Deep - Learning - Based Automatic Segmentation and Quantitative Measurement of Normal Spleen in Chinese Adults

Objective： To perform quantitative measurements of the spleen's morphological properties and to develop a 3D U-Net-based segmentation and automatic model measurement of the normal spleen using CT images. Additionally, the study intends to examine age-related changes in the normal spleen's volume as well as the enhancement characteristics of the normal spleen on contrast-enhanced CT scans. Materials and Methods： A total of 2856 images of spleens(dataset 1) were collected from four public sources and randomly split into three groups: training (2292 images), validation (280 images), and test (284 images). The segmentation efficiency for the spleen was evaluated by the Dice similarity coefficient (DSC), volume similarity (VS), Hausdorff distance (HD), and average HD. Another dataset of 3490 normal spleen CT images (dataset 2) was obtained for external validation, including 862 non-contrast images, 744 arterial phase images, 947 portal venous phase images, and 937 delayed phase images. Then, 947 portal venous phase image series of inpatient contrast-enhanced abdominal CT examinations with normal spleen were included for analysis and grouped by every 10-year gap. A 3D U-Net-based segmentation model was used to predict spleen labels, followed by manual label modifications as appropriate. Quantitative parameters of the spleen (volume, CT value, and diameter) were then analyzed. Results： In dataset 1, the testing dataset (N = 284) showed segmentation performance with a Dice Similarity Coefficient (DSC) of 0.982 [0.975–0.988], Volume Similarity (VS) of 0.995 [0.991,0.998] , Hausdorff Distance (HD) of 3.047 [2.578, 4.653] mm, and Average HD of 0.014 [0.009, 0.019] mm. In dataset 2, the distribution ranges of the three-dimensional diameters (x, y, z) of the spleen were as follows: for males, 9.20 [8.40 - 9.95] cm (median [interquartile range]), 9.50 ± 1.97 cm (mean ± standard deviation), and 9.40 ± 1.91 cm; for females, 9.50 ± 1.09 cm, 8.92 ± 1.83 cm, and 8.64 ± 1.67 cm. The distribution range of the spleen volume was 213.74 [158.4, 284.88] cm³ for males and 163.7 [125.99, 217.72] cm³ for females. In the enhanced scan images of the spleen, it was found that the CT values of the spleen in the arterial phase, portal venous phase, and delayed phase were all higher in females than in males. With the increase in age, the spleen volume of males showed a trend of first increasing and then decreasing, reaching its peak at the age of 28–37, with a peak value of 273.75 ± 92.96 cm³. The spleen volume of females gradually decreased with the increase of age, with the peak value of 225.01 ± 65.53 cm³. In all age groups, the spleen volume of females was smaller than that of males. Conclusion： The spleen segmentation tool based on deep learning can segment the spleen on CT images and measure its normal diameter, volume, and CT value accurately and effectively.