Sky-Net: A Deep Learning Approach to Predicting Lung Function Decline in Sufferers of Idiopathic Pulmonary Fibrosis

Abstract

Idiopathic Pulmonary Fibrosis (IPF) is a kind of Interstitial Lung Disease (ILD) that can be recognized by observing an atypical formation and accumulation of fibrotic tissue in the lungs. The lung's alveolar structure is damaged; as a result, people afflicted with IPF experience increasingly restricted lung capacity as time progresses. Diagnosis of this disease is typically performed by analyzing the patient's computed tomography (CT) scans and measuring their Forced Vital Capacity (FVC) using a Spirometer. However, the absence of an apparent cause of IPF restricts the ability of doctors to accurately diagnose the patient. Furthermore, IPF progression in patients is highly volatile and unpredictable, which means that one patient's health could deteriorate significantly quicker compared to another. Taking the problems mentioned above into account, in this paper a 3-layer ResNet machine learning model is proposed that determines the rate of lung function decline of sufferers from IPF. Proposed model is applied on the “OSIC Pulmonary Fibrosis Progression” dataset publicly available on Kaggle, and compare it against various state-of-the-art models and winning Kaggle entries.