Classification of Cervical Cancer Using an Autoencoder and Cascaded Multilayer Perceptron

Cervical cancer is the most frequent and potent form of cancer in women. The complications caused by it can be avoided if it is detected and treated promptly. In oncology, artificial intelligence has improved the prediction accuracy in the preliminary stages of cervical cancer. In this work, a novel machine learning-based approach is introduced to predict and categorize the healthy and anomalous cervical cells. Initially, an adaptive median filter is utilized to eliminate the noise artefacts in the pap smears from the Herlev dataset. The auto-encoder (AE) is employed to extract the features and reduce the dimension of features to progress the training process. Consequently, a cascaded multilayer perceptron (c-MLP) is employed for the classification of the normal and abnormal cervical cells. The c-MLP was trained using the Bayesian Regulation algorithm to generate the best classification accuracy of 97.63%. As a result, the classification using the c-MLP is more accurate and effective for classifying healthy and malignant cervical cells than the classic ML classifiers. The proposed ML-based framework progresses the overall accuracy range by 3.61%, 4.40%, 5.07%, and 4.66% better than SVM, XGB, RF, and DT, respectively.