Dependence of the performance quality of neural networks on the characteristics of training data when working with thyroid ultrasound images

E.A. Ekhlakov, K.V. Tsyguleva, M.E. Dunaev, S.M. Zakharova

Abstract


The purpose of this work is to test the hypothesis that the performance of neural network models for the detection and segmentation of nodular formations in thyroid ultrasound images is practically independent of the number of analyzed images of one patient obtained at the same time. Two deep architectures were used for verification: YOLOv5 when solving the detection problem and DeepLabV3 when solving the segmentation problem. During the experiments, video loops (frame sequences) of thyroid ultrasound were used, containing more than 7000 images of transverse and longitudinal projections of 166 patients. The performance of deep architectures was evaluated both at the stage of their training and validation, and at the stage of their testing. According to the results of the experiments, it was found that an increase in the number of similar images in loops with a constant number of patients under study does not affect the operation of deep architectures.


Full Text:

PDF (Russian)

References


Safronov D.A., Katser Yu.D., Zaytsev K.S. Finding anomalies with autoencoders // International Journal of Open Information Technologies, 2022. T. 10, Issue. 8 p. 39-45.

Dyuldin E.V., Zaytsev K.S. Application of deep learning to identify and classify DGA domains // International Journal of Open Information Technologies, 2022. T. 10, Issue. 8 p. 3-10.

Junying Chena, , Haijun Youa, Kai Li A review of thyroid gland segmentation and thyroid nodule segmentation methods for medical ultrasound images // Computer Methods and Programs in Biomedicine, v.185, March 2020, 105329 https://www.sciencedirect.com/science/article/pii/S0169260719308454

Pengju Deng, Xiaohong Han, Xi Wei, Luchen Chang Automatic classification of thyroid nodules in ultrasound images using a multi-task attention network guided by clinical knowledge // Computer Methods and Programs in Biomedicine, October 2022, 105329

https://www.sciencedirect.com/science/article/pii/S0010482522008800

Xinyu Zhanga, Vincent CS. Leea, Jia Ronga & etc. Deep convolutional neural networks in thyroid disease detection: A multi-classification comparison by ultrasonography and computed tomography // Computer Methods and Programs in Biomedicine, v. 220 (2022) June 2022, 106823 https://www.sciencedirect.com/science/article/pii/S016926072200205X

Yasaman Sharifi, Mohamad Amin Bakhshali, Toktam Dehghani & etc. Deep learning on ultrasound images of thyroid nodules // Biocybernetics and Biomedical Engineering v. 41, Iss 2, April-June, p. 636-655 https://www.sciencedirect.com/science/article/abs/pii/S0208521621000152

Tessler, F. N., Middleton, W. D., & Grant, E. G. (2018). Thyroid Imaging Reporting and Data System (TI-RADS): A User’s Guide. Radiology, 287(1), 29–36. https://doi.org/10.1148/radiol.2017171240

Timo Luddecke, Alexander Ecker; The Role of Data for One-Shot Semantic Segmentation; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2021, pp. 2653-2658

Redmon, J., Divvala, S., Girshick, R., & Farhadi, A. (2015). You Only Look Once: Unified, Real-Time Object Detection. http://arxiv.org/abs/1506.02640

Ananev, V. V., Skorik, S. N., Shaklein, V. V., Avetisyan, A. A., Teregulov, Y. E., Turdakov, D. Y., Gliner, V., Schuster, A., & Karpulevich, E. A. (2021). Assessment of the impact of non-architectural changes in the predictive model on the quality of ECG classification. Proceedings of the Institute for System Programming of the RAS, 33(4), 87–98. https://doi.org/10.15514/ispras-2021-33(4)-7.

Luca, A. R., Ursuleanu, T. F., Gheorghe, L., Grigorovici, R., Iancu, S., Hlusneac, M., & Grigorovici, A. (2022). Impact of quality, type and volume of data used by deep learning models in the analysis of medical images. In Informatics in Medicine Unlocked (Vol. 29). Elsevier Ltd. https://doi.org/10.1016/j.imu.2022.100911

Chen, L.-C., Papandreou, G., Schroff, F., & Adam, H. (2017). Rethinking Atrous Convolution for Semantic Image Segmentation. http://arxiv.org/abs/1706.05587


Refbacks

  • There are currently no refbacks.


Abava  Кибербезопасность IT Congress 2024

ISSN: 2307-8162