Radiomics Feature Activation Maps as a New Tool for Signature Interpretability

Abstract

IntroductionIn the field of personalized medicine, radiomics has shown its potential to support treatment decisions. However, the limited feature interpretability hampers its introduction into the clinics. Here, we propose a new methodology to create radiomics feature activation maps, which allows to identify the spatial-anatomical locations responsible for signature activation based on local radiomics. The feasibility of this technique will be studied for histological subtype differentiation (adenocarcinoma versus squamous cell carcinoma) in non-small cell lung cancer (NSCLC) using computed tomography (CT) radiomics.Materials and MethodsPre-treatment CT scans were collected from a multi-centric Swiss trial (training, n=73, IIIA/N2 NSCLC, SAKK 16/00) and an independent cohort (validation, n=32, IIIA/N2/IIIB NSCLC). Based on the gross tumor volume (GTV), four peritumoral region of interests (ROI) were defined: lung_exterior (expansion into the lung), iso_exterior (expansion into lung and soft tissue), gradient (GTV border region), GTV+Rim (GTV and iso_exterior). For each ROI, 154 radiomic features were extracted using an in-house developed software implementation (Z-Rad, Python v2.7.14). Features robust against delineation variability served as an input for a multivariate logistic regression analysis. Model performance was quantified using the area under the receiver operating characteristic curve (AUC) and verified using five-fold cross validation and internal validation. Local radiomic features were extracted from the GTV+Rim ROI using non-overlapping 3x3x3 voxel patches previously marked as GTV or rim. A binary activation map was created for each patient using the median global feature value from the training. The ratios of activated/non-activated patches of GTV and rim regions were compared between histological subtypes (Wilcoxon test).ResultsIso_exterior, gradient, GTV+Rim showed good performances for histological subtype prediction (AUCtraining=0.68–0.72 and AUCvalidation=0.73–0.74) whereas GTV and lung_exterior models failed validation. GTV+Rim model feature activation maps showed that local texture feature distribution differed significantly between histological subtypes in the rim (p=0.0481) but not in the GTV (p=0.461).ConclusionIn this exploratory study, radiomics-based prediction of NSCLC histological subtypes was predominantly based on the peritumoral region indicating that radiomics activation maps can be useful for tracing back the spatial location of regions responsible for signature activation.

Context-Aware Saliency Guided Radiomics: Application to Prediction of Outcome and HPV-Status from Multi-Center PET/CT Images of Head and Neck Cancer

“…Vuong et al [ 34 ] created a radiomics feature activation map in CT images of non-small-cell lung cancer by comparing whether the feature value from a small patch was higher than the population median feature value from the entire tumor or not, aiming to track the spatial location of regions responsible for signature activation. They found that the texture feature GLSZM_zone size non-uniformity normalized was more activated on the adjacent region of the tumor.…”

Development and External Validation of a Simple-To-Use Dynamic Nomogram for Predicting Breast Malignancy Based on Ultrasound Morphometric Features: A Retrospective Multicenter Study

“…Finally, when compared to other quantification-based nomograms ( 14 – 17 ), the predictor variables in our nomogram are easily accessed and interpreted. In general, lack of interpretability is one of the major barriers to successful translation of predictive models from research to clinical practice, particularly for data-driven precision medicine ( 20 ). From a clinical perspective, interpretability is critical for winning the trust of physicians, developing a robust decision-making system, and overcoming regulatory concerns ( 48 ).…”

“…These models have been developed to mine high-throughput quantitative image features fusing image pixels and morphology through machine learning methods to improve cancer diagnosis and prognosis ( 18 ). However, to varying degrees, reproducibility of quantification features derived from image pixels is sensitive to image preprocessing ( 19 ), particularly for US technology, which has the distinct inherent characteristic of operator- and device-dependent, not to mention that such pixel-based features often lack interpretability ( 20 ). This may lead to limitations in usability for real end-users, impeding their large-scale clinical applications.…”

Voxel‐wise supervised analysis of tumors with multimodal engineered features to highlight interpretable biological patterns

“…Beaumont et al 19 used a random forest approach to predict local recurrence from baseline images thanks to locally calculated features and voxel-wise ground truths. Vuong et al 20 investigated patch-based radiomics with binary activation for tracing the spatial location of regions responsible for a given classification. To the best of our knowledge, although engineered radiomics is largely used especially when datasets are not amenable to deep radiomics, no approach has been proposed to quantitatively map, at the voxel level, the output of a model based on engineered radiomic features.…”

“…used a random forest approach to predict local recurrence from baseline images thanks to locally calculated features and voxel‐wise ground truths. Vuong et al 20 . investigated patch‐based radiomics with binary activation for tracing the spatial location of regions responsible for a given classification.…”