Reproducibility and reliability of renal image segmentation and radiomics: the AI4PKD project

Project “AI-based methods to improve stratification of patients affected by Polycystic Kidney Disease using multiparametric MRI” (AI4PKD), funded within the PRIN 2022 program, aims at discovering MRI-based biomarkers to identify ADPKD stage, based on robust and reliable AI-based methods for the automatic analysis of Magnetic Resonance Images (MRI) from patients with Autosomal Dominant Polycystic Kidney Disease (ADPKD). This work presents the first two activities of the project, focused on developing automatic segmentation tools and radiomic analysis pipelines that may guarantee reproducibility and reliability of the analyses, which are essential for trustable and reliable application in the clinical field. On the one hand, we quantified the impact of segmentation and scan-rescan variability on radiomics reproducibility using T2-weighted kidney MRI scans. On the other hand, we assessed the reliability of radiomic features extracted from anatomical and diffusion MRI images of ADPKD patients. These analyses allowed us to identify the best setting to ensure robustness and reliability of the analyses for future extraction of the biomarker