Prediction of Relapse of Acute Myeloid Leukemia based on Chest CT Features using a Nomogram

To develop a novel model for predicting relapse risk in patients with acute myeloid leukemia (AML) based on chest CT features.

A total of 131 patients from Wuhan Tongji Hospital and 30 patients from Wuhan Central Hospital with AML received standardized treatment and chest CT. The automatic segmentation of 51 organs was completed with TotalSegmentator. Follow-up on the survival and recurrence status of patients was performed to obtain the relapse-free survival (RFS) after CT examination. Feature selection was performed based on reproducibility, significant differences between patients with short and long RFS cases, and Least Absolute Shrinkage and Selection Operator (LASSO) regression. Cox proportional hazards regression was performed based on selected features. A nomogram was constructed and was to predict outcomes in the external validation cohort. Concordance index (C-index) and time-dependent area under the curve (AUC) were calculated for the model.

RFS and 5457 features were obtained for each case. After False discovery rate (FDR) control, the short and long RFS cases differed in 27 features. By using LASSO-Cox and initial Cox proportional hazards regression, 5 features were used in the final Cox model. In the external cohort, the Cindex of the prediction model was 0.71. The time-dependent AUCs were 0.70, 0.71, 0.80, and 0.85, respectively, for the time points of 10, 20, 30, and 40 months.

Minimum residual disease (MRD) is closely related to relapse of AML. In the study, some CT features seemed to correlate with RFS of AML. A possible explanation is that these features reflect MRD in different organs. The nomogram displays the numerical correlations between variables through geometric shapes, and converts predictive indicators into visual scale segments based on regression models. The column chart used in this study helps to calculate the risk quickly.

A total of 5 CT features from the heart, pulmonary venous system, T8, T11 vertebra, and adrenal gland were useful in predicting AML RFS. A nomogram was useful and convenient in calculating the risk.