Probabilistic Prediction of Cancer Using Nuclei Morphometry

In this study, the aim is to develop a probabilistic model for predicting oral cancer followed by contour based shape features extraction of basal cell nuclei. Especially, oral submucous fibrosis (OSF) having malignant potentiality is one of the leading causes of death in Indian subcontinent. Generally, the oral onco-pathologists screen a patient at early stage by examining the morphometric features of the basal cell nuclei, at which OSF originates. In most of the situations, subjective evaluation leads to biased diagnosis especially in early detection. In this perspective, there is a great demand of objective evaluation simultaneously with subjective information as per clinical acumen to automate a diagnostic tool for early detection of disease. In view of this, the morphological features are estimated followed by the extraction of contours of basal cell nuclei of OSF images using histogram approach. The features are here found to be statistically significant in discriminating normal (n = 19) and OSF (n = 19) classes. Then, a multivariate logistic regression is fitted with the formulation of suitable decision rule for disease probability estimation. As an advantage, the oral oncologist will have the provision of selecting a suitable threshold for effective decision-making. However, the predictive model is good fitted (-2 Log-Likelihood = 14.483 and chi-square = 38.197) whereas the overall prediction accuracy is 94.74%.