基于注意力机制与多任务的肺部疾病诊断方法

doi:10.19678/j.issn.1000-3428.0068786

摘要/Abstract

摘要：

肺部疾病存在种类多、病灶区域小的特点, 现有数据集也存在数据量小的问题, 导致模型效果不理想。为提高诊断效果, 提出一种基于多任务联合注意力机制的肺部诊断网络(ASNet)。基于U-Net构建多任务诊断网络, 在原有病灶分割任务基础上加入病理分类任务, 加强任务之间的联系, 以分割任务为辅, 提升分类任务准确率; 提出多尺度挤压激励模块, 加强空间和通道之间的信息融合; 引入一种轴向注意力机制, 强调全局上下文信息和位置信息缓解由于医疗数据匮乏引起的欠拟合问题; 设计自适应多任务混合损失函数, 实现分割和分类任务损失权重的均衡。在自建数据集上的实验结果表明, 提出网络在病灶分割任务上的Dice系数、特异性(SP)、灵敏度(SE)、HD距离和准确率的平均结果分别为81.1%、99.0%、84.1%、24.6 mm和97.5%, 优于SAUNet++、SwinUnet等其他先进分割网络; 在病理分类任务上比MobileNetV2网络在精确率、召回率和准确率指标上分别提升了2.0、1.8和1.7百分点, 明显提升了在分类和分割上的精度, 对小目标病灶分割效果更佳, 其在合理的参数量下更适用于协助肺部疾病诊断。

关键词: 深度学习, 信息融合, 疾病诊断, 多任务联合学习, 注意力机制, 医学图像分割

Abstract:

There are many types of lung diseases and small lesion areas. Furthermore, existing datasets have limited data volumes, resulting in unsatisfactory model results. To improve the diagnostic effect, a lung diagnosis Network(ASNet) based on multitask joint attention is proposed. A multitask diagnostic network is constructed using U-Net, with pathological classification tasks integrated into the original lesion segmentation tasks. This integration strengthens the connection between tasks, allowing for supplementation via segmentation tasks and improving the accuracy of classification tasks. A multiscale squeeze excitation module is proposed to enhance spatial and information fusion between channels, along with an axial attention mechanism to emphasize global context and location information. This approach aims to mitigate the underfitting problem caused by limited medical data. Additionally, an adaptive multitask hybrid loss function is designed to achieve segmentation and classification tasks with loss weighted equilibrium. Detailed experiments are conducted using a self-built dataset. The experimental results on this self-built dataset indicate that the proposed network achieves an average Dice coefficient, Specificity(SP), Sensitivity(SE), Hausdorff Distance(HD), and accuracy for the focal segmentation task of 81.1%, 99.0%, 84.1%, 24.6 mm, and 97.5%, respectively. Superior to SAUNet++, SwinUnet and other advanced segmentation networks. Compared to the MobileNetV2 network, the accuracy rate, recall rate, and accuracy index for pathological classification tasks are improved by 2.0, 1.8, and 1.7 percentage points, respectively. These improvements significantly enhance the accuracy of classification and segmentation, leading to better segmentation of small target lesions, thus making the method more suitable for assisting in lung disease diagnosis with a reasonable number of parameters.

Key words: deep learning, information fusion, disease diagnosis, multi-task joint learning, attention mechanism, medical image segmentation

刘兆伟, 方艳红, 郑明宇, 锁斌. 基于注意力机制与多任务的肺部疾病诊断方法[J]. 计算机工程, 2025, 51(1): 332-342.

LIU Zhaowei, FANG Yanhong, ZHENG Mingyu, SUO Bin. Lung Disease Diagnosis Method Based on Attention Mechanism and Multi-tasking[J]. Computer Engineering, 2025, 51(1): 332-342.

https://www.ecice06.com/CN/Y2025/V51/I1/332

图/表 14

图1 本文模型结构

Fig.1 Structure of the proposed mode

图2 CBAM模块结构

Fig.2 CBAM module structure

图3 多尺度挤压激励模块结构

Fig.3 Multi-scale squeeze excitation module structure

图4 膨胀卷积

Fig.4 Dilated convolution

图5 轴向注意力结构

Fig.5 Axial attention structure

图6 数据集分布

Fig.6 Dataset distribution

图7 不同组件在分割效果上的可视化展示

Fig.7 Visual presentation of different components in segmentation effect

图8 多分类分割实验的混淆矩阵

Fig.8 Confusion matrix of multi-class segmentation experiments

图9 不同模型分割结果可视化

Fig.9 Visualization of segmentation results of different models

参考文献 27

1	曹玉红, 徐海, 刘荪傲, 等. 基于深度学习的医学影像分割研究综述. 计算机应用, 2021, 41 (8): 2273- 2287.
	CAO Y H , XU H , LIU S A , et al. Review of deep learning-based medical image segmentation. Journal of Computer Applications, 2021, 41 (8): 2273- 2287.
2	FISCHBACH F , KNOLLMANN F , GRIESSHABER V , et al. Detection of pulmonary nodules by multislice computed tomography: improved detection rate with reduced slice thickness. European Radiology, 2003, 13 (10): 2378- 2383. doi: 10.1007/s00330-003-1915-7
3	ROBERTS M , DRIGGS D , THORPE M , et al. Common pitfalls and recommendations for using machine learning to detect and prognosticate for COVID-19 using chest radiographs and CT scans. Nature Machine Intelligence, 2021, 3, 199- 217. doi: 10.1038/s42256-021-00307-0
4	SHAN F, GAO Y, WANG J. Lung infection quantification of COVID-19 in CT images with deep learning[EB/OL]. [2023-09-30]. https://arxiv.org/abs/2003.04655.
5	HASAN A M , AL-JAWAD M M , JALAB H A , et al. Classification of COVID-19 coronavirus, pneumonia and healthy lungs in CT scans using Q-deformed entropy and deep learning features. Entropy, 2020, 22 (5): 517. doi: 10.3390/e22050517
6	WANG J , BAO Y M , WEN Y F , et al. Prior-attention residual learning for more discriminative COVID-19 screening in CT images. IEEE Transactions on Medical Imaging, 2020, 39 (8): 2572- 2583. doi: 10.1109/TMI.2020.2994908
7	CHEN L C , ZHU Y K , PAPANDREOU G , et al. Encoder-decoder with atrous separable convolution for semantic image segmentation. Berlin, Germany: Springer, 2018.
8	FAN D P , ZHOU T , JI G P , et al. Inf-net: automatic COVID-19 lung infection segmentation from CT images. IEEE Transactions on Medical Imaging, 2020, 39 (8): 2626- 2637. doi: 10.1109/TMI.2020.2996645
9	冉智强. 基于深度学习的新冠肺炎CT图像辅助诊断研究[D]. 重庆: 重庆理工大学, 2022.
	RAN Z Q. Research on COVID-19 CT image aided diagnosis based on deep learning[D]. Chongqing: Chongqing University of Technology, 2022. (in Chinese)
10	DOSOVITSKIY A, BEYER L, KOLESNIKOV A, et al. An image is worth 16×16 words: transformers for image recognition at scale[EB/OL]. [2023-09-30]. https://arxiv.org/abs/2010.11929.
11	CAO H , WANG Y Y , CHEN J , et al. Swin-Unet: Unet-like pure transformer for medical image segmentation. Berlin, Germany: Springer, 2023.
12	YAN K , TANG Y B , PENG Y F , et al. MULAN: multitask universal lesion analysis network for joint lesion detection, tagging, and segmentation. Berlin, Germany: Springer, 2019.
13	AYAAN H, ABDULLAH-AL-ZUBAER I, ADAM W, et al. MultiMix: sparingly supervised, extreme multitask learning from medical Images[EB/OL]. [2023-09-30]. https://arxiv.org/abs/2010.14731v2.
14	ZHU M Z, LI H T, CHEN H, et al. SegPrompt: boosting open-world segmentation via category-level prompt learning[C]//Proceedings of IEEE/CVF International Conference on Computer Vision. Washington D. C., USA: IEEE Press, 2023: 235-248.
15	PATIL D , DEORE S G . Medical image segmentation: a review. International Journal of Computer Science & Mobile, 2021, 66, 22- 27.
16	SZEGEDY C, VANHOUCKE V, IOFFE S, et al. Rethinking the lnception architecture for computer vision[EB/OL]. [2023-09-30]. https://arxiv.org/abs/1512.00567.
17	RONNEBERGER O, FISCHER P, BROX T. U-Net: convolutional networks for biomedical image segmentation[C]//Proceedings of International Conference on Medical Image Computing and Computer-Assisted Intervention. Berlin, Germany: Springer International Publishing, 2015: 234-241.
18	HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C]//Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City, USA: IEEE Press, 2018: 684-686.
19	WOO S, PARK J, LEE J Y, et al. CBAM: convolutional block attention module[C]//Proceedings of European Conference on Computer Vision. Berlin, Germany: Springer, 2018: 3-19.
20	HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]//Proceedings of IEEE Conference on Computer Vision and Pattern Recognition. Washington D. C., USA: IEEE Press, 2016: 770-778.
21	JEYA M J V, POOJAN OZA, et al. Medical transformer: gated axial-attention for medical image segmentation[EB/OL]. [2023-09-30]. https://arxiv.org/abs/2102.10662v2.
22	ZHANG Y , YANG Q . A survey on multi-task learning. IEEE Transactions on Knowledge and Data Engineering, 2022, 34 (12): 5586- 5609. doi: 10.1109/TKDE.2021.3070203
23	ZHAO J Y, ZHANG Y C, HE X H, et al. COVID CT-dataset: a CT scan dataset about COVID-19[EB/OL]. [2023-09-30]. https://arxiv.org/abs/2003.13865v1.
24	JOAO P, GUILHERME A, CARLOS F, et al. LNDB: a lung nodule database on computed tomography[EB/OL]. [2023-09-30]. https://arxiv.org/abs/1911.08434.
25	SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[EB/OL]. [2023-09-30]. https://arxiv.org/abs/1409.1556.
26	GAO H, ZHUANG L, LAURENS V D M. Densely connected convolutional networks[EB/OL]. [2023-09-30]. https://arxiv.org/abs/1608.06993v5.
27	MARK S, ANDREW H, MENGLONG Z, et al. MobileNetV2: inverted residunecks[EB/OL]. [2023-09-30]. https://arxiv.org/abs/1801.04381v4.

[1]	罗旭东, 袁笛, 常晓军, 何震宇. 基于不确定性启发图像增强的水下目标跟踪[J]. 计算机工程, 2025, 51(1): 11-19.
[2]	周宇, 谢威, 邝得互, 江健民. 基于三元自注意力的视频快照压缩成像重建[J]. 计算机工程, 2025, 51(1): 20-30.
[3]	胡升龙, 陈彬, 张开华, 宋慧慧. 场景结构知识增强的协同显著性目标检测[J]. 计算机工程, 2025, 51(1): 31-41.
[4]	周雪阳, 傅启明, 陈建平, 陈延明, 陆悠, 王蕴哲. 基于证据和图推理的文档级关系抽取方法: 以医学关系为例[J]. 计算机工程, 2025, 51(1): 106-117.
[5]	喻勇涛, 孙奥, 李昂, 朱琳琳. 基于孪生网络的分类器输出重复性优化方法[J]. 计算机工程, 2025, 51(1): 118-127.
[6]	肖超恩, 李子凡, 张磊, 王建新, 钱思源. 基于Transformer模型与注意力机制的差分密码分析[J]. 计算机工程, 2025, 51(1): 156-163.
[7]	张会影, 圣文顺. 基于标记适应的人脸年龄识别优化算法[J]. 计算机工程, 2025, 51(1): 174-181.
[8]	杨红菊, 吉昌. 学习驱动的图像压缩算法研究[J]. 计算机工程, 2025, 51(1): 190-197.
[9]	王晓路, 汶建荣. 基于运动-时间感知的人体动作识别方法[J]. 计算机工程, 2025, 51(1): 216-224.
[10]	胡涌涛, 黄洪琼. 结合特征融合和通道注意力的多分支换装行人重识别[J]. 计算机工程, 2025, 51(1): 225-234.
[11]	火久元, 苏泓瑞, 武泽宇, 王婷娟. 基于改进YOLOv8的道路交通小目标车辆检测算法[J]. 计算机工程, 2025, 51(1): 246-257.
[12]	郑雅洲, 刘万平, 黄东. 一种基于注意力机制的BERT-CNN-GRU检测方法[J]. 计算机工程, 2025, 51(1): 258-268.
[13]	王骞, 张俊华, 王泽彤, 李博. X2S-Net: 基于双平面X线片的脊柱三维重建[J]. 计算机工程, 2025, 51(1): 277-286.
[14]	易鹏, 杨晔, 严仕嘉. 基于MPCNN模型的sEMG快速迁移学习的手势识别应用研究[J]. 计算机工程, 2025, 51(1): 304-311.
[15]	刘钟, 唐宏, 王宁喆, 朱传润. 融合RNN与稀疏自注意力的文本摘要方法[J]. 计算机工程, 2025, 51(1): 312-320.

选择文件类型/文献管理软件名称

选择包含的内容