Speaker
Description
The post-surgical gauze retention can lead to serious complications and necessitate additional surgery for its removal. Due to data scarcity, the research on gauze segmentation on real-world surgical data remains underexplored. This work presents first investigation of gauze segmentation on real-surgical data. We use prevalently used segmentation architectures, including CNN-based, transformers, and hybrid architectures, to provide a proof-of-concept for gauze segmentation in robot-assisted minimally invasive abdominal surgeries. We use an in-house dataset prepared at a University Hospital Bonn (UKB) which reflects realistic surgical setting and extensive diversity in spatial, morphological, and visual attributes of three different gauze categories. Besides, we investigate the influence of auto-tracked segmentation masks, which incorporate in-domain knowledge but inferior quality annotations to address the bottleneck of data scarcity and further optimize the performance. Our work demonstrates the significance of utilizing real surgical data in overcoming the challenge reported in existing works of poor gauze detection in high blood presence. Besides, the models show robustness to variability in spatial and contextual attributes reflected by the comparable performance on out-of-domain simulated data.
