Segmentation and classification of cervix lesions by pattern and texture analysis

This work aims at automated segmentation of major lesions observed in early stages of uterine cervical cancer. Automated segmentation reduces subjective variability and cost in current manual evaluation methods used to determine the biopsy locations for diagnosis. Two different methods, a non-convex optimisation approach and mathematical morphological approach, are used to segment the aceto-white region. Within this region other abnormalities, such as mosaic patterns, are classified by fuzzy c-means using a textural feature obtained from skeletonised vascular structures. These vascular structures are extracted by a series of morphological operations. Minimisation of uncertainties for degraded images is also discussed.