Classification and Analysis of Squamous Cell for Cervical Cytology

Classification and Analysis of Squamous Cell for Cervical Cytology

Pap Smear Test

Cervical cancer is the third most diagnosed cancer and the fourth leading cause of cancer death among women worldwide accounting for 9% of all malignancies among females in 2008. More than 85% of the cervical cancer cases occur in developing countries where public health infrastructure does not support Papanicolaou testing. In countries with developed healthcare systems, widespread cervical screening programmers, aimed at detecting precancerous changes that can then be treated to prevent invasive cancer, have significantly reduced the number of deaths from the disease. The Papanicolaou (Pap) smear, is the primary screening test for cervical cancer. It has been largely responsible for diagnosing cancerous and precancerous lesions in many developed countries.

Automated screening

Automated screening machines can analyses Pap smear slides in a short time without fatigue, providing consistent and objective classification results. The rationale for automated screening is to improve the limitations of the conventional Pap smear test in the following ways:

• Increase the sensitivity and specificity of the Pap smear test;
• Decrease the workload of technicians and pathologists;
• Reduce the cost for cervical cancer screening programmers; and
• Lower the probability of incidence of cervical cancer and the mortality rate from the disease.

Aim and objectives

The aim of this project was to fully explore the structural approach to chromatin pattern description and to evaluate the efficacy of the features derived from it for discriminating between normal and abnormal Pap slides. The project had the following objectives:

• To develop a robust algorithm for detecting and segmenting cell nuclei in digitized Pap smear images obtained using bright field microscopy;
• To develop structural texture features that quantitatively characterize the pattern (arrangement, size, shape, etc.) of the nuclear chromatin;
• To determine the most discriminatory subset of features for discriminating between normal and abnormal slides using real clinical data; and
• To evaluate the performance of a classifier(s), based on the selected features, using real clinical data.

Scope

The proposed cervical screening approach has the following steps:

• Scanning the Pap-stained slide using a light microscope coupled with a CCD camera with multiple objectives.
• Locating and segmenting the free-lying cell nuclei in each EDF image, and performing artefact rejection to make sure only nuclei-like objects are retained;
• Segmenting the nuclear chromatin inside each nucleus into texture primitives (blobs);
• Extracting features from this structural model to quantitatively characterize the chromatin pattern;
• Deriving slide-based features from the features in 6 in order to classify the slide as normal or abnormal.

Preprocessing

The image preprocessing stage is required when segmenting the cell for background extraction. Every captured image exhibits a certain percentage of noise and may have low contrast. Overall, most methods start by reducing noise and increasing contrast. Median filters are traditionally used to reduce noise, where the gray level of every pixel is replaced by the median of the intensity levels of the pixel neighborhood.

Segmentation

This chapter deals specially with the problem of accurately and robustly segmenting the cervical cell nuclei in digitized light microscopy images of Pap smears.

Feature Extraction

This section deals with the problem of quantitative characterization of chromatin texture and presents a set of novel structural texture features to describe nuclear chromatin patterns in cells on a conventional Pap smear. These features are derived from a segmentation of the chromatin into blob-like primitives. The proposed set of features are, in particular, derived from statistics of morphometric features and contextual features computed for these blobs.

Feature Analysis and Selection

This section presents an evaluation of the performance of the proposed structural chromatin texture features. In particular, it presents an investigation of the most discriminatory subset of features, from among the proposed features and a wide range of features drawn from the literature, for discriminating between normal and abnormal Pap smears. The section presents the details of the two experiments carried out in this project. The first is a feature selection experiment performed to obtain the most discriminatory subset of features. The second experiment is to evaluate the performance of a variety of classifiers built using the feature subset obtained in the first experiment to discriminate between the normal and abnormal slides.

Classification

In this step we need to assign an object to a specific category from a variety of different categories based on some special characteristics of that object. As a case in point, in this project we want to categorize a particular Pap slide as either normal or abnormal. In computer science, these kind of situations are described as classification problems.

Normal: columnar epithelial, parabasal squamous epithelial, intermediate squamous epithelial, superficial squamous epithelial.

Abnormal: mild squamous non-keratinizing dysplasia, moderate squamous non-keratinizing dysplasia, severe squamous non-keratinizing dysplasia.

Collaborator:

Dr.Mansoor Fatehi 
Director, Medical Imaging Informatics Research Center at Tehran
Chairman of the Board, Sorenahealth, Iranian Modern Health Strategies

Dr. Ali Sadeghitabar

Doctorate in Clinical Laboratory Science
Iranian Association of Clinical Laboratory Doctors