A combined CNN-LSTM and LSTM-QRNN model for prediction of Idiopathic Pulmonary Fibrosis Progression using CT Scans and Clinical Data

Idiopathic Pulmonary Fibrosis (IPF), which causes scarred tissues and lung function damage over time, is a serious progressive lung disease. In addition, this chronic disease is irreversible, with unknown cures and unknown causes, so it is difficult to treat and becomes a challenge faced by doctors and others. Furthermore, Forced Vital Capacity (FVC) can assess the progression of lung function and it can assist to detect the disease in the early stage, so doctors have more time to give appropriate treatment and patients have more opportunities to increase survival time. Thus, the hybrid model convolutional neural network - long short-term memory (CNN-LSTM) and long short-term memory - quantile regression neural network (LSTM-QRNN) have been presented in this paper to predict FVC values by using CT scan images and clinical data. The experiment results show that the model also achieved the better modified Laplace Log Likelihood score in the private leader-board in Kaggle OSIC11https://www.kaggle.com/competitions/osic-pulmonary-fibrosis-progression dataset.