Deep Fuzzy Cognitive Map methodology for Non-Small Cell Lung Cancer diagnosis based on Positron Emission Tomography imaging

Non-Small Cell Lung Cancer (NSCLC) constitutes the major cause of cancer deaths worldwide in both men and women, accounting for approximately 85% of the total lung cancers. Early characterization of Solitary Pulmonary Nodules (SPNs) enables early treatment and can increase the survival rate. The present study focuses on classifying 2D SPN representations from a PET/CT scanner into two possible classes, namely benign and malignant SPN s. In this regard, 2D images were acquired from the scanner, where the SPN is alienated. Also, the SUVmax and SPN diameter data were included. A pre-trained RBG convolutional neural network processed the images to predict the class of each instance. The predictions were handled by a Fuzzy Cognitive Map with Particle Swarm Optimization in combination with the SUVmax and SPN diameter to provide a final decision. The proposed model, DeepFCM, has been evaluated with robust metrics like accuracy, loss, sensitivity, specificity, and precision and extracted 94.71±2.99, 0.05, 92.06, 96.98, 91.91, accordingly. DeepFCM provides explainability and transparency by modeling a complex system in concepts and providing the importance of each feature.