Breast DCE-MRI: lesion classification using dynamic and morphological features by means of a multiple classifier...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Accurate classification of breast lesions in MRI is crucial for diagnosis.
• Both lesion morphology and dynamic enhancement patterns are vital for differentiating benign from malignant cases.
• Current methods may benefit from integrated feature analysis.

Purpose of the Study:

• To develop and evaluate a multiple classifier system (MCS) for breast lesion classification.
• To combine morphological and dynamic features from dynamic contrast-enhanced MRI (DCE-MRI) for improved accuracy.
• To compare the MCS performance against individual classifiers and pathological classification.

Main Methods:

• A multiple classifier system (MCS) was developed, integrating two classifiers: one for morphological features and one for dynamic features.
• Feature selection was performed on an initial set of 54 morphological and 98 dynamic features.
• The system was trained and tested on 48 histologically proven breast lesions (26 malignant, 22 benign), with automatic and manual segmentation options.

Main Results:

• The MCS achieved 91.7% accuracy on the testing set using automatic segmentation.
• Combining decision tree (morphological) and Bayesian (dynamic) classifiers yielded optimal results.
• The MCS demonstrated significant accuracy improvements of 12.5% and 31.3% over individual classifiers.

Conclusions:

• A multiple classifier system effectively optimizes breast lesion classification accuracy.
• Integrating morphological and dynamic features in an MCS enhances diagnostic performance in DCE-MRI.
• The proposed MCS offers a promising tool for radiologists in breast lesion analysis.