VIDEO: Automated tumor contact surface area scores concur with human-generated scores

Editor’s note: This is an automatically generated transcript. Please notify editor@healio.com if there are concerns regarding accuracy of the transcription.

And finally, I would like to highlight an abstract presented, again, by Dr. Wood, who kind of combines the two ideas that I talked about which are the AI generation of nephrometry scores and the post-op GFR prediction. It talks about the tumor contact surface area, which is another validated nephrometry scoring system that requires measurement of multiple dimensions of cross-sectional imaging. And again, it has been demonstrated to be predictive of the GFR following partial nephrectomy, however, not widely adopted and used.

So, an AI-generated test, a score, was developed and studied in the KiTS challenge data. A deep neural network approach was used for the segmentation and later generated the measurements and the tumors' CSA scores. The human scores were tabulated by personnel who was completely blinded to the AI scores. Then, the ability to predict the ipsilateral GFR preservation was compared between the AI and the human-generated scores. Again, there was a significant agreement between the two scores. And what was interesting here is that on univariate linear regression analysis, both the AI and the human-generated CSA similarly predicted ipsilateral GFR preservation.

However, when the CSA score was incorporated into a multivariate model which included age, gender, BMI, diabetes status, and ischemia time, only the AI-generated tumor CSA remained a significant predictor of ipsilateral GFR preservation. So, the fully automated tumor CSA calculations are not inferior, and on the contrary, may be superior to human-generated CSA calculations in predicting post-operative GFR after a partial nephrectomy. So, finally, I would like to say that the goal of all of our research at the Cleveland Clinic, especially concerning the use of AI in kidney cancer management, is to validate our results in bigger cohorts and ultimately, automatically deliver meaningful scores to clinicians from a pre-operative CT scan only in the easiest way possible in order to help the decision making in kidney cancer management.