An Effective Implementation of Activation Functions in Nodule Classification

The categorization of lung nodules is crucial for promptly identifying and evaluating lung-related illnesses. This research explores the efficient utilization of activation functions in classifying nodules within CT scan images. The study compares model performances with both standard benchmarks and a custom-built 3D Convolutional Neural Network (3D CNN). The effectiveness of these models is assessed by comparing their performance metrics, such as accuracy and ROC curve (receiver operating characteristic curve). Our custom build model is a fine tuned model in the aspect of model architecture, data normal-ization data augmentation and other hyper tuning parameters except activation functions. Model performance will be evaluated on the basis of accuracy on the implementation of activation functions - Tanh, ReLU (Rectified Linear Unit), softmax, Leaky ReLU (Leaky Rectified Linear Unit) etc.