Automatic Breast Mass Segmentation and Classification Using Subtraction of Temporally Sequential Digital Mammograms.

Area of Science:

• Radiology and Medical Imaging
• Machine Learning in Healthcare
• Breast Cancer Diagnostics

Background:

• Breast cancer is a leading global health concern, with mammography crucial for early detection.
• Accurate classification of breast masses on mammograms is challenging due to image quality and tissue variations.
• Computer-Aided Diagnosis (CAD) systems aim to improve radiologist accuracy in breast abnormality detection.

Purpose of the Study:

• To develop and evaluate an automated system for breast mass segmentation and classification using temporal mammogram subtraction and machine learning.
• To enhance the accuracy of breast mass detection and improve the classification of masses as benign or suspicious.

Main Methods:

• Utilized subtraction of temporally sequential digital mammograms for mass analysis.
• Employed machine learning algorithms, including Neural Networks, for automated segmentation and classification.
• Evaluated performance on a dataset of 320 mammograms from 80 patients with radiologist-annotated mass locations.

Main Results:

• Achieved 99.9% accuracy in mass detection using Neural Networks.
• Improved classification accuracy of masses (benign vs. suspicious) from 92.6% to 98% compared to state-of-the-art temporal analysis.
• Demonstrated a statistically significant improvement (p < 0.05) in diagnostic performance.

Conclusions:

• Subtraction of temporally consecutive mammograms is an effective method for improving breast mass diagnosis.
• The proposed algorithm shows potential for developing advanced automated breast cancer CAD systems.
• This technology could significantly impact patient prognosis through earlier and more accurate breast cancer detection.