Stratification of thyroid nodules by Eu-TIRADS categories using transfer learning of convolutional neural networks
https://doi.org/10.14341/ket12724
Abstract
The article describes a method for assessing the malignancy potential of thyroid nodules and their stratification according to the European Thyroid Imaging And Reporting Data System (Eu-TIRADS) scale based on ultrasound diagnostic images using an artificial intelligence system. The method is based on the use of transfer learning technology for multi-parameter models of convolutional neural networks and their subsequent fine tuning. It was shown that even on a small dataset consisting of 1129 thyroid ultrasound images classified by 5 Eu-TIRADS categories, the application of the method provides high training accuracy (Accuracy: 0.8, AUC: 0.92). This makes it possible to introduce and use this technology in clinical practice as an additional tool (‘second opinion’) for an objective assessment of the risk of malignancy in thyroid nodules for the purpose of their further selection for fine needle biopsy.
About the Authors
E. N. Fartushnyi
I.M. Sechenov First Moscow State Medical University
Russian Federation
Competing Interests:
Russian Federation
Eduard N. Fartushnyi; eLibrary SPIN: 9763-4896
Moscow
Competing Interests:
none
Yu. P. Sytch
I.M. Sechenov First Moscow State Medical University
Russian Federation
Competing Interests:
Russian Federation
Yulia P. Sytch - MD, PhD; eLibrary SPIN: 3406-0978.
8-2 Trubetskaya str., Moscow 119991
Competing Interests:
none
I. E. Fartushnyi
I.M. Sechenov First Moscow State Medical University
Russian Federation
Competing Interests:
Russian Federation
Igor E. Fartushnyi
Moscow
Competing Interests:
none
K. A. Koshechkin
I.M. Sechenov First Moscow State Medical University
Russian Federation
Competing Interests:
Russian Federation
Konstantin A. Koshechkin - PhD; eLibrary SPIN: 1709-1219.
Moscow
Competing Interests:
none
G. S. Lebedev
I.M. Sechenov First Moscow State Medical University
Russian Federation
Competing Interests:
Russian Federation
Georgy S. Lebedev - PhD; eLibrary SPIN: 2297-6877.
Moscow
Competing Interests:
none
References
1. Chernikov RA, Vorobjov SL, Slepzov IV, et al. Nodular goiter (epidemiology and diagnostics). Clinical and experimental thyroidology. 2013;9(2):29-35. (In Russ.). doi: https://doi.org/10.14341/ket20139229-35
2. Russ G, Bonnema SJ, Erdogan MF, et al. European Thyroid Association guidelines for ultrasound malignancy risk stratification of thyroid nodules in adults: The EU-TIRADS. Eur Thyroid J. 2017;6(5):225-237. doi: https://doi.org/10.1159/000478927
3. Sych YP, Fadeev VV, Fisenko EP, Kalashnikova M. Reproducibility and interobserver agreement of different Thyroid Imaging and Reporting Data Systems (TIRADS). Eur Thyroid J. 2021;10(2):161-167. doi: https://doi.org/10.1159/000508959
4. Tran B, Vu G, Ha G, et al. Global evolution of research in artificial intelligence in health and medicine: A bibliometric study. J Clin Med. 2019;8(3):360. doi: https://doi.org/10.3390/jcm8030360
5. Song J, Chai YJ, Masuoka H, et al. Ultrasound image analysis using deep learning algorithm for the diagnosis of thyroid nodules. Medicine (Baltimore). 2019;98(15):e15133. doi: https://doi.org/10.1097/MD.0000000000015133
6. Platforma mashinnogo obuchenija TensorFlow (In Russ.). Доступно по: https://www.tensorflow.org/
Supplementary files
|
|
1. Fig. 1. Trained models of convolutional neural networks in the TensorFlow repository and their characteristics (https://www.tensorflow.org/) [6]. | |
| Subject | ||
| Type | Исследовательские инструменты | |
View
(878KB)
|
Indexing metadata ▾ | |
|
|
2. Fig. 2. Transfer learning model architecture for Eu-TIRADS classification with trained layers of the Xсeption model. | |
| Subject | ||
| Type | Исследовательские инструменты | |
View
(152KB)
|
Indexing metadata ▾ | |
|
|
3. Fig. 3. Distribution of initial sizes of ultrasound images in the labeled dataset. | |
| Subject | ||
| Type | Исследовательские инструменты | |
View
(223KB)
|
Indexing metadata ▾ | |
|
|
4. Fig. 4. Normalized images (example 1, 4, 5 classes). | |
| Subject | ||
| Type | Исследовательские инструменты | |
View
(279KB)
|
Indexing metadata ▾ | |
|
|
5. Fig. 5. RGB profile of the ultrasound image in the sample. | |
| Subject | ||
| Type | Исследовательские инструменты | |
View
(345KB)
|
Indexing metadata ▾ | |
|
|
6. Fig. 6. Original and augmented ultrasound image | |
| Subject | ||
| Type | Исследовательские инструменты | |
View
(171KB)
|
Indexing metadata ▾ | |
|
|
7. Fig. 7. Distribution by categories of Eu-TIRADS in the labeled dataset | |
| Subject | ||
| Type | Исследовательские инструменты | |
View
(77KB)
|
Indexing metadata ▾ | |
|
|
8. Fig. 8. Examples of the thyroid echograms distribution by Eu-TIRADS categories. | |
| Subject | ||
| Type | Исследовательские инструменты | |
View
(282KB)
|
Indexing metadata ▾ | |
|
|
9. Fig. 9. Indicators of learning. The diagnostic accuracy of the model (accuracy) (blue graph) increases rapidly with each training epoch | |
| Subject | ||
| Type | Исследовательские инструменты | |
View
(97KB)
|
Indexing metadata ▾ | |
|
|
10. Fig. 10. Matrix of errors. Model classes (“Fact” and “Predict” axes) correspond to the following Eu-TIRADS categories: class “0” - Eu-TIRADS 1, class “1” - Eu-TIRADS 2, class “2” - Eu-TIRADS 3, class "3" - Eu-TIRADS 4, class "4" - Eu-TIRADS 5. | |
| Subject | ||
| Type | Исследовательские инструменты | |
View
(98KB)
|
Indexing metadata ▾ | |
|
|
11. Rice. 11. ROC-curve (curve of errors of the constructed model). In the “legend”, the numbers indicate the model classes that correspond to the following Eu-TIRADS categories: class “0” - Eu-TIRADS 1, class “1” - Eu-TIRADS 2, class “2” - Eu-TIRADS 3, class “3 "- Eu-TIRADS 4, class "4" - Eu-TIRADS 5. | |
| Subject | ||
| Type | Исследовательские инструменты | |
View
(230KB)
|
Indexing metadata ▾ | |
Review
For citations:
Fartushnyi E.N., Sytch Yu.P., Fartushnyi I.E., Koshechkin K.A., Lebedev G.S. Stratification of thyroid nodules by Eu-TIRADS categories using transfer learning of convolutional neural networks. Clinical and experimental thyroidology. 2022;18(2):17-26. (In Russ.) https://doi.org/10.14341/ket12724
Views:
1354
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License (CC BY-NC-ND 4.0).
Email this article 
