Machine Learning Approaches for Pap-Smear Diagnosis: An Overview

Abstract

This chapter is a typical example of usage of Computational Intelligence Techniques-CI-Techniques (Machine Learning-Artificial Intelligence) in medical data analysis problems, such as optimizing the Pap-Smear or Pap-Test diagnosis. Pap-Smear or Pap-Test is a method for diagnosing Cervical Cancer (4th leading cause of female cancer and 2nd common female cancer in the women aged 14–44 years old), invented by Dr. George Papanicolaou in 1928 (Bruni et al. in Human papillomavirus and related diseases in the world [1]; Marinakis and Dounias in The Pap Smear Benchmark, Intelligent and Nature Inspired Approaches in Pap Smear Diagnosis, Special Session Proceedings of the NISIS—2006 Symposium [2]). According to Pap-Smear, specialized doctors collect a sample of cells from specific areas of cervical, observe (using microscope) specific cells of the above cell-sample and classify these cells into 2 general (Normal and Abnormal cells) and 7 individual categories/classes: Superficial squamous epithelial, Intermediate squamous epithelial, Columnar epithelial, Mild squamous non-keratinizing dysplasia, Moderate squamous non-keratinizing dysplasia, Severe squamous non-keratinizing dysplasia and Squamous cell carcinoma in situ intermediate. The ideal aim of this classification process is the early diagnosis of cervical cancer. Pap-Test was a time-consuming process and with considerable errors of observation, resulting in diagnosis with a high degree of uncertainty. Considering these problems, Data Analysis researchers in collaboration with specialized doctors have presented several successful approaches to the Pap-Smear diagnosis optimization problem using Computational Intelligence (CI) Techniques, whose results are acceptable to the medical community and have room for improvement. An equivalent effort was made by the researchers and students of Department of Automation of the Technical University of Denmark for the first time in 1999 is the cornerstone of further Pap-Smear data analysis using CI-Techniques, which was then continued until nowadays by researchers and students of the Management and Decision Engineering Laboratory (MDE-Lab) of Technical Department of Financial and Management Engineering of the University of the Aegean (http://mde-lab.aegean.gr/downloads ). This research focuses on the approach of the Pap-Smear Classification Problem with the use of CI-Techniques and has as an ideal goal the contribution of Artificial Intelligence to the optimization of medical diagnoses. In addition, the research conducted aims at diagnosing cervical cancer both at an early stage and at an advanced stage, improving the Pap-Smear classification process as well as the process of Feature Selection. The aforementioned research was based on two databases, called Old Data and New Data which consist of 500 and 917 single cell patterns respectively, described by 20 features. These data were collected by qualified doctors and cyto-technicians from the Department of Pathology of the Herlev University Hospital and are available at the web-page of MDE-Lab (http://mde-lab.aegean.gr/downloads ). The CI-Techniques, that were used to build classifiers, both for the 2-class classification problem and for the 7-class classification problem, are machine learning algorithms, such as Adaptive Network-based Fuzzy Inference Systems, Artificial Neural Networks, k-Nearest Neighbor approaches, etc. Also, various algorithms, such as Fuzzy C-Means, Tabu Search, Ant Colony etc., were used to improve the operating processes, such as the training process or the feature selection approach. All these CI-Techniques are briefly presented in Sect. 3. Finally, the results of the application of the above CI-Techniques are presented in Sect. 4, and prove very satisfactory, fully accepted by the doctors of this particular field of medicine and with room for further improvement.