Multiparametric ultrasound image-derived analysis for monitoring breast cancer response to chemotherapy

A major challenge in monitoring neoadjuvant chemotherapy (NAC) is how to detect and accurately quantify tumor response at an early stage of therapy. Based on the principle of ultrasound (US) scattering from different classes of tissue, the H-scan analysis has evolved. Preclinical results have demonstrated the ability of H-scan US imaging to detect early tumor response to chemotherapy in animal models. In this research, we compared the role of both B-scan and H-scan US imaging integrated with texture analysis, envelope statistics, and tumor shape markers for assessing breast cancer response to NAC in human subjects. Twenty-two patients underwent US scans before and after 10 % NAC using a clinical US scanner equipped with a 9L-D transducer. After data acquisition, a series of Gaussian filters were employed on the US data, to determine a close match to the localized tissue scattering function, and then the matched filter order were color-coded to form H-scan US images. The filters were centered equally between 3 and 9 MHz. Other parameters, such as Nakagami, texture, and tumor convexity features, were also assessed. All possible combinations of US parameters underwent dimensionality reduction using principal component analysis and then first component was used for predictive analysis through support vector machines, and an area under the curve (AUC) was reported. The data classification indicated that using optimized US combination yielded an higher AUC when H-scan US images utilized rather than B-scan US images (0.87 vs. 0.76).