Detection and removal of artifacts in cervical Cytology images using Support Vector Machine

Abstract

Cervical cancer is a leading cause of cancer-related deaths in women worldwide which kills more than 288,000 of them each year. 80% of these deaths occur in the developing countries like India, where there are no well established screening programmes. For implementing an effective and optimal mass screening the quantity of false positive rates should be controlled. The artifacts present in a massive order which are similar in size and shape to abnormal cells would cause the misclassification of cytology images in the screening process. This increases the false positive rate which can hinder the mass screening. Therefore the elimination of these artifacts plays a key role in designing a proper Classification strategy for the malignancy detection from Cytology images. This paper, deals with a new Pattern Recognition strategy on detection and removal of artifacts in cervical cytology images using Support Vector Machine (SVM).

Automatic Detection of Cervical Cancer Cells by a Two-Level Cascade Classification System

“…Morphologic features were used to describe the shape of the nucleus, which was important to distinguish between different cell types. According to the different morphology of the nucleus, we extracted 28 morphologic features, including area [ 12 , 15 , 29 ], Circularity, Distance, Sigma, Sides, Roundness, Convexity [ 29 ], I a (centroid coordinates of x -axis), I b (centroid coordinates of y -axis), M 11 , M 02 , M 20 , Compactness [ 29 ], Count-Length [ 30 ], Diameter [ 30 ], Radius [ 29 , 30 ], Rectangularity [ 30 ], Anisometry [ 31 ], Bulkiness [ 32 ], and Structure-Factor [ 33 ].…”

“…In another study [ 11 ], the premalignant stage was further divided into CIN1 (carcinoma in situ 1), CIN2, and CIN3. Rajesh Kumar et al [ 12 ] classified the cervical cells into two types of cells, normal and abnormal cervical cells. Sarwar et al [ 13 ] divided the cells into three normal cells (superficial squamous epithelial, intermediate squamous epithelial, and columnar epithelial), and four abnormal cells (mild squamous nonkeratinizing dysplasia, moderate squamous nonkeratinizing dysplasia, severe squamous nonkeratinizing dysplasia, and moderate squamous cell carcinoma in situ).…”

“…In regard to the acquirement of cell features, most of the researchers used multidimensional features to classify the cells [ 12 , 14 – 16 ]. Some authors analyzed four parameters: area, integrated optical density (IOD), eccentricity, and Fourier coefficients [ 12 ]. Other authors used 16 features: area of nucleus, area of cytoplasm, nuclear gray level, cytoplasm's gray level, and so forth [ 14 ].…”

Developmental Deconvolution for Classification of Cancer Origin

“…Machine learning classifiers have shown promise as new tools when applied to image processing in radiology and histopathology (4)(5)(6). However, image classifiers only detect visual features and are sometimes subject to artifacts (7,8). Classifiers that use molecular features, such as gene expression, have great potential to aid in diagnosis through capturing nonvisual information, and recent approaches have demonstrated value in combining visual and molecular features for classification (9,10).…”