Mueller polarimetric microscopic images analysis based classification of breast cancer cells

Abstract

Mueller polarimetric imaging is considered a potentially powerful technique for probing the microstructural information in the biomedical field. In this study, the Mueller matrix microscopy was adopted to characterize the microstructures of breast cancer cells exhibiting different receptor proteins expressions. To be specific, four types of breast cancer cells were selected, and a suitable method was developed for cell sample preparation to capture clear cell polarimetric images. Subsequently, convolutional neural network was utilized to classify breast cancer cells with different input datasets types, and Mueller matrix elements images achieved the optimal accuracy of 88.3% (10.1% higher than that of ordinary optical images). The proposed technique demonstrated the potential application of Mueller polarimetric images to classify unstained cells harvested from breast cancer cytological biopsies. Furthermore, by immunofluorescence experiments and cytochalasin B treatment, this study verified that the polarization imaging can effectively show the intracellular localization and content of fibrous actin, which is critical to tumorigenesis and metastasis. It was thus indicated that Mueller matrix imaging can also help study the pathological process of breast cancer by displaying fibrous actin variations.