Comparison of methods for texture analysis of H-scan ultrasound images from breast cancer patients undergoing neoadjuvant chemotherapy

H-scan ultrasound (US) imaging is a tissue characterization technique that depicts the relative size of acoustic scatterers. A total of nice patients diagnosed with locally advanced breast cancer and scheduled to receive neoadjuvant chemotherapy (NAC) were enrolled in this study. At baseline and again after completion of 10 and 30% neoadjuvant chemotherapy, imaging was performed using a clinical US scanner (Logiq E9, GE Healthcare) equipped with an 9L-D linear array transducer. Beamformed radiofrequency (RF) data was acquired at a center frequency of 9 MHz and saved for offline analysis. Two digital filters were constructed using 2
^nd
and 8
^th
-order Gaussian-weighted Hermite polynomial functions and convolved in parallel with each RF data sequence to measure the relative strength of the backscattered US signals. The signal envelope for each filter output was then calculated and color-coded to form the final H-scan US image. Textural analysis and eight features derived from the H-scan US images were used to assess various spatial relationships between pixel intensities. H-scan US image and texture parameters were linearly combined using support vector machines to quantify the ability to separate tumor response. Pathology reports indicated that all cancers that responded to NAC exhibited reduced cellularity content. In addition to changes in mean H-scan US image intensity, texture feature maps also progressively changed during the NAC treatment period suggesting differences in US scattering from responding and nonresponding tumor types.