A Kidney Dynamic Ultrasound Image Segmentation Method Based on STDC Network

Le résumé fourni par la source

Ultrasound dynamic images of the kidney serve as crucial tools in renal diagnosis, dynamically displaying the anatomical structure and pathological information of the kidney. However, traditional ultrasound image segmentation heavily relies on the experience of ultrasound doctors, leading to inaccuracies due to repetition and redundancy. In this paper, we present a deep learning method based on Short-Term Dense Concatenate network (STDC network) for kidney dynamic ultrasound (KDU) images segmentation. STDC network adopts Short Term Dense Concatenate as the basic module. In the decoder, the learning of spatial information is integrated into the low-level layer through a single stream approach. Finally, fuse the low-level features and deep features to predict the final segmentation results. Experiments based on our self-collected dataset shows that our method achieves 0.966 in term of Jaccard, 0.983 in term of Precision and 0.982 in term of Recall in the segmentation task. Furthermore, we compare our network with other classic real-time segmentation models, showing that our method outperforms in both accuracy and speed.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.