Indian Journal of Science and Technology

Abstract

Objective: Cancer diagnostic using clinical pathology have been proved as a standard method in which histologist/ pathologist examines biopsy sample for cell morphology and tissue distribution. Pathologist detects random growth and random placements in tissue samples. These diagnostics are very subjective and based on experience/knowledge base of pathologists. This work presents the use of 2D Autoregressive And Moving Average (ARMA) model in computer assisted automatic cancer detection. Analysis: ARMA model parameters have been considered for representing entire histopathology image. These features have further used for analysis and classification. Parameter estimation has been carried out by Yule walker Least Square (LS) method. Histology images have been classified into healthy and malignant images according to ARMA parameters. K- Fole cross validation has been performed with Linear Kernel support vector machine classifier for classification. Findings: As an outcomes of this experimentation, it is proven that ARMA model parameters works as an excellent discriminating features. These ARMA features are capable of extracting hidden information of the underlying cancer decease. This study also presents the role of neighborhood pixel in image analysis and classification. Improvement: This work have described innovative way of using ARMA features in histopathology imagery and can be implemented in computer assisted diagnosis.
Keywords: Autoregressive Model and Moving Average (ARMA), Markov Random Field model (MRF), Model Based Study, Quarter Plane (QP), Support Vector Machine Minimum (SVM), Texture Analysis, Yule Walker Least Square (YWLS)