Texture Analysis of Breast Cancer Cells in Microscopic Images Using Critical Exponent Analysis Method

Abstract To explore application of fractal analysis to study texture features of microscopic images, a critical exponent analysis (CEA) method is proposed to improve classification ability of histological structures in microscopic breast cancer images based on one-dimensional (1D) sequences. Fractal analysis is commonly a mathematical tool for handling with a complex system. A method of estimating fractal dimension (FD) has been found to be useful for an analysis of various medical images. The CEA has been established as an important tool for detecting the FD parameter of the self-affine series information. To reduce computational complexity, two-dimensional (2D) images are firstly preprocessed to form two 1D sequences, including horizontal and vertical landscapes, and then their complexity and self-affinity are detected using the CEA. Subsequently, the FDs at different image orientations are analyzed. Results from both horizontal and vertical landscape sequences indicate that a region of stromal cells has the FD value higher than a region of cancer cells and a region of lymphocytes; in contrast, a region of lymphocytes has the FD value lower than other two regions. Results of the p values obtained from analysis-of-variance of FDs from three regions of histological structures indicate that the difference between mean FDs of three regions from vertical landscape is statistically more significant than that of three regions from horizontal landscape. Texture features computed from the CEA method can be useful for solving the classification of breast cancer cell from microscopic images that is difficult to classify if a colour feature and a spatial feature, notably shape, of cells are similar to each other.