Answer Happy

13. One hospital asks for your advice to do lung nodule classification with deep CNNs. The hospital has a considerable large connection of CT images of lung nodules and their labels (i.e., benign or cancerous). Which strategy will you mostly suggest considering both performance and efficiency for this task? A. Use a pretrained model (such as VGG16, ResNet) as feature extractor. The features then are used as an input to a network with two full-connected layers. B. Fine-tune all the convolutional layers in a pretrained model (such as VGG16, ResNet) and train two fully-connected layers. C. Fine-tune a few last convolutional layers in a pretrained model (e.g., VGG16, ResNet) and train two fully-connected layers. D. Define a middle-size CNN architecture, and train it on this dataset from scratch.