723: AI-Based Renal Lesion Profiling for Personalized Surgery- Daniele Amparore
This is one of the entries accepted for adjudication for the 2026 KS Innovation Awards, sponsored by the Vattikuti Foundation. Posting here does not imply being a Finalist.
Here is the Abstract:
Al-Based Renal Lesion Profiling for Personalized Surgery
Presenting Author: Daniele Amparore, MD, PhD, FEBU
Division of Urology, Department of Surgery
INOC – Istituto Nazionale Oncologico Candiolo, IRCCS
Candiolo (TO), Italy
Phone: +39 333 3352758
Fax: +39 011 9026244
E-mail: daniele.amparore@unito.it
Contributors: 2F. Piramide, 2A. Quarà, 1G. Bignante, 1A. La Gatta, 1M. Ortenzi, 2S. De Cillis,
3A. Simeri, 1E. Checcucci, 2C. Fiori, 1F. Porpiglia
1Department of Surgery National Oncology Institute of Candiolo (INOC) – IRCCS Candiolo, Turin,
Italy
2Department of Urology, San Luigi Gonzaga Hospital, University of Turin, Italy
3Department of Mathematics and Computer Science, University of Calabria, Rende, Cs, Italy
Purpose
3D virtual models (3DVMs) are increasingly used in robotic-assisted partial nephrectomy (RAPN), improving surgical planning and precision. However, current models provide mainly morphological information and limited insights into tumor biology. We developed an artificial intelligence (AI)- based system integrating imaging and clinical data to generate a biological and morphological tumor profile.
Methods
A computer vision system was trained on preoperative metadata and DICOM CT-scan images to predict renal tumor characteristics. The AI integrates four CT phases, tracking the same anatomical
voxel across the study and analyzing contrast enhancement dynamics, tissue composition, and structural features. The system generates a tumor report including predicted biological behavior and probability across three clinical risk categories. The report is incorporated into 3D virtual models for surgical planning. The model was trained on 100 patients and tested on 20 patients, achieving an AUC of 0.84 and 0.79, respectively.
Results
Fifteen patients underwent AI-assisted RAPN. Median tumor size was 47.7 mm, with 60% classified as cT1b. The AI system generated detailed tumor profiling, predicted malignancy in 93.3% of cases, and classified tumors into risk categories. Final histopathology showed clear cell carcinoma (53.3%), papillary carcinoma (20%), angiomyolipoma (13.3%), and oncocytoma
(13.3%). Agreement between AI predictions and pathology was 80% for malignancy identification and 73% for histological subtype.
Conclusion
Our AI-based computer vision system provides a detailed analysis of renal tumor features, enhancing standard imaging with biologically relevant information. In this preliminary experience, The system is a promising step toward personalized surgical planning in RAPN.