果冻视频

The 果冻视频's (BIT) research team has innovatively proposed a feature difference-optimized segmentation framework based on semi-supervised federated learning - Federated Client-Distilled model (FedCD). The paper on their groundbreaking research has been published in the top international journal in the field of medical image processing, IEEE Transactions on Medical Imaging (IEEE TMI).

Vein intelligent recognition technology is an important technical means to achieve precision and intelligence in venipuncture, with significant application value in clinical medicine, intelligent diagnostic equipment, and other fields. However, in actual scenarios, veins are affected by factors such as tissue occlusion and individual differences, making it difficult for traditional medical imaging segmentation methods to achieve precise segmentation. In response to the challenges of vein visualization and recognition faced by intelligent venipuncture, the BIT research team innovatively proposed the FedCD.

This groundbreaking research effectively addresses the key challenges of vein feature diversity and insufficient medical data annotation in venipuncture applications. It successfully overcomes the issue of model bias caused by the heterogeneity of medical data distribution across institutions, providing a reliable theoretical framework and technical support for intelligent vein segmentation.

Paper Details: Ning Shen, Tingfa Xu, Shiqi Huang, Zhenxiang Chen, and Jianan Li. "Dynamic Client Distillation for Semi-supervised Federated Learning in A Realistic Scenario [J]" IEEE Transactions on Medical Imaging, Early Access.