Body Composition Assessment Of Prognosis Of Patients With Small Cell Lung Cancer And Construction Of A Deep Learning Model For CT Body Composition Segmentation

Objective:Deep learning model was used to automatically segment human composition at the level of the fourth thoracic vertebra in chest CT of patients with small cell lung cancer,and the correlation analysis between human composition indexes and survival of patients with small cell lung cancer was conducted,and the role of human composition analysis in survival prediction of patients with small cell lung cancer was explored.Methods:Patients diagnosed with small cell lung cancer at the China-Japan Union Hospital of Jilin University between January 2016 and December 2019 were selected for retrospective analysis.ITK-SNAP was used to the segment pectoralis major,pectoralis minor,subcutaneous fat and mediastinal fat of 230 patients,and their areas were measured by Image J.Meanwhile,the deep learning model was trained and the segmentation accuracy of the model was evaluated without considering the characteristics of tumors.Sixty-nine patients were evaluated for association between body composition and survival in small cell lung cancer patients.Ka Plan-Meier survival analysis and Log-rank test were used to compare the survival relationship,and Cox proportional risk model was used for multivariate analysis.Results:1.In patients with small cell lung cancer,the overall survival of patients with high pectoral muscle area was significantly higher than that of patients with low thoracic muscle area（HR=0.824,95% CI 0.511～1.329,P=0.044）,and the difference was statistically significant.Multivariate analysis showed that breast muscle area,mediastinal fat area,age at diagnosis,and clinical stage were independent factors affecting the prognosis of patients with small cell lung cancer（HR=1.725,95% CI1.054～2.822,P=0.030;HR=1.749,95% CI 1.067～2.868,P=0.027;HR=2.285,95% CI 1.337～3.907,P=0.003）.2.Log-rank test results showed that OS of high and low muscle area groups was statistically significant in patients with small cell lung cancer（median 11 vs 7,95%CI8.148-13.852 vs 4.682-9.318,P=0.034）.In patients with small cell lung cancer,the OS of time limit and extensive stage was statistically significant（median 12 vs 7,95%CI 6.170-17.830 vs 4.596-9.404,P=0.001）.OS of low age group and high age group in small cell lung cancer patients was statistically significant（median 11 vs 6,95%CI 9.280-12.720 vs 4.480-7.520,P=0.011）.3.In the test set of the U-Net++ model,the Dice coefficient and Jaccard index of pectoralis major segmentation were 0.86 and 0.78,while the Dice coefficient and Jaccard index of pectoralis minor segmentation were 0.80 and 0.69.Conclusion:1.Pectoral muscle area has an independent prediction effect on OS in patients with small cell lung cancer,and is an independent risk factor for OS in patients with small cell lung cancer.2.BMI,subcutaneous fat area and mediastinal fat area have no effect on small cell lung cancer and OS,and have no predictive effect on small cell lung cancer patients.3.Automatic segmentation of pectoral muscle can be realized by using U-net++.