P1.20 - A Deep Learning Segmentation Approach for Lung Ultrasound Scoring Classification
- Event
- 2025 ICU PADERBORN - 9th International Congress on Ultrasonics
2025-09-21 - 2025-09-25
Paderborn - Band
- Posters
- Chapter
- Posters
- Author(s)
- M. Muñoz, J. Camacho - Spanish National Research Council, Madrid (Spain), X. Han, L. Demi - University of Trento, Trento (Italy), T. Perrone - IRCCS San Matteo, Pavia (Italy), A. Smargiassi, R. Inchingolo - Fondazione Policlinico Universitario Agostino Gemelli IRCCS, Rome (Italy), Y. Tung Chen - Hospital Universitario La Paz, Madrid (Spain), A. Trueba Vicente - Hospital de Emergencias Enfermera Isabel Zendal, Madrid (Spain)
- Pages
- 352 - 355
- DOI
- 10.5162/Ultrasonic2025/P1.20
- ISBN
- 978-3-910600-08-9
- Price
- free
Abstract
This work presents a Deep Learning method to automate the interpretation of lung ultrasound (LUS), aiming to reduce diagnostic subjectivity. A segmentation model was developed to first identify key artifacts, such as vertical artifacts or consolidations, and then calculate a corresponding severity score. Its performance was benchmarked against a classification model across two video datasets. The segmentation model achieved comparable accuracy to the traditional classification method. Furthermore, the approach proved robust to variations in the ultrasound probe’s orientation.