Danny Sherwinter, MD1, Andrew Miesse2, Matthew Eschbach2, Amit Vasanji3. 1Maimonides Medical Center, 2Covidien, 3Image IQ
BACKGROUND: Artificial neural networks (ANN) are computer learning models designed to replicate the way humans process information and learn. ANN’s are increasingly being used in complex image recognition applications where noise tolerance and the ability to rapidly generate responses to novel and incomplete stimuli are required.
Staple line dehiscence and bleeding are the most common complications after laparoscopic sleeve gastrectomy (LSG). Mismatch between tissue thickness and staple height may contribute to these complications yet no technology exists to measure gastric wall thickness (GWT) or guide the surgeon to the correct staple cartridge size.
METHODS: Demographics such as sex, age, BMI and weight was collected on 10 consecutive subjects undergoing LSG for weight loss. A commercially available bougie with integrated LED lights arrayed along its length was employed for sleeve sizing. The locations of the LEDs in the stomach were marked and video captured. Following extraction of the gastric specimen a spring-loaded caliper was used to accurately measure the GWT at points corresponding to each LED position.
A custom image processing algorithm was created to extract 4, intensity-based parameters, for each LED position in the acquired videos (36-48 parameters/patient). An ANN was subsequently trained on 60% of the patient cohort using 12 analysis parameters per data point and correlated with the measured GWT.
RESULTS: Applying the resulting trained network on the remaining cohort (40%), GWT could be predicted with 83% accuracy based solely on light transmittance through the gastric wall. Furthermore, by grouping LED data predictions and measured thickness by subject there was a 100% consensus match for tissue thickness classification.
CONCLUSIONS: By combining a commercially available lighted bougie, together with advanced image processing software and ANN, a highly accurate noninvasive technique for predicting GWT was developed.
This example highlights how the power of artificial intelligence can be harnessed to augment surgical decision making and opens the door to countless additional surgical applications for this exciting new technology.
Presented at the SAGES 2017 Annual Meeting in Houston, TX.
Abstract ID: 84283
Program Number: ET005
Presentation Session: Emerging Technology Session
Presentation Type: Podium