An accurate computer aided diagnostic (CAD) system is very significant and critical for early detection of lung cancer. A new framework for lung nodule classification is proposed in this paper using different imaging markers from one computed tomography (CT) scan. Texture and shape features are combined together to show the main discriminative characteristics between malignant and benign pulmonary nodules. 7th-Order Markov Gibbs random field, (MGRF), is implemented to give a good description of the nodule’s appearance by involving the spatial data. A Various-views Marginal Aggregation Curvature Scale Space (MACSS) and the primitive geometrical properties are used to indicate the nodule’s shape complexity. Eventually, all these modeled descriptors are combined using a stacked autoencoder and softmax classifier to give the final diagnosis. Our system has been validated using 727 samples from the Lung Image Database Consortium. Our diagnosis framework’s accuracy, sensitivity, and specificity were 94.63%, 93.86%, 94.78% respectively, showing that our system serves as an important clinical assistive tool.