Deep learning based CT images for lung function prediction in patients with chronic obstructive pulmonary disease
AI-Powered Lung Function Prediction for COPD
This study developed a deep learning (DL)-based multimodal feature fusion model to accurately estimate pulmonary function test (PFT) parameters from chest CT images in COPD patients. By combining CT image features, radiomics, and clinical data, the model achieved high accuracy in predicting FEV1, FVC, FEV1/FVC, FEV1%, and FVC%. The model demonstrated strong correlations with measured PFT values and showed potential for real-time prediction and opportunistic screening, especially for patients unable to undergo traditional PFTs. The Grad-CAM visualization technique revealed that the model focuses on lung tissue surrounding bronchi and lower lung fields, correlating with airway obstruction and emphysema distribution.
Executive Impact
Transforming COPD Diagnosis & Management
Deploying this AI model could significantly improve early diagnosis and personalized management of COPD, reducing diagnostic delays and facilitating timely interventions. It offers a non-invasive alternative for PFT assessment from existing CT scans, potentially impacting millions of undiagnosed patients globally.
0.84
Pearson Correlation (r) for FEV1
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
The core of our multimodal feature fusion model consists of three parts: an independent feature transformation layer, a gated network layer for dynamic feature weight generation, and a goal-specific prediction head layer for regression. This architecture, specifically an MLP-based approach with a DenseNet encoder for CT image feature extraction, allows for the integration of multi-source data (CT images, demographics, radiomics) to capture complex pathological changes in COPD. The use of residual learning adapts to different PFT metric dimensions, enhancing overall robustness and accuracy. This ensures that the model can handle the multi-dimensional and spatially heterogeneous nature of COPD pathology more effectively than previous methods.
The model's predictive performance was evaluated using Mean Absolute Error (MAE), Mean Squared Error (MSE), and Pearson correlation coefficient (r). For FEV1, MAE was 0.34, MSE 0.20, and r 0.84. For FVC, MAE 0.42, MSE 0.31, and r 0.81. FEV1/FVC showed MAE 6.64, MSE 0.73, and r 0.77. FEV1% yielded MAE 13.42, MSE 3.01, and r 0.73, while FVC% had MAE 13.33, MSE 2.97, and r 0.61. Classification performance for high-risk populations also showed high AUCs (e.g., FEV1 AUC 0.94, FVC AUC 0.91) and accuracies. The Bland-Altman plots confirmed good consistency between predicted and actual values for all PFT parameters. These results highlight the model's stability and generalization ability in predicting critical lung function parameters.
This AI model provides significant clinical value by enabling quantitative assessment of spatial distribution characteristics associated with COPD from CT images. It can identify abnormal lung areas correlated with PFT results, establishing a mapping between structural changes and overall function. This offers a new approach for opportunistic screening, particularly beneficial for patients already undergoing chest CT for other reasons or those unable to complete traditional PFTs. The model supplements, rather than replaces, PFTs, aiding in refined phenotypic analysis and risk assessment by incorporating imaging features related to both emphysema and small airway disease, reflecting the multifactorial nature of COPD pathophysiology. This contributes to earlier detection and better-informed treatment strategies.
0.84
Pearson Correlation (r) for FEV1
Multimodal Feature Fusion Workflow
Chest CT Imaging & PFT Data Collection
→
Image Preprocessing & Feature Extraction (DenseNet)
→
Clinical & Radiomics Data Integration
→
Multilayer Perceptron (MLP) Fusion Model
→
PFT Parameter Prediction (FEV1, FVC, etc.)
→
Clinical Validation & Interpretation (Grad-CAM)
| Feature | Our AI Model | Traditional PFT |
|---|---|---|
| Data Source |
|
|
| Accessibility |
|
|
| Insights |
|
|
| Speed & Efficiency |
|
|
Accelerated COPD Diagnosis
Scenario: A 72-year-old patient presented with mild, non-specific respiratory symptoms, but was unable to complete a full traditional PFT due to a co-existing cardiac condition. Standard chest CT was performed for an unrelated issue.
Outcome: Our AI model analyzed the existing CT images, integrating them with limited clinical data, and accurately predicted the patient's FEV1% at 65% and FEV1/FVC at 60%, indicating early-stage COPD (GOLD 1). This provided a crucial early diagnosis without requiring additional patient effort, allowing for timely intervention and disease management. Traditional methods would have either delayed diagnosis or missed it entirely until symptoms progressed.
Estimate Your Enterprise AI ROI
Understand the potential time and cost savings this AI solution can bring to your organization based on key operational metrics.
Estimated Annual Savings
$0
Estimated Annual Hours Reclaimed
0
Your AI Implementation Roadmap
A structured approach to integrating cutting-edge AI into your operations, ensuring seamless adoption and measurable results.
Data Preparation & Annotation
Gathering and meticulously annotating diverse CT image datasets from various patient populations to ensure robust model training and validation.
Model Customization & Training
Tailoring the deep learning architecture to specific institutional needs, followed by iterative training and fine-tuning with local data for optimal performance.
Integration with Clinical Workflows
Seamlessly integrating the AI prediction model into existing PACS and EMR systems for real-time analysis and reporting during routine CT scans.
Pilot Deployment & Validation
Conducting a pilot program in a clinical setting to validate the model's accuracy, efficiency, and clinical utility with real-world patient cases, gathering feedback for refinement.
Scalable Rollout & Monitoring
Expanding the AI solution across the enterprise, establishing continuous monitoring for performance, data drift, and ensuring ongoing regulatory compliance and ethical use.
Ready to Transform Your Enterprise with AI?
Schedule a personalized consultation with our AI specialists to explore how this technology can be tailored to your organization's unique needs.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat