Cervical Cancer Diagnosis System using Convolutional Neural Network ResidualNet

Abstract

Cervical cancer is a deadly disease attacking women. It represents 6.6% of all female cancers. The stadium of cervical cancer is determined based on the presence of carcinoma. The cervical cancer classification system can be used to help medical workers to analyze the stadium of cervical cancer. In this study, cervical cancer stages were divided into five classes, namely, normal cervix, stadium I, stadium II, stadium III, and stadium IV based on colposcopy images. The proposed method is one of deep learning methods, that is convolutional neural network (CNN) using deep residual network (ResidualNet) architecture. This study compared ResidualNet-18, ResidualNet-50, and ResidualNet-101 models and some conventional methods. The comparison results show that ResidualNet is more accurate than conventional methods. From the experiment, based on the accuracy value and elapsed time, ResidualNet-50 is worth using for cervical cancer classification. The result of this evaluation is higher than the maximum achievement of the ResidualNet-18 architecture. In addition, the elapsed time of the classification process using the ResidualNet-50 architecture with the accuracy, sensitivity, and specificity values reaching 100% is shorter than ResidualNet-101, which is 4397 s.

Parallel Ensemble Feature Subset Selection based Deep Learning Approach for Imbalanced High Dimensional Cancer Datasets

“…Ensemble feature selections [2][3][4][5] are popular due to their accuracy and stability [6,7]. Deep Learning [8][9][10][11][12][13][14] is a recent field of study in machine learning that has had remarkable success in high-level data abstraction and representation. Instead of using a single shallow "fat" structure, it employs numerous levels of non-linear operations to address the relevant machine learning problems.…”