Deep learning for C-reactive protein prediction

Neural networks have been extensively utilised to perform biosignal prediction over the last two decades. This study proposes a long short-term memory-based recurrent neural network, to predict future state in a C-reactive protein (CRP) time series from a given cancer patient. Experiments are conducted to demonstrate that a LSTM-based RNN is capable of CRP time series forecasting using data obtained from ten patients with melanoma. Since CRP is biomarker of immune system activity, the ability to forecast can potentially guide clinical decisions in cancer treatments.