Deep Neural Network using Hadoop for Cervical Cell Classification

Abstract

Deep learning is a concept close to AI technique connecting neural networks. Several industries are poised to be revolutionized by deep learning. In this field, variety of data including images, sounds, and text are analysed through representing and abstracting information on multiple levels. The main objective of this work is to implement a ResNet like architecture in Hadoop for analysing cervical cancer big data to reduce processing time. Novel contribution includes implementing a distributed approach utilizing HDFS and MapReduce Frameworks in order to train ResNet-like neural networks in the field of cervical cell cancer classification. A big data platform called Hadoop is used to evaluate its implementation and performance. The proposed architecture applied to a preprocessed image of pap smear slide for classification and it yields good result for both 2-class (99.9% Specificity, 99.8% Accuracy, 99.7% h-mean and 99.1% Sensitivity) and 5-class classification problem (99.9% Specificity, 99.1% Accuracy, 99.6% h-mean and 99.1% Sensitivity).