AUTOMATED SEGMENTATION OF ULTRASOUND MEDICAL IMAGES USING THE ATTENTION U-NET MODEL
Creators
Description
Abstract
The article deals with the method of automated semantic segmentation of ultrasound medical images using the Attention U-Net deep learning model. The advantages of using Attention blocks in neural network architectures for segmentation tasks are analyzed. To test the described algorithms, the Breast Ultrasound Images training dataset was chosen. The method described in the article allows for automating the process of detecting and preliminary classification of breast tumors based on the analysis of ultrasound images. As a result of training the Attention U-Net model, the Mean IOU value of 49.2% was obtained on the test set. The network can automatically classify the detected neoplasm as benign or malignant with an F1 Score of 0.87. The results indicate the prospects of using the Attention U-Net model in the tasks of analyzing ultrasound medical images. Ways to further improve the considered method are proposed.
Other
References:
1. Theoretical foundations of ultrasonic non-destructive testing / R. M. Galagan. - Kyiv: Igor Sikorsky Kyiv Polytechnic Institute, 2019. - 263 p. (in Ukrainian).
2. Steshenko Y. V. Automation of the process of segmentation of images of metal surface defects using the neural network U-Net / Y.V. Steshenko, A.S. Momot, A.G. Protasov O.V. Muraviov
3. Ultrasound examination of the mammary glands. Step by step. From simple to complex: textbook / A.N. Sencha. - Kyiv: MEDpress-inform, 2018. 184 p.
4. Steger C. Machine vision algorithms and applications / Carsten Steger, Markus Ulrich, Christian Wiedemann: Wiley & Sons, Incorporated, John, 2017. - 516 p.
5. Zhou, Xiangrong, et al. // Deep learning of the sectional appearances of 3D CT images for anatomical structure segmentation based on an FCN voting method // Medical physics, - 2017. - pp. 5221-5233.
6. Long J. Fully convolutional networks for semantic segmentation [Electronic resource] / Jonathan Long, Evan Shelhamer, Trevor Darrell // 2015 IEEE conference on computer vision and pattern recognition (CVPR), Boston, MA, USA, 7-12 June 2015.
7. Semantic Segmentation with Extended DeepLabv3 Architecture [Electronic resource] // IEEE Xplore.
8. Ronneberger, Olaf, Philipp Fischer, and Thomas Brox: Convolutional networks for biomedical image segmentation // International Conference on Medical image computing and computer-assisted intervention. - Springer, Cham, - 2015. - pp. 155-170.
9. Manjunath, R. V.; Kwadiki, Karibasappa. // Modified U-NET on CT images for automatic segmentation of liver and its tumor. Biomedical Engineering Advances, - 2022. - pp. 103-110.
10. Ronneberger, Olaf, Philipp Fischer, and Thomas Brox: Convolutional networks for biomedical image segmentation // International Conference on Medical image computing and computer-assisted intervention. - Springer, Cham, - 2015. - pp. 155-170.
11. Attention U-Net Based Adversarial Architectures for Chest X-ray Lung Segmentation [Electronic resource] // arXiv.org.
12. Oktay, Ozan, et al. Attention u-net: Learning where to look for the pancreas. // arXiv preprint. - 2018. - pp. 99-119.
13. Al-Dhabyani W, Gomaa M, Khaled H, Fahmy A. A dataset of breast ultrasound images. Data in Brief. 2020 Feb; 28:104863.
14. Momot A. Deep learning automated data analysis of security infrared cameras / A. Momot, I. Skladchykov // Slovak international scientific journal. - 2021. - No. 52. - pp. 13-16.
15. Momot A. Analysis of the influence of the threshold level of binarisation on the efficiency of segmentation of images of surface defects in steel by the U-Net network / A. Momot, Y. Steshenko // XXII International Scientific and Technical Conference "Instrumentation: State and Prospects", 16-17 May 2023, Igor Sikorsky Kyiv Polytechnic Institute, Kyiv, Ukraine: conference proceedings.
Files
NJD_128-56-60.pdf
Files
(429.3 kB)
| Name | Size | Download all |
|---|---|---|
|
md5:948da0477ac644c8e236ca7fab892f11
|
429.3 kB | Preview Download |