Automatic Pectoral Muscle Removal and Microcalcification Localization in Digital Mammograms

Area of Science:

• Medical Imaging
• Radiology
• Artificial Intelligence in Healthcare

Background:

• Microcalcification (MCC) clusters in mammograms are vital early indicators of breast cancer.
• Accurate detection of MCCs is critical for timely breast cancer diagnosis.

Purpose of the Study:

• To develop and evaluate a novel methodology for mammogram preprocessing and MCC detection.
• To enhance the accuracy of identifying potential breast cancer indicators.

Main Methods:

• Mammogram preprocessing involved automatic artifact deletion and pectoral muscle removal using region-growing segmentation and polynomial contour fitting.
• Microcalcification detection utilized a convolutional neural network for region-of-interest (ROI) classification, complemented by morphological operations and wavelet reconstruction to minimize false positives (FPs).

Main Results:

• The methodology achieved high accuracy in breast segmentation (99% on mini-MIAS, 97% on UTP) and pectoral segmentation (95% on mini-MIAS, 91% on UTP).
• MCC detection demonstrated a sensitivity of 82% (mini-MIAS) and 78% (UTP), with low FP rates per image (3.27 on mini-MIAS, 0.74 on UTP).

Conclusions:

• The proposed preprocessing method surpasses existing techniques for breast segmentation and shows promising results in pectoral muscle removal.
• The MCC detection module demonstrated superior test accuracy in identifying ROIs with MCCs compared to alternative methods.