A machine learning approach for preoperatively assessing pulmonary function with computed tomography in patients with...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonary Medicine

Background:

• Accurate pulmonary function assessment is crucial for lung cancer patients, especially pre-surgery.
• It aids in monitoring, predicting surgical impact, and optimizing recovery.
• Deep learning on CT scans is explored for this assessment.

Purpose of the Study:

• To evaluate a deep learning approach for assessing pulmonary function using CT scans in lung cancer patients.
• To compare machine learning model outcomes with established clinical criteria for pulmonary function.

Main Methods:

• 188 pathologically confirmed lung cancer patients were included.
• Radiomics features were extracted from airway, lobe, and whole lung regions using automated software.
• A logistic regression model was trained and tested for pulmonary function assessment.

Main Results:

• The model showed good performance in predicting forced vital capacity (FVC) and maximum vital capacity (VCmax) from lobe ROIs (r=0.714 for FVC, r=0.687 for VCmax).
• Accuracy was also demonstrated using airway and whole lung ROIs for VCmax and FVC prediction.
• Results indicate strong correlations between model predictions and clinical criteria (P<0.001).

Conclusions:

• Preoperative CT scans combined with deep learning and radiomics can effectively evaluate pulmonary function in lung cancer patients.
• This imaging-based approach offers clinicians reliable indicators for pulmonary status.
• The method has the potential to improve preoperative risk assessment and surgical planning.