Breast lesion detection and characterization with 3D features

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Oncology

Background:

• Automated Breast Ultrasound (ABUS) is a valuable tool for breast cancer screening.
• The large data volumes generated by ABUS can be challenging for manual analysis.
• Automated analysis can enhance diagnostic efficiency and accuracy.

Purpose of the Study:

• To develop and evaluate a fully automatic algorithm for lesion detection and characterization in ABUS 3D volumes.
• To compare the effectiveness of various region descriptors for lesion analysis.
• To assess the algorithm's performance in differentiating malignant lesions from other masses.

Main Methods:

• Implementation of a generic algorithm pipeline for lesion detection and characterization.
• Computation of region descriptors on multiple feature images at candidate lesion locations.
• Utilizing Random Forests classifier for evaluating candidate region descriptors.
• Categorization of detected lesions as malignant or other masses (e.g., cysts).

Main Results:

• Achieved an Area Under the Curve (AUC) of 92.6% for lesion detection.
• Achieved an AUC of 89% for lesion characterization (malignant vs. other masses).
• The approach obviates the need for explicit lesion segmentation by using region descriptors.

Conclusions:

• The developed automated algorithm pipeline demonstrates high efficacy for lesion detection and characterization in ABUS.
• This automated approach can significantly improve the efficiency and accuracy of breast cancer screening using ABUS.
• The findings support the integration of automated analysis into clinical workflows for enhanced breast lesion diagnosis.