Surgical reports aimed at understanding inrobot-assisted surgery can contribute to documenting entry tasks and post-operative analysis . Despite the impressive outcome, the deep learningmodel degrades the performance when applied to different domains encountering domain shifts . In this work, we proposeclass-incremental domain adaptation (CIDA) with a multi-layer transformer-basedmodel to tackle the new classes and domain shift in the target domain to generate surgical reports . The code is publicly available athttps://github.com/XuMengyaAmy/CIDACaptioning. We observe that domain invariant feature learning and the well-calibratednetwork improves the surgical report generation performance in both source and target domain under domain shift and unseen classes in the manners of one-shot learning . We also adopt label smoothing to calibrate prediction probability and obtain better feature representation with both feature extractor and captioning model . The proposed techniques are empirically evaluated by using using the datasets of nephrectomy operations and transoral robotic surgery and transalogical robotic surgery .

Author(s) : Mengya Xu, Mobarakol Islam, Chwee Ming Lim, Hongliang Ren

Links : PDF - Abstract

Code :
Coursera

Keywords : domain - surgical - feature - surgery - shift -

Leave a Reply

Your email address will not be published. Required fields are marked *