2019/02/24 by S. M. Kamrul Hasan, Hasan, S. M. Kamrul, Cristian A. Linte +1 · 2 citations
Engineering · Medicine · #Anatomy and Medical Technology #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Medical Imaging and Analysis #Surgical Simulation and Training
paper · pdf · doi:10.48550/arxiv.1902.08994
openalex publication_date 2019/02/24 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Conventional therapy approaches limit surgeons' dexterity control due to\nlimited field-of-view. With the advent of robot-assisted surgery, there has\nbeen a paradigm shift in medical technology for minimally invasive surgery.\nHowever, it is very challenging to track the position of the surgical\ninstruments in a surgical scene, and accurate detection & identification of\nsurgical tools is paramount. Deep learning-based semantic segmentation in\nframes of surgery videos has the potential to facilitate this task. In this\nwork, we modify the U-Net architecture named U-NetPlus, by introducing a\npre-trained encoder and re-design the decoder part, by replacing the transposed\nconvolution operation with an upsampling operation based on nearest-neighbor\n(NN) interpolation. To further improve performance, we also employ a very fast\nand flexible data augmentation technique. We trained the framework on 8 x 225\nframe sequences of robotic surgical videos, available through the MICCAI 2017\nEndoVis Challenge dataset and tested it on 8 x 75 frame and 2 x 300 frame\nvideos. Using our U-NetPlus architecture, we report a 90.20% DICE for binary\nsegmentation, 76.26% DICE for instrument part segmentation, and 46.07% for\ninstrument type (i.e., all instruments) segmentation, outperforming the results\nof previous techniques implemented and tested on these data.\n