Article Preview
TopIntroduction
Cytopathology is one of the most recent medical disciplines. By this discipline, the disease study and the diagnosis are based on the examination of free cells or small tissue fragments, by a microscope. Cytopathology was founded in 1928 by G. Papanicolaou (Papanicolaou, 1928). This specialty became popular through the worldwide known Pap test, a test used extensively as a screening method, for the detection of precancerous cervical lesions and therefore for preventing cervical cancer (Diamantis, Magiorkinis, & Koutselini, 2014; Papanicolaou & Traut, 1941). From the early years, cytopathology (Diamantis, Beloukas, Kalogeraki, & Magiorkinis, 2013; Naylor, 2000; Williams & Rosenthal, 1993) was used to investigate diseases affecting other (except than cervix) organs, such as: thyroid, body fluids (e.g. cerebrospinal, pericardial, peritoneal pleural, etc.). Cytopathology examines almost the total range of body organs and systems. The significant advantage of cytopathology practice is the fact that biological material is extracted in a non-invasive or minimally invasive manner, for example, cells are extracted via a brush, a spatula or a syringe with a fine needle. Thus, anesthesia is not usually required, and the patient is in a rather comfortable status.
Cytopathology is a discipline that requires highly skilled and trained professionals. However, the diagnostic procedures usually do not involve quantifications and can be relatively subjective and moreover require a lot of experience. However, cell measurements and measurements of cell formations are nowadays possible with the use of image analysis software; thus, technologically is possible to move towards a more objective approach. Such an image analysis, however, has a barrier, due to the high number of variables and the number of collected data. Machine learning and modeling approaches, such as the logistic regression approach, proposed in this paper, among others, are highly suitable for application in cytopathology and pathology in general (Dybowski & Gant, 1995) since they can introduce a systematic and objective approach via data evaluation
In the arena of gastric cytopathology, it is widely accepted that the endoscopy is a useful method for the investigation of gastric lesions. Moreover, it allows the application of bioptic procedures for the histological diagnosis of benign and malignant diseases. The combined application of cytology and histology for diagnosis could increase the accuracy of such bioptic procedures (Waldron et al., 1990) to more than 90%. Nevertheless, the cytological approach of gastric brush smears has a second-order diagnostic significance because of the difficulties in distinguishing benign cells with severe regenerative alterations from well-differentiated cancer cells. That results in relatively low sensitivity for the gastric cytopathology.
In order for the diagnostic accuracy to be increased, the application of nuclear morphometric and densitometric data has been investigated (Boon, Kurver, Baak, & Thompson, 1981; Karakitsos et al., 1998; Karakitsos et al., 2000; Karakitsos et al., 1997; Karakitsos et al., 1996). The analysis of such data can be also performed by multivariate mathematical methods such as discriminant analysis, which has been applied many times in various aspects in modern medicine (Han, Li, & Han, 2003; Marchevsky, Tsou, & Laird-Offringa, 2004; McHugh, Shinn, & Kay, 2000; Mijovic & Mihailovic, 2002), as well as logistic regression (Zygouris et al., 2014). Both methods can classify cell nuclei, according to their characteristics, as benign or malignant through a statistical model.
Although various machine learning methods have been applied in gastric cytopathology, the logistic regression approach has not yet been investigated. The aim of this study was to evaluate the potential of the logistic regression approach in the classification of benign from malignant gastric nuclei in routine prepared gastric cytological smears. In addition, we aimed to develop a patient classification scheme based on the classification results of the logistic regression on individual cell nuclei.