Dr. Abdallah highlights an AI model with superior predictive accuracy in renal masses

In this video, Nour Abdallah, MD, highlights findings from the study, “The AI-R.E.N.A.L. + score surpasses the human expert-generated R.E.N.A.L. score in predicting oncologic outcomes of renal tumors.” The findings were presented at the 2023 American Urological Association Annual meeting held in Chicago, Illinois. Abdallah is a research fellow at Cleveland Clinic Glickman Urological and Kidney Institute in Cleveland, Ohio.

Video Transcript:

What were the notable findings?

What we did is to compare 3 scores. The first is an AI generated score using continuous variables which we named AI+ score, an AI generated renal score using categorical variables, and the human generated traditional renal score using categorical variables. What we found is that AI+ score using continuous variables had the highest predictive ability of all oncologic outcomes that were measured, which were the presence of malignancy, high stage, high grade, and pathological tumor necrosis. It was also able to predict an approach of a partial over a radical nephrectomy.

We also compared the importance of the individual components of the AI+ score. It was not surprising to us to find that the size, which is the “R” components being the maximum diameter of the tumor, is the most important predictor of outcomes as it was associated with statistically significant odds ratio relative to the outcomes.

What are some of the implications of these findings?

This is all still in its infancy, whether the segmentations of the kidneys or also the AI generated R.E.N.A.L. score, but we feel that they can offer some very important benefits both on the clinical and research level. On the clinical level, physicians are limited by time, and when you offer them an AI generated score, you are gaining time. So, it's easily generated within seconds or even less. You are generating a score that has a high predictive accuracy, so you're increasing the accuracy of the information that you're transmitting to patients.

On the research level, it's very important that when you use AI, you are getting rid of the interobserver variability. So you are increasing the accuracy and the consistency of the data that you are generating, which makes it more robust. Furthermore, when you have time on your side, you don't need a lot of manpower. So, you can generate the score to AI to a higher volume of data. This is also beneficial to the AI model itself, because the more you feed the AI model, the more it's going to increase its accuracy.

This transcription has been edited for clarity.