Automatic 2D Lung Nodule Patch Classification using Deep Neural Networks

Automatic lung cancer diagnosis is considered one of the most demanded tasks in recent years which could aid the specialists in inspecting the cancerous nodules. The automatic extraction of discriminative patterns and features from the 2D nodule patches is one of the challenging problems in lung cancer diagnosis. This paper proposes an efficient deep neural networkbased approach for automatic pulmonary nodule classification from chest CT images. The key idea of this neural network-based approach is to understand the generic discriminative pattern from the 2D patches in a supervised manner. In this paper various deep neural networks such as RNN, LSTM and CNN have been implemented to classify the 2D nodule patches. The experimental results show that the RNN is not suitable for learning patterns from 2D image data, whereas the LSTM and CNN models obtain the AUC of 0.912 and 0.944 respectively. CNN based classifiers gain 25% more accuracy as compared with the RNN model.