Page 117 - 南京医科大学自然版

第46卷第8期 冯 婕，渠 强，雷心仪，等. 冠状动脉造影及介入治疗患者辐射剂量的影响因素分析及预测模型
2026年8月 构建［J］. 南京医科大学学报（自然科学版），2026，46（8）：1218-1227，1268 ·1223 ·

架最小直径、手术时间和透视时间。随机森林+最 三酯、HDL⁃C、LVDd、LVEF、手术时间和透视时间。
优子集回归模型纳入的候选变量为年龄、BMI、甘油 随机森林+LASSO回归模型纳入的候选变量为BMI、

表2 不同变量筛选和建模策略下辐射剂量预测模型的构建及性能比较
Table 2 Development and performance comparison of radiation dose prediction models using different variable selection
and modeling strategies

Training set Testing set
Algorithm Initially included variables Finally included variables
RMSE MAE R 2 RMSE MAE R 2
Random forest algorithm Age，BMI，SBP，neurologic symptoms， - - - - - - -
TC，TG，HDL⁃C，LDL⁃C，eGFR，
hs⁃cTnT，LVDd，LVDs，LVEF，minimum
stent diameter，procedure time，fluo⁃
roscopy time
Random forest algorithm+ BMI，TG，HDL⁃C，LVDd，LVDs，mini⁃ BMI，HDL ⁃ C，LVDd，proce⁃ 289.5 168.7 0.38 174.1 139.7 0.51
univariate linear regression mum stent diameter，procedure time， dure time，fluoroscopy time
fluoroscopy time
Random forest algorithm+ Age，BMI，TG，HDL⁃C，LVDd，LVEF， Age，BMI，TG，HDL⁃C，LVDd， 282.8 168.8 0.41 170.1 136.1 0.55
best subset regression procedure time，fluoroscopy time LVEF，procedure time，fluoros⁃
copy time
Random forest algorithm + BMI，neurologicsymptoms，TG，HDL⁃C， BMI，neurologic symptomsyes， 287.5 169.7 0.39 175.9 141.4 0.50
LASSO regression LVDd，procedure time，fluoroscopy HDL⁃C，LVDd，procedure time，
time fluoroscopy time
BMI：body mass index；eGFR：estimated glomerular filtration rate；HDL⁃C：high⁃density lipoprotein cholesterol；hs⁃cTnT：high⁃sensitivity cardiac
troponin T；LASSO：least absolute shrinkage and selection operator；LDL⁃C：low⁃density lipoprotein cholesterol；LVDd：left ventricular end⁃diastolic
diameter；LVDs：left ventricular end⁃systolic diameter；LVEF：left ventricular ejection fraction；MAE：mean absolute error；RMSE：root mean square
error；SBP：systolic blood pressure；TC：total cholesterol；TG：triglyceride.

Fluoroscopy time Fluoroscopy time
Procedure time eGFR
HDL⁃C HDL⁃C
LVDd Procedure time
BMI SBP
hs⁃cTnT LVDd
TG BMI
Minimum luminal area stenosis LVDs
TC TG
LDL⁃C hs⁃cTnT
LVDs Age
Neurologic symptoms LVEF
Number of conventional balloons Lp（a）
Total stent length Heart rate
Minimum stent diameter TC
Number of diseased coronary arteries CK⁃MB
Tube voltage DBP
Other atypical symptoms LDL⁃C
Lp（a） Glucose
ECG abnormalities Minimum luminal area stenosis
DBP Special cardiac examinations
Beta⁃blocker Tube voltage
SBP Duration of CAD
Cigarettes per day Cigarettes per day
CCB Number of coronary stents
Heart rate Total stent length
Number of drug⁃coated balloons Neurologic symptoms
CK⁃MB Number of treated coronary vessels
Special cardiac examinations Minimum stent diameter
Coronary artery calcification Palpitation
0 2.5 5.0 7.5 10.0 12.5 0 5 10 15 20 25 30 35
Mean decrease in accuracy（%） Increase in node purity（×10 ）
5
A：Variable importance measured by mean decrease in accuracy. B：Variable importance measured by increase in node purity.
图2 随机森林模型中辐射剂量潜在影响因素的重要性排序
Figure 2 Variable importance of potential predictors of radiation dose based on the random forest model