The Application of Artificial Intelligence for Renal Stone Volume Determination: Ready for Prime Time? - Andrei D. Cumpanas
June 7, 2023
In a research presentation from Andrei Cumpanas, significant strides are shown in using artificial intelligence to accurately determine renal stone volume. Traditional linear measurements prove inaccurate due to stones' irregular shapes, leading the team to train an AI neural network for kidney segmentation and CT-based stone volume calculation. Cumpanas reports an impressive advancement in the accuracy of their AI algorithm, rising from a Dice Score of 0.56 to 0.96 within a year, and a Pearson correlation coefficient of 0.99, proving the AI's efficacy and consistency in determining stone volume.
Biographies:
Andrei D. Cumpanas, LIFT Research Fellow, Department of Urology, University of California, Irvine, CA
Biographies:
Andrei D. Cumpanas, LIFT Research Fellow, Department of Urology, University of California, Irvine, CA
Read the Full Video Transcript
Andrei Cumpanas: Dear viewers, my name is Andrei Cumpanas. I'm a LIFT research scholar here at University of California Irvine's Department of Urology, and today I'm going to be presenting the results of our study, entitled The Application of Artificial Intelligence for Renal Stone Volume Determination. Are we ready for prime time?
Given the irregular shape of most renal stones, linear measurements either individually or as part of an ellipsoid formula have proven inaccurate in depicting the true stone volume. As such, we sought to train an artificial intelligence neural network to automate kidney segmentation and CT-based stone volume calculation. To assess the accuracy of our AI algorithm against a radiologist-determined ground truth, both a Pearson correlation coefficient and a Dice Score were calculated.
In our prior work, comprising of 219 CT scans published in 2021, the Dice Score for our AI algorithm was 0.56. One year later, with further refinements, the algorithm achieved a nearly perfect spatial overlap between the AI-determined stone volume and the ground truth, with a Dice Score of 0.96 and a Pearson correlation coefficient of 0.99. In conclusion, our AI algorithm convolutional neural network proved to be an accurate, efficient and consistent tool for determining stone volume. Thank you.
Andrei Cumpanas: Dear viewers, my name is Andrei Cumpanas. I'm a LIFT research scholar here at University of California Irvine's Department of Urology, and today I'm going to be presenting the results of our study, entitled The Application of Artificial Intelligence for Renal Stone Volume Determination. Are we ready for prime time?
Given the irregular shape of most renal stones, linear measurements either individually or as part of an ellipsoid formula have proven inaccurate in depicting the true stone volume. As such, we sought to train an artificial intelligence neural network to automate kidney segmentation and CT-based stone volume calculation. To assess the accuracy of our AI algorithm against a radiologist-determined ground truth, both a Pearson correlation coefficient and a Dice Score were calculated.
In our prior work, comprising of 219 CT scans published in 2021, the Dice Score for our AI algorithm was 0.56. One year later, with further refinements, the algorithm achieved a nearly perfect spatial overlap between the AI-determined stone volume and the ground truth, with a Dice Score of 0.96 and a Pearson correlation coefficient of 0.99. In conclusion, our AI algorithm convolutional neural network proved to be an accurate, efficient and consistent tool for determining stone volume. Thank you.