Fuzzy C-Means Clustering: A Review of Applications in Breast Cancer Detection

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Breast cancer diagnosis relies on mammograms, but human interpretation introduces errors and is resource-intensive.
• Automated mammogram analysis offers a promising supplement to traditional detection methods.
• Standard fuzzy c-means clustering (FCM) struggles with non-spherical tumour structures due to Euclidean distance limitations.

Purpose of the Study:

• To review fuzzy c-means clustering (FCM) and its Mahalanobis-distance-based variant (FCM-M) for image segmentation.
• To explore modifications in distance functions and centroid initialization to improve FCM performance.
• To develop and evaluate optimized FCM algorithms for breast tumour segmentation in mammograms.

Main Methods:

• Review of fuzzy c-means clustering (FCM), FCM with Mahalanobis distance (FCM-M), and three centroid initialization algorithms.
• Implementation and comparative analysis of these algorithms for image segmentation tasks.
• Development of a Python package for optimized FCM algorithms, intended for GitHub release.

Main Results:

• Centroid initialization algorithms significantly enhance the performance of basic FCM.
• Fuzzy c-means clustering with Mahalanobis distance (FCM-M) demonstrates superior clustering accuracy compared to standard FCM.
• FCM-M provides more precise outlining of tumour structures in mammograms.

Conclusions:

• Optimized FCM algorithms, especially FCM-M with appropriate initialization, improve the accuracy of breast tumour segmentation.
• Enhanced automated analysis of mammograms can aid in early breast cancer detection.
• The developed Python package aims to make these advanced segmentation tools accessible.