Computer Vision-Based Microcalcification Detection in Digital Mammograms Using Fully Connected Depthwise Separable...

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Microcalcification clusters in mammograms are key indicators of breast cancer.
• Detecting these microcalcifications is challenging due to their size, location, and breast density.
• Current computer-aided detection (CAD) systems require enhancement for early and accurate breast cancer prediction.

Purpose of the Study:

• To develop and evaluate a computer-vision-based CAD system for detecting and classifying microcalcification clusters in mammograms.
• To improve the true-positive rate and reduce false positives in breast cancer detection.
• To classify microcalcification clusters as malignant or benign using a deep learning approach.

Main Methods:

• A computer-vision-based framework utilizing a Fully Convolutional Depthwise Separable Convolutional Neural Network (FC-DSCNN) was proposed.
• The method involved image preprocessing, augmentation, RGB to grayscale conversion, and microcalcification region segmentation.
• Microcalcification regions of interest (ROIs) were classified into malignant and benign categories.

Main Results:

• The proposed FC-DSCNN CAD system achieved high performance on large mammogram datasets (DDSM and PINUM).
• The system demonstrated a high true-positive ratio (0.97 with 2.35 false positives and 0.99 with 2.45 false positives per image).
• The method outperformed traditional and previous approaches in detecting and classifying microcalcifications.

Conclusions:

• The developed computer-vision-based FC-DSCNN CAD system effectively detects and classifies microcalcification clusters in mammograms.
• This approach enhances classification performance by automatically managing noise and contrast.
• The system shows significant potential for improving early breast cancer diagnosis and supporting radiologists.