JU INSIGHT Computer-Generated R.E.N.A.L. Nephrometry Scores Yield Comparable Predictive Results to Those of Human-Expert Scores in Predicting Oncologic and Perioperative Outcomes

By: Nicholas Heller, BSc; Resha Tejpaul, BSc; Fabian Isensee, PhD; Tarik Benidir, MD, MSc; Martin Hofmann, MD; Paul Blake, MD; Zachary Rengal, MD; Keenan Moore, BSc; Niranjan Sathianathen, MD; Arveen Kalapara, MD; Joel Rosenberg, MD; Sarah Peterson, BSc; Edward Walczak, MD, MBA; Alexander Kutikov, MD; Robert G. Uzzo, MD; Diego Aguilar Palacios, MD; Erick M. Remer, MD; Steven C. Campbell, MD, PhD; Nikolaos Papanikolopoulos, PhD; Christopher J. Weight, MD, MSc | Posted on: 01 May 2022

Heller N, Tejpaul R, Isensee F et al: Computer-generated R.E.N.A.L. nephrometry scores yield comparable predictive results to those of human-expert scores in predicting oncologic and perioperative outcomes. J Urol 2021; https://doi.org/10.1097/JU.0000000000002390.

Figure. Schematic diagram demonstrating AI-generated segmentation, subsequent coding to fully automated R.E.N.A.L. scores and its comparison to a human-generated R.E.N.A.L. score.

Study Need and Importance

Kidney cancer surgeons have long sought an unambiguous way to define kidney tumor complexity to aid in surgical planning and prediction of perioperative outcomes. Ten years ago, nephrometry scores were introduced and have provided valuable information for medical decision making; however, their widespread adoption, especially outside of academic centers, has been modest because of the time and expertise required to generate such scores. We hypothesized that a computer could generate a nephrometry score automatically.

What We Found

Deep-learning computer vision, a subfield of machine learning or artificial intelligence (AI), has shown promise in classifying and segmenting images. We created an international challenge within the deep-learning community, named KiTS19, to identify high-quality deep-learning algorithms that would segment renal computerized tomography scans and delineate renal masses. From these segmented images, we created a fully automated R.E.N.A.L. (for radius, exophytic/endophytic, nearness of tumor to collecting system, anterior/posterior, location relative to polar line) nephrometry score (AI-score) and compared its ability to predict meaningful perioperative and oncologic outcomes to human-expert–generated R.E.N.A.L. scores (H-score; see figure). We found a significant agreement between H-scores and AI-scores (Lin’s p=0.59). AI-scores performed similarly to H-scores in their ability to predict oncologic outcomes including presence of malignancy, high-grade disease, high-stage disease or necrosis (p <0.05). AI-scores were also helpful in predicting surgical approach, such as nephron sparing surgery and use of minimally invasive surgery (p <0.05). Meaningful perioperative outcomes were predicted by both AI- and H-scores including estimated blood loss, perioperative transfusion requirements (p <0.05) and change in estimated glomerular filtration rate after surgery (p <0.001).

“Deep-learning computer vision, a subfield of machine learning or artificial intelligence (AI), has shown promise in classifying and segmenting images.”

Limitations

Potential limitations include that a single center and a single phase of imaging were used. Furthermore, the AI-score could not be calculated in <2% of patients.

Interpretation for Patient Care

We created a fully automated, unambiguous kidney tumor complexity score that functions as well as scores created by human experts in predicting patient outcomes. We anticipate that the AI-score may serve as a support to patient–physician decision making.