• Home

• Members

• Areas
• Projects
• Contracts
• Seminars
• Colleagues

• Books
• Papers
• Conferences and workshops
• Technical reports
• Theses

• Medicina (modular)
• Master AI
• Doctorado

• Job offers
• Links
• Contact

Breast Ultrasound Training

Web application: BreastUSTraining

Introduction

BreastUSTraining is a web application developed for the project Cost-effective screening for breast cancer and RHD.

We have created this site to be a link between students and experts in breast ultrasound imaging from all around the world. Here you can describe tumors, learn from your results, and compare yourself with other experts. The descriptions and images you upload will be stored and used for research purposes only. To this end, we are asking you to describe some ultrasound images, either from the available database or from your own uploaded images. The nodules of the dataset in the web application were taken from a publicly available dataset [1]. The descriptions will be done using the BI-RADS 5th edition, to make things quicker, the descriptions will be conducted as a test.

This website provides a comprehensive way to practice and enhance your understanding of breast ultrasound imaging. Your contributions will help expand the knowledge base for both students and seasoned professionals in this field.

The application has been created as part of the doctoral thesis of Mikel Carrilero-Mardones, supervised by Jorge Pérez-Martín and Francisco Javier Díez, at the Universidad Nacional de Educación a Distancia (UNED) and has been supported by grant PID2019-110686RB-I00 from the Spanish Government and grant PEJ-2021-AI/TIC-23268 from the Community of Madrid. Ana Delgado Laguna, as an expert radiologist, has helped Mikel in the development of the web, testing the results, the models, and the interactions with the users. Special thanks to Manuela Parras Jurado and Dominica Dulnik Bucka, who also helped in the process of describing the first tumours to have an initial dataset.

What is in this for me?

• You will have access to the artificial model described in the paper "Deep learning for describing breast ultrasound images with BI-RADS terms" . This model will help you to crop your own images to get the nodules you want to describe. Once you have described a nodule, you can compare your results with the AI. You can also request a personalised AI model after describing 150 tumours. This may take approximately an hour to an hour and a half. The more tumours you describe, the more the model will learn from you, allowing updates every 150 tumours.
• You will be able to compare a tumour description with the rest of the experts. After describing a tumour, you can compare your results with a panel of experts who have also described that tumour.
• You can analyse your results statistically.
  • Analyse your intracorrelation. If you describe tumours more than once (over a period of at least 2 weeks), you will be able to see your intracorrelation on the results page, measured using Cohen's kappa metric, indicating how well you agree with yourself.
  • Database in tables and graphs. You will have access to organised tables and graphs displaying the importance of each BI-RADS characteristic in malignancy classification.

We hope that you will enjoy this website and that it will be an opportunity for you to learn from others and from yourself.

Thank you for helping us with this project!

Bibliography:

[1] Al-Dhabyani W, Gomaa M, Khaled H, Fahmy A. Dataset of breast ultrasound images. Data in Brief. 2020 Feb;28:104863. DOI: 10.1016/j.dib.2019.104863.

[2] M. Carrilero-Mardones, M. Parras-Jurado, A. Nogales, J. Perez-Martin, F. J. Diez, Deep learning for describing breast ultrasound images with bi-rads terms, Journal of Imaging Informatics in Medicine, p. 1-15, 2024.