Feature data processing: Making medical data fit deep neural networks

Abstract With the rapid development of artificial intelligence technology, deep neural networks (DNNs), especially convolutional neural networks (CNNs) and recurrent neural networks (RNNs), have been widely used. However, the current application fields of DNN are severely limited. For example, CNN is mainly used to process image format data, and RNNs are often used to process voice and text format data. How to use DNNs to process general medical data, especially feature data, is gradually becoming a popular research topic. To effectively improve the current situation of blindly using DNN on feature data, this paper proposes a practical, systematic feature data processing system (FDPS). Using multiple data processing methods, the processed data are more suitable for analysis using DNN. Some constructive advice on choosing a better DNN model architecture suitable for training such data is also provided. Then, network traffic data of the Internet of Medical Things (IoMT) are used as an example to verify the effectiveness of the proposed system separately using CNN and RNN. The experimental results show that the proposed approach can use fewer training data and that a simpler model architecture achieves better performance compared with other existing methods. To the best of our knowledge, this paper is the first to clearly define feature data, propose a detailed and systematic processing method to make it suitable for DNN training, and provide some specific suggestions on how to choose an appropriate DNN model architecture. Furthermore, the system can provide an effective reference for the promotion of DNN applications, especially for the security and analysis of IoMT.