Classification of breast tumor models with a prototype microwave imaging system.

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Machine Learning

Background:

• Accurate breast tumor size and shape assessment is crucial for cancer diagnosis and staging.
• Heterogeneity in breast tissue presents challenges for imaging and classification.

Purpose of the Study:

• To classify breast tumor models of varying sizes and shapes using ultra-wideband radar microwave imaging.
• To evaluate the effectiveness of machine learning algorithms for benign vs. malignant tumor classification.
• To investigate the impact of breast phantom heterogeneity and skin artifacts on classification performance.

Main Methods:

• Created a database of 26 tumor models (13 benign, 13 malignant) using tissue-mimicking materials.
• Imaged tumor models within homogeneous and heterogeneous breast phantoms using a monostatic microwave imaging system (1-6 GHz).
• Employed Principal Component Analysis for feature extraction and tuned Naïve Bayes, decision trees, and k-nearest neighbours classifiers.

Main Results:

• The k-nearest neighbours (kNN) classifier achieved the highest accuracy at 96.2% for benign vs. malignant tumor classification.
• kNN outperformed decision trees (DT) and Naïve Bayes (NB) classifiers.
• Classification accuracy remained high even when considering synthetic skin response.

Conclusions:

• Successfully classified benign and malignant tumor models using a microwave imaging system and machine learning.
• Demonstrated a methodology with potential for assessing breast tumor shape, aiding diagnosis and staging.
• Optimized classification algorithms and antenna positions for improved diagnostic accuracy.