Microcalcification detection based on wavelet domain hidden markov tree model: study for inclusion to computer aided...

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Statistical Modeling

Background:

• Digital mammography is crucial for breast cancer screening.
• Accurate detection of microcalcifications (MCs) and microcalcification clusters (MCCs) is vital for early diagnosis.
• Computer-aided diagnosis (CAD) systems can enhance radiologist performance.

Purpose of the Study:

• To evaluate a statistical modeling approach using wavelet domain hidden Markov trees for microcalcification detection in digital mammograms.
• To integrate this model into a computer-aided diagnostic prompting system.
• To assess the system's performance in segmenting and classifying microcalcifications and microcalcification clusters.

Main Methods:

• Utilized wavelet domain hidden Markov trees for statistical modeling of digital mammograms.
• Employed a maximum likelihood classifier with a weighting technique for image segmentation.
• Incorporated spatial filtering for single microcalcification (MC) and microcalcification cluster (MCC) detection.
• Applied contrast filtering to the Digital Database for Screening Mammography (DDSM) dataset prior to spatial filtering.

Main Results:

• The system demonstrated high accuracy in detecting MC clusters from the mini-MIAS database (92.5%-100% true positives with 2-3 false positives per image).
• For the DDSM dataset, the system achieved up to 98% true positive detection with 3.3% false positives.
• Contrast filtering significantly improved classification accuracy.

Conclusions:

• The developed statistical modeling approach is effective for detecting microcalcification clusters in digital mammograms.
• The computer-aided diagnostic system shows promise in improving the accuracy and efficiency of mammogram analysis.
• The system's performance suggests its potential utility in assisting radiologists with breast cancer screening.