Article Preview
TopIntroduction
The endometrium is a human system with inherent complexity because it undergoes cyclic regeneration under estrogen effects during each menstrual cycle. These effects, as well as sex steroids, oncogene products, growth factors and various peptides (Murphy, Murphy, & Friesen, 1987; Shyamala & Ferenczy, 1981) may cause premalignant and malignant transformation of the endometrium. Early diagnosis is crucial because it is associated with patient management and therapy. However, and despite that in the last decades the incidence of endometrial cancer has increased (Jemal et al., 2006), there is still not a global and well-established screening method for the early detection of endometrial cancer. Moreover, there is no automated procedure.
Artificial Intelligence (AI) techniques are not new in medicine (Adams, Bello, & Dumancas, 2015; Darcy, Louie, & Roberts, 2016; Foran, Chen, & Yang, 2011; Grabe et al., 2010; Karakitsos et al., 1996; Luo et al., 2015; Seffens, Evans, Minority Health, & Taylor, 2015; Su, Xu, He, & Song, 2016). During the last decades, numerous applications have been reported; these involve the use of classical statistical models (Karakitsos et al., 2004), as well as more advanced techniques, such as neural networks. In the field of oncology related medical disciplines there are numerous efforts (Cruz & Wishart, 2006; Hassan, Ruusuvuori, Latonen, & Huttunen, 2015; Karakitsos et al., 1997; Kourou, Exarchos, Exarchos, Karamouzis, & Fotiadis, 2015; Pouliakis et al., 2016; Simões et al., 2014). However, the literature related to endometrial cytological material evaluation by AI and especially publications reporting the comparison of machine learning performance are rather poor.
In this article, a methodology aiming to classify endometrial lesions based on cell nuclei morphometry data will be presented. Initially, the morphometrical features and their biological relation will be described and then the construction of machine learning systems classifying individual nuclei will be outlined. Furthermore, according to the nuclei classification results, a second stage classifier aiming to categorize individual cases will be constructed. The validation of the results will be based on the histological result, which will be considered our gold standard. The applied Machine Learning Algorithm (MLA) was the Learning Vector Quantizer (LVQ) ANN and was selected because it has the capability to provide solutions by creating clusters. The applied classifier was compared with the performance of the cytological outcome.