Analysis of elastographic and B-mode features at sonoelastography for breast tumor classification

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Breast tumor classification relies on accurate imaging techniques.
• Ultrasound elastography provides tissue stiffness information.
• Distinguishing benign from malignant breast masses remains a clinical challenge.

Purpose of the Study:

• To assess the accuracy of neural network analysis of elastographic features for classifying breast tumors.
• To compare the diagnostic performance of elastographic and B-mode ultrasound features.
• To evaluate the combined utility of both feature sets in tumor classification.

Main Methods:

• Sonoelastography was performed on 181 solid breast masses.
• Manual segmentation allowed computation of five elastographic and six B-mode features.
• A neural network classifier was trained and validated using these features.
• Statistical analysis included Student's t test and ROC curve analysis.

Main Results:

• Elastographic features (mean, median, mode) showed significantly higher Area under ROC curve (Az) values (0.83-0.87) compared to B-mode features (0.54-0.69).
• The neural network achieved 86.2% accuracy using elastographic features and 82.3% using B-mode features.
• Combining elastographic and B-mode features with neural network analysis yielded the highest accuracy (90.6%) and Az (0.92).

Conclusions:

• Sonoelastography, analyzed by neural networks, shows significant potential for improving breast tumor classification accuracy.
• Elastographic features are more discriminative than B-mode features for differentiating benign and malignant breast tumors.
• The integration of elastographic and B-mode features offers a powerful tool for enhanced ultrasound-based breast lesion assessment.