Models of Applied Computational Intelligence in Breast Cancer Prediction

Autor: Leal, Marcelle Rosa Ribeiro
Přispěvatelé: Silva, Robson Mariano da, Teixeira, Rafael Bernardo, Benac, Marcos Azevedo, Gol?alves, Reinaldo Bellini
Jazyk: portugalština
Rok vydání: 2018
Předmět:
Zdroj: Biblioteca Digital de Teses e Dissertações da UFRRJ
Universidade Federal Rural do Rio de Janeiro (UFRRJ)
instacron:UFRRJ
Popis: Submitted by Leticia Schettini (leticia@ufrrj.br) on 2021-07-19T15:30:39Z No. of bitstreams: 1 2018 - Marcelle Rosa Ribeiro Leal.pdf: 2364927 bytes, checksum: 0bdb119a0bf329dbe4d1118b12a8881c (MD5) Made available in DSpace on 2021-07-19T15:30:39Z (GMT). No. of bitstreams: 1 2018 - Marcelle Rosa Ribeiro Leal.pdf: 2364927 bytes, checksum: 0bdb119a0bf329dbe4d1118b12a8881c (MD5) Previous issue date: 2018-05-22 Breast cancer is the second most frequent neoplasm in the world. According to data from the National Cancer Institute (INCA), 52,680 new cases were diagnosed in Brazil in 2014, an increase of 22% compared to the year 2013. It accounts for approximately 39% of women's deaths cancer patients. For an accurate diagnosis, a lot of experience is required, and especially that the classification of the clinical staging of the tumor (stage of the cancer) is correct. Thus, it is necessary to develop integrated systems that, combined with the experience of the professionals of the area, make it possible to carry out the precise diagnosis in the detection of breast cancer. The objective of the present study is to apply the RNA and SVM techniques to assist in the diagnostic interpretation of microcalcifications detected in screening mammography. The set used in this study consists of 569 data, from patients with suspected breast cancer obtained from the Institute of Radiology of Erlangen-Nuremberg University from 2003 to 2006. The database has clinical information about lightning, texture, perimeter, area, smoothness, compactness, concavity, concave, symmetry and fractal dimension. The data were divided into two groups: the training set consisting of 75% of the mammographic examination samples and the independent test set, with 25% of the remaining samples. The techniques developed were implemented using software R. According to the analysis of the results it was possible to evidence the promising performance of the SVM, which obtained in its best simulation an accuracy above 98%, in relation to the values of False Negatives The best value obtained was 1.96%. However, the model using the MLP Neural Networks presented in its best simulation an accuracy of over 96% and in relation to the values of False Negatives, the best value obtained was also 2%, and therefore its relevant use. There was a statistically significant difference at the level of 95% (p-value
Databáze: OpenAIRE
Externí odkaz: