Fibro-CoSANet

Fibro-CoSANet predicts prognosis of idiopathic pulmonary fibrosis (IPF) by modeling decline in forced vital capacity (FVC) from computed tomography (CT) images and demographic data using a multi-modal deep learning approach.

Key Features:

• End-to-end multi-modal learning: Integrates CT images and demographic information within a single trainable model.
• Data modalities: Uses computed tomography (CT) imaging paired with patient demographic data as inputs.
• Model architecture: Implements a convolutional neural network (CNN) framework for feature extraction and prediction.
• Attention mechanism: Incorporates a stacked attention layer to focus feature extraction on relevant data regions.
• Prediction target: Predicts decline in forced vital capacity (FVC) to estimate IPF prognosis.
• Validation dataset: Evaluated on the OSIC Pulmonary Fibrosis Progression Dataset.
• Evaluation metric: Reported a modified Laplace log-likelihood score of -6.68.
• Deep learning approach: Leverages deep learning to jointly process complex medical imaging and demographic features.

Scientific Applications:

• IPF prognosis: Estimates disease progression in idiopathic pulmonary fibrosis via FVC decline prediction.
• Clinical decision support: Provides quantitative prognostic estimates to inform treatment planning.
• Research on disease dynamics: Supports investigation of progression patterns and personalized treatment strategies in pulmonary fibrosis.

Methodology:

End-to-end multi-modal convolutional neural network integrating CT images and demographic information, augmented with a stacked attention layer; validated on the OSIC Pulmonary Fibrosis Progression Dataset and evaluated using the modified Laplace log-likelihood (-6.68).