Vincent Dong

PurposeBreast cancer risk depends on an accurate assessment of breast density due to lesion masking. Although governed by standardized guidelines, radiologist assessment of breast density is still highly variable. Automated breast density assessment tools leverage deep learning but are limited by model robustness and interpretability.ApproachWe assessed the robustness of a feature selection methodology (RFE-SHAP) for classifying breast density grades using tissue-specific radiomic features extracted from raw central projections of digital breast tomosynthesis screenings (nI=651, nII=100). RFE-SHAP leverages traditional and explainable AI methods to identify highly predictive and influential features. A simple logistic regression (LR) classifier was used to assess classification performance, and unsupervised clustering was employed to investigate the intrinsic separability of density grade classes.ResultsLR classifiers yielded cross-validated areas under the receiver operating characteristic (AUCs) per density grade of [A: 0.909±0.032, B: 0.858±0.027, C: 0.927±0.013, D: 0.890±0.089] and an AUC of 0.936±0.016 for classifying patients as nondense or dense. In external validation, we observed per density grade AUCs of [A: 0.880, B: 0.779, C: 0.878, D: 0.673] and nondense/dense AUC of 0.823. Unsupervised clustering highlighted the ability of these features to characterize different density grades.ConclusionsOur RFE-SHAP feature selection methodology for classifying breast tissue density generalized well to validation datasets after accounting for natural class imbalance, and the identified radiomic features properly captured the progression of density grades. Our results potentiate future research into correlating selected radiomic features with clinical descriptors of breast tissue density.

Objective: Diabetic macular edema (DME) and retinal vein occlusion (RVO) are the leading causes of visual impairments across the world. Vascular endothelial growth factor (VEGF) stimulates breakdown of blood-retinal barrier that causes accumulation of fluid within macula. Anti-VEGF therapy is the first-line treatment for both the diseases; however, the degree of response varies for individual patients. The main objective of this work was to identify the (i) texture-based radiomics features within individual fluid and retinal tissue compartments of baseline spectral-domain optical coherence tomography (SD-OCT) images and (ii) the specific spatial compartments that contribute most pertinent features for predicting therapeutic response. Methods: A total of 962 texture-based radiomics features were extracted from each of the fluid and retinal tissue compartments of OCT images, obtained from the PERMEATE study. Top-performing features selected from the consensus of different feature selection methods were evaluated in conjunction with four different machine learning classifiers: Linear Discriminant Analysis (LDA), Quadratic Discriminant Analysis (QDA), Random Forest (RF), and Support Vector Machine (SVM) in a cross-validated approach to distinguish eyes tolerating extended interval dosing (non-rebounders) and those requiring more frequent dosing (rebounders). Results: Combination of fluid and retinal tissue features yielded a cross-validated area under receiver operating characteristic curve (AUC) of 0.78±0.08 in distinguishing rebounders from non-rebounders. Conclusions: This study revealed that the texture-based radiomics features pertaining to IRF subcompartment were most discriminating between rebounders and non-rebounders to anti-VEGF therapy. Clinical Impact: With further validation, OCT-based imaging biomarkers could be used for treatment management of DME patients.