Attention Networks for Improving Surgical Tool Classification in Laparoscopic Videos
Attention Networks for Improving Surgical Tool Classification in Laparoscopic Videos
TLDR
The Squeeze and Excitation and Convolutional Block Attention Module were employed and evaluated for improving surgical tool classification in laparoscopic videos and the advantage of both attention modules to the tool classification task was explicate.
摘要
: Deep learning approaches have been extensively developed to promote intelligent applications, such as surgical tool detection in surgical videos, inside the operating rooms (ORs). However, robustness and high-performance accuracy are demanding for such high-risk applications. In this paper, the Squeeze and Excitation (SE) and Convolutional Block Attention Module (CBAM) were employed and evaluated for improving surgical tool classification in laparoscopic videos. Experimental results explicate the advantage of both attention modules to the tool classification task. The SE and CBAM achieved mean average precision (mAP) of 88.38% and 88.40%, respectively, compared to 86.35% achieved by the base CNN model.
