Classification of Lung nodules using Convolutional long short-term Neural Network

Lung nodule classification is a key process for early prediction of lung cancer. Automatic diagnosis of lung nodules from computed tomography images decreases the risk of misdiagnosis. Deep neural networks have recently made phenomenal progress in analyzing medical images. A convolutional long short-term (ConvLSTM) based classification model is proposed to distinguish malignant from benign nodules. The proposed model captures important morphological features found in the series of CT images using the Convolutional long short-term memory network, which is capable of understanding linearly related spatial features from three dimensional images. The efficiency of the proposed classifier is assessed by the accuracy, precision, sensitivity, and area under the curve (AUC). The proposed model outperforms LSTM and CNN classifiers with an average accuracy of 93%. As the ConvLSTM based classifier forgets the irrelevant data and keeps the essential information required for identifying cancer nodules, it could distinguish the malignant and benign nodules better than the LSTM and CNN model.