Background: The prediction of postoperative respiratory function is necessary in identifying patients that are at greater risk of complications. There are not enough studies on the effect of the diaphragm on postoperative respiratory function prediction in lung cancer surgical patients. The objective of this study is to estimate the precision of machine learning methods in the prediction of respiratory function in the immediate postoperative period and how diaphragm function contributes to that prediction.
Materials and methods: Our prospective study included 79 patients who underwent lung cancer surgery. Diaphragm function was estimated by its mobility measured both ultrasonographically and radiographically and by noninvasive muscle strength tests. We present a new machine learning multilayer regression metamodel, which predicts FEV1 for each patient based on preoperative measurements.
Results: The proposed regression models are specifically trained to predict FEV1 in the immediate postoperative period and were proved to be highly accurate (mean absolute error in the range from 8 to 11%). Predictive models based on resected segments give two to three times less precise results. Measured FEV1 was 44.68% ± 14.07%, 50.95% ± 15.80%, and 58.0%1 ± 14.78%, and predicted postoperative (ppo) FEV1 was 43.85% ± 8.80%, 50.62% ± 9.28%, and 57.85% ± 10.58% on the first, fourth, and seventh day, respectively. By interpreting the obtained model, the diaphragm contributes to ppoFEV1 13.62% on the first day, 10.52% on the fourth, and 9.06% on the seventh day.
Conclusion: The machine learning metamodel gives more accurate predictions of postoperative lung function than traditional calculations. The diaphragm plays a notable role in the postoperative FEV1 prediction.
Keywords: Diaphragm; Lung cancer; Machine learning; Prediction; Respiratory function.
© 2023. The Author(s).