A Decision-Support Tool for Renal Mass Classification

Abstract

We investigate the viability of statistical relational machine learning algorithms for the task of identifying malignancy of renal masses using radiomics-based imaging features. Features characterizing the texture, signal intensity, and other relevant metrics of the renal mass were extracted from multiphase contrast-enhanced computed tomography images. The recently developed formalism of relational functional gradient boosting (RFGB) was used to learn human-interpretable models for classification. Experimental results demonstrate that RFGB outperforms many standard machine learning approaches as well as the current diagnostic gold standard of visual qualification by radiologists.

Keywords: Clinical decision support; Multiphase CT; Radiomics; Renal mass; Statistical relational learning.

Fig. 1

RFGB learns tree models that can be easily explained (left), while SVM models (right) are far harder to interpret. RFGB’s decision can be explained using intuitive comparisons and conjunctions (AND), while the SVM learns a non-linear function. Both models use radiomics features: SQV and HOM. CECT phases: corticomedullary (C) and nephrographic (N) are also captured. Illustrative example only

Fig. 2

RFGB visualized

Fig. 3

Accuracy of various ML models (averaged over 10 runs) in renal mass classification compared to RFGB (bold numbers over the bars are significantly different from RFGB at p = 0.05)

Fig. 4

F-measures of various ML models (averaged over 10 runs) in renal mass classification compared to RFGB (bold numbers over the bars are significantly different from RFGB at p = 0.05). The F-measure is the harmonic mean of precision and recall

Fig. 5

Area under the receiver operator curve (AUC-ROC) of various ML models (averaged over 10 runs) in renal mass classification compared to RFGB (bold numbers over the bars are significantly different from RFGB at p = 0.05). The AUC-ROC is a measure of the probability that a classifier will rank a randomly chosen positive example (malignant) higher than a randomly chosen negative example (benign)