Convexified Coupled Active Contour Segmentation of Clue Cells for Assessing Bacterial Vaginosis with Immunofluorescence Microscopy

Clue cells are vaginal epithelial cells heavily covered with bacteria. Presence of 20% or more clue cells is a critical criterion in the diagnosis of bacterial vaginosis (BV), which is associated with an increased risk of HIV infection. Clue cell segmentation is a prerequisite task towards the enumeration of clue cells. We adopt a two-band immunofluorescence microscope to capture the high-resolution images of clue cell candidates for segmentation. We present a convexified coupled active contour (CCAC) model that takes the advantage of the cell structure and background relationship across the bands, enabling jointly segmenting the boundaries of the epithelial cell, its nucleus, and distinct bacteria. Convexification is formulated on top of the dual band, region-based level set segmentation framework using characteristic functions. This approach provides a global optimal solution and does not require good initializations. Efficacy of the method is demonstrated on clinical data, showing robust segmentation results in a wide range of scales. Our method achieves a high sensitivity of 98.6%.