Archives

  • 2019-07
  • 2019-08
  • 2019-09
  • 2019-10
  • 2019-11
  • 2020-03
  • 2020-07
  • 2020-08
  • In a Pap smear analysis

    2020-08-30

    In a Pap smear analysis, a feature can be defined as a piece of in-formation that is relevant for solving the computational task related to the Pap smear image analysis. Features extracted from the Pap smear images can be broadly classified as structural or textural features [28] and these include:
    ⁃ Size and shape: This contains morphometric features that express the overall size and shape of a cell. Examples include position and orientation dependent features, geometric features (area, perimeter, longest and shortest diameter), contour features (curvature, bending energy, convex hull, elliptic deviation, and Fourier descriptors) and invariant moment features [29].
    ⁃ Intensity. These features use the absolute intensity values in the image. Some of the intensity features include the largest/lowest density and different region intensity features. ⁃ Texture: Textural features help to obtain quantifiable measures of overall local density variability within an object of interest. Examples of texture measures are gradient image features, Laplace image features, flat texture features, topological gradients, run-length and co-occurrence features.
    ⁃ Structure: With structural features, each chromatin particle in the cell is considered to be an object. Features are extracted by de-scribing the relationships between these objects. Examples of structure features include the nearest neighbourhood graph, the minimum spanning tree graph and the convex hull.
    Due to the importance of feature extraction to any automated cer-vical cancer screening system, a number of cell features including the nucleus area, nucleus perimeter, nucleus roundness, cytoplasm area, and nucleus to cytoplasm ratio, have previously been utilized to help facilitate cervical cancer classification [17,30–32].
    Feature selection involves evaluating and optimizing the feature space used for the actual classification. Adding more features to a set will not always lead to a better separability, but could instead introduce
    noise to the different classes; hence all features used in classification should add information, which increases the separability between the different classes [33]. Feature selection techniques include:
    ⁃ Multivariate statistics. This encompasses a number of procedures that can be used to analyze more than one statistical variable at a time. Examples include principal component analysis (PCA) and linear discriminant analysis (LDA) [34].
    ⁃ Genetic Algorithms. These use a model of genetic evolution to at-tempt to find an optimal solution to some kind of classification problem [35]. ⁃ Clustering methods. These involve unsupervised and supervised classification of samples into clusters. The two most common clus-tering methods are k-means and hierarchical clustering. ⁃ Bayesian methods. These are based on Bayes' theorem [19].
    ⁃ Artificial Neural Networks. These try to mimic the way the ALLN computes information [36].
    ⁃ Support Vector Machines. These aim to separate multiple clusters with a set of unique hyperplanes that have the greatest margin to the edge of each cluster [37].
    The aim of any automated cervical cancer screening system is to determine whether a sample contains any evidence of cancer. The most common method for classification involves analyzing all cells using selected features and then classifying each cell as normal or suspicious [38]. Another approach is to mimic the classification methodology used by cytotechnologists [39]. This involves analyzing the sample based on several factors such as patterns in cell distribution, the frequency of cells and cell clusters, the occurrence of degenerated cells and cyto-plasm, and the abundance of bare nuclei. A number of researchers have developed classification techniques for automated diagnosis of cervical cancer from Pap smear images.
    J. Su et al. [40] proposed a method for automatic detection of cervical cancer from Pap smear images using a two-level cascade in-tegration system of two classifiers. The results showed that the re-cognition rates for abnormal cervical cells were 92.7% and 93.2% when C4.5 classifier or logical regression classifier was used individually; while the recognition rate was significantly higher (95.6%) when the two-level cascade integrated classifier system was used.
    M. Sharma et al. [41] used the K-Nearest-Neighbors (KNN) method to classify the stage of cervical cancer from Pap smear images. A clas-sification accuracy of 82.9% with 5-fold cross-validation was achieved. R. Kumar et al. [42] proposed a framework for automated detection and classification of cervical cancer from microscopic biopsy images using biologically interpretable features. The K-nearest neighbor method was used for cervical cancer classification. Performance measures for ac-curacy, specificity and sensitivity of 92%, 94% and 81% were obtained. T. Chankong et al. [43] presented a method for automatic cervical cancer cell segmentation and classification using the fuzzy C-means (FCM) clustering technique. Validation with Artificial Neural Networks (ANN) yielded accuracies of 93.78% and 99.27% for the 7-class and 2-class problems, respectively. J. Talukdar et al. [44] presented a fuzzy clustering based image segmentation of Pap smear images of cervical cancer cells using the Fuzzy C-Means (FCM) Algorithm. Two random numbers were utilized to form the membership matrix for each pixel to guide clustering. Promising results were obtained using the pixel level segmentation. M. Sreedevi et al. [45] presented an algorithm based on an iterative thresholding method for segmentation of Pap smear images and classification of cervical cells as normal or abnormal, based on the area parameter of the nucleus. The features of the nucleus were ex-tracted using regional properties, and cells were classified as normal if the nucleus area was less than 1635 mm, and classified as abnormal otherwise. A sensitivity of 95% and specificity of 90% was achieved.