Predicting the Stage of Non-small Cell Lung Cancer with Divergence Neural Network Using Pre-treatment Computed Tomography

Abstract

Determining the stage of non-small cell lung cancer (NSCLC) is important for treatment and prognosis. Staging includes a professional interpretation of imaging, thus we aimed to build an automatic process with deep learning (DL). We proposed an end-to-end DL method that uses pre-treatment computer tomography images to classify the early- and advanced-stage of NSCLC. DL models were developed and tested to classify the early- and advanced-stage using training (n = 58), validation (n = 7), and testing (n = 17) cohorts obtained from public domains. The network consists of three parts of encoder, decoder, and classification layer. Encoder and decoder layers are trained to reconstruct original images. Classification layers are trained to classify early- and advanced-stage NSCLC patients with a dense layer. Other machine learning-based approaches were compared. Our model achieved accuracy of 0.8824, sensitivity of 1.0, specificity of 0.6, and area under the curve (AUC) of 0.7333 compared with other approaches (AUC 0.5500 ─ 0.7167) in the test cohort for classifying between early- and advanced-stages. Our DL model to classify NSCLC patients into early-stage and advanced-stage showed promising results and could be useful in future NSCLC research.