Automatic cell nuclei segmentation and classification of breast cancer histopathology images

• Combine wavelet decomposition and multi-scale region-growing for locating of ROI.
• Combine morphologic operation and CSS corner detection for segmentation of cells.
• Propose wrapper feature selection algorithm making samples good separability.
• Study selected features′ correlation and importance on classification schemes.
• Can detect cancer from histopathological normal-appearing cells.

Breast cancer is the leading type of malignant tumor observed in women and the effective treatment depends on its early diagnosis. Diagnosis from histopathological images remains the "gold standard" for breast cancer. The complexity of breast cell histopathology (BCH) images makes reliable segmentation and classification hard. In this paper, an automatic quantitative image analysis technique of BCH images is proposed. For the nuclei segmentation, top-bottom hat transform is applied to enhance image quality. Wavelet decomposition and multi-scale region-growing (WDMR) are combined to obtain regions of interest (ROIs) thereby realizing precise location. A double-strategy splitting model (DSSM) containing adaptive mathematical morphology and Curvature Scale Space (CSS) corner detection method is applied to split overlapped cells for better accuracy and robustness. For the classification of cell nuclei, 4 shape-based features and 138 textural features based on color spaces are extracted. Optimal feature set is obtained by support vector machine (SVM) with chain-like agent genetic algorithm (CAGA). The proposed method was tested on 68 BCH images containing more than 3600 cells. Experimental results show that the mean segmentation sensitivity was 91.53% (±4.05%) and specificity was 91.64% (±4.07%). The classification performance of normal and malignant cell images can achieve 96.19% (±0.31%) for accuracy, 99.05% (±0.27%) for sensitivity and 93.33% (±0.81%) for specificity.