Identifying Quantitative and Explanatory Tumor Indexes from Dynamic Contrast Enhanced Ultrasound

Contrast-enhanced ultrasound (CEUS) has been one of the most promising imaging techniques in tumor differential diagnosis since the real-time view of intra-tumor blood microcirculation. Existing studies primarily focus on extracting those discriminative imaging features whereas lack medical explanations. However, accurate quantitation of some clinical experience-driven indexes regarding intra-tumor vascularity, such as tumor infiltration and heterogeneity, still faces significant limitations. To tackle this problem, we present a novel scheme to identify quantitative and explanatory tumor indexes from dynamic CEUS sequences. Specifically, our method mainly comprises three steps: 1) extracting the stable pixel-level perfusion pattern from dynamic CEUS imaging using an improved stable principal component pursuit (SPCP) algorithm; 2) performing local perfusion variation comparison by the proposed Phase-constrained Wasserstein (PCW) distance; 3) estimating three clinical knowledge-induced tumor indexes, i.e. infiltration, regularity, and heterogeneity. The effectiveness of this method was evaluated on our collected CEUS dataset of thyroid nodules, and the resulting infiltration and heterogeneity index with \(p < 0.05\) between different pathological types validated the efficacy of this quantitation scheme in thyroid nodule diagnosis.