Publication Details
Overview
 
 
 

Contribution to journal

Abstract 

Convolutional neural networks have been widely used to detect and classify variousobjects and structures in computer vision and medical imaging. Access to large sets of annotateddata is commonly a prerequisite for achieving good performance. Before the deep learning era,systems based on handcrafted features were employed, which typically required less annotated databut also reached inferior performance. In this work, we investigate the benefit of combining deeplearning using a convolutional neural network (CNN), with handcrafted features for lung noduledetection from CT imaging. We investigate three fusion strategies with increasing complexity,and evaluate their performance for varying amounts of training data. Our results indicate thatcombining handcrafted features with a 3D CNN approach significantly improves lung noduledetection performance in comparison to an independently trained CNN model, regardless of thefusion strategy. Comparatively larger increases in performance were obtained when less trainingdata was available. The fusion strategy in which features are combined with a CNN using a singleend-to-end training scheme performed best overall, allowing to reduce training data by 33% to 43%,while maintaining performance. Among the investigated handcrafted features, those that describethe relative position of the

Reference