Markov random field for tumor detection in digital mammography

Area of Science:

• Medical Imaging
• Artificial Intelligence in Radiology
• Biomedical Engineering

Background:

• Digital mammography is crucial for early breast cancer detection.
• Accurate segmentation and classification of suspicious regions are key challenges.
• Existing methods may struggle with subtle or small masses.

Purpose of the Study:

• To develop and evaluate a novel algorithm for automated tumor detection in digital mammography.
• To improve sensitivity and specificity in identifying malignant masses.
• To assess the algorithm's performance on minimal cancer detection.

Main Methods:

• Image segmentation using adaptive thresholding and a modified Markov random field (MRF) model.
• Classification of segmented regions using a fuzzy binary decision tree with radiographic and density features.
• Evaluation using free-response receiver operating characteristic (FROC) curves on a dataset of 95 mammograms (50 normal, 45 abnormal).

Main Results:

• Achieved 90% sensitivity for detecting various malignant masses with an average of two false positive alarms per image.
• Demonstrated high efficacy in detecting minimal cancers (masses ≤10 mm), with 94% sensitivity and 1.5 false alarms per image for 16 such cases.
• Extensive parameter analysis was conducted to optimize the algorithm for clinical application.

Conclusions:

• The proposed two-step algorithm (segmentation and classification) shows significant promise for accurate tumor detection in digital mammography.
• The method is particularly effective for identifying subtle and small malignant masses, potentially improving early diagnosis rates.
• Further optimization and clinical validation are warranted for widespread adoption in breast cancer screening.