(B-74) Prediction of response to neoadjuvant chemotherapy in breast cancer using H-scan ultrasound images and texture analysis

Introduction:: Neoadjuvant chemotherapy (NAC) is a standard treatment for patients diagnosed with locally advanced breast cancer (LABC). However, monitoring NAC response at an early of therapy remains a major challenge. A new ultrasound (US)-based tissue characterization method termed H-scan US, which is based on the classification theory for tissue scattering, may provide the answer. In this research, we explored use of H-scan US and different image texture analysis methods for assessing the breast cancer response to NAC in human subjects.

Materials and Methods:: Patients scheduled to undergo NAC for LABC provided written informed consent and participated in this IRB-approved study at two clinical sites. In this ongoing clinical study, 12 patients underwent US exams at 0, 10, and 30% completion of NAC. At each time point, data was obtained using a clinical US scanner (Logiq E9, GE HealthCare) equipped with an 9L-D transducer. Beamformed radiofrequency (RF) data was acquired at 5 MHz. Two digital filters were constructed using 2nd and 8th-order Gaussian-weighted Hermite polynomial functions and convolved in parallel with the RF data to measure the relative strength of the backscattered US signals. The signal envelope for each filter output was color coded to form the final H-scan US image. Texture analysis methods were then applied to the H-scan US images to assess spatial relationships between neighboring pixel intensities, namely, contrast (CON), homogeneity (HOM), sum average (SA), sum variance (SV), energy (ENE), difference variance (DV), entropy (ENT), and correlation (COR). All tumor locations were identified by a Radiologist. Support vector machines (SVM) were used to classify the tumor response into non-responder (NR), partial responder (PR), and complete responder (CR), using the mean of the H-scan US images and texture maps. Finally, a leave-one-out analysis was performed to compute parameters such as NR prediction rate (specificity), PR+CR prediction rate (sensitivity), and area under the receiver operating characteristics curve (AUC).

Results, Conclusions, and Discussions:: Pathology reports categorized 2 patients as NR, 3 as PR, and 7 as CR. Within 30% completion of NAC, results reveal that H-scan US imaging can distinguish tumor response with a sensitivity and specificity of 90.0 % and 100 %, respectively (AUC = 0.95). However, use of the H-scan US image-derived CON, SV and DV texture features increased sensitivity and specificity to 100% (AUC = 1.0). The remaining features achieved AUC values in the 0.90 to 0.95 range. Overall, H-scan US imaging with texture analysis is a promising new approach for monitoring breast cancer response to NAC.

Acknowledgements (Optional): :

References (Optional): :