Gesture Recognition Network for Non-Contact Manipulation of Medical Images

During intraoperative imaging operations, due to the fundamental conflict between strict sterile protection requirements and traditional mouse interaction methods, we urgently need a contact-free way to manipulate medical images. To address this need, we have specifically constructed a medical glove...

Full description

Saved in:
Bibliographic Details
Published inInternational Conference on Systems and Informatics pp. 1 - 7
Main Authors Cai, Wuxin, Mo, Jianqing, Yu, Qiushuo, Chen, Guangzhong, Qin, Kun
Format Conference Proceeding
LanguageEnglish
Published IEEE 14.12.2024
Subjects
Accuracy
Attention mechanisms
Biomedical imaging
gesture dataset
gesture interaction
Gesture recognition
medical image manipulation
object detection network
Protection
Real-time systems
Refining
Robustness
Surgery
YOLO
Online AccessGet full text

Cover

More Information
Summary:During intraoperative imaging operations, due to the fundamental conflict between strict sterile protection requirements and traditional mouse interaction methods, we urgently need a contact-free way to manipulate medical images. To address this need, we have specifically constructed a medical glove gesture dataset called GDMGW, aiming to enhance the robustness and generalization ability of the gesture recognition models. To meet the accurate and real-time interaction requirements during surgery, we have made several targeted improvements based on the efficient one-stage target detection algorithm, YOLOv5s: optimizing the anchor box mechanism, incorporating an attention mechanism, and adjusting the loss function. These improvements increase gesture recognition accuracy and enhance the interaction experience. Experimental results demonstrate that these proposed improvements can significantly boost the model's overall performance, thus better meeting the demand for contact-free manipulation of medical images.
ISSN:2689-7148
DOI:10.1109/ICSAI65059.2024.10893747