Related Experiment Video
Updated: Sep 11, 2025
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
Machine Learning for Monoclonal Gammopathies of Renal Significance Risk Stratification Using Clinical and Pathology
Giorgio Cazzaniga1, Giulia Capitoli2,3, Raffaella Barretta1
1Department of Medicine and Surgery, Pathology, University of Milano-Bicocca, IRCCS Fondazione San Gerardo dei Tintori, Monza, Italy.
Introduction:
The prompt detection of monoclonal gammopathies of renal significance (MGRS) is clinically relevant for the initiation of chemotherapy. Although clinical and laboratory data can suggest the presence of MGRS, renal biopsy still represents the gold standard, despite not always being performed or eventually postponed because it is not deemed useful or informative. In this retrospective study, machine learning (ML) is used to build a tool to assist the prebiopsy risk stratification of MGRS and reinforce the rationale for the histological examination.
Methods:
The study included a total of 258 patients with monoclonal gammopathy of undetermined significance, of which 168 MGRS cases (65%) and 90 non-MGRS cases (35%) based on the final renal biopsy result.
Results:
Patients with MGRS were more frequently female (48.2% vs. 28.9%, P = 0.004) and had more frequent Bence Jones proteinuria (73.2% vs. 41.1%, P < 0.001) than non-MGRS cases, with amyloidosis being the most common diagnosis (62%). The ML model achieved an accuracy of 0.79 (95% confidence interval [CI]: 0.67-0.88) and an area under the receiver operating characteristic curve of 0.80 (95% CI: 0.65-0.93) in the validation set. The model is available as a free desktop and Android mobile app (MGRS Interactive Resource for AI-Guided Evaluation, MIRAGE).
Conclusion:
The ML-based MGRS risk stratification tool can help in the selection of patients with higher probability to have MGRS on renal biopsy, aiming to direct those patients to a complete histological characterization of the disease.
