On The Integration of CT-Derived Features for Accurate Detection of Lung Cancer

Lung cancer is one of the unsafe maladies that reason enormous disease passing around the world. Early and accurate detection of lung cancer is the main conceivable approach to enhance patients' survival rate. In this paper, we proposes a new framework for pulmonary nodule diagnosis using various features extracted from a single computed tomography (CT) scan. The proposed system fuse texture and shape features to get an accurate diagnosis for the extracted lung nodules. 3D Local Binary Pattern (LBP) and higher-order Markov Gibbs random field (MGRF) models are utilized to model the texture appearance due to their capability to give a precise description for the spatial non-uniformity in the texture of the nodules. Spherical Harmonic expansion and some basic geometric features are utilized to model the shape features due to their capability to give a full description of the shape complexity of the nodules. Finally, all the modeled features are fused and fed to a stacked autoencoder to differentiate between the malignant and benign nodules. Our framework is evaluated using 727 nodules which are selected from the Lung Image Database Consortium (LIDC) dataset, and achieved classification accuracy, sensitivity, and specificity of 92.66%, 95.70%, and 90.40% respectively.