Interstitial lung disease diagnosis and prognosis using an AI system integrating longitudinal data

Area of Science:

• Radiology
• Pulmonology
• Artificial Intelligence in Medicine

Background:

• Accurate interstitial lung disease (ILD) diagnosis requires integrating radiologic, pathological, and clinical data.
• Managing ILD necessitates regular computed tomography (CT) and lung function tests for monitoring progression and treatment response.
• Classifying ILD subtypes and predicting survival dynamically is challenging due to disease complexity and irregular patient data.

Purpose of the Study:

• To develop a deep learning system for diagnosing five ILD subtypes using multimodal data.
• To create a transformer model for predicting 3-year patient survival rates in ILD.
• To assist clinicians in ILD diagnosis, classification, and prognosis prediction.

Main Methods:

• Utilized RadImageNet pretrained models for ILD diagnosis.
• Employed a transformer model for 3-year survival rate prediction.
• Integrated clinical history and CT scan data for a multimodal deep learning approach.

Main Results:

• The system can diagnose five types of ILD.
• The system can dynamically predict a patient's 3-year survival rate.
• Deep learning aids in classifying ILD patients and predicting disease prognosis.

Conclusions:

• A deep learning system can effectively diagnose ILD subtypes using multimodal data.
• The proposed model assists clinicians in dynamic prognosis prediction for ILD patients.
• This AI-driven approach enhances ILD patient management and clinical decision-making.