Page 103 - 南京医科大学自然版
P. 103

第46卷第7期               王   添,许正刚,吴怀玉,等. 时序影像组学预测肝癌免疫联合治疗疗效[J].
                  2026年7月                  南京医科大学学报(自然科学版),2026,46(7):1046-1054,1063                    ·1049 ·


                                           表1 融合预测模型的16个最优特征及其LASSO系数
                         Table 1 The sixteen optimal features of the combined prediction model and their LASSO coefficients

                 No.               Feature name               Feature type       Data source      LASSO coefficient
                 01     AP_wavelet⁃LHL_GLCM_ClusterShade    Wavelet⁃Texture  Arterial phase           -0.214
                 02     AP_shape_Sphericity                 Shape            Arterial phase           -0.186
                 03     PVP_firstorder_Skewness             First⁃order      Portal⁃venous phase      -0.172
                 04     PVP_GLSZM_LargeArea                 Texture          Portal⁃venous phase      -0.158
                 05     DP_GLRLM_RunEntropy                 Texture          Delayed phase            -0.143
                 06     ITH_Habitat_SpatialEntropy          ITH              Multi⁃phase voxel⁃level  -0.238
                 07     ITH_SubRegion_FeatureHeterogeneity  ITH              Multi⁃phase voxel⁃level  -0.201
                 08     ITH_EcologicalDiversityIndex        ITH              Multi⁃phase voxel⁃level  -0.177
                 09     TSR_longitudinal_GLCMContrast_slope  TSR             Longitudinal dynamics    -0.195
                 10     TSR_longitudinal_firstorder_Mean_slope  TSR          Longitudinal dynamics    -0.184
                 11     TSR_shape_Volume_ShrinkageRate      TSR              Longitudinal dynamics    -0.213
                 12     DR_Delta_wavelet⁃HLL_GLCM_Correlation  DR            Last⁃baseline delta      -0.167
                 13     DR_Delta_shape_SurfaceVolumeRatio   DR               Last⁃baseline delta      -0.149
                 14     CTId_composite_score                Clinical lab     Composite lab indices    -0.226
                 15     CTId_Log_AFP                        Clinical lab     Serology                 -0.181
                 16     CTId_PIVKA⁃Ⅱ_ratio                  Clinical lab     Serology                 -0.165
                   AP:arterial phase;PVP:portal⁃venous phase;DP:delayed phase;ITH:intratumoral heterogeneity;TSR:time⁃series radiomics;DR:delta⁃ra⁃
                diomics;CTId:clinical test index density.
                自多期相传统影像组学、ITH、TSR、DR 与 CTId 的贡                   0.987),全面领先于单期相模型(动脉期0.712、门静
                献分布相对均衡;若把多期相联合、纵向动态与异                            脉期 0.738、延迟期 0.706)、ITH 单一模型(0.752)、
                质性三类特征一并纳入考察,共占13个(81.3%),这                       TSR 单一模型(0.743)与传统临床模型(0.685),各
                一比例直观反映出“动态+异质”维度在模型判别中                           组 DeLong 检验均达统计学意义(P 均<0.05,表 3,图
                的主导作用。在7∶3划分的训练集/验证集上,融合                          1)。校准曲线显示模型预测概率与实际发生率吻
                模型对治疗进展的训练集 AUC 为 0.852(95%CI:                    合良好,Hosmer⁃Lemeshow 检验训练集 P=0.642、验

                0.738~0.966),验证集 AUC 为 0.821(95%CI:0.629~         证集 P=0.508(图 2)。DCA 进一步呈现,在 0.15~

                                                    表2 不同风险组基线特征比较
                                  Table 2 Comparison of baseline characteristics among different risk groups

                             Variable              Low⁃risk(n=10)  Intermediate⁃risk(n=24)  High⁃risk(n=28)  P
                Age[years,M(P25,P75)]             57(54,61)        55(51,60)00.         62(60,67)          <0.001
                Male[n(%)]                        7(70.0)/3(30.0)  16(66.7)/8(33.3)     23(82.1)            0.422
                HBV infection[n(%)]               9(90.0)/1(10.0)  17(70.8)/7(29.2)     21(75.0)            0.489
                Solitary/Multifocal tumor[n(%)]   10(100.0)/0(0.0)  19(79.2)/5(20.8)    6(21.4)/22(78.6)   <0.001
                Maximum tumor diameter[cm,M(P25,P75)] 2.4(2.2,3.3)  7.5(6.7,8.3)00      14.2(12.8,16.6)    <0.001
                Portal vein tumor thrombus[n(%)]  1(10.0)/9(90.0)  7(29.2)/17(70.8)     17(60.7)            0.007
                Child⁃Pugh(A/B)[n(%)]             9(90.0)/1(10.0)  20(83.3)/4(16.7)     15(53.6)/13(46.4)   0.022
                TNM(Ⅰ-Ⅱ/Ⅲ-Ⅳ)[n(%)]                8(80.0)/2(20.0)  14(58.3)/10(41.7)    3(10.7)/25(89.3)   <0.001
                AFP[ng/mL,M(P25,P75)]             70.5(5.4,117.8)  165.5(131.8,400.9)   960.5(321.2,2 518.9)  <0.001
                PIVKA⁃Ⅱ[mAU/mL,M(P25,P75)]        268.6(102.1,418.0)  273.0(52.9,602.6)0  1 156.8(18.8,4 399.5) <0.001
                ALT[U/L,M(P25,P75)]               33.5(26.8,39.2)  37.8(29.9,47.7)0     62.9(47.7,82.7)    <0.001
                AST[U/L,M(P25,P75)]               38.2(33.3,42.5)  42.9(31.3,52.6)0     58.5(33.9,65.8)    <0.001
                ALB(g/L,x ± s)                    40.1 ± 3.1       35.8 ± 4.1           33.0 ± 5.3         <0.001
                TACE combination[n(%)]            10(100.0)/0(0.0)  17(70.8)/7(29.2)    19(67.9)            0.122
                Targeted therapy combination[n(%)]  7(70.0)/3(30.0)  16(66.7)/8(33.3)   24(85.7)            0.250
   98   99   100   101   102   103   104   105   106   107   108