Predicting Pulmonary Fibrosis Progression Using EfficientNet-LSTM Model: A Hybrid Approach

The progressive lung condition known as idiopathic pulmonary fibrosis is defined by lung tissue scarring, which impairs lung function. Traditional models fail to accurately capture both structural change on CT scans and temporal trends in clinical variables and thus cannot be trusted for predictive accuracy. A good prediction method is needed to improve patient monitoring and treatment planning. This paper suggests a hybrid deep learning architecture that combines EfficientNet for the extraction of fibrosis features from DICOM CT scans with long short term memory networks to track the evolution of the disease over time. Compared to single CNN or LSTM models, the combination of spatial and temporal data increases the precision of FVC prediction. The models are trained and validated on the OSIC Pulmonary Fibrosis Progression dataset, which shows better performance and less prediction error. The results show how deep learning can be used to improve the prognosis of IPF and provide a more accurate method for monitoring the condition and planning the treatment.