CSDA-UNet: A ConvNeXt and Swin Dual Attention U-Net for Tracheal Cartilage Segmentation in Emergency Ultrasound
Abstract
the target is small, low contrast, and affected by speckle noise. Excessive model complexity may overfit small ultrasound datasets when it
lacks task appropriate inductive bias. We propose CSDA-UNet, which combines a ConvNeXt V2 encoder, a window based Swin Transformer
V2 bottleneck, and a dual attention gate for noise aware skip refinement. The model was evaluated on 1000 cervical ultrasound images from
150 patients. On the held out evaluation set, CSDA-UNet achieved Dice 0.8002 +/- 0.1442, IoU 0.6874 +/- 0.1529, precision 0.8240 +/- 0.1606,
recall 0.8110 +/- 0.1574, and the lowest HD95 among the evaluated methods. Paired Wilcoxon testing showed higher Dice and IoU than four
baselines after false discovery rate correction, although effect sizes were small. These results support the feasibility of task biased automated
tracheal cartilage segmentation, while external validation remains necessary.
Keywords
Full Text:
PDFReferences
[1] Cook TM, Woodall N, Frerk C. Major complications of airway management in the UK: results of the Fourth National Audit Project of
the Royal College of Anaesthetists and the Diffi cult Airway Society. Part 1: Anaesthesia. Br J Anaesth 2011;106(5):61731.
[2] Frerk C, Mitchell VS, McNarry AF, et al. Diffi cult Airway Society 2015 guidelines for management of unanticipated diffi cult intubation
in adults. Br J Anaesth 2015;115(6):82748.
[3] Kristensen MS. Ultrasonography in the management of the airway. Acta Anaesthesiol Scand 2011;55(10):115573.
[4] You-Ten KE, Siddiqui N, Teoh WH, Kristensen MS. Point-of-care ultrasound (POCUS) of the upper airway. Can J Anaesth
2018;65(4):47384.
[5] Chou HC, Tseng WP, Wang CH, et al. Tracheal rapid ultrasound exam (T.R.U.E.) for confi rming endotracheal tube placement during
emergency intubation. Resuscitation 2011;82(10):127984.
[6] Lin J, Bellinger R, Shedd A, et al. Point-of-care ultrasound in airway evaluation and management: A comprehensive review. Diagnostics
(Basel) 2023;13(9):1541.
[7] Ronneberger O, Fischer P, Brox T. U-Net: Convolutional networks for biomedical image segmentation. In: MICCAI 2015. p. 23441.
[8] Oktay O, Schlemper J, Folgoc LL, et al. Attention U-Net: Learning where to look for the pancreas. arXiv preprint arXiv:1804.03999,
2018.
[9] Zhou Z, Siddiquee MMR, Tajbakhsh N, et al. UNet++: A nested U-Net architecture for medical image segmentation. IEEE Trans Med
Imaging 2020;39(6):185667.
[10] Dosovitskiy A, Beyer L, Kolesnikov A, et al. An image is worth 16 16 words: Transformers for image recognition at scale. In: ICLR
2021.
[11] Chen J, Lu Y, Yu Q, et al. TransUNet: Transformers make strong encoders for medical image segmentation. arXiv preprint
arXiv:2102.04306, 2021.
[12] Cao H, Wang Y, Chen J, et al. Swin-Unet: Unet-like pure Transformer for medical image segmentation. In: ECCV Workshops 2022. p.
20518.
[13] Azad R, Arimond R, Aghdam EK, et al. DAE-Former: Dual Attention-Guided Efficient Transformer for Medical Image Segmentation.
In: Predictive Intelligence in Medicine, PRIME 2023. p. 8395. doi:10.1007/978-3-031-46005-0_8.
[14] Huang X, Deng Z, Li D, et al. MISSFormer: An effective transformer for 2D medical image segmentation. IEEE Trans Med Imaging
2023;42(5):148494.
[15] Hu J, Shen L, Sun G. Squeeze-and-Excitation Networks. In: CVPR 2018. p. 713241.
[16] Woo S, Park J, Lee JY, et al. CBAM: Convolutional block attention module. In: ECCV 2018. p. 319.
[17] Woo S, Debnath S, Hu R, et al. ConvNeXt V2: Co-designing and scaling ConvNets with masked autoencoders. In: CVPR 2023. p.
1613342.
[18] Lin TY, Goyal P, Girshick R, et al. Focal loss for dense object detection. In: ICCV 2017. p. 29808.
DOI: http://dx.doi.org/10.70711/pmr.v3i8.9543
Refbacks
- There are currently no refbacks.
