Predicting Pathological Complete Response to Neoadjuvant Therapy in Breast Cancer Using Contrast-Enhanced Ultrasound: A Retrospective Study

2. Materials and Methods

2.1 Study Population

This study retrospectively enrolled 212 consecutive breast cancer patients who received NAT at the First Affiliated Hospital of Sun Yat-sen University between May 2017 and April 2022. Inclusion criteria were: (1) biopsy-confirmed breast cancer before NAT; (2) completion of NAT followed by surgery for gross pathological assessment; and (3) baseline and at least one follow-up CEUS examination performed at either the early (after 2 cycles of NAT), middle (after 4/5/6 cycles of NAT), or pre-operative stage after completion of NAT. Exclusion criteria included: (1) primary breast lesion excision before NAT; (2) occult breast cancer; (3) prior traditional Chinese medicine or endocrine therapy; (4) recurrent breast cancer; (5) incomplete imaging data; and (6) treatment discontinuation for over six months. Ethical approval was obtained from our institution's Ethics Committee.

2.2 Neoadjuvant Treatment Regimen

Patients received one of the following NAT regimens: (1) 6 cycles of taxanes with or without carboplatin and anti-HER2 targeted therapy (T (CbHP)); (2) 4 cycles of anthracyclines and cyclophosphamide followed by 4 cycles of taxanes with or without anti-HER2 targeted therapy (EC-T (HP)); and (3) other treatment protocols. Early CEUS assessment was performed after 2 cycles of all regimens. Middle CEUS assessment was performed after 4 cycles for the T(CbHP) regimen and after 5 and 6 cycles for the EC-T(HP) regimen. Other chemotherapy regimens did not include a middle CEUS evaluation. A final pre-operative CEUS examination was performed after NAT completion (Figure 2-1).

2.3 Pathological Criteria

Pathological complete response (pCR) was defined as the absence of invasive cancer in the breast, permitting ductal carcinoma in situ, irrespective of axillary lymph node status [38]. HER2 positivity was defined as immunohistochemistry 3+ or positive in situ hybridization [45]. ER and PR positivity was defined as ≥1% positive cells on immunohistochemical staining [46].

2.4 Instrumentation and Methods

CEUS examinations were performed using Siemens ACUSON Sequoia Redwood (L10-4 linear array probe), Mindray Resona 7 (L9-3 linear array probe), Canon Aplio i900 TUS-A1900 (i18LX5 linear array probe), or Philips iU22 (L9-3 linear array probe) ultrasound systems. Real-time contrast-enhanced imaging was performed with a mechanical index (MI) < 0.15. SonoVue (Bracco, Italy), a lyophilized sulfur hexafluoride microbubble contrast agent (59mg, mean diameter 2.5μm; 90% <6μm, 99% <11μm), was reconstituted with 5ml saline and agitated for 30 seconds.

Three experienced breast imaging specialists (over 10 years of experience) conducted the ultrasound examinations. A blinded reader, unaware of clinical-pathological information, interpreted the images. Patients were positioned supine with arms abducted to expose the breast and axilla. Grayscale ultrasound evaluated lesion number, location, morphology, size, echogenicity, and margins. Color Doppler ultrasound assessed blood flow using the Adler grading system [47]: grade 0, no flow; grade I, minimal flow (1-2 punctate or thin rod-shaped signals); grade II, moderate flow (one major vessel or multiple small vessels); and grade III, abundant flow (over four vessels). CEUS was performed on sections with rich vascularity and partial normal gland tissue. Gentle pressure ensured optimal probe-breast contact. Enhancement level (high or no abnormal high enhancement), uniformity (uniform or non-uniform), necrosis (perfusion defects within the lesion), and crab-like branching of the enhancement margin were recorded.

Maximum diameter (MD) and anteroposterior diameter (APD) were measured on grayscale and CEUS images. For single lesions, both MD and APD were recorded. If the lesion was too large for accurate MD measurement, only APD was recorded. For multifocal lesions [3, 48] (two or more lesions in the same quadrant, inter-lesion distance <5cm), MD was the sum of all lesions, and APD was recorded for the largest lesion. For multicentric lesions [3, 48] (two or more lesions in different quadrants, inter-lesion distance >5cm), the largest lesion's MD and APD were recorded.

Tumor response was evaluated using the reduction rate relative to baseline at early, middle, and post-NAT stages. Reduction rate = (baseline size - early/middle/post-NAT size) / baseline size. Recorded parameters included grayscale maximum diameter reduction rate (BUS MDRR), grayscale anteroposterior diameter reduction rate (BUS APDRR), CEUS maximum diameter reduction rate (CEUS MDRR), and CEUS anteroposterior diameter reduction rate (CEUS APDRR).

Clinical and baseline pathology data included age, menopausal status, clinical TNM stage, NAT regimen, anti-HER2 therapy use, breast pathology type, Ki-67 index, ER status, PR status, and HER2 status.

2.5 Statistical Methods

Normally distributed continuous variables were compared using t-tests, while skewed data were compared using Wilcoxon rank-sum tests. Categorical variables were analyzed using chi-square or Fisher's exact tests, as appropriate. Univariate analyses were performed to assess differences between pCR and non-pCR groups for all variables. Variables with p<0.10 were included in a backward elimination multivariate logistic regression analysis to construct three models: clinical-pathological, ultrasound, and combined. Model discrimination was assessed using the area under the curve (AUC) and 95% confidence intervals (CI). Calibration was assessed using calibration slope, calibration plot, and the Hosmer-Lemeshow test. A slope of 1 indicated perfect calibration, while values <1 suggested potential overfitting. The Hosmer-Lemeshow test (p>0.05) indicated good model fit, suggesting no significant difference between predicted and observed values. Internal validation was performed using the bootstrap method. The early multifactorial model included only patients with complete data (n=133). For the middle and post-NAT multifactorial models, MDRR was excluded due to missing data, and only APDRR was included. Analyses were performed using SPSS 25.0, MedCalc 20.0.10, and R version 4.1.2.