The value of quantitative artificial intelligence measurement in evaluation of CT dynamic changes for COVID-19
10.3760/cma.j.cn112149-20200330-00474
- VernacularTitle:人工智能定量测量对新型冠状病毒肺炎患者胸部CT炎性病灶动态变化的评估价值
- Author:
Dan DU
;
Yuanliang XIE
;
Hui LI
;
Shengchao ZHAO
;
Yi DING
;
Pei YANG
;
Bin LIU
;
Jianqing SUN
;
Xiang WANG
- From:
Chinese Journal of Radiology
2021;55(3):250-256
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To investigate the value of artificial intelligence (AI)-assisted quantitative measurement in evaluation of the dynamic changes of CT for COVID-19 pneumonia.Methods:The clinical and chest CT dynamic imaging data of 99 patients with confirmed COVID-19 pneumonia who were hospitalized in Wuhan Central Hospital Affiliated to Tongji Medical College of Huazhong University of Science and Technology from January 15, 2020 to March 10, 2020 were retrospectively analyzed. According to the definitive diagnosis, the 99 patients were classified into common ( n=36), severe ( n=33) and critical ( n=30) type, the CT imaging findings of each type were analyzed, including CT basic signs, total volume of pneumonia lesions and percentage of pneumonia lesions of the total lung volume (volume ratio). AI software was used to quantitatively evaluate the dynamic changes of chest CT images. The quantitative indicators included CT peak time of lesions, total volume of lesions peak, volume ratio of lesions peak, maximum growth rate of total volume and maximum growth rate of volume ratio. Kruskal-Wallis rank sum test was used to compare the difference of quantitative indexes between the 3 types, and χ 2 test or Fisher exact probability test was used to compare the difference of qualitative indexes between the 3 types. Sequence measurement and scatter plots were used to show the evolution trend of the volume ratio of the three types of COVID-19 pneumonia lesions. The ROC curve was used to analyze the value of the volume ratio of pneumonia lesions and its maximum growth rate in predicting the conversion of common pneumonia to severe or critical pneumonia. Results:There were statistically significant differences in age and gender distribution among patients with common, severe and critical COVID-19 ( P<0.05), the age of severe and critical types were significantly higher than that of common type ( P<0.01). Compared with common [2.5 (1.0, 5.0) d] and critical type[2.5 (1.0, 4.0) d], the time from onset to the first chest CT scan of severe type was prolonged [5.0 (2.5, 8.0) d, P<0.01]. There were statistically significant differences in involvement of multiple lung lobes (20 cases, 29 cases, 25 cases, χ2=10.403, P=0.006) in patients with common, severe and critical COVID-19 at the first scan, the incidence of the involvement of multiple lung lobes in severe and critical types was significantly higher than that of common type ( P=0.002). The volume ratios of patients with common, severe and critical COVID-19 at the first scan were statistically significant [1.0% (0.2%, 4.7%), 9.30% (1.63%, 26.83%), 2.10% (0.64%, 8.61%), Z=14.236, P=0.001], and the volume ratio of severe type was significantly higher than that of common type ( P<0.001), there was no statistically significant difference between common type and critical type ( P=0.062). Follow-up CT showed that the pneumonia lesions showed a dynamic transformation of progress and recovery, and it was seen that the coexistence of multiphase lesions. The trend line in the scatter plot of the three types of COVID-19 pneumonia lesions showed that the lesions in the advanced stage developed from less to more. The lesion peak volume ratios of the common, severe and critical types were 9.75% (4.83%, 13.18%), 29.80% (23.99%, 42.36%) and 61.81% (43.73%, 72.82%), respectively, the difference was statistically significant ( Z=74.147, P<0.001). The maximum growth rates of lesion volume ratio were 1.27% (0.50%, 1.81%)/d, 4.39% (3.16%, 5.54%)/d and 6.02% (4.77%, 9.96%)/d, respectively, the difference was statistically significant ( Z=52.453, P<0.001). The peak times of lesions were 12.0 (9.0, 15.0) d, 13.0 (10.0, 16.0) d and 16.5 (12.0, 25.0)d, respectively, the difference was statistically significant ( Z=9.524, P=0.009). Taking the volume ratio of pneumonia lesion 22.60% and the maximum growth rate of the volume ratio 1.875%/d as the boundary value, the sensitivity of diagnosing common type to severe or critical type was 92.10% and 96.83%, and the specificity was 100% and 80.56%, respectively. The area under the curve was 0.987 and 0.925, respectively. Conclusions:The lesions of COVID-19 pneumonia show a similar parabolic change on CT imaging. The use of AI technology to dynamitcally and accurately measure the CT pneumonia lesion volume ratio is helpful to evaluate the severity of the disease and predict the development trend of the disease. Patients with a rapid growth of volume ratio are more likely to become severe or critical type.
