Character-Based Convolutional Grid Neural Network for Breast Cancer Classification

Abstract

According to the World Health Organization (WHO)1, an early detection of cancer greatly increases the chances of making the right decision in a successful treatment plan. Over the last decade, the increasing world-wide demand for early detection of breast cancer at the hospitals has resulted in necessity of new research avenues. The traditional domain knowledge based diagnostic method requires hand-crafted features, which are not only time consuming, but also corpus dependent. In this paper, we propose a novel neural net-work architecture for the disease classification that relies only on character level representations. Our model leverages subword information through a convolutional neural net-work (CNN) and a residual network over characters, whose output is given to a Grid long short-term memory (Grid-LSTM) recurrent neural network language model (RNN-LM). It's truly an end-to-end model requiring no task-specific feature engineering or data pre-processing beyond pre-trained character embedding on unlabeled corpora. Thus, it can be easily applied to a wide range of NLP tasks in different domains. We evaluate our module using NCBI disease dataset for classification tasks, finding that it can achieve a state-of-the-art performance with minimal computational cost.