Computerized detection of breast masses in digitized mammograms

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Radiology

Background:

• Mammography is crucial for breast cancer screening.
• Accurate detection of malignant masses is essential for early diagnosis.
• Reducing false positives in mammographic interpretation remains a challenge.

Purpose of the Study:

• To develop and evaluate a novel system for detecting malignant masses on mammograms.
• To assess the efficacy of an iris filter and adaptive thresholding for region segmentation.
• To investigate the utility of a backpropagation neural network for reducing false positives.

Main Methods:

• An iris filter was applied to mammograms at various scales.
• Suspicious regions were segmented using an adaptive threshold.
• Features including iris filter output, gray level, texture, contour, and morphology were extracted.
• A backpropagation neural network classifier was trained for false positive reduction.
• The system was evaluated on two independent datasets using free-response receiver operating characteristic analysis.

Main Results:

• The system achieved a sensitivity of 88% for lesion-based evaluation.
• A sensitivity of 94% was obtained for case-based evaluation.
• Performance was evaluated at 1.02 false positives per image.

Conclusions:

• The proposed system demonstrates significant potential for detecting malignant masses in mammograms.
• The method shows promise as a supplementary tool for radiologists in mammographic screening.
• The integration of iris filter, feature extraction, and neural network classification contributes to improved detection accuracy.