Breast Tumour Classification Using Ultrasound Elastography with Machine Learning: A Systematic Scoping Review

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Ultrasound elastography quantifies tissue stiffness, aiding breast cancer screening alongside B-mode ultrasound.
• Machine learning (ML) and deep learning (DL) enhance computer-aided diagnosis for automated segmentation and tumor classification in ultrasound elastography.

Purpose of the Study:

• To review the application of machine learning models in ultrasound elastography for breast tumor classification.

Main Methods:

• A systematic review of four databases (PubMed, Web of Science, CINAHL, EMBASE) identified 13 eligible articles.
• Studies utilized either shear-wave elastography (SWE) or strain elastography (SE).
• Methods included traditional computer vision workflows (segmentation, feature extraction, classification with SVMs) and deep learning models (CNNs).

Main Results:

• All reviewed studies achieved sensitivity ≥ 80%; however, only 50% reached acceptable specificity ≥ 95%.
• Deep learning models did not consistently outperform traditional computer vision workflows.
• Significant inconsistencies were noted in reporting crucial details like dataset specifics, cross-validation, and overfitting prevention.

Conclusions:

• Current machine learning applications in ultrasound elastography for breast tumor classification show promise but require standardized reporting and methodology.
• Future research should explore advanced deep learning architectures, such as attention mechanisms, and online training for improved performance and adaptability.