A multimodal feature disentanglement model for lymphadenopathy diagnosis based on BUS and CDFI ultrasound videos: a retrospective, prospective, multicenter study - PubMed
4 hours ago
- #Deep Learning
- #Ultrasound
- #Lymphadenopathy
- Developed a deep learning model for diagnosing lymphadenopathy (LA) using B-mode ultrasound (BUS) and color Doppler flow imaging (CDFI) videos.
- Conducted a retrospective and prospective study involving 7371 patients from six centers across China.
- Extracted 147,420 key frames from BUS and CDFI videos for model training and validation.
- Integrated patient clinical information to enhance diagnostic performance.
- Model achieved high AUCs in internal (0.956), retrospective external (0.928), and prospective external (0.912) validation cohorts.
- Improved junior radiologists' diagnostic accuracy significantly with model assistance (AUC from 0.739 to 0.891 in retrospective cohort, 0.767 to 0.899 in prospective cohort).
- Demonstrated potential as a noninvasive, efficient tool for clinical decision-making in LA diagnosis.