Fetal multi-anatomical structure detection in ultrasound images helps sonographers make more accurate diagnoses by revealing the relationships between anatomical structures.
Deep learning has improved the detection of these structures, but challenges remain when dealing with rare diseases due to data scarcity, prompting interest in few-shot learning techniques.
The proposed TKR-FSOD method enhances fetal anatomical structure detection using a Topological Knowledge Reasoning Module and a Discriminate Ability Enhanced Feature Learning Module, outpacing existing methods with a significant performance improvement.