Segmentation of Cell Objects in IIF Images using Split-Bregman Global Convex Optimization

In this work, an attempt has been made to segment the cell objects in homogeneous pattern in Indirect Immunofluorescence (IIF) images. Homogenous pattern is indicative of the several autoimmune diseases but it is most commonly observed in lupus or Sjogren’s syndrome. The IIF images for this study are obtained from a publicly available dataset. The images are preprocessed using various intensity normalization techniques such as Histogram Equalization (HE), Adaptive Histogram Equalization (AHE), Local Contrast Enhancement (LCE) and Intensity Adjustment (IA). The preprocessing methods are validated by obtaining image intensity profile of the preprocessed images. Comparison of intensity profile indicates that the IA technique has better contrast enhancement than the other three methods. After preprocessing, segmentation is carried out in the intensity adjusted IIF images. Three different segmentation techniques namely Watershed Segmentation (WS), Super pixel based Fast-Fuzzy C-Means Segmentation (SPFFCMS) and Split-Bregman Global Convex Segmentation (SP-GCS) are implemented. The segmentation results are validated with cell object extraction in the binary images. Results indicate that SP-GCS method has extracted more cell objects than WS and FFCMS algorithm. The binary masks extracted from SP-GCS method are then multiplied with the original images to obtain the segmented regions. The images could be further utilized for classification of homogenous pattern in IIF imaging. From this study, it can be concluded that IA method has better contrast enhancement for homogenous pattern and SP-GCS method has better segmentation results. Hence, this study could assist the clinicians in the identification of homogenous pattern for the diagnosis of lupus.