Artificial Intelligence Deep Learning Ultrasound Discrimination of Cosmetic Fillers

    September 2025 in “ Journal of Ultrasound in Medicine ”
    Ximena Wortsman, Manuel Lozano, Francisco Javier Rodríguez‐Gómez, Yessenia Valderrama, Gabriela Ortiz‐Orellana, Luciana Zattar-Ramos, Francisco de Cabo, Eliza Porciuncula Justo Ducati, Rosa Sigrist, Cláudia Borges Fontan Câmara, Juliana Paulos de Rezende, Claudia González, Leonie Schelke, Julia Diva Zavariz, Patricia Barrera, Peter J. Velthuis
    The study explored the use of AI, specifically deep learning with YOLO architecture, to discriminate cosmetic fillers on ultrasound images. An international team of 14 physicians from 6 countries compiled a dataset of 1432 images, including various fillers like HA, PMMA, CaHA, and SO. The AI model showed strong classification performance, with an average accuracy of 0.92. YOLOv11 performed exceptionally well in detecting HA and SO, with F1 scores of 0.96 and 0.94, respectively. However, the model was less consistent in identifying CaHA and PMMA, with F1 scores around 0.83. The study concludes that while AI can reliably distinguish HA and SO, further work is needed to improve the accuracy for CaHA and PMMA.
    Discuss this study in the Community →

    Research cited in this study

    2 / 2 results