Accurate Prediction of Pulmonary Fibrosis Progression Using EfficientNet and Quantile Regression: A High Performing Approach

Abstract

Pulmonary fibrosis (PF) is a chronic lung disease characterized by the formation of scar tissue in the lungs, leading to difficulty breathing and a reduced ability to oxygenate the blood. The progression of PF is difficult to predict, and current methods of diagnosis and treatment are often ineffective. In this study, we propose to use EfficientNet, utilizing a cutting-edge convolutional neural network (CNN) architecture and quantile regression (QR) to predict the progression of PF in patients. Our approach includes analyzing data from the OSIC dataset, the biggest publicly accessible dataset containing medical imaging, patient demographics, and lab results. The analyzed data was trained on an EfficientNet model and QR to predict the progression of the disease, as well as estimate the uncertainty of the predictions. The performance of the model was evaluated using Laplace-Log-Likelihood. The results demonstrate that the proposed approach outperforms existing literature in predicting pulmonary fibrosis progression, with the highest score (-6.64). This approach has the potential to aid in the development of new treatments for this disease.