Multiscale connected chain topological modelling for microcalcification classification.

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Biomedical Engineering

Background:

• Mammographic microcalcification clusters are key indicators in breast cancer diagnosis.
• Existing computer-aided diagnosis (CAD) systems primarily rely on morphology and texture analysis.
• A need exists for novel classification methods that capture spatial relationships within microcalcification clusters.

Purpose of the Study:

• To introduce a novel method for classifying mammographic microcalcification clusters based on topology and connectivity.
• To evaluate the effectiveness of a multiscale topological approach in distinguishing benign from malignant clusters.
• To compare the proposed method against existing techniques focusing on morphology and texture.

Main Methods:

• Developed a multiscale approach analyzing the morphological relationship of connectivity between microcalcifications.
• Generated connected chains between nearest microcalcifications at each scale.
• Extracted graph connectivity features to estimate topological structure for classification.

Main Results:

• Achieved classification accuracies of 86.47% (DDSM), 90.00% (MIAS-manual), 82.50% (MIAS-auto), and 76.75% (OPTIMAM) using KNN.
• Demonstrated that topological/connectivity modeling is suitable for microcalcification cluster analysis.
• Linked topological connectivity and distribution to clinical understanding of spatial distribution.

Conclusions:

• The proposed multiscale topological/connectivity modeling is appropriate for microcalcification cluster analysis and classification.
• Topological features provide valuable insights into the spatial distribution of microcalcifications, aiding diagnosis.
• This novel approach offers a promising alternative to morphology- and texture-based CAD systems.