CT image segmentation of lung nodules based on channel residual nested U structure

doi:10.11996/JG.j.2095-302X.2023050879

Abstract

Abstract:

Early diagnosis and treatment are pivotal in elevating the chances of lung cancer survival. Early-stage lung cancer often manifests through lung nodules. However, their heterogeneity poses a challenge in their detection of lung nodules via computed tomography, subsequently diminishing the accuracy of segmentation results. To improve the completeness and accuracy of lung nodule segmentation results, a 3D channel residual nested U-network (CR U²Net) was proposed for lung nodule segmentation. The shallow information processing U-structure (SIPU) was proposed to address the challenge of managing the interference of noise information while simultaneously incorporating key lesion details within shallow features. To enhance the interaction across different layers of feature information, and to enrich feature expression and transfer, the Channel Residual structure was introduced in conjunction with the nested U-structure to extract and optimize feature information. Acknowledging the spatial detail information found in shallow features and the semantic abstraction in deep features, the channel extrusion U-structure (CEU) was designed to effectively fuse features at different semantic levels. By integrating the proposed modules into UNet, a lung nodule segmentation model based on nested U-structures was constructed. The proposed model was trained on the Lung Image Database Consortium and Image Database Resource Initiative (LIDC-IDRI) dataset. And chieved the best Dice Similarity Coefficient performance, reaching 83.83%. This outperformed UNet, UNet++, and PCAMNet networks by 3.98%, 1.96%, and 1.26%, respectively. In addition, ablation experiments were conducted to evaluate the structural validity of the proposed CR U²Net, demonstrating that each module within the proposed segmentation algorithm contributes to achieving optimal performance while adhering to acceptable parameter and computational constraints.

Key words: deep neural network, lung nodule segmentation, channel residual structure, nested U structure, channel extrusion module

CLC Number:

TP391

JIANG Wu-jun, ZHI Li-jia, ZHANG Shao-min, ZHOU Tao. CT image segmentation of lung nodules based on channel residual nested U structure[J]. Journal of Graphics, 2023, 44(5): 879-889.

Figures/Tables 14

References 35

[1]	YU H, LI J Q, ZHANG L X, et al. Design of lung nodules segmentation and recognition algorithm based on deep learning[J]. BMC Bioinformatics, 2021, 22(5): 1-21. DOI
[2]	KIDO S, KIDERA S, HIRANO Y, et al. Segmentation of lung nodules on CT images using a nested three-dimensional fully connected convolutional network[J]. Frontiers in Artificial Intelligence, 2022, 5: 782225. DOI URL
[3]	AGNES S A, ANITHA J. Efficient multiscale fully convolutional UNet model for segmentation of 3D lung nodule from CT image[J]. Journal of Medical Imaging: Bellingham, Wash, 2022, 9(5): 052402.
[4]	WANG Z R, MEN J R, ZHANG F C. Improved V-Net lung nodule segmentation method based on selective kernel[J]. Signal, Image and Video Processing, 2023, 17(5): 1763-1774. DOI
[5]	ZHOU Z X, GOU F F, TAN Y L, et al. A cascaded multi-stage framework for automatic detection and segmentation of pulmonary nodules in developing countries[J]. IEEE Journal of Biomedical and Health Informatics, 2022, 26(11): 5619-5630. DOI URL
[6]	WANG Y F, ZHOU C, CHAN H P, et al. Hybrid U-Net-based deep learning model for volume segmentation of lung nodules in CT images[J]. Medical Physics, 2022, 49(11): 7287-7302. DOI URL
[7]	SONG J D, HUANG S C, KELLY B, et al. Automatic lung nodule segmentation and intra-nodular heterogeneity image generation[J]. IEEE Journal of Biomedical and Health Informatics, 2021, 26(6): 2570-2581. DOI URL
[8]	TYAGI S, TALBAR S N. CSE-GAN: a 3D conditional generative adversarial network with concurrent squeeze-and- excitation blocks for lung nodule segmentation[J]. Computers in Biology and Medicine, 2022, 147: 105781. DOI URL
[9]	许正玺, 张少敏, 支力佳, 等. 三维多尺度嵌套U结构CT影像肺结节检测[J]. 中国图象图形学报, 2022, 27(3): 797-811.
	XU Z X, ZHANG S M, ZHI L J, et al. Detection of pulmonary nodules in three-dimensional multiscale nested U-structure computed tomography images[J]. Journal of Image and Graphics, 2022, 27(3): 797-811. (in Chinese)
[10]	QIN X B, ZHANG Z C, HUANG C Y, et al. U²-Net: going deeper with nested U-structure for salient object detection[J]. Pattern Recognition, 2020, 106: 107404. DOI URL
[11]	CHEN Q, XIE W, ZHOU P, et al. Multi-crop convolutional neural networks for fast lung nodule segmentation[J]. IEEE Transactions on Emerging Topics in Computational Intelligence, 2022, 6(5): 1190-1200. DOI URL
[12]	XU W X, XING Y, LU Y T, et al. Dual encoding fusion for atypical lung nodule segmentation[C]// 2022 IEEE 19th International Symposium on Biomedical Imaging. New York: IEEE Press, 2022: 1-5.
[13]	HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2016: 770-778.
[14]	MA N N, ZHANG X Y, ZHENG H T, et al. ShuffleNet V2: practical guidelines for efficient CNN architecture design[M]// Computer Vision - ECCV 2018. Cham: Springer International Publishing, 2018: 122-138.
[15]	ARMATO S G I, MCLENNAN G, BIDAUT L, et al. The lung image database consortium (LIDC) and image database resource initiative (IDRI): a completed reference database of lung nodules on CT scans[J]. Medical Physics, 2011, 38(2): 915-931. PMID
[16]	SETIO A A A, TRAVERSO A, DE BEL T, et al. Validation, comparison, and combination of algorithms for automatic detection of pulmonary nodules in computed tomography images: the LUNA16 challenge[J]. Medical Image Analysis, 2017, 42: 1-13. DOI PMID
[17]	ZHOU Z W, SIDDIQUEE M M R, TAJBAKHSH N, et al. UNet: redesigning skip connections to exploit multiscale features in image segmentation[J]. IEEE Transactions on Medical Imaging, 2020, 39(6): 1856-1867. DOI URL
[18]	MILLETARI F, NAVAB N, AHMADI S A. V-net: fully convolutional neural networks for volumetric medical image segmentation[C]// 2016 Fourth International Conference on 3D Vision. New York: IEEE Press, 2016: 565-571.
[19]	肖毅, 谢珺, 谢刚, 等. 融合注意力特征的多任务肺结节检测和分割[J]. 计算机工程与设计, 2022, 43(9)2525-2532.
	XIAO Y, XIE J, XIE G, et al. Multi-task lung nodule detection and segmentation fused with attention features[J]. Computer Engineering and Design, 2022, 43(9)2525-2532. (in Chinese)
[20]	钟思华, 王梦璐, 郭兴明, 等. 基于改进VNet的肺结节分割方法研究[J]. 仪器仪表学报, 2020, 41(9): 206-215.
	ZHONG S H, WANG M L, GUO X M, et al. Study on the improved VNet network based pulmonary nodule segmentation method[J]. Chinese Journal of Scientific Instrument. 2020, 41(9): 206-215. (in Chinese)
[21]	ÇIÇEK Ö, ABDULKADIR A, LIENKAMP S S, et al. 3D U-net: learning dense volumetric segmentation from sparse annotation[M]// Medical Image Computing and Computer- Assisted Intervention - MICCAI 2016. Cham: Springer International Publishing, 2016: 424-432.
[22]	MEHTA S, MERCAN E, BARTLETT J, et al. Y-net: joint segmentation and classification for diagnosis of breast biopsy images[M]// Medical Image Computing and Computer Assisted Intervention - MICCAI 2018. Cham: Springer International Publishing, 2018: 893-901.
[23]	ZHENG H, QIN Y L, GU Y, et al. Refined local-imbalance-based weight for airway segmentation in CT[M]// Medical Image Computing and Computer Assisted Intervention - MICCAI 2021. Cham: Springer International Publishing, 2021: 410-419.
[24]	TURELLA F, BREDELL G, OKUPNIK A, et al. High-resolution segmentation of lumbar vertebrae from conventional thick slice MRI[M]// Medical Image Computing and Computer Assisted Intervention - MICCAI 2021. Cham: Springer International Publishing, 2021: 689-698.
[25]	LIANG D, LIU J, WANG K Q, et al. Position-prior clustering-based self-attention module for knee cartilage segmentation[M]// Lecture Notes in Computer Science. Cham: Springer Nature Switzerland, 2022: 193-202.
[26]	WANG W X, CHEN C, DING M, et al. TransBTS: multimodal brain tumor segmentation using transformer[M]// Medical Image Computing and Computer Assisted Intervention - MICCAI 2021. Cham: Springer International Publishing, 2021: 109-119.
[27]	HATAMIZADEH A, TANG Y C, NATH V, et al. UNETR: transformers for 3D medical image segmentation[C]// 2022 IEEE/CVF Winter Conference on Applications of Computer Vision. New York: IEEE Press, 2022: 1748-1758.
[28]	HUANG J J, LI H F, LI G B, et al. Attentive symmetric autoencoder for brain MRI segmentation[M]// Lecture Notes in Computer Science. Cham: Springer Nature Switzerland, 2022: 203-213.
[29]	SAVIC M, MA Y H, RAMPONI G, et al. Lung nodule segmentation with a region-based fast marching method[J]. Sensors, 2021, 21(5): 1908. DOI URL
[30]	FU X L, ZHENG J Y, MAI J Y, et al. A coarse-to-fine morphological approach with knowledge-based rules and self-adapting correction for lung nodules segmentation[C]// 2022 IEEE International Conference on Image Processing. New York: IEEE Press, 2022: 1696-1700.
[31]	LEE K, ZUNG J, LI P, et al. Superhuman accuracy on the SNEMI3D connectomics challenge[EB/OL]. [2023-08-8]. http://arxiv.org/abs/1706.00120.
[32]	WOO S, PARK J, LEE J Y, et al. CBAM: convolutional block attention module[M]// Computer Vision - ECCV 2018. Cham: Springer International Publishing, 2018: 3-19.
[33]	HU J, SHEN L, ALBANIE S, et al. Squeeze-and-excitation networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020, 42(8): 2011-2023. DOI PMID
[34]	LI X, WANG W H, HU X L, et al. Selective kernel networks[C]// 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2020: 510-519.
[35]	ZHANG H, WU C R, ZHANG Z Y, et al. ResNeSt: split-attention networks[C]// 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. New York: IEEE Press, 2022: 2735-2745.

方法		数据集	输入尺寸	样本数量	交叉验证	DSC	样本选取策略
传统肺结节分割方法	文献[29]	模拟数据 LIDC-IDRI	-	108 82	否	93.30 90.10	管电流：30~197 mA，切片厚度：0.625 mm 管电流：40~582 mA，切片厚度：0.625~3.000 mm
传统肺结节分割方法	文献[30]	LIDC-IDRI LC015	-	2 651 1 186	否	69.90 76.00	-
基于深度学习的肺结节分割方法	文献[20]	LIDC-IDRI	64×64×32	728	否	88.18	切片厚度≤2.5 mm、肺结节直径≥3 mm
	文献[7]	LIDC-IDRI 私有	256×256×3	2 635 3 200	5-折交叉	82.05 81.61	依据LIDC结节尺寸报告由至少2名经验丰富的医生筛选
	文献[1]	LIDC-IDRI	48×48×16	1 074	否	80.50	至少3名医生进行标注
	文献[3]	LIDC-IDRI	64×64×64	-	否	83.00	随机选取300个CT文件
	文献[8]	Luna16 ILND	64×64×32	835 200	否	80.74 76.36	切片厚度≤2.5 mm、肺结节直径≥3 mm 丢弃切片数量太少或数据不完整的CT
	文献[5]	LIDC-IDRI	64×64×64	1 979	否	86.75	肺结节直径≥3 mm
	文献[6]	Luna16	64×64×32	1 086	否	79.60	切片厚度≤2.5 mm、肺结节直径≥3 mm
	文献[4]	LIDC-IDRI	64×64×64	2 885	否	75.00	7 mm肺结节直径≤45 mm
	文献[19]	LIDC-IDRI	-	1 131	否	80.89	肺结节直径≥3 mm
	本文方法	LIDC-IDRI	64×64×64	1 186	5-折交叉	83.83	切片厚度≤2.5 mm、肺结节直径≥3 mm

方法		数据集	输入尺寸	样本数量	交叉验证	DSC	样本选取策略
传统肺结节分割方法	文献[29]	模拟数据 LIDC-IDRI	-	108 82	否	93.30 90.10	管电流：30~197 mA，切片厚度：0.625 mm 管电流：40~582 mA，切片厚度：0.625~3.000 mm
传统肺结节分割方法	文献[30]	LIDC-IDRI LC015	-	2 651 1 186	否	69.90 76.00	-
基于深度学习的肺结节分割方法	文献[20]	LIDC-IDRI	64×64×32	728	否	88.18	切片厚度≤2.5 mm、肺结节直径≥3 mm
	文献[7]	LIDC-IDRI 私有	256×256×3	2 635 3 200	5-折交叉	82.05 81.61	依据LIDC结节尺寸报告由至少2名经验丰富的医生筛选
	文献[1]	LIDC-IDRI	48×48×16	1 074	否	80.50	至少3名医生进行标注
	文献[3]	LIDC-IDRI	64×64×64	-	否	83.00	随机选取300个CT文件
	文献[8]	Luna16 ILND	64×64×32	835 200	否	80.74 76.36	切片厚度≤2.5 mm、肺结节直径≥3 mm 丢弃切片数量太少或数据不完整的CT
	文献[5]	LIDC-IDRI	64×64×64	1 979	否	86.75	肺结节直径≥3 mm
	文献[6]	Luna16	64×64×32	1 086	否	79.60	切片厚度≤2.5 mm、肺结节直径≥3 mm
	文献[4]	LIDC-IDRI	64×64×64	2 885	否	75.00	7 mm肺结节直径≤45 mm
	文献[19]	LIDC-IDRI	-	1 131	否	80.89	肺结节直径≥3 mm
	本文方法	LIDC-IDRI	64×64×64	1 186	5-折交叉	83.83	切片厚度≤2.5 mm、肺结节直径≥3 mm

方法		评估指标
方法		PRE	SEN	DSC	mIoU	Param	FLOPs
UNet		82.16	81.99	79.85	84.00	16.32	237.01
YNet		80.27	83.87	79.53	83.71	33.28	297.05
UNet++		83.49	84.32	82.27	85.37	9.64	74.54
WingsNet		83.02	84.79	82.44	85.42	1.47	38.55
ReconNet		82.94	82.22	80.46	84.34	4.08	59.64
PCAMNet		82.83	85.10	82.57	85.51	9.44	12.30
采用Transformer技术的3D分割模型	TransBTS	83.22	78.93	78.67	83.05	30.95	32.68
	Unetr	83.57	79.96	79.34	83.49	92.34	21.49
	ASA	82.17	84.57	81.50	84.86	85.29	52.86
本文方法		84.15	86.03	83.83	86.40	44.70	101.72

方法		评估指标
方法		PRE	SEN	DSC	mIoU	Param	FLOPs
UNet		82.16	81.99	79.85	84.00	16.32	237.01
YNet		80.27	83.87	79.53	83.71	33.28	297.05
UNet++		83.49	84.32	82.27	85.37	9.64	74.54
WingsNet		83.02	84.79	82.44	85.42	1.47	38.55
ReconNet		82.94	82.22	80.46	84.34	4.08	59.64
PCAMNet		82.83	85.10	82.57	85.51	9.44	12.30
采用Transformer技术的3D分割模型	TransBTS	83.22	78.93	78.67	83.05	30.95	32.68
	Unetr	83.57	79.96	79.34	83.49	92.34	21.49
	ASA	82.17	84.57	81.50	84.86	85.29	52.86
本文方法		84.15	86.03	83.83	86.40	44.70	101.72

方法	En₁/De₁	En₂/De₂	En₃/De₃	En₄/De₄	瓶颈层	Params	FLOPs	PRE	SEN	DSC	mIoU
Base	RSU₇	RSU₆	RSU₅	RSU₄	RSU_4F	51.52	133.32	83.91	84.41	82.77	85.67
Base_SIPU	SIPU(U₇)	RSU₆	RSU₅	RSU₄	RSU_4F	51.52	133.30	84.99	83.64	82.96	85.78
Base_Uall	SIPU	U₆	U₅	U₄	U₄F	51.52	133.30	83.36	85.57	83.08	85.87