Automated Detection of Malignant Cells Based on Structural Analysis and Naive Bayes Classifier

Abstract

Breast cancer is the second most common cancer in all over the world. The treatment of breast cancer is possible if the problem is properly identified. To solve this problem it needs such type of automated system that detects it in early stage. This paper presents a new method for detection of malignant cells and their classification in breast cytology images. The proposed method is divided into five phases. In the first phase, image is pre-processed for contrast enhancement followed by noise removal. During the second phase, the image is segmented into foreground and background regions. In third phase total numbers of cells are counted. Using naïve Bayes classifier all cells are classifieds into malignant and benign on the basis of size and shape features of the nucleus. The experiments were performed on local data set provided by the pathology department, Lady Reading Hospital (LRH) Peshawar, Pakistan. The results produced by the proposed technique were validated by a team of senior pathologist of the pathology department LRH. The proposed technique was evaluated individually on the dataset for malignant cells detection and was found to be effective. Experimental results show that scheme has accuracy up to 98.49% by naïve byes classifier with multiple cross validations. Results show that the scheme has improved the performance of malignant cells detection and classification. It is also capable to classify images of Fine Needle Aspiration Biopsy (FNAB) slides with high accuracy.