Quantitative ultrasound spectroscopy and a kernel-based metric in clinical cancer response monitoring

Abstract

In this study, a metric based on Hilbert-Schmidt independence criterion (HSIC) is introduced in conjunction with quantitative ultrasound (QUS) spectroscopy methods for cancer response monitoring in locally advanced breast cancer (LABC) patients receiving neoadjuvant chemotherapy. Midband fit spectral parametric maps were computed using QUS radiofrequency data, which were obtained from 56 LABC patient before treatment and at three different times during the course of chemotherapy, i.e., on weeks 1, 4, and 8. Histograms of intensities were computed using 2D parametric maps to represent the images. Subsequently, the baseline features, i.e., the features extracted from “pre-treatment” parametric maps, were compared with those extracted from the parametric maps during the course of treatment using a kernel-based metric as an indication of chemotherapy effectiveness. As a result, dissimilarity measures were obtained between “pre-” and “during-treatment” images, which were used in a supervised learning paradigm to estimate whether a patient is a responder or a non-responder. High accuracy, sensitivity, and specificity were obtained on weeks 1 and 4, which demonstrated that the proposed system can effectively discriminate between the two patient populations early after start of treatment.