Effective Screening and Classification of Cervical Precancer Biopsy Imagery

Microscope images of biopsy samples of cervical precancers conventionally discriminated by histopathology, the current "gold standard" for cancer detection, showed that their correlation properties are segregated into different classes. The correlation domains clearly indicate increasing cellular clustering in different grades of precancer compared with their normal counterparts. This trend indicates the probability of pixel distribution of the corresponding tissue images. Because the cell density is not uniform in the higher grades, the skewness (asymmetry of a distribution), kurtosis (sharpness of a distribution), entropy (randomness), and standard deviation are affected. A combination of these parameters effectively improves the diagnosis and quantitatively classifies the normal and all the three grades of precancerous cervical tissue sections significantly. Thus, the statistical analysis of microscope images is a promising approach for early stage tumor detection and quantitative classification of precancerous grades; this can effectively supplement the qualitative analysis by the pathologist.