Automated quantitative evaluation of thymic involution and hyperplasia on plain chest CT

Abstract

Objective: To establish an automatic method to quantify thymic involution and hyperplasia based on plain chest computed tomography (CT). Methods: We defined the thymic region for quantification (TRQ) as the target region. We manually segmented the TRQ in 135 CT studies, followed by construction of segmentation neural network (NN) models based on the data. We developed the estimator of thymic volume (ETV), a measure of the thymic tissue volume in the segmented TRQ. The Hounsfield unit (HU) value and volume of the TRQ were measured, and the ETV was calculated in each CT study from 853 healthy subjects. We investigated how these measures were related to the age and sex using quantile additive regression models. We defined the ETV z-score, an age- and sex-adjusted version of ETV, to distinguish between subjects with thymic hyperplasia (18 cases) and healthy subjects. A receiver operating characteristic (ROC) curve analysis was conducted. Results: A significant correlation between the NN-segmented and manually segmented TRQ was seen for both the HU value and volume of the TRQ (r = 0.996 and r = 0.986 respectively). The ETV could detect age-related decline in the thymic tissue volume (p < 0.001). No statistically significant difference was detected between male and female subjects (p = 0.19). The ETV was significantly higher in the thymic hyperplasia group as compared with that in the healthy control group (p < 0.001). The ETV z-score could distinguish between subjects with thymic hyperplasia and healthy subjects, with the ROC curve analysis revealing an area under the curve (AUC) of 0.88 (95% CI: 0.75-1.0). Conclusion: Our method enabled robust quantification of thymic involution and hyperplasia. The results were consistent with the trends found in previous studies.