Deep Learning Model for Endometrium Segmentation in Transvaginal Ultrasound (TVUS) Images

DOI

10.2991/978-94-6463-566-9_6How to use a DOI?

Keywords

TVUS images; endometrium; Transformer; dice coefficient; segmentation

Abstract

Manual analysis of endometrial thickness from transvaginal ultrasound (TVUS) images can lead to inconsistent interpretations due to varying expertise among medical professionals. Accurate endometrial thickness measurement is crucial for successful in vitro fertilization (IVF) treatments and increased pregnancy rates. Few studies have proposed deep-learning methods for endometrium segmentation in TVUS images, which are challenging due to their noisiness and blurriness. This study explores the application of a Transformer network for endometrium segmentation in TVUS images, aiming to leverage its ability to capture global context and long-range correlations. The dataset, collected from Hospital Canselor Tuanku Muhriz (HCTM), consists of 25 images, with 20 for training and 5 for testing. Images were pre-processed using MATLAB for cropping, removing text and symbols, masking, and resizing. Networks were trained using Python in Visual Studio Code with optimal parameters of 600 iterations, a batch size of 8, and a learning rate of 0.0001. Performance was evaluated using the dice coefficient. Results showed that U-Net outperformed the Transformer, achieving a dice coefficient of 0.977 compared to 0.956 on the test dataset. Despite a small difference of 2.20%, U-Net demonstrated better segmentation due to the limited training data, favoring its simpler architecture. Conversely, the Transformer network relies on global context and attention, requiring more data for effective feature extraction. Further research is needed to enhance the Transformer network’s performance with small datasets.

Copyright

© 2024 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Qurratu’aini Thaqifah Ithani
AU  - Siti Salasiah Mokri
AU  - Noraishikin Zulkarnain
AU  - Mohd Faizal Bin Ahmad
PY  - 2024
DA  - 2024/11/01
TI  - Deep Learning Model for Endometrium Segmentation in Transvaginal Ultrasound (TVUS) Images
BT  - Proceedings of the  International Conference on Advanced Technology and Multidiscipline (ICATAM 2024)
PB  - Atlantis Press
SP  - 66
EP  - 80
SN  - 2352-5401
UR  - https://doi.org/10.2991/978-94-6463-566-9_6
DO  - 10.2991/978-94-6463-566-9_6
ID  - Ithani2024
ER  -