Parallel computing for real time gauze detection in laparoscopy images

Gauze Detection and Segmentation in Minimally Invasive Surgery Video Using Convolutional Neural Networks

“…Gauze detection algorithms not only allow for greater accuracy and reliability in the determination of the operation phase, but also enable the development of other more immediate applications. In not-robotized laparoscopic surgery, it is feasible to track the gauze by processing the captured video signal provided by the endoscope in a transparent and unattended way for the health personnel [ 19 , 20 , 21 ]. Automated gauze tracking relieves medical staff from routine counting and control tasks to avoid the inadvertent retention of these items, a medical error that occurs rarely, but it can cause serious complications in the patient’s health [ 22 , 23 , 24 ].…”

“…In other circumstances, under steady imaging conditions, gauze texture features could be modelled explicitly with hand-crafted descriptors, expecting good results in classification. However, previous work in surgical gauze detection in images [ 19 , 20 , 21 ] shows that it is challenging to achieve good robustness with traditional methods for feature extraction in these complex situations and they always require a great deal of programming effort. In uncontrolled environments where imaging conditions fluctuate and when the intra-class variability of the features is important, it is known that Convolutional Neural Networks (CNNs) are more robust and deliver superior results than traditional descriptors [ 21 , 25 ] with the only shortcoming of their computational burden.…”