Persistent Homology-Based Machine Learning Method for Filtering and Classifying Mammographic Microcalcification...

Area of Science:

• Medical Imaging
• Computational Mathematics
• Biomedical Data Analysis

Background:

• Microcalcifications in mammograms are key indicators for early breast cancer detection.
• Image noise and dense tissues complicate accurate microcalcification classification.
• Current preprocessing methods can degrade image quality and detail.

Purpose of the Study:

• To introduce a new filtering and feature extraction technique using persistent homology (PH).
• To improve the classification accuracy of microcalcifications in mammograms.
• To address challenges posed by image noise and data complexity in cancer detection.

Main Methods:

• Applied persistent homology (PH) for filtering and feature extraction, operating on topological diagrams rather than raw image matrices.
• Utilized PH-derived diagrams to differentiate significant image features from noise.
• Vectorized filtered diagrams using PH features for input into machine learning models.

Main Results:

• Supervised machine learning models were trained on the MIAS and DDSM datasets.
• The efficacy of PH-extracted features in discriminating benign from malignant microcalcifications was evaluated.
• Optimal PH filtering levels were determined to enhance classification performance.

Conclusions:

• Persistent homology provides an effective method for noise reduction and feature extraction in mammogram analysis.
• Appropriate PH filtering and feature selection significantly improve classification accuracy for early breast cancer detection.
• This approach offers a promising alternative to traditional image preprocessing techniques.