Deep learning model demonstrates high accuracy in surgical instrument recognition

September 29, 2022

2 min read

Kitaguchi reports no relevant financial disclosures. Please see the study for all other authors’ relevant financial disclosures.

We were unable to process your request. Please try again later. If you continue to have this issue please contact

A deep learning-based instance segmentation model was highly accurate in simultaneously recognizing eight types of surgical instruments in laparoscopic colorectal surgical videos, according to a study published in JAMA Network Open.

“Deep learning-based [computer vision] technology has been applied to various tasks in laparoscopic surgery, such as anatomical structures, surgical action, and step or phase recognition,” Daichi Kitaguchi, MD, of the surgical device innovation office and department of colorectal surgery at the National Cancer Center Hospital East in Kashiwa, Japan, and colleagues wrote. “Applying this technology to surgical instrument recognition can enable the development of robotic camera holders with automated real-time instrument tracking during surgery, and it can be used for surgical skill assessment, augmented reality and depth enhancement during [minimally invasive surgery].”

Source: Adobe Stock.
Source: Adobe Stock.

Kitaguchi and colleagues conducted a quality improvement study at a single institution and included 337 laparoscopic colorectal surgical videos recorded between April 1, 2009, and Dec. 31, 2021. The researchers performed deep-learning based instance segmentation for eight surgical instruments. This method recognizes images of objects individually and pixel-by-pixel instead of through bounding-box object detection.

Of the 337 surgical videos, pixel-by-pixel annotation was manually performed for 81,760 labels on 38,628 randomly selected static images of surgical instruments. The eight instruments included were surgical shears, a spatula-type electrode, atraumatic universal forceps with grooves, dissection forceps, endoscopic clip appliers, staplers, grasping forceps and a suction/irrigation system.

Findings were evaluated by average precision, which is calculated from the area under the precision-recall curve and based on the number of true-positive, false-positive and false-negative results.

According to the researchers, the mean average precisions of the instance segmentation were 90.9% for three instruments, 90.3% for four instruments, 91.6% for six instruments and 91.8% for eight instruments.

“We developed an instance segmentation model that can simultaneously recognize 8 types of surgical instruments that are frequently used in laparoscopic colorectal operations,” Kitaguchi and colleagues wrote. “In this quality improvement study, the recognition accuracy of the model was high and was maintained even when the number of types of surgical instruments to be recognized increased.”

They added: “Surgical instrument recognition is an important fundamental technology for various surgical research and development areas. The proposed model can be applied not only to the proposed automatic surgical progress monitoring in this study but also for future computer-assisted surgery realization or surgical automation.”

Source link

Back to top button