Breast Cancer Detection Using Multimodal Time Series Features From Ultrasound Shear Wave Absolute Vibro-Elastography.

Area of Science:

• Medical Imaging
• Biophysics
• Oncology

Background:

• Shear wave absolute vibro-elastography (S-WAVE) involves applying mechanical excitation to tissue and acquiring ultrasound data.
• Accurate classification of breast tissue lesions as malignant or benign is crucial for effective patient management.

Purpose of the Study:

• To evaluate the potential of S-WAVE for classifying breast tissue lesions.
• To introduce and validate a novel processing pipeline for S-WAVE data analysis.

Main Methods:

• A new processing pipeline was developed using bi-spectral and Wigner spectrum features from RF time series, combined with B-mode and elasticity image features.
• Feature reduction was performed using Random Forest permutation importance and Quadratic Mutual Information, reducing 377 features to 20.
• Classification was conducted using Support Vector Machines and Random Forest classifiers with leave-one-patient-out and Monte Carlo cross-validations on data from 40 patients.

Main Results:

• The developed S-WAVE pipeline achieved high classification performance: Area Under Curve of 95%±1.45%, 95% sensitivity, and 93% specificity.
• These results surpass previously reported state-of-the-art performance for S-WAVE breast cancer classification.
• The study also analyzed the impact of feature selection and lesion contour variations on classification accuracy.

Conclusions:

• Time-series analysis of vibrated tissue using S-WAVE shows significant potential for breast lesion classification, even without explicit elasticity computation.
• The proposed novel directions in elasticity imaging for tissue classification warrant further investigation with larger patient cohorts.