Proceedings Article | 12 March 2024
KEYWORDS: Tissues, Surgery, Diffuse reflectance spectroscopy, Robots, Muscles, Robotic surgery, Colorectal cancer, Robotics, Biomedical optics, Cancer, Artificial intelligence, Machine learning, Near infrared spectroscopy, Real time optical diagnostics, Automatic target recognition
Intestinal anastomosis is a surgical procedure to reconnect two portions of the intestine and restore bowel continuity after removal of pathological elements. Anastomosis is required in nearly all the >330,000 bowel resections performed annually in the USA and England. Anastomotic leakage (AL) has been reported to happen in 2.6%-19% of cases depending on multiple factors such as the definition, location and type of anastomosis, as well as the cohort under investigation. Mortality rates due to AL vary between 10%-20%. AL occurs primarily when tissues are not healed after suture. Healing is affected by the restoration of local blood flow as well as matching tissue layers during suture. Ensuring this healing requires millimetric accuracy and consistency from surgeons despite the movement due to patient breathing and other factors, making anastomosis the most challenging step in gastrointestinal surgery. Current robotic surgery has advanced to enable laparoscopic surgery without human help. However, the potential of robotic surgery to lower AL rates is still hindered by the missing tissue identification due to the lack of tactile feedback by surgeons, which does not allow matching tissue layers during suture. Tissue identification can be performed by using diffuse reflectance spectroscopy with the aim of excluding fat and including only mucosa and muscle tissues in sutures. In this study, we have classify fat, mucosa and muscle tissues based on 3125 DRS measurements of freshly excised ex vivo specimens of 47 patients. DRS measurements were performed with fiber-optic probes of 630-μm source detector distance (SDD; probe 1) and 2500-μm SDD (probe 2) to measure tissue layers from ~0.5-1mm and from ~0.5-2 mm deep, respectively. By using probe 1 and 5-fold cross-validation of quadratic support vector machine (SVM) models, we obtained true positive rates of (97.9 ±1.8) % for fat tissues, (95.5 ±1.3) % for mucosa, and (95.5 ±1.1) % for muscle. Similarly for probe 2, we achieved true positive rates of (96.7 ±1.3) % for fat tissues, (95.7 ±1.0) % for mucosa, and (94.5 ±1.9) % for muscle.
