Decision Support System for Classification Medullary Thyroid Cancer

Abstract

Due to the complex, heterogeneous and mimic morphological features of medullary thyroid cancer (MTC). It becomes often difficult to diagnose MTC at early stage. Since histopathological complex patterns of cancerous cells and tissues requires a huge effort to classify. Therefore thyroid cancer classification has become one of the significant research area area(s) of Machine Learning. We propose a decision support system to classify initial variation of morphological appearance of nuclei by using Convolutional Neural Networks (CNNs). The system comprises over three major layers, where image preprocessing techniques are used at top layer along with feature selection techniques. Classification model constructed by using CNNs at the second layer and result visualization described at third layer. Due to the unavailability of datasets for medullary thyroid cancer in literature, this research uses real-world datasets consisting upon 20GB cytological medical images and the approximated classification accuracy is measured about 99.00%. Malignant and non - malignant cells are visualized to assist the doctors in better way.