轻量化的多尺度注意力脊柱侧弯筛查方法

doi:10.3778/j.issn.1002-8331.2309-0192

摘要/Abstract

摘要： 近年来，深度学习越来越多地应用于脊柱侧弯筛查技术研究，并且取得了突出的成效。为了解决脊柱侧弯筛查的精度和效率不高，无法满足大规模脊柱侧弯筛查需要的问题，设计了一种轻量化的多尺度注意力卷积神经网络，对ResNet50进行改进，取得了较好的筛查效果。提出了一种多尺度残差特征提取模块，使用不同大小的卷积核，提取不同尺度的信息；使用三个残差块并在残差块中使用一种混合注意力机制，关注通道和空间两方面的信息，增强特征提取能力；将普通卷积替换成一种深度混洗卷积，在精度损失不多的情况下，提高网络效率；提出了一种多层次特征融合模块，将多个层次信息进行特征融合，提取更加多样化的特征信息。实验证明，相比ResNet50总体准确率提高了11.19个百分点，测试时长减少了2?s。

关键词: 脊柱侧弯, 深度学习, 多尺度特征, 注意力机制

Abstract: In recent years, deep learning has been increasingly applied to the research of scoliosis screening technology and has achieved outstanding results. In order to solve the problem that the accuracy and efficiency of scoliosis screening are not high enough to meet the needs of large-scale scoliosis screening, a lightweight multiscale attention convolutional neural network is designed to improve ResNet50, and better screening results are achieved. A multiscale residual feature extraction module is proposed, which uses convolution kernels of different sizes to extract information at different scales. Three residual blocks are used and a hybrid attention mechanism is used in the residual blocks, which focuses on both channel and spatial information and enhances the feature extraction. The ordinary convolution is replaced by a depthwise shuffle convolution, which improves the efficiency of the network with little loss of accuracy; and a multilevel feature fusion module, which fuses multiple levels of information to extract more diverse feature information. The experiment proves that the overall accuracy is improved by 11.19 percentage points and the test duration is reduced by 2 seconds compared with ResNet50.

Key words: scoliosis, deep learning, multiscale feature, attention mechanism

郝子强, 唐颖, 田芳, 张岩, 詹伟达. 轻量化的多尺度注意力脊柱侧弯筛查方法[J]. 计算机工程与应用, 2025, 61(3): 286-294.

HAO Ziqiang, TANG Ying, TIAN Fang, ZHANG Yan, ZHAN Weida. Lightweight Multiscale Attention Scoliosis Screening Method[J]. Computer Engineering and Applications, 2025, 61(3): 286-294.

参考文献

张俊华. 多尺度特征融合的脊柱X线图像分割方法[J]. 计算机工程与应用, 2021, 57(8): 214-219.
ZHAO Y, ZHANG J H. Multi-scale feature fusion method for spinal X-ray image segmentation[J]. Computer Engineering and Applications, 2021, 57(8): 214-219.
[2] WEINSTEIN S L, DOLAN L A, WRIGHT J G, et al. Effects of bracing in adolescents with idiopathic scoliosis[J]. New England Journal of Medicine, 2013, 369(16): 1512-1521.
[3] SUNG S, CHAE H W, LEE H S, et al. Incidence and surgery rate of idiopathic scoliosis: a nationwide database study[J]. International Journal of Environmental Research and Public Health, 2021, 18(15): 8152.
[4] 刘晓民, 王哲, 郭伟, 等. 基于掩膜分割的Cobb角测量方法[J]. 北京工业大学学报, 2021, 47(11): 8.
LIU X M, WANG Z, GUO W, et al. Cobb angle measurement base on mask segmentation[J]. Journal of Beijing University of Technology, 2021, 47(11): 8.
[5] NEGRINI S, DONZELLI S, AULISA A G, et al. 2016 SOSORT guidelines: orthopaedic and rehabilitation treatment of idiopathic scoliosis during growth[J]. Scoliosis and Spinal Disorders, 2018, 13(1): 1-48.
[6] KOROVESSIS P, ZACHARATOS S, KOUREAS G, et al. Comparative multifactorial analysis of the effects of idiopathic adolescent scoliosis and scheuermann kyphosis on the self-perceived health status of adolescents treated with brace[J]. European Spine Journal, 2007, 16: 537-546.
[7] LENKE L G, BETZ R R, HARMS J, et al. Adolescent idiopathic scoliosis: a new classification to determine extent of spinal arthrodesis[J]. The Journal of Bone and Joint Surgery, 2001, 83(8): 1169-1181.
[8] STOKES I A F. Analysis and simulation of progressive adolescent scoliosis by biomechanical growth modulation[J]. European Spine Journal, 2007, 16: 1621-1628.
[9] LANGENSIEPEN S, SEMLER O, SOBOTTKE R, et al. Measuring procedures to determine the cobb angle in idiopathic scoliosis: a systematic review[J]. European Spine Journal, 2013, 22: 2360-2371.
[10] 魏旋旋, 黄子健, 曹乐, 等. 采用轻量级姿态估计网络的脊柱侧弯筛查方法[J]. 智能系统学报, 2023, 18(5): 1039-1046.
WEI X X, HUANG Z J, CAO L, et al. Scoliosis screening method using lightweight pose estimation network[J]. CAAI Transactions on Intelligent Systems, 2023, 18(5): 1039-1046.
[11] LAFAGE V, SCHWAB F, PATEL A, et al. Pelvic tilt and truncal inclination: two key radiographic parameters in the setting of adults with spinal deformity[J]. Spine, 2009, 34(17): E599-E606.
[12] LODER R T, URQUHART A, STEEN H, et al. Variability in cobb angle measurements in children with congenital scoliosis[J]. The Journal of Bone and Joint Surgery. British Volume, 1995, 77(5): 768-770.
[13] ALLEN S, PARENT E, KHORASANI M, et al. Validity and reliability of active shape models for the estimation of cobb angle in patients with adolescent idiopathic scoliosis[J]. Journal of Digital Imaging, 2008, 21: 208-218.
[14] HUANG C, TANG H, FAN W, et al. Fully-automated analysis of scoliosis from spinal x-ray images[C]//Proceedings of the 2020 IEEE 33rd International Symposium on Computer-Based Medical Systems (CBMS), 2020: 114-119.
[15] RAMIREZ L, DURDLE N G, RASO V J, et al. A support vector machines classifier to assess the severity of idiopathic scoliosis from surface topography[J]. IEEE Transactions on Information Technology in Biomedicine, 2006, 10(1): 84-91.
[16] YANG J, ZHANG K, FAN H, et al. Development and validation of deep learning algorithms for scoliosis screening using back images[J]. Communications Biology, 2019, 2(1): 390.
[17] XU Z, OUYANG J, GAO Q, et al. 2D photogrammetry image of adolescent idiopathic scoliosis screening using deep learning[C]//Proceedings of the International Symposium on Bioinformatics Research and Applications. Cham: Springer Nature Switzerland, 2022: 330-342.
[18] SZEGEDY C, LIU W, JIA Y, et al. Going deeper with convolutions[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2015: 1-9.
[19] ZHANG M, PANG K, GAO C, et al. Multi-scale aerial target detection based on densely connected inception ResNet[J]. IEEE Access, 2020, 8: 84867-84878.
[20] HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 770-778.
[21] GEHRING J, AULI M, GRANGIER D, et al. Convolutional sequence to sequence learning[C]//Proceedings of the International Conference on Machine Learning, 2017: 1243-1252.
[22] CHOLLET F. Xception: deep learning with depthwise separable convolutions[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017: 1251-1258.
[23] ZHANG X, ZHOU X, LIN M, et al. ShuffleNet: an extremely efficient convolutional neural network for mobile devices[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 6848-6856.
[24] 贾天豪, 彭力, 戴菲菲. 引入残差学习与多尺度特征增强的目标检测器[J]. 计算机科学与探索, 2023, 17(5): 1102-1111.
JIA T H, PENG L, DAI F F. Object detector with residual learning and multi-scale feature enhancement[J]. Journal of Frontiers of Computer Science and Technology, 2023, 17(5): 1102-1111.
[25] HORNG M H, KUOK C P, FU M J, et al. Cobb angle measurement of spine from X-ray images using convolutional neural network[J]. Computational and Mathematical Methods in Medicine, 2019 (9): 1-18.
[26] HAN K, WANG Y, TIAN Q, et al. GhostNet: more features from cheap operations[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 1580-1589.
[27] HOWARD A, SANDLER M, CHU G, et al. Searching for MobileNetV3[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision, 2019: 1314-1324.
[28] TAN M, LE Q. EfficientNet: rethinking model scaling for convolutional neural networks[C]//Proceedings of the International Conference on Machine Learning, 2019: 6105-6114.
[29] HUANG G, LIU Z, VAN DER MAATEN L, et al. Densely connected convolutional networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017: 4700-4708.

[1]	李小童, 马素芬, 生慧, 魏国辉, 李欣桐. 基于深度学习的肺部CT图像病灶区域分割研究综述[J]. 计算机工程与应用, 2025, 61(4): 25-42.
[2]	董甲东, 郭庆虎, 陈琳, 桑飞虎. 深度学习中单阶段金属表面缺陷检测算法优化综述[J]. 计算机工程与应用, 2025, 61(4): 72-89.
[3]	谢永华, 卓安南. 基于注意力解码和连续性监督的裂缝检测方法[J]. 计算机工程与应用, 2025, 61(4): 122-129.
[4]	肖立中, 殷晨旭. 融合预训练模型与注意力的事件抽取方法[J]. 计算机工程与应用, 2025, 61(4): 130-140.
[5]	雷景生, 章志豪, 钱小鸿, 王巍然, 杨胜英. 改进YOLOX的轻量级多方向车牌检测算法[J]. 计算机工程与应用, 2025, 61(4): 230-240.
[6]	袁姮, 颜廷昊, 张晟翀. 两阶段特征迁移图像去雾算法[J]. 计算机工程与应用, 2025, 61(4): 241-252.
[7]	赵磊, 李栋. PMM-YOLO：多尺度特征融合的交通标志检测算法[J]. 计算机工程与应用, 2025, 61(4): 262-271.
[8]	张锴, 贾涛. 结合知识图谱和小目标改进的RCNN电力杆塔部件识别方法[J]. 计算机工程与应用, 2025, 61(4): 299-309.
[9]	张相胜, 程嘉宝, 顾斌杰. 基于旋转框定位的拆垛箱体目标检测[J]. 计算机工程与应用, 2025, 61(4): 323-330.
[10]	蒋悦晗, 陈俊杰, 李洪均. 基于骨骼图神经网络的人体行为识别综述[J]. 计算机工程与应用, 2025, 61(3): 34-47.
[11]	李泽慧, 张琳, 山显英. 三维卷积神经网络方法改进及其应用综述[J]. 计算机工程与应用, 2025, 61(3): 48-61.
[12]	廖宁生, 曹天秀, 刘科言, 徐猛, 朱秘, 古宇轩, 王朋飞. 复合特征与多尺度融合的无人机小目标检测算法[J]. 计算机工程与应用, 2025, 61(3): 111-120.
[13]	徐壮, 钱育蓉, 颜丰. GCW-YOLOv8n：轻量级安全帽佩戴检测算法[J]. 计算机工程与应用, 2025, 61(3): 144-154.
[14]	余本功, 石中玉. 深层注意力和两阶段融合的图文情感对比学习方法[J]. 计算机工程与应用, 2025, 61(3): 223-233.
[15]	李志媛, 刘祎, 张鹏程, 张丽媛, 任时磊, 芦婧, 桂志国. AWTV和高斯注意力引导的LDCT图像去噪网络[J]. 计算机工程与应用, 2025, 61(3): 253-263.