Article Preview
TopIntroduction
According to the Brazilian data from the National Cancer Institute (INCA), breast cancer was responsible for about 28% of new cases of cancer in 2018. Breast cancer is, then, the second type of cancer that most affects women in Brazil. In this country, 59,700 new cases of the disease were estimated for 2019 (Instituto Nacional de Câncer, 2017). Another worrying fact is the increased occurrence in young women (Borchartt et al., 2013). This growth may be partly because of the increased incidence due to a greater exposure of women to risk factors resulting from the urbanization process and changes in lifestyle (Porter, 2008). The situation is aggravated by the aging population that occurred intensely in Brazil (Victora et al., 2011). Due to this large number of influences, early detection of the disease is critical since the sooner the disease is detected, the better are the treatments. Therefore, the early detection provides better chances of cure of the patient, which would lead to a decrease in the fatality rate due to this type of cancer (Borchartt et al., 2013). According to Lessa and Marengoni (2016), the chances of cure dramatically reduce if the disease is not diagnosed in the early stages (Lessa & Marengoni, 2016).
Nowadays, mammography is the gold standard for breast cancer diagnosis. However, this exam has some limitations, such as the deficiency in detecting the disease in the case of dense breasts, which are breasts mainly made of glandular tissue. This type of breast is the most predominant in young patients. In addition, mammography has high rates of false positives and exposures patient to ionizing radiation. The exposure to ionizing radiation may even increase the chances of developing the disease (Borchartt et al., 2013). Thus, there is a need for other methodologies to support the diagnosis of this type of cancer (Leles, 2015).
Infrared thermography is a fast, inexpensive and noninvasive screening technique that aims to record the radiation emitted by the patient's skin surface. The temperature variations of cancerous tissue in relation to healthy neighboring tissue are due to the angiogenesis process. Through this process, the lesion stimulates the creation of new blood vessels for its nutrition. With more vessels feeding the lesion, the temperature of the region is higher than the temperature of healthy regions (Dourado Neto, 2014). Keyserlingk et al. (1998) reported in their paper that sensitivity for detecting ductal carcinoma-type cancer is significantly improved when combining mammography to thermal imaging (Keyserlingk et al., 1998). The sensitivity when combining these two techniques was 95% (Andrade, Paiva & Correa, 2017). Thus, due to the facts explained above, thermography is an attractive technique to help to provide an early detection of breast cancer.
Tumor classification systems used today are detailed, complex and usually hard to be used by pathologists, so, the specialists do not feel comfortable using them (Ferreira, Oliveira & Martinez, 2011). However, it is proven that the combination of specialized professionals with digital image analysis methods applied to breast thermography can contribute to the improvement of breast cancer diagnosis, prognosis and treatment (Bandyopdhyay, 2010). As a consequence, intelligent systems are being developed as a decision support tool in many health areas. These systems may help pathologists to perform a more objective, accurate and uniform classification of the lesions. Thus, minimizing the limitations imposed by the existing classification systems, and thus accelerating the work of this professional (Ferreira, Oliveira & Martinez, 2011).
This technological advance has contributed to the generation and storage of an amount of data that is constantly increasing at a faster rate than we are able to process. From this, several areas have been dedicated to the research and the proposal of methods and processes to treat these data (Andrade, Santana & Santos, 2018; Azevedo et al., 2015; Souza et al., 2019). The features selection plays a critical role in this process, since it may overcome an important issue in machine learning: the computational cost. Features selection is often performed as a preprocessing step. Its purpose is to select the most important features, since the non-relevant or redundant features may reduce the accuracy, as well as increase the computational cost (Lee, 2005; Souza, 2017). This leads to the following issue: How many and which features are enough or necessary to describe a particular problem? For many problems, we can use evolutionary algorithms to find the best possible solution. This work aimed to evaluate the performance of the most relevant features subsets, selected by Genetic Algorithm (GA) and Particle Swarm Optimization (PSO), in the problem of breast lesion classification in thermographic images.