Interpretability of radiomics models is improved when using feature group selection strategies for predicting molecular and clinical targets in clear-cell renal cell carcinoma: insights from the TRACERx Renal study

Abstract

Background The aim of this work is to evaluate the performance of radiomics predictions for a range of molecular, genomic and clinical targets in patients with clear cell renal cell carcinoma (ccRCC) and demonstrate the impact of novel feature selection strategies and sub-segmentations on model interpretability. Methods Contrast-enhanced CT scans from the first 101 patients recruited to the TRACERx Renal Cancer study (NCT03226886) were used to derive radiomics classification models to predict 20 molecular, histopathology and clinical target variables. Manual 3D segmentation was used in conjunction with automatic sub-segmentation to generate radiomics features from the core, rim, high and low enhancing sub-regions, and the whole tumour. Comparisons were made between two classification model pipelines: a Conventional pipeline reflecting common radiomics practice, and a Proposed pipeline including two novel feature selection steps designed to improve model interpretability. For both pipelines nested cross-validation was used to estimate prediction performance and tune model hyper-parameters, and permutation testing was used to evaluate the statistical significance of the estimated performance measures. Further model robustness assessments were conducted by evaluating model variability across the cross-validation folds. Results Classification performance was significant (p < 0.05, H0:AUROC = 0.5) for 11 of 20 targets using either pipeline and for these targets the AUROCs were within ± 0.05 for the two pipelines, except for one target where the Proposed pipeline performance increased by > 0.1. Five of these targets (necrosis on histology, presence of renal vein invasion, overall histological stage, linear evolutionary subtype and loss of 9p21.3 somatic alteration marker) had AUROC > 0.8. Models derived using the Proposed pipeline contained fewer feature groups than the Conventional pipeline, leading to more straightforward model interpretations without loss of performance. Sub-segmentations lead to improved performance and/or improved interpretability when predicting the presence of sarcomatoid differentiation and tumour stage. Conclusions Use of the Proposed pipeline, which includes the novel feature selection methods, leads to more interpretable models without compromising prediction performance. Trial registration NCT03226886 (TRACERx Renal)

Evaluating Outcome Prediction via Baseline, End-of-Treatment, and Delta Radiomics on PET-CT Images of Primary Mediastinal Large B-Cell Lymphoma

“…Determining the interpretability of radiomics features is somewhat subjective; for instance, features linked to tumor shape can be easily comprehended by non-experts, while those rooted in voxel grey-level distributions necessitate specialist knowledge, and texture-related features are challenging to understand, even for image processing experts [ 61 ]. Based on our findings, ( Table S1 ), the most predictive Delta radiomics features in our study are: PET features including GLSZM (Normalized grey level non-uniformity), GLCM (difference average, correlation, normalized inverse difference moment), GLRLM (grey level non-uniformity), Intensity histogram (root mean square and uniformity), local intensity based (coefficient of variation, intensity peak discretized volume sought), CT features including GLSZM (zone size entropy), GLCM (Sum Entropy, Zone size entropy), local intensity based (intensity peak discretized volume sought), and the innermost RIM features in PET image.…”

A CT-based radiomics classification model for the prediction of histological type and tumour grade in retroperitoneal sarcoma (RADSARC-R): a retrospective multicohort analysis

“… 25 We applied a recently developed pipeline designed to improve model interpretability compared with standard radiomics pipelines. 16 Furthermore, our novel sub-segmentation method attempted to address the often heterogeneous radiological phenotypes of retroperitoneal sarcoma seen on CT and our method using fixed thresholds for sub-region segmentation makes it more likely to be reproducible. 26 …”

“…A recently developed machine learning pipeline was used, which is designed to discover models that are easier to interpret 16 ( appendix pp 3–5, 16 ). The pipeline is based on a nested cross-validation structure, where the outer cross-validation provides performance estimates (ie, area under the receiver operator curve [AUROC]), and the inner cross-validation is used for tuning variable optimisation.…”

“…Further details are in the appendix (pp 3–5) . 16 The pipeline included a step that removed correlated features by comparing pairwise feature correlations (Spearman correlation) with a threshold, and this threshold was optimised, as described in the appendix (pp 3–5) .…”

Clear Cell Renal Cell Carcinoma: A Test Bench for Investigating Tumor Complexity

“…Magnetic resonance imaging (MRI) and positron emission tomography (PET)/CT using different radiolabeled molecules such as 18 F-fluorodeoxyglucose, 124 I-cG250, radiolabeled prostate-specific membrane antigen, and 11 C-acetate, together with a computational approach to CT images, are unveiling some data; however, in the opinion of the authors, these data still require standardization and external validation before their integration into clinical practice. In this sense, recent research proposes two new feature selection strategies in the interpretation of radiomics studies to predict molecular and clinical targets in CCRCC [17].…”