Image-based morphometric scoring systems such as the RENAL and PADUA scores are useful to evaluate the complexity of partial nephrectomy for renal cell carcinoma (RCC). The main aim of this study was to develop a new imaging software to enable an automatic detection and a 3D visualization of RCC from CT angiography (CTA) and to address the feasibility to use it to evaluate the features of the RENAL and the PADUA scores. A training dataset of 210 patients CTA-scans manually segmented was used to train a deep learning algorithm to develop the automatic detection and 3D-visualization of RCC. A trained operator blindly assessed the RENAL and PADUA scores on a testing dataset of 41 CTA from patients with RCC using a commercialized semi-automatic software (ground truth) and the new automatic software. Concordance between the two methods was evaluated. The median PADUA score was 9 (7-11) and the renal score was 8 (5.5-9). The automatic software enabled to automatically detect the tumoral kidney and provided a 3D-visualization in all cases, with a computational time less than 20 seconds. Concordances for staging the anatomical features of the RENAL scores were respectively: 87.8% for radius, 85.4% for exophytic rate, 82.9% for location to the polar lines and 92.7% for the antero-posterior location. For the PADUA scores, concordances were 90.2% for tumor size, 85.4% for exophytic rate, 87.8% for polar location and 100% for renal rim. By enabling an automatic 3D-visualization of tumoral kidney, this software could help to calculate morphometric scores, save time and improve reproducibility for clinicians.
Read full abstract