Multisequence combined magnetic resonance imaging radiomics model to noninvasively predict nuclear grade of clear cell renal cell carcinoma: interpretable model development.

Objective: The nuclear grade of clear cell renal cell carcinoma directly relates to prognosis and is usually determined through invasive methods like biopsy or surgery. This study aimed to predict the nuclear grade of clear cell renal cell carcinoma using a noninvasive method: multisequence magnetic resonance imaging-based radiomics analysis.

Methods: A total of 42 clear cell renal cell carcinomas (29 low grade, 13 high grade) were included in the study. T2, fat-suppressed T2, noncontrast T1, corticomedullary phase, nephrographic phase, excretory phase, and apparent diffusion coefficient sequences of patients were used for radiomics analysis. Inter-observer agreement was assessed for these sequences, and following reproducibility analysis and feature selection, three new groups were formed: noncontrast enhancement, contrast enhancement, and combined groups, with different combinations of features extracted from these sequences. As a result, seven different sequences and three different groups constituted 10 classification groups. An extreme gradient boosting model was used for classification, employing 10-fold cross-validation.

Results: Radiomics features from corticomedullary phase and nephrographic phase sequences showed excellent inter-observer agreement, with Pearson correlation coefficient values of 0.88 for corticomedullary phase and 0.90 for nephrographic phase. The study included 42 clear cell renal cell carcinomas with a mean age of 60.8 years. Individually, the corticomedullary phase sequence achieved the highest area under the curve and accuracy values (0.88 and 0.85), followed by the apparent diffusion coefficient sequence (0.87 and 0.79). In the combined sequence group, the contrast enhancement group showed the highest area under the curve and accuracy (0.93 and 0.87), ranking highest across the entire study.

Conclusion: Multisequence magnetic resonance imaging radiomics has great potential to predict the nuclear grade of clear cell renal cell carcinoma and guide the treatment plan noninvasively.