Small Renal Masses: Developing a Robust Radiomic Signature

Abstract

(1) Background and (2) Methods: In this retrospective, observational, monocentric study, we selected a cohort of eighty-five patients (age range 38-87 years old, 51 men), enrolled between January 2014 and December 2020, with a newly diagnosed renal mass smaller than 4 cm (SRM) that later underwent nephrectomy surgery (partial or total) or tumorectomy with an associated histopatological study of the lesion. The radiomic features (RFs) of eighty-five SRMs were extracted from abdominal CTs bought in the portal venous phase using three different CT scanners. Lesions were manually segmented by an abdominal radiologist. Image analysis was performed with the Pyradiomic library of 3D-Slicer. A total of 108 RFs were included for each volume. A machine learning model based on radiomic features was developed to distinguish between benign and malignant small renal masses. The pipeline included redundant RFs elimination, RFs standardization, dataset balancing, exclusion of non-reproducible RFs, feature selection (FS), model training, model tuning and validation of unseen data. (3) Results: The study population was composed of fifty-one RCCs and thirty-four benign lesions (twenty-five oncocytomas, seven lipid-poor angiomyolipomas and two renal leiomyomas). The final radiomic signature included 10 RFs. The average performance of the model on unseen data was 0.79 ± 0.12 for ROC-AUC, 0.73 ± 0.12 for accuracy, 0.78 ± 0.19 for sensitivity and 0.63 ± 0.15 for specificity. (4) Conclusions: Using a robust pipeline, we found that the developed RFs signature is capable of distinguishing RCCs from benign renal tumors.

Keywords: benign; characterization; kidney cancer; malignant; oncocytoma; radiomics; renal cell carcinoma; small renal masses.

Figure 1

Segmentation (in red) of a benign (a) and a malignant (b) small renal mass (SRM) hardly distinguishable on portal venous phase CT images.

Figure 2

Whole analysis pipeline to develop the radiomics signature, including CT images segmentation, ROI analysis with Slicer Software, features selection and the model training and validation. (a) Generic overview of the whole pipeline, from CT image segmentation to model training and testing, (b) Detailed focus on the two steps of the machine learning method, i.e., the feature selection and the model training. The employed classifier was kNN: each patient was represented by a point in the feature space and it was classified by the algorithm based on its fist k neighbors and on the Euclidean distance with respect to each of them. In the example of figure (b), the patient marked with ★ is compared to the three (k = 3) closer patients which could have either a benign (marked with •) or a malign (marked with •) lesion.

Figure 3

Heatmap showing the correlation matrix among radiomic features.

Figure 4

Most scored features after 100 rounds of MCCV. The cumulative score was calculated as the sum of the scores of all MCCV rounds.

Figure 5

Optimization of model parameters.

Figure 6

PCA (2) + 7−NN model explainability: patients have been represented as points in a 2D features space and class membership has been proven based on neighbors’ points and their reciprocal distances.