The Multi-Task Deep Model-Based Pneumonia Detection

Abstract

Abstract As a common lung disease, pneumonia affects millions of people worldwide each year and is often detected during physical examination using chest X-ray images (CXRs), which are usually diagnosed by radiologists. This time-consuming task often leads to fatigue-based diagnostic error and cannot be performed in countries or areas lack of radiologists. In this work, we proposed a multi-task learning model: Mt-pnet. We learn from the experience of imaging doctors in the diagnosis of pneumonia, let the network pay attention to the patient’s gender and age information in the diagnosis of pneumonia. The combined information is helpful for the network to learn robust image features for pneumonia diagnosis. At the same time, we also found that using multi-task model is helpful to improve the recall of patients with pneumonia, which has important clinical significance.