Application of Semi-supervised Fuzzy Clustering Based on Knowledge Weighting and Cluster Center Learning to Mammary...

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Mammography is crucial for breast cancer diagnosis.
• Accurate segmentation of lesions in mammograms aids radiologists and reduces workload.
• Obtaining large, labeled medical image datasets is challenging, necessitating unsupervised or semi-supervised methods.

Purpose of the Study:

• To develop an advanced semi-supervised fuzzy clustering algorithm for improved mammography image segmentation.
• To address the limitations of traditional unsupervised and semi-supervised clustering algorithms in handling complex mammographic data.
• To enhance the accuracy of segmenting both large and small lesion areas in mammograms.

Main Methods:

• A novel semi-supervised fuzzy clustering algorithm named WSFCM_V was developed.
• WSFCM_V incorporates knowledge weighting and cluster center learning strategies.
• Three learning modes were proposed: knowledge weighting for cluster centers, Euclidean distance weights for unlabeled samples, and learning from labeled sample set cluster centers.

Main Results:

• The WSFCM_V algorithm demonstrated superior performance compared to existing unsupervised and semi-supervised clustering algorithms on real breast images.
• WSFCM_V achieved the best evaluation index values in comparative experiments.
• The algorithm showed higher segmentation accuracy for both large lesion regions (tumors) and small lesion areas (calcifications).

Conclusions:

• The proposed WSFCM_V algorithm significantly improves the accuracy of breast cancer lesion segmentation in mammography.
• WSFCM_V effectively overcomes the limitations of traditional clustering methods by leveraging prior knowledge and enhanced learning strategies.
• This advancement holds significant clinical value for radiologists in diagnosing breast cancer more efficiently and accurately.