Deep Learning for Preoperative Pulmonary Assessment in Thoracic CT

Preoperative pulmonary function tests are crucial in assessing perioperative complications or mortality risks and providing decision support for thoracic surgery. However, traditional pulmonary function assessment methods have significant limitations, including long testing durations, difficulties in patient cooperation, high false-negative rates, and numerous contraindications. Thus, our study optimized the final model based on 1500 single inspiratory phase CTs by transferring model parameters trained on 500 dual-phase respiratory CTs, enhancing its predictive capabilities for pulmonary function. This adjustment suits real-world application demands, offering more convenient, comprehensive, and personalized preoperative pulmonary function assessment support. Our study optimized the final model based on 1500 single inspiratory phase CTs by transferring model parameters trained on 500 dual-phase respiratory CTs, enhancing its predictive capabilities for pulmonary function. This adjustment suits real-world application demands, offering more convenient, comprehensive, and personalized preoperative pulmonary function assessment support.

• Single inspiratory phase computed tomography. (Other): Utilizing deep learning technology in conjunction with single inspiratory phase computed tomography images to accurately predict the pulmonary function indicators of preoperative thoracic surgery patients.
• Respiratory dual-phase computed tomography. (Other): Utilizing deep learning technology in conjunction with respiratory dual-phase computed tomography images to accurately predict the pulmonary function indicators of preoperative thoracic surgery patients.

Primary outcomes

• Mean Absolute Error（MAE） [2 years]: Used to assess the discrepancy between pulmonary function predictions made by the deep learning algorithm and actual results obtained from pulmonary function tests (measured with a spirometer).

Secondary outcomes

• Concordance Correlation Coefficient（CCC） [2 years]: Used to assess the discrepancy between pulmonary function predictions made by the deep learning algorithm and actual results obtained from pulmonary function tests (measured with a spirometer).

Sex: All, ages: 18 Years to 75 Years

Inclusion Criteria:

* (1) Signing of the informed consent form;
* (2) Male or female, aged 18-75 years;
* (3) Undergoing elective thoracic surgery;
* (4) Good preoperative pulmonary function cooperation and complete reporting;
* (5) Preoperative chest single/dual phase CT scans without significant artefacts and with complete imaging;
* (6) The interval between preoperative pulmonary function and single/dual phase CT scans does not exceed one month.

Exclusion Criteria:

* (1) Poor preoperative pulmonary function cooperation or missing reports;
* (2) Preoperative chest single/dual phase CT scans exhibit significant artefacts or image omission;
* (3) The interval between preoperative pulmonary function and single/dual phase CT scans exceeds one month;
* (4) Complication with severe respiratory disorders (such as lung transplantation, pneumothorax, giant bullae, etc.);
* (5) Coexisting with other severe functional impairments;
* (6) Patients with obstructive lesions such as airway or esophageal stenosis;
* (7) Height beyond the predicted equation range (Female < 1.45m; Male < 1.55m);
* (8) Medication use before pulmonary function testing that does not meet the cessation guidelines;
* (9) Pulmonary function report quality graded D-F.

• Jianxing He, MD, Principal investigator, Department of Cardiothoracic Surgery, the First Affiliated Hospital of Guangzhou Medical College