CN-121982318-A - Scoliosis screening method, system, electronic equipment and storage medium

CN121982318ACN 121982318 ACN121982318 ACN 121982318ACN-121982318-A

Abstract

The application discloses a scoliosis screening method, a system, electronic equipment and a storage medium, wherein the method comprises the steps of obtaining a back image of a subject to obtain a binary outline mask and a key point set; according to the binary contour mask, region enhancement processing is carried out on the back image through the learnable gating parameters, an enhanced image is obtained, a key point set is divided into a plurality of channels according to semantic categories, spline curve fitting is carried out on key points of each channel respectively, a multi-channel dense geometric canvas is generated, a texture feature vector and a geometric feature vector are obtained respectively according to the enhanced image and the multi-channel dense geometric canvas, attention fusion is carried out on the texture feature vector and the geometric feature vector, a fusion feature vector is obtained, the fusion feature vector is input into a condition ordinal regression prediction classification head, and a scoliosis screening result is output. The application improves the anti-interference capability, the accuracy and the clinical suitability of screening through soft gating enhancement, feature fusion and ordinal regression classification.

Inventors

Ju Fujiao
GUI JIAWEI
ZHU SHAOTAO
DONG MINGJIE

Assignees

北京工业大学

Dates

Publication Date: 20260505
Application Date: 20260126

Claims (10)

1. A method for screening scoliosis, characterized by comprising the following steps: acquiring a back image of a subject, and obtaining a binary outline mask corresponding to the back image and a key point set extracted from the back image; according to the binary outline mask, carrying out region enhancement processing on the back image through a learnable gating parameter to obtain an enhanced image, dividing the key point set into a plurality of channels according to semantic categories, and respectively carrying out spline curve fitting on key points of each channel to generate a multi-channel dense geometric canvas; respectively obtaining texture feature vectors and geometric feature vectors according to the enhanced image and the multi-channel dense geometric canvas, and performing attention fusion on the texture feature vectors and the geometric feature vectors to obtain fusion feature vectors; and inputting the fusion feature vector into a condition ordinal regression prediction classification head, and outputting a scoliosis screening result.
2. The scoliosis screening method according to claim 1, wherein the performing region enhancement processing on the back image according to the binary contour mask by using a learnable gating parameter to obtain an enhanced image includes: ; In the formula, the term "-" means element level multiplication, For the binary profile mask of the back region, Alpha is a learnable gating parameter as a Sigmoid function; -providing the back image; is the enhanced image.
3. The scoliosis screening method according to claim 1, wherein the spline curve fitting is performed on the keypoints of each channel, respectively, to generate a multi-channel dense geometric canvas, which comprises: ; wherein: Respectively performing spline curve fitting on the key points of each channel to obtain spline fitting curves; is a semantic category; Is a constant for controlling the spatial broadening of the gaussian response; is a pixel point To fit a curve Is the shortest euclidean distance of (c).
4. The scoliosis screening method of claim 1, wherein the obtaining texture feature vectors and geometric feature vectors from the enhanced image and the multi-channel dense geometric canvas, respectively, comprises: Performing multi-scale feature extraction on the enhanced image through a deep learning network to obtain a first feature map and corresponding first feature and second feature map and corresponding second feature, wherein the first feature comprises local fine granularity information, and the second feature comprises global semantic information; Constructing a first weight vector along the height dimension and the width dimension of the first feature map according to the first feature, and fusing the first weight vector with the first feature map to obtain a first reinforced feature; Constructing a second weight vector along the height dimension and the width dimension of the second feature map according to the second feature, and fusing the second weight vector with the second feature map to obtain a second reinforced feature; aggregating the first reinforcing feature and the second reinforcing feature to obtain the texture feature vector; inputting the multi-channel dense geometric canvas into a convolutional neural network, wherein the input channels of the convolutional neural network are equal to the channels of the multi-channel dense geometric canvas, and outputting to obtain implicit characteristic vectors; inputting the key point set into the convolutional neural network, and outputting to obtain an explicit feature vector; And splicing the implicit characteristic vector and the explicit characteristic vector to obtain the geometric characteristic vector.
5. The scoliosis screening method of claim 4, wherein the explicit feature vector includes 24-dimensional clinical geometry descriptors relating to horizontal distance statistics, curve length asymmetry, local curvature, and midline deviation.
6. The scoliosis screening method according to claim 1, wherein the performing attention fusion on the texture feature vector and the geometric feature vector to obtain a fused feature vector includes: aligning the texture feature vector and the geometric feature vector through feature mode projection, and splicing according to channel dimensions after alignment to obtain a mode feature vector; inputting the modal feature vector into an attention sensing network to obtain a dynamic weight coefficient of the texture feature vector and a dynamic weight coefficient of the geometric feature vector; And performing attention fusion on the aligned texture feature vector and the geometrical feature vector according to the corresponding dynamic weight coefficient to obtain a fusion feature vector.
7. The scoliosis screening method according to claim 1, wherein inputting the fusion feature vector into a conditional ordinal regression prediction classification head to output a scoliosis screening result includes: The condition ordinal number regression prediction classification head divides scoliosis into a plurality of stages of tasks according to severity, generates corresponding number of sub-tasks, inputs the fusion feature vector into the condition ordinal number regression prediction classification head, and outputs the condition probability corresponding to the sub-tasks; And obtaining the scoliosis screening result comprising the multi-stage task according to the conditional probability.
8. A scoliosis screening system, characterized by comprising the following steps: the acquisition module is used for acquiring a back image of a subject, obtaining a binary outline mask corresponding to the back image and a key point set extracted from the back image; the processing module is used for carrying out region enhancement processing on the back image through the learnable gating parameters according to the binary outline mask to obtain an enhanced image, dividing the key point set into a plurality of channels according to semantic categories, and respectively carrying out spline curve fitting on the key points of each channel to generate a multi-channel dense geometric canvas; And the execution module is used for inputting the fusion feature vector into a condition ordinal regression prediction classification head and outputting a scoliosis screening result.
9. An electronic device comprising at least one processing unit and at least one storage unit, wherein the storage unit stores a computer program that, when executed by the processing unit, causes the processing unit to perform the method of any of claims 1-7.
10. A storage medium storing a computer program executable by an electronic device, which when run on the electronic device causes the electronic device to perform the method of any one of claims 1-7.

Description

Scoliosis screening method, system, electronic equipment and storage medium Technical Field The application relates to the technical field of artificial intelligence model coordination, in particular to a scoliosis screening method, a scoliosis screening system, electronic equipment and a storage medium. Background Adolescent Idiopathic Scoliosis (AIS) is the most common type of structural scoliosis in the adolescent population, meaning a complex three-dimensional deformity of one or more segments of the spine that curves laterally with rotation of the vertebral body, whose etiology is not yet fully defined, and may be related to various factors such as genetics, skeletal development, endocrine regulation, and the like. The clinical diagnosis and treatment core of AIS is early discovery, early diagnosis and early intervention, and screening is used as a key link of early identification, and the efficiency and accuracy of the AIS directly influence the subsequent treatment effect and the prognosis of patients. With the deep application of artificial intelligence technology in the medical field, a non-invasive screening scheme based on computer vision gradually becomes a research hot spot, and aims to solve the pain points of strong subjectivity, radiation risk, low efficiency and the like of the traditional screening method. The conventional AI auxiliary screening technology is mostly based on human back images, key semantic points such as shoulders, waists and the like are extracted through a human body posture detection algorithm, and then lateral bending risk assessment is carried out by combining image features. On the one hand, the back region in the original image is not effectively separated from the background and non-back tissues (head and limbs), the background noise is easy to interfere with the extraction of key features, so that the recognition accuracy of a model on slight lateral bending is insufficient, on the other hand, the extracted key points are in discrete distribution and are easy to be influenced by shooting angles, posture changes and positioning noise, continuous and stable back structure characterization is difficult to form, and the back asymmetry morphological features caused by the lateral bending of the spine cannot be accurately depicted, so that the reliability and clinical applicability of the screening result are influenced. Disclosure of Invention In order to overcome the defects in the prior art, the application provides a scoliosis screening method, a scoliosis screening system, electronic equipment and a storage medium. In a first aspect, to achieve the above object, the present application provides a scoliosis screening method, including: Acquiring a back image of a subject, and obtaining a binary outline mask corresponding to the back image and a key point set extracted from the back image; according to the binary contour mask, region enhancement processing is carried out on the back image through the learnable gating parameters, an enhanced image is obtained, a key point set is divided into a plurality of channels according to semantic categories, spline curve fitting is carried out on key points of each channel respectively, and a multi-channel dense geometric canvas is generated; Respectively obtaining texture feature vectors and geometric feature vectors according to the enhanced image and the multi-channel dense geometric canvas, and carrying out attention fusion on the texture feature vectors and the geometric feature vectors to obtain fusion feature vectors; and inputting the fusion feature vector into a condition ordinal regression prediction classification head, and outputting a scoliosis screening result. Further, according to the binary profile mask, performing region enhancement processing on the back image through the learnable gating parameters to obtain an enhanced image, including: ; In the formula, the term "-" means element level multiplication, Is a binary profile mask for the back region,Alpha is a learnable gating parameter as a Sigmoid function; is a back image; to enhance the image. Further, spline curve fitting is performed on the key points of each channel respectively, and a multi-channel dense geometric canvas is generated, which comprises: ; wherein: Respectively performing spline curve fitting on key points of each channel to obtain spline fitting curves; is a semantic category; Is a constant for controlling the spatial broadening of the gaussian response; is a pixel point To fit a curveIs the shortest euclidean distance of (c). Further, according to the enhanced image and the multi-channel dense geometric canvas, respectively obtaining texture feature vectors and geometric feature vectors, including: performing multi-scale feature extraction on the enhanced image through a deep learning network to obtain a first feature map and corresponding first feature and second feature map and corresponding second feature, wherein the first feat