT device.

[0014] Optionally, the nasal septum detection model comprises a preprocessing layer, a backbone layer, a bottleneck layer, a detection layer and an output layer, and the nasal CBCT image is input into the nasal septum detection model to perform target detection on the nasal CBCT image to obtain a nasal septum area image corresponding to the nasal CBCT image, comprising:

[0015] The nasal CBCT image is input into the preprocessing layer inside the nasal septum detection model to perform image dimension adjustment to obtain a preprocessed image;

[0016] The preprocessed image is input into the backbone layer to perform feature extraction to obtain first, second and third granularity features;

[0017] The first, second and third granularity features are input into the bottleneck layer to obtain first, second and third aggregated features after multi-scale aggregation;

[0018] The first, second and third aggregated features are input into the detection layer to perform target detection using first, second and third detection heads to obtain first, second and third detection boxes, and first, second and third confidence degrees corresponding to the first, second and third detection boxes, respectively;

[0019] The first, second and third detection boxes are input into the output layer, the first, second and third detection boxes are input into a non-maximum suppression function to sort the detected candidate boxes according to the confidence degrees, a target box is selected from the sorted candidate boxes according to the intersection over union of the candidate boxes, and the target box is used to crop the nasal septum area image from the nasal CBCT image.

[0020] Optionally, the backbone layer comprises a first convolutional layer module, a first C3K2 feature extraction module, a SPPF feature extraction module, and a C2PSA feature enhancement module; the inputting the preprocessed image into the backbone layer for feature extraction to obtain a first granularity feature, a second granularity feature, and a third granularity feature comprises:

[0021] The preprocessed image is input into the backbone layer, sequentially subjected to convolution operation by the first convolutional layer module, feature extraction by the first C3K2 feature extraction module, feature extraction by the SPPF feature extraction module, and feature enhancement by the C2PSA feature enhancement module, and then the first granularity feature, the second granularity feature, and the third granularity feature are output.

[0022] Optionally, the bottleneck layer comprises an upsampling module, a splicing module, a second C3K2 feature extraction module, and a second convolutional layer module; the inputting the first granularity feature, the second granularity feature, and the third granularity feature into the bottleneck layer for multi-scale aggregation to obtain a first aggregated feature, a second aggregated feature, and a third aggregated feature comprises:

[0023] The first granularity feature, the second granularity feature, and the third granularity feature are input into the bottleneck layer, subjected to upsampling by the upsampling module, feature splicing by the splicing module, convolution operation by the second convolutional layer module, and feature extraction by the second C3K2 feature extraction module, and then the first aggregated feature, the second aggregated feature, and the third aggregated feature are obtained.

[0024] Optionally, the deviated nasal septum classification model comprises an input layer, a first convolutional layer, a bottleneck layer module, a second convolutional layer, an average pooling layer, and a fully connected layer; the deviated nasal septum result comprises a deviated nasal septum probability; the inputting the nasal septum region image into the deviated nasal septum classification model to classify the nasal septum region image to obtain a deviated nasal septum result comprises:

[0025] Within the deviated nasal septum classification model, the nasal septum region image is input into the input layer for dimension adjustment to obtain a standardized region image;

[0026] The standardized region image is input into the first convolutional layer for convolution operation to obtain a preliminary feature;

[0027] The preliminary feature is input into the bottleneck layer module to learn spatial features to obtain spatial features;

[0028] The spatial features are input into the second convolutional layer for convolution operation to obtain a feature map;

[0029] inputting the feature map into the average pooling layer to compress into a single vector to obtain a feature vector;

[0030] inputting the feature vector into the full connection layer to obtain the deviation of the nasal septum probability.

[0031] Optionally, the bottleneck layer module comprises a plurality of bottleneck sub-layers; each bottleneck sub-layer comprises: a first convolutional sub-layer, a first batch normalization layer, a first activation function layer, a first depth separable convolutional layer, a second batch normalization layer, a second activation function layer, a second convolutional sub-layer and a third batch normalization layer; the step of inputting the preliminary feature into the bottleneck layer module to learn spatial features to obtain spatial features comprises:

[0032] the preliminary feature sequentially passes through the first convolutional sub-layer, the first batch normalization layer, the first activation function layer, the first depth separable convolutional layer, the second batch normalization layer, the second activation function layer, the second convolutional sub-layer and the third batch normalization layer of each bottleneck sub-layer to learn spatial features to obtain the spatial features.

[0033] Optionally, the nasal septum region image and the deviation of the nasal septum result are sent to a film review terminal to display the nasal septum region image and the deviation of the nasal septum result on the film review terminal, comprising:

[0034] in the case of the deviation of the nasal septum result, the nasal CBCT image, the nasal septum region image and the deviation of the nasal septum result are sent to the film review terminal to display the nasal CBCT image, the nasal septum region image and the deviation of the nasal septum result on the film review terminal according to a set display format.

[0035] The present disclosure also provides a nasal septum deviation identification device, comprising:

[0036] an acquisition module configured to acquire a nasal CBCT image taken by a cone beam CT device;

[0037] a region of interest detection module configured to input the nasal CBCT image into a nasal septum detection model to perform target detection on the nasal CBCT image to obtain a nasal septum region image corresponding to the nasal CBCT image;

[0038] a nasal septum deviation classification module configured to input the nasal septum region image into a nasal septum deviation classification model to perform nasal septum deviation classification on the nasal septum region image to obtain a deviation of the nasal septum result;

[0039] A result display module is configured to send the nasal septum region image and the nasal septum deviation result to a film review terminal, so that the film review terminal displays the nasal septum region image and the nasal septum deviation result.

[0040] The embodiments of the present disclosure also provide an electronic device, comprising:

[0041] a processor;

[0042] a memory for storing processor-executable instructions;

[0043] The processor is configured to execute the instructions to implement the steps in the above method.

[0044] The technical solutions provided by the embodiments of the present disclosure at least bring the following beneficial effects:

[0045] In some embodiments of the present disclosure, a nasal CBCT image photographed by a cone beam CT device is acquired; the nasal CBCT image is input into a nasal septum detection model, target detection is performed on the nasal CBCT image, and a nasal septum region image corresponding to the nasal CBCT image is obtained; the nasal septum region image is input into a nasal septum deviation classification model, nasal septum deviation classification is performed on the nasal septum region image, and a nasal septum deviation result is obtained; the nasal septum region image and the nasal septum deviation result are sent to a film review terminal, so that the film review terminal displays the nasal septum region image and the nasal septum deviation result; the present disclosure can automatically review the three-dimensional CBCT image with the help of the nasal septum detection model and the nasal septum deviation classification model, and the review efficiency of the nasal CBCT image is improved compared with the existing manual review method; the automatic review is performed in an artificial intelligence manner, human interference factors are reduced, and the review accuracy is improved.

[0046] It should be understood that the foregoing general description and the following detailed description are only exemplary and explanatory, and cannot limit the present disclosure. BRIEF DESCRIPTION OF DRAWINGS

[0047] The accompanying drawings incorporated in and forming a part of the specification illustrate embodiments consistent with the present disclosure and serve to explain the principles of the present disclosure, and do not constitute an undue limitation on the present disclosure.

[0048] Figure 1 A flowchart of a nasal septum deviation identification method provided for an exemplary embodiment of the present disclosure;

[0049] Figure 2 A structure diagram of a nasal septum detection model provided for an embodiment of the present disclosure;

[0050] Figure 3 An architecture diagram of a YOLOv11 model provided for an exemplary embodiment of the present disclosure;

[0051] Figure 4 A schematic diagram of a C2PSA feature enhancement module provided for an exemplary embodiment of the present disclosure;

[0052] Figure 5 A structural schematic diagram of a PSA provided for an exemplary embodiment of the present disclosure;

[0053] Figure 6 A schematic diagram of a SPPF feature extraction module provided for an exemplary embodiment of the present disclosure;

[0054] Figure 7 A structural schematic diagram of a deviated nasal septum classification model provided for an exemplary embodiment of the present disclosure;

[0055] Figure 8 An architecture diagram of a MobileNet model provided for an exemplary embodiment of the present disclosure;

[0056] Figure 9 A structural schematic diagram of a Bottleneck provided for an embodiment of the present disclosure;

[0057] Figure 10 A schematic diagram of a display interface provided for an exemplary embodiment of the present disclosure;

[0058] Figure 11 A schematic diagram of a deviated nasal septum recognition system provided for an exemplary embodiment of the present disclosure;

[0059] Figure 12 A structural schematic diagram of a deviated nasal septum recognition device provided for an exemplary embodiment of the present disclosure;

[0060] Figure 13 A structural schematic diagram of an electronic device provided for an exemplary embodiment of the present disclosure. DETAILED DESCRIPTION

[0061] In order to make the ordinary person in the art better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be described clearly and completely below in conjunction with the drawings.

[0062] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily have to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The implementation described in the following exemplary embodiments does not represent all implementations consistent with the present disclosure. Rather, they are only examples of devices and methods consistent with some aspects of the present disclosure.

[0063] It should be noted that the user information involved in the present disclosure includes but is not limited to user equipment information and user personal information; the collection, storage, use, processing, transmission, provision and disclosure of the user information in the present disclosure all comply with the provisions of relevant laws and regulations and do not violate public order and good customs.

[0064] At present, the CBCT image is mainly read by artificial method, and the reading efficiency is low; the result is disturbed by human factors, and the reading accuracy is low.

[0065] In order to solve the above technical problems, in some embodiments of the present disclosure, a nasal CBCT image photographed by a cone beam CT device is obtained; the nasal CBCT image is input into a nasal septum detection model, the nasal CBCT image is subjected to target detection, and a nasal septum region image corresponding to the nasal CBCT image is obtained; the nasal septum region image is input into a nasal septum deviation classification model, the nasal septum region image is subjected to nasal septum deviation classification, and a nasal septum deviation result is obtained; the nasal septum region image and the nasal septum deviation result are sent to a reading terminal, so that the reading terminal displays the nasal septum region image and the nasal septum deviation result; by means of the nasal septum detection model and the nasal septum deviation classification model, the present disclosure can automatically read the three-dimensional CBCT image, which improves the reading efficiency of the nasal CBCT image compared with the existing artificial reading method; the artificial intelligence method is used for automatic reading, which reduces human interference factors and improves the reading accuracy.

[0066] The technical solutions provided by the embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0067] Figure 1 A flowchart of a nasal septum deviation identification method provided by an exemplary embodiment of the present disclosure is shown. The present disclosure also provides a computer program product including computer programs / instructions, which, when executed by a processor, implement the steps shown in Figure 1 The steps shown in the figure are as follows:

[0068] S101: Obtain a nasal CBCT image photographed by a cone beam CT device;

[0069] S102: Input the nasal CBCT image into a nasal septum detection model, perform target detection on the nasal CBCT image, and obtain a nasal septum region image corresponding to the nasal CBCT image;

[0070] S103: Input the nasal septum region image into a nasal septum deviation classification model, perform nasal septum deviation classification on the nasal septum region image, and obtain a nasal septum deviation result;

[0071] S104: send the nasal septum region image and the nasal septum deviation result to the film reviewing terminal, so that the film reviewing terminal displays the nasal septum region image and the nasal septum deviation result.

[0072] It should be noted that the present disclosure does not limit the type of computer program product, which can be a software product or a hardware product. The software product is a product storing the program code corresponding to the present disclosure, and the hardware product is, for example, a storage medium or an electronic device storing the program code corresponding to the present disclosure.

[0073] Another aspect of the present disclosure also provides a nasal septum deviation identification method, which includes the above steps S101-S104.

[0074] In some embodiments of the present disclosure, a nasal CBCT image taken by a cone beam CT device is obtained. One implementable way is to receive a nasal CBCT image uploaded by a film reviewing terminal. The film reviewing terminal receives the nasal CBCT image sent by the cone beam CT device in response to an image uploading operation. It should be noted that the film reviewing terminal can be a terminal device used in a hospital CT reporting room.

[0075] For example, the nasal part of a patient is photographed by using a cone beam CT device to obtain a nasal CBCT image; the cone beam CT device uploads the nasal CBCT image to a film reviewing terminal; the film reviewing terminal receives the nasal CBCT image; the application interface of the film reviewing terminal includes an uploading control; the film reviewing terminal uploads the nasal CBCT image to a server in response to a triggering operation of the uploading control; and the server receives the nasal CBCT image.

[0076] Before using the nasal septum detection model, the nasal septum detection model needs to be trained first. The training process of the nasal septum detection model is described below.

[0077] The training process of the nasal septum detection model is as follows: first, a large number of nasal CBCT image samples are collected, and different subject objects in various scenes are covered as much as possible to improve the sample coverage. Then, the nasal septum region on the training sample can be labeled to obtain the actual distribution of the nasal septum region on the nasal CBCT image sample. Then, the labeled training sample can be input into a neural network model. Inside the neural network model, the training sample can be feature extracted and calculated according to the model parameters, and the nasal septum region detection result is output by the output layer of the neural network model. Then, the loss function layer of the neural network model can calculate the loss function according to the difference between the output layer output nasal septum region and the real nasal septum region on the training sample. If the loss function does not meet the set requirements, the model parameters can be adjusted for further iteration training. When the loss function of the neural network model meets the set requirements, the training of the nasal septum detection model is completed.

[0078] It should be noted that the nasal septum detection model includes a preprocessing layer, a backbone layer, a bottleneck layer, a detection layer, and an output layer.

[0079] In some embodiments of the present disclosure, the nasal CBCT image is input into the nasal septum detection model, the target detection of the nasal CBCT image is performed, and the nasal septum region image corresponding to the nasal CBCT image is obtained.

[0080] Figure 2 A structure diagram of a nasal septum detection model provided by an embodiment of the present disclosure is shown in FIG. 1. As shown in FIG. 1, the nasal septum detection model includes a preprocessing layer, a backbone layer, a bottleneck layer, a detection layer, and an output layer. Figure 2As shown, the nasal septum detection model comprises a preprocessing layer, a backbone layer, a bottleneck layer, a detection layer and an output layer. Inside the nasal septum detection model, the nasal CBCT image is input into the preprocessing layer for image dimension adjustment to obtain a preprocessed image; the preprocessed image is input into the backbone layer for feature extraction to obtain first, second and third granularity features; the first, second and third granularity features are input into the bottleneck layer, and after multi-scale aggregation, first, second and third aggregated features are obtained; the first, second and third aggregated features are input into the detection layer, and the first, second and third detection heads are used for target detection to obtain first, second and third detection boxes, and first, second and third confidence scores corresponding to the first, second and third detection boxes, respectively; the first, second and third detection boxes are input into the output layer, and the first, second and third detection boxes are input into a non-maximum suppression function to sort the detected candidate boxes according to the confidence scores; a target box is selected from the sorted candidate boxes according to the intersection-over-union of the candidate boxes; and the target box is used to crop a nasal septum region image from the nasal CBCT image. Wherein, after the nasal septum detection model receives an input image, the image features are extracted through the processing of network structures such as convolution layers and pooling layers, and the bounding box containing the nasal septum position information and the confidence score are output. The ROI region of the nasal septum (i.e. the nasal septum region image) is labeled on the original image in the form of a bounding box, and the corresponding confidence score is output, which represents the reliability of the detection result.

[0081] In the above embodiment, the backbone layer comprises a first convolution layer module, a first C3K2 feature extraction module, an SPPF feature extraction module and a C2PSA feature enhancement module; the preprocessed image is input into the backbone layer for feature extraction to obtain first, second and third granularity features. One realizable way is that the preprocessed image is input into the backbone layer, and after the convolution operation of the first convolution layer module, the feature extraction of the first C3K2 feature extraction module, the feature extraction of the SPPF feature extraction module and the feature enhancement of the C2PSA feature enhancement module are sequentially performed, the first, second and third granularity features are output.

[0082] In the above embodiment, the bottleneck layer comprises: an upsampling module, a splicing module, a second C3K2 feature extraction module and a second convolutional layer module; the first granularity feature, the second granularity feature and the third granularity feature are input into the bottleneck layer, and after multi-scale aggregation, the first aggregated feature, the second aggregated feature and the third aggregated feature are obtained, comprising: the first granularity feature, the second granularity feature and the third granularity feature are input into the bottleneck layer, and after upsampling by the upsampling module, feature splicing by the splicing module, convolution operation by the second convolutional layer module and feature extraction by the second C3K2 feature extraction module, the first aggregated feature, the second aggregated feature and the third aggregated feature are obtained.

[0083] In an optional embodiment, the nasal septum detection model is a YOLOv11 model. The YOLOv11 algorithm is an advanced real-time target detection system, which is optimized to have the characteristics of fast speed and high precision, and is suitable for processing complex medical image data. The YOLOv11 model directly predicts the bounding box and class probability from the complete image through single network forward propagation, which greatly improves the detection efficiency compared with traditional R-CNN methods. The model training uses a labeled CBCT image dataset, each image accurately labels the position of the nasal septum, ensuring that the model can accurately learn the features of the nasal septum, so as to accurately detect the ROI in actual application.

[0084] Figure 3 An architecture diagram of a YOLOv11 model is provided for the exemplary embodiments of the present disclosure. As shown in Figure 3 The YOLOv11 model comprises a backbone layer, a bottleneck layer and a detection layer. As shown in Figure 3 First, the nasal CBCT image is adjusted to 3x640x640 and input into the backbone layer; after convolution operation by the first convolutional layer module, feature extraction by the first C3K2 feature extraction module, feature extraction by the SPPF feature extraction module and feature enhancement by the C2PSA feature enhancement module, three kinds of granularity image features are output: the first granularity feature I1∈R 64×80×80 , the second granularity feature I2∈R 128×40×40 and the third granularity feature I3∈R 256×20×20Among them, C3K2 feature extraction is an important feature extraction component in YOLOv11 model, which is based on the improved design of traditional C3 module. It provides stronger feature extraction capability by combining variable convolution kernel and channel separation strategy. SPPF feature extraction is a spatial pyramid pooling fast module that performs pooling at multiple scales. C2PSA feature enhancement module enhances the spatial attention in the feature map and improves the model's focus on important parts of the image. Through spatial pooling features, the model can more effectively focus on specific areas of interest. Then, I1, I2 and I3 are input to the bottleneck layer, and the features of the three granularities are transmitted to the detection layer for prediction after multi-scale aggregation. YOLOv11 uses multi-scale detection heads to detect targets of different sizes, outputting three different scale detection boxes and their confidence. The three scale detection boxes are input to the non-maximum suppression function, and the detected candidate boxes are sorted by confidence. Then, starting from the candidate box with the highest confidence, it is compared with other candidate boxes one by one for intersection over union (IoU). If the IoU of two candidate boxes exceeds the set threshold, it is considered that the two candidate boxes overlap, and the candidate box with lower confidence is suppressed. Through this process, overlapping candidate boxes can be effectively reduced, and the candidate box most likely to contain the nasal septum is retained. Finally, based on the final target box, the nasal septum region image is cropped from the originally input nasal CBCT image.

[0085] Figure 4 A schematic diagram of a C2PSA feature enhancement module is provided for an exemplary embodiment of the present disclosure. As shown in Figure 4 , the C2PSA feature enhancement module includes a convolutional layer, a split module, two PSAs, a concatenation module, and a convolutional layer connected in series.

[0086] Figure 5 A structural schematic diagram of a PSA is provided for an exemplary embodiment of the present disclosure. As shown in Figure 5 , the PSA includes an attention module, a concatenation module, a feedforward neural network, two convolutional layers, and a concatenation module connected in series.

[0087] Figure 6 A schematic diagram of a SPPF feature extraction module is provided for an exemplary embodiment of the present disclosure. As shown in Figure 6 , the SPPF feature extraction module includes a convolutional layer, three max-pooling layers, a concatenation module, and a convolutional layer connected in series.

[0088] Before using the nasal septum deviation classification model, the nasal septum deviation classification model needs to be trained first. The training process of the nasal septum deviation classification model is described below.

[0089] The training process of the nasal septum deviation classification model is as follows: first, a large number of nasal septum region image samples are collected, and different subject objects in various scenes are covered as much as possible to improve the sample coverage. Then, whether the training sample has nasal septum deviation is labeled. Then, the labeled training sample is input into a neural network model. Inside the neural network model, feature extraction and calculation of the training sample can be performed according to the model parameters, and the nasal septum deviation result is output by the output layer of the neural network model. Then, the loss function layer of the neural network model can calculate the loss function according to the difference between the nasal septum deviation result output by the output layer and the true nasal septum deviation result on the training sample. If the loss function does not meet the set requirement, the model parameters can be adjusted for further iteration training. When the loss function of the neural network model meets the set requirement, the training of the nasal septum deviation classification model is completed.

[0090] It should be noted that the nasal septum deviation classification model includes an input layer, a first convolutional layer, a bottleneck layer module, a second convolutional layer, an average pooling layer, and a fully connected layer.

[0091] In an implementable manner, the nasal septum region image is input into the nasal septum deviation classification model to classify the nasal septum region image to obtain the nasal septum deviation result.

[0092] Figure 7 A structural diagram of a nasal septum deviation classification model is provided for the exemplary embodiments of the present disclosure. As shown in Figure 7 The nasal septum deviation classification model includes an input layer, a first convolutional layer, a bottleneck layer module, a second convolutional layer, an average pooling layer, and a fully connected layer. The nasal septum deviation result includes a nasal septum deviation probability. The nasal septum region image is input into the nasal septum deviation classification model to classify the nasal septum region image to obtain the nasal septum deviation result. In an implementable manner, the nasal septum region image is input into the input layer inside the nasal septum deviation classification model to adjust the dimensions of the nasal septum region image to obtain a standardized region image; the standardized region image is input into the first convolutional layer to perform convolution operation to obtain preliminary features; the preliminary features are input into the bottleneck layer module to learn spatial features to obtain spatial features; the spatial features are input into the second convolutional layer to perform convolution operation to obtain a feature map; the feature map is input into the average pooling layer to be compressed into a single vector to obtain a feature vector; and the feature vector is input into the fully connected layer to obtain the nasal septum deviation probability. The ROI image (i.e., the nasal septum region image) is input into the nasal septum deviation classification model, the model extracts image features through multiple convolution, pooling, and activation operations, and performs classification prediction to obtain the nasal septum deviation result. The nasal septum deviation result can be the probability of the existence of nasal septum deviation in the ROI image; or whether the ROI image has nasal septum deviation.

[0093] In one optional embodiment, the bottleneck layer module includes multiple bottleneck sub-layers; each bottleneck sub-layer includes: a first convolutional sub-layer, a first batch of normalized layers, a first activation function layer, a first depthwise separable convolutional layer, a second batch of normalized layers, a second activation function layer, a second convolutional sub-layer, and a third batch of normalized layers; preliminary features are input into the bottleneck layer module to learn spatial features, thereby obtaining spatial features. One possible implementation is that the preliminary features sequentially pass through the first convolutional sub-layer, the first batch of normalized layers, the first activation function layer, the first depthwise separable convolutional layer, the second batch of normalized layers, the second activation function layer, the second convolutional sub-layer, and the third batch of normalized layers in each bottleneck sub-layer to learn spatial features, thereby obtaining spatial features.

[0094] In one alternative embodiment, the nasal septum deviation classification model is the MobileNet model. Three different CNN architectures—ResNet, EfficientNet, and MobileNet—are pre-evaluated to determine the most suitable model for the nasal septum deviation classification task. ResNet: Utilizing its deep network structure and residual connections, it effectively captures complex features in the input data and has good recognition capabilities for subtle anatomical changes in medical images. Model performance is optimized by adjusting parameters such as network depth and width. EfficientNet: Employing a composite scaling method, it uniformly expands the network width, depth, and resolution, resulting in better accuracy and efficiency compared to manually scaled models. Transfer learning is used to further improve the model's performance on the nasal septum deviation classification task. MobileNet: Designed for mobile and embedded applications, it features a lightweight architecture, enabling real-time processing and low-power operation. By decomposing standard convolutions into depthwise separable convolutions, the number of parameters and computational cost are reduced while maintaining good classification performance. Transfer learning and fine-tuning techniques are employed to adapt it to the nasal septum deviation classification task. The three models were trained and evaluated separately. Cross-validation and other methods were used to compare their performance on metrics such as classification accuracy, AUC, and AUPR. Finally, the MobileNet model was selected for the prediction of nasal septum deviation.

[0095] Figure 8 An architecture diagram of a MobileNet model provided for an exemplary embodiment of this disclosure. Figure 8 As shown, the nasal septum image output by the YOLOv11 module is adjusted to a uniform size of 224×224×3 before being input into MobileNet for classification prediction. First, it undergoes a standard 3×3 convolution (Conv2D) operation for initial feature extraction; then, it passes through multiple bottleneck layers to learn spatial features. Figure 9A structural schematic diagram of a Bottleneck provided by an embodiment of the present disclosure. Among them, 1x1Conv2D is used to increase the number of channels, and 3x3 depthwise convolution (Depthwise Conv) is used for spatial feature extraction. Then, 1x1Conv2D is used for dimension reduction processing, and finally, the input of the current layer is spliced and output, which is used to enhance the feature information; after the bottleneck layer processing, a 1x1Conv2D is used for the generation of the final feature map; the feature map is compressed into a single vector through average pooling, and then input into the fully connected layer, which is used for the output of the classification task: the probability of deviation of the nasal septum.

[0096] In some embodiments of the present disclosure, for each slice image of the three-dimensional nasal CBCT image, target detection is performed to obtain a plurality of nasal CBCT images, and whether the nasal septum deviation exists in the nasal CBCT image is output, and the nasal septum deviation existing in the nasal CBCT image is displayed. The nasal septum region image and the nasal septum deviation result are sent to the film review terminal for the film review terminal to display the nasal septum region image and the nasal septum deviation result. One realizable way is that in the case that the nasal septum deviation result is that the nasal septum deviation exists, the nasal CBCT image, the nasal septum region image and the nasal septum deviation result are sent to the film review terminal for the film review terminal to display the nasal CBCT image, the nasal septum region image and the nasal septum deviation result according to the set display format.

[0097] Figure 10 A schematic diagram of a display interface provided by an exemplary embodiment of the present disclosure is shown. As shown in Figure 10 , the nasal CBCT image, the nasal septum region image and the nasal septum deviation result are displayed in the format shown in the figure by using the graphical user interface shown in the figure.

[0098] In order to better illustrate the present disclosure, Figure 11 A schematic diagram of a nasal septum deviation identification system provided by an exemplary embodiment of the present disclosure is shown. As shown in Figure 11 , the nasal septum deviation identification system of the present disclosure includes a region of interest detection module, a nasal septum deviation classification module and a result display module.

[0099] Among them, the region of interest detection module: uses the YOLOv11 target detection algorithm to process the input CBCT image, and accurately identifies the ROI image containing the nasal septum.

[0100] The nasal septum deviation classification module: classifies the ROI image based on the convolutional neural network (CNN) algorithm, and judges whether the nasal septum deviation exists.

[0101] The result display module: displays the classification result to the user in the form of probability, and displays the ROI image slices suspected to have nasal septum deviation in the CBCT one by one.

[0102] In the computer program product or the nasal septum deviation identification method embodiment of the present disclosure, a nasal CBCT image photographed by a cone beam CT device is acquired; the nasal CBCT image is input into a nasal septum detection model, target detection is performed on the nasal CBCT image, and a nasal septum region image corresponding to the nasal CBCT image is obtained; the nasal septum region image is input into a nasal septum deviation classification model, nasal septum deviation classification is performed on the nasal septum region image, and a nasal septum deviation result is obtained; and the nasal septum region image and the nasal septum deviation result are sent to a film reading terminal, so that the film reading terminal displays the nasal septum region image and the nasal septum deviation result; by means of the nasal septum detection model and the nasal septum deviation classification model, the present disclosure can automatically read the three-dimensional CBCT image, and the reading efficiency of the nasal CBCT image is improved compared with the existing manual reading method; the automatic reading is performed in an artificial intelligence manner, human interference factors are reduced, and the reading accuracy is improved.

[0103] Figure 12 A structure diagram of a nasal septum deviation identification device 120 provided for an exemplary embodiment of the present disclosure is shown in FIG. 12. As shown in FIG. 12, the nasal septum deviation identification device 120 includes an acquisition module 1201, a region of interest detection module 1202, a nasal septum deviation classification module 1203, and a result display module 1204. Figure 12

[0104] The acquisition module 1201 is configured to acquire a nasal CBCT image photographed by a cone beam CT device.

[0105] The region of interest detection module 1202 is configured to input the nasal CBCT image into a nasal septum detection model, perform target detection on the nasal CBCT image, and obtain a nasal septum region image corresponding to the nasal CBCT image.

[0106] The nasal septum deviation classification module 1203 is configured to input the nasal septum region image into a nasal septum deviation classification model, perform nasal septum deviation classification on the nasal septum region image, and obtain a nasal septum deviation result.

[0107] The result display module 1204 is configured to send the nasal septum region image and the nasal septum deviation result to a film reading terminal, so that the film reading terminal displays the nasal septum region image and the nasal septum deviation result.

[0108] Optionally, when acquiring the nasal CBCT image photographed by the cone beam CT device, the acquisition module 1201 is configured to:

[0109] receive a nasal CBCT image uploaded by the film reading terminal; wherein the nasal CBCT image is an image uploaded by the film reading terminal after receiving the nasal CBCT image sent by the cone beam CT device in response to an image uploading operation.​

[0110] Optionally, the nasal septum detection model comprises a preprocessing layer, a backbone layer, a bottleneck layer, a detection layer and an output layer, and the region of interest detection module 1202 is configured to, when performing target detection on the nasal CBCT image to obtain a nasal septum region image corresponding to the nasal CBCT image, input the nasal CBCT image into the nasal septum detection model.

[0111] Inside the nasal septum detection model, the nasal CBCT image is input into the preprocessing layer to perform image dimension adjustment to obtain a preprocessed image.

[0112] The preprocessed image is input into the backbone layer to perform feature extraction to obtain first, second and third granularity features.

[0113] The first, second and third granularity features are input into the bottleneck layer to obtain first, second and third aggregated features after multi-scale aggregation.

[0114] The first, second and third aggregated features are input into the detection layer to perform target detection using first, second and third detection heads to obtain first, second and third detection boxes, and a first confidence corresponding to the first detection box, a second confidence corresponding to the second detection box and a third confidence corresponding to the third detection box.

[0115] The first, second and third detection boxes are input into the output layer, and the first, second and third detection boxes are input into a non-maximum suppression function to sort the detected candidate boxes according to the confidence. The target box is selected from the sorted candidate boxes according to the intersection over union of the candidate boxes. The nasal septum region image is cropped from the nasal CBCT image using the target box.

[0116] Optionally, the backbone layer comprises a first convolutional layer module, a first C3K2 feature extraction module, an SPPF feature extraction module and a C2PSA feature enhancement module, and the region of interest detection module 1202 is configured to, when performing feature extraction on the preprocessed image to obtain first, second and third granularity features, input the preprocessed image into the backbone layer.

[0117] The preprocessed image is input into the backbone layer, sequentially subjected to convolution operation by the first convolutional layer module, feature extraction by the first C3K2 feature extraction module, feature extraction by the SPPF feature extraction module and feature enhancement by the C2PSA feature enhancement module, and the first, second and third granularity features are output.

[0118] Optionally, the bottleneck layer comprises: an up-sampling module, a concatenation module, a second C3K2 feature extraction module and a second convolutional layer module; the region of interest detection module 1202, when inputting the first-granularity feature, the second-granularity feature and the third-granularity feature into the bottleneck layer, obtaining the first-aggregated feature, the second-aggregated feature and the third-aggregated feature after multi-scale aggregation, is configured to:

[0119] inputting the first-granularity feature, the second-granularity feature and the third-granularity feature into the bottleneck layer, performing up-sampling through the up-sampling module, performing feature concatenation through the concatenation module, performing convolutional operation through the second convolutional layer module, and performing feature extraction through the second C3K2 feature extraction module, to obtain the first-aggregated feature, the second-aggregated feature and the third-aggregated feature.

[0120] Optionally, the deviated nasal septum classification model comprises: an input layer, a first convolutional layer, a bottleneck layer module, a second convolutional layer, an average pooling layer and a fully connected layer; the deviated nasal septum result comprises: a deviated nasal septum probability; the deviated nasal septum classification module 1203, when inputting the nasal septum region image into the deviated nasal septum classification model to classify the nasal septum region image to obtain the deviated nasal septum result, is configured to:

[0121] inputting the nasal septum region image into the input layer to adjust the dimension of the nasal septum region image to obtain a standardized region image;

[0122] inputting the standardized region image into the first convolutional layer to perform convolutional operation to obtain a preliminary feature;

[0123] inputting the preliminary feature into the bottleneck layer module to learn spatial features to obtain spatial features;

[0124] inputting the spatial features into the second convolutional layer to perform convolutional operation to obtain a feature map;

[0125] inputting the feature map into the average pooling layer to compress into a single vector to obtain a feature vector;

[0126] inputting the feature vector into the fully connected layer to obtain the deviated nasal septum probability.

[0127] Optionally, the bottleneck layer module comprises a plurality of bottleneck sub-layers; each bottleneck sub-layer comprises: a first convolutional sub-layer, a first batch normalization layer, a first activation function layer, a first depthwise separable convolutional layer, a second batch normalization layer, a second activation function layer, a second convolutional sub-layer and a third batch normalization layer; the deviated nasal septum classification module 1203, when inputting the preliminary feature into the bottleneck layer module to learn spatial features to obtain spatial features, is configured to:

[0128] The preliminary features sequentially pass through a first convolution sublayer, a first batch normalization layer, a first activation function layer, a first depth separable convolution layer, a second batch normalization layer, a second activation function layer, a second convolution sublayer, and a third batch normalization layer of each bottleneck sublayer to learn spatial features, so as to obtain the spatial features.

[0129] As to the apparatus in the above-described embodiments, the specific manners in which the respective modules perform operations have been described in detail in the embodiments related to the method, and thus will not be described in detail here.

[0130] Figure 13 A structural schematic diagram of an electronic device is provided for an exemplary embodiment of the present disclosure. As shown in the figure, Figure 13 The electronic device includes a memory 131 and a processor 132. In addition, the electronic device also includes a power supply component 133 and a communication component 134.

[0131] The memory 131 is configured to store computer programs and can be configured to store other various data to support operations on the electronic device. Examples of these data include instructions for operating any application or method on the electronic device.

[0132] The memory 131 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0133] The communication component 134 is configured to perform data transmission with other devices.

[0134] The processor 132 can execute computer instructions stored in the memory 131, to acquire a nasal CBCT image photographed by a cone beam CT device; input the nasal CBCT image into a nasal septum detection model, perform target detection on the nasal CBCT image, and obtain a nasal septum region image corresponding to the nasal CBCT image; input the nasal septum region image into a nasal septum deviation classification model, perform nasal septum deviation classification on the nasal septum region image, and obtain a nasal septum deviation result; and send the nasal septum region image and the nasal septum deviation result to a film review terminal, so that the film review terminal displays the nasal septum region image and the nasal septum deviation result.

[0135] Correspondingly, the present disclosure also provides a computer readable storage medium storing a computer program. When the computer readable storage medium stores the computer program and the computer program is executed by one or more processors, the one or more processors are caused to perform Figure 1 The steps in the method embodiment.

[0136] The communication component in the above-described Figure 13 is configured to facilitate wired or wireless communication between a device where the communication component is located and other devices. The device where the communication component is located can access a wireless network based on a communication standard, such as WiFi, a 2G, 3G, 4G / LTE, 5G, or the like cellular communication network, or a combination thereof. In an exemplary embodiment, the communication component receives a broadcast signal or a broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component further includes a Near Field Communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on Radio Frequency Identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0137] The power component in the above-described Figure 13 provides power to various components of a device where the power component is located. The power component can include a power management system, one or more power sources, and other components associated with generating, managing, and distributing power to a device where the power component is located.

[0138] The above-described electronic device further includes a display screen and an audio component.

[0139] The display screen includes a screen, which can include a Liquid Crystal Display (LCD) and a Touch Panel (TP). If the screen includes a touch panel, the screen can be implemented as a touch screen to receive an input signal from a user. The touch panel includes one or more touch sensors to sense a touch, a slide, and a gesture on the touch panel. The touch sensor can not only sense a boundary of a touch or a slide action, but also detect a duration and a pressure associated with a touch or a slide operation.

[0140] The audio component can be configured to output and / or input audio signals. For example, the audio component includes a microphone (MIC) that is configured to receive an external audio signal when the device where the audio component is located is in an operational mode, such as a call mode, a recording mode, and a voice recognition mode. The received audio signal can be further stored in a memory or transmitted via the communication component. In some embodiments, the audio component further includes a speaker to output audio signals.

[0141] In the device, the equipment and the storage medium embodiments of the present disclosure, the nasal CBCT image photographed by the cone beam CT equipment is acquired; the nasal CBCT image is input into a nasal septum detection model, target detection is performed on the nasal CBCT image, and a nasal septum region image corresponding to the nasal CBCT image is obtained; the nasal septum region image is input into a nasal septum deviation classification model, nasal septum deviation classification is performed on the nasal septum region image, and a nasal septum deviation result is obtained; the nasal septum region image and the nasal septum deviation result are sent to a film reading terminal, so that the film reading terminal displays the nasal septum region image and the nasal septum deviation result; by means of the nasal septum detection model and the nasal septum deviation classification model, the present disclosure can automatically read the three-dimensional CBCT image, and the reading efficiency of the nasal CBCT image is improved compared with the existing manual reading mode; the automatic reading is performed in an artificial intelligence mode, human interference factors are reduced, and the reading accuracy is improved.

[0142] Those skilled in the art will understand that the embodiments of the present disclosure can be provided as a method, a system or a computer program product. Therefore, the present disclosure can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Moreover, the present disclosure can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0143] The present disclosure is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present disclosure. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device that implements the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that performs the functions specified in one or more flows or blocks.

[0144] These computer program instructions can also be stored in a computer-readable memory that can direct the computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a product including instruction apparatus, which implements the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that performs the functions specified in one or more flows or blocks.

[0145] These computer program instructions can also be loaded into a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 Figure 1

[0146] In one typical configuration, the computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0147] The memory can include non-persistent memory and / or persistent memory, both of which can be volatile and / or non-volatile. Non-persistent memory can include, for example, random access memory (RAM), which can be static RAM (SRAM) or dynamic RAM (DRAM), and non-persistent memory can also include registers within the processor(s). Persistent memory can include, for example, read-only memory (ROM), programmable ROM (PROM), erasable programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory, just to name a few. The memory is an example of computer readable media.

[0148] Computer readable media includes permanent and non-permanent, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile discs (DVDs) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer readable media does not include transitory media, such as modulated data signals and carrier waves.

[0149] ​​It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0150] The above are merely specific embodiments of this disclosure, enabling those skilled in the art to understand or implement this disclosure. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this disclosure. Therefore, this disclosure is not to be limited to these embodiments, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A computer program product, comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, the following steps are performed: Acquire CBCT images of the nose taken by a cone-beam computed tomography (CBCT) device; The nasal CBCT image is input into the nasal septum detection model, and target detection is performed on the nasal CBCT image to obtain the nasal septum region image corresponding to the nasal CBCT image. The nasal septum detection model includes: a preprocessing layer, a backbone layer, a bottleneck layer, a detection layer, and an output layer. The backbone layer includes a first convolutional layer module, a first C3K2 feature extraction module, an SPPF feature extraction module, and a C2PSA feature enhancement module. The step of inputting the nasal CBCT image into the nasal septum detection model, performing target detection on the nasal CBCT image, and obtaining the nasal septum region image corresponding to the nasal CBCT image includes: Inside the nasal septum detection model, the nasal CBCT image is input into the preprocessing layer to adjust the image dimensions and obtain a preprocessed image. The preprocessed image is input into the backbone layer, and then sequentially passes through the first convolutional layer module for convolution operation, the first C3K2 feature extraction module for feature extraction, the SPPF feature extraction module for feature extraction, and the C2PSA feature enhancement module for feature enhancement. Finally, the first granularity feature, the second granularity feature, and the third granularity feature are output. The nasal septum region image is input into the nasal septum deviation classification model to classify the nasal septum deviation in the nasal septum region image and obtain the nasal septum deviation result. The nasal septum region image and the nasal septum deviation result are sent to the review terminal so that the review terminal can display the nasal septum region image and the nasal septum deviation result. The nasal septum deviation classification model includes: an input layer, a first convolutional layer, a bottleneck layer module, a second convolutional layer, an average pooling layer, and a fully connected layer.

2. The computer program product according to claim 1, characterized in that, The acquisition of nasal CBCT images captured by cone-beam CT equipment includes: The system receives the nasal CBCT image uploaded by the image review terminal; wherein the nasal CBCT image is an image uploaded by the image review terminal after responding to the image upload operation by receiving the nasal CBCT image sent by the cone-beam CT device.

3. The computer program product according to claim 1, characterized in that, The step of inputting the nasal CBCT image into the nasal septum detection model, performing target detection on the nasal CBCT image, and obtaining the nasal septum region image corresponding to the nasal CBCT image further includes: The first granularity feature, the second granularity feature, and the third granularity feature are input into the bottleneck layer, and after multi-scale aggregation, the first aggregation feature, the second aggregation feature, and the third aggregation feature are obtained. The first aggregated feature, the second aggregated feature, and the third aggregated feature are input into the detection layer, and the first detection head, the second detection head, and the third detection head are used to perform target detection to obtain a first detection box, a second detection box, and a third detection box, as well as a first confidence level corresponding to the first detection box, a second confidence level corresponding to the second detection box, and a third confidence level corresponding to the third detection box. The first detection box, the second detection box, and the third detection box are input into the output layer. The first detection box, the second detection box, and the third detection box are input into the non-maximum suppression function. The detected candidate boxes are sorted according to their confidence scores. The target box is selected from the sorted candidate boxes based on the intersection-union ratio of the candidate boxes. The nasal septum region image is cropped from the nasal CBCT image using the target box.

4. The computer program product according to claim 3, characterized in that, The bottleneck layer includes: an upsampling module, a splicing module, a second C3K2 feature extraction module, and a second convolutional layer module; the step of inputting the first granular feature, the second granular feature, and the third granular feature into the bottleneck layer, and performing multi-scale aggregation to obtain the first aggregated feature, the second aggregated feature, and the third aggregated feature includes: The first granular feature, the second granular feature, and the third granular feature are input into the bottleneck layer. After upsampling by the upsampling module, feature concatenation by the splicing module, convolution operation by the second convolutional layer module, and feature extraction by the second C3K2 feature extraction module, the first aggregated feature, the second aggregated feature, and the third aggregated feature are obtained.

5. The computer program product according to claim 1, characterized in that, The nasal septum deviation result includes: nasal septum deviation probability; the step of inputting the nasal septum region image into the nasal septum deviation classification model, classifying the nasal septum region image for nasal septum deviation, and obtaining the nasal septum deviation result includes: Within the nasal septum deviation classification model, the nasal septum region image is input into the input layer, and the dimensionality of the nasal septum region image is adjusted to obtain a standardized region image. The standardized region image is input into the first convolutional layer for convolution operation to obtain preliminary features; The preliminary features are input into the bottleneck layer module to learn spatial features, thereby obtaining spatial features; The spatial features are input into the second convolutional layer for convolution operation to obtain a feature map; The feature map is input into the average pooling layer and compressed into a single vector to obtain the feature vector. The feature vector is input into the fully connected layer to obtain the probability of nasal septum deviation.

6. The computer program product according to claim 5, characterized in that, The bottleneck layer module includes multiple bottleneck sub-layers; each bottleneck sub-layer includes: a first convolutional sub-layer, a first batch normalization layer, a first activation function layer, a first depthwise separable convolutional layer, a second batch normalization layer, a second activation function layer, a second convolutional sub-layer, and a third batch normalization layer; the step of inputting the preliminary features into the bottleneck layer module to learn spatial features and obtain spatial features includes: The preliminary features are sequentially passed through the first convolutional sub-layer, the first batch of normalized layers, the first activation function layer, the first depthwise separable convolutional layer, the second batch of normalized layers, the second activation function layer, the second convolutional sub-layer, and the third batch of normalized layers of each bottleneck sub-layer to learn spatial features, thereby obtaining the obtained spatial features.

7. The computer program product according to claim 1, characterized in that, The nasal septum region image and the nasal septum deviation result are sent to the review terminal for the review terminal to display the nasal septum region image and the nasal septum deviation result, including: If the nasal septum deviation result indicates that nasal septum deviation exists, the nasal CBCT image, the nasal septum region image, and the nasal septum deviation result are sent to the review terminal so that the review terminal can display the nasal CBCT image, the nasal septum region image, and the nasal septum deviation result according to the set display format.

8. A nasal septum deviation detection device, characterized in that, include: The acquisition module is used to acquire CBCT images of the nose taken by cone-beam CT equipment; The region of interest detection module is used to input the nasal CBCT image into the nasal septum detection model, perform target detection on the nasal CBCT image, and obtain the nasal septum region image corresponding to the nasal CBCT image; The nasal septum deviation classification module is used to input the nasal septum region image into the nasal septum deviation classification model, perform nasal septum deviation classification on the nasal septum region image, and obtain the nasal septum deviation result. The result display module is used to send the nasal septum region image and the nasal septum deviation result to the review terminal, so that the review terminal can display the nasal septum region image and the nasal septum deviation result. The nasal septum detection model includes: a preprocessing layer, a backbone layer, a bottleneck layer, a detection layer, and an output layer. The backbone layer includes a first convolutional layer module, a first C3K2 feature extraction module, an SPPF feature extraction module, and a C2PSA feature enhancement module. The region of interest detection module is also used for: Inside the nasal septum detection model, the nasal CBCT image is input into the preprocessing layer to adjust the image dimensions and obtain a preprocessed image. The preprocessed image is input into the backbone layer, and then sequentially passes through the first convolutional layer module for convolution operation, the first C3K2 feature extraction module for feature extraction, the SPPF feature extraction module for feature extraction, and the C2PSA feature enhancement module for feature enhancement. Finally, the first granularity feature, the second granularity feature, and the third granularity feature are output. The nasal septum deviation classification model includes: an input layer, a first convolutional layer, a bottleneck layer module, a second convolutional layer, an average pooling layer, and a fully connected layer.

9. An electronic device, characterized in that, include: processor; Memory for storing the computer program product according to any one of claims 1-7; The processor is configured to execute the computer program / instructions to implement the steps.

Images