Page 85 - 南京医科大学学报自然科学版
P. 85

第44卷第2期        刘婷婷,林佳璐,娄鉴娟,等. 多参数MRI影像组学评估浸润性乳腺癌HER⁃2表达状态的临床
                  2024年2月                应用价值[J]. 南京医科大学学报(自然科学版),2024,44(2):218-227                    ·223 ·


                                              表4 最终筛选出的26个多参数影像组学特征
                                      Table 4 Final selection of 26 multi⁃parametric radiomics features

                  MRI sequence  features category               Radiomics features                     Coef
                  DCE2           First order  DCE2_wavelet_LLH_firstorder_Skewness                 -3.420175E-02
                                 First order  DCE2_wavelet_LLL_firstorder_Minimum                  4.090850E-01
                                  Shape      DCE2_original_shape_Sphericity                        -8.281829E-02
                                  Shape      DCE2_original_shape_SurfaceVolumeRatio                1.264466E+00
                                  GLRLM      DCE2_lbp_3D_2_glrlm_RunLengthNonUniformity            4.018018E-05
                                  GLRLM      DCE2_wavelet_LLL_glrlm_GrayLevelNonUniformity         8.337503E-02
                                  GLRLM      DCE2_wavelet_HLH_glrlm_LowGrayLevelRunEmphasis        2.207430E-07
                                  GLRLM      DCE2_wavelet_LHH_glrlm_LongRunLowGrayLevelEmphasis    2.640173E-02
                                  GLRLM      DCE2_wavelet_HLH_glrlm_HighGrayLevelRunEmphasis       -1.729161E-01
                                  GLSZM      DCE2_wavelet_HHH_glszm_SmallAreaLowGrayLevelEmphasis  -1.456218E-01
                                  GLDM       DCE2_wavelet_LHH_gldm_SmallDependenceLowGrayLevelEmphasis  -2.777829E-01
                  DCE4           First order  DCE4_wavelet_LLH_firstorder_Skewness                 -2.930163E-01
                                  GLRLM      DCE4_wavelet_LHH_glrlm_GrayLevelNonUniformityNormalized  -5.542897E-02
                                  GLRLM      DCE4_wavelet_LHH_glrlm_GrayLevelVariance              9.834993E-06
                                  GLSZM      DCE4_wavelet_LLH_glszm_SmallAreaEmphasis              3.128872E-02
                                  GLSZM      DCE4_wavelet_LLH_glszm_SmallAreaLowGrayLevelEmphasis  1.006800E-04
                  DWI            First order  DWI_wavelet_HHL_firstorder_Skewness                  -1.826650E-01
                                  GLSZM      DWI_wavelet_HLH_glszm_SizeZoneNonUniformityNormalized  -1.946239E-01
                                  GLSZM      DWI_wavelet_LHL_glszm_LargeAreaLowGrayLevelEmphasis   7.234353E-02
                                  NGTDM      DWI_wavelet_HHL_ngtdm_Contrast                        1.083605E-01
                  ADC            First order  ADC_lbp-3D_m1_firstorder_90Percentile                1.147517E-01
                                  NGTDM      ADC_wavelet_LLL_ngtdm_Strength                        -2.584432E-01
                  TIRM           First order  TIRM_lbp_3D_k_firstorder_Median                      -1.534631E-02
                                  Shape      TIRM_original_shape_Elongation                        -2.008755E-01


                          DCE2_wavelet. HHH_glsam_Small Area Low Gray Level Emphasis
                               DCE2_wavelet. HLH_glrim_High Gray Level Run Emphasis
                                           DWI_wavelet. HHL_firstorder_Skewness
                           DWI_wavelet. HLH_glsam_Size Zone Non Uniformity Normalized
                     feature                  ADC_wavelet. LLL_ngtdm_strength
                                                TIRM_original_shape_Elongation
                      DCE2_wavelet. LHH_gldm_Small Dependence Low Gray Level Emphasis
                                          DCE4_wavelet. LHH_firstorder_Skewness
                                           DCE2_wavelet. LLL_firstorder_Minimum
                                        DCE2_onginal_shape_Surtace Volume Ratio
                                                                        -0.2  0.0  0.2  0.4  0.6  0.8  1.0  1.2
                                                                                        Coef
                                    The x⁃axis represents the weight values,and the y⁃axis represents the radiomics features.
                                                   图4 按照权重排序的前10个特征
                                               Figure 4  Top10 features ranked by weights

                间的差异没有统计学意义(P 均>0.05)。测试集的                        现HER⁃2的扩增和过表达,HER⁃2状态是乳腺癌症
                决策曲线表明多参数模型在大范围阈值概率下优                             疾病进展和死亡的独立预测因子,HER⁃2过表达的
                于其他模型,临床获益度更高(图6)。                                浸润性乳腺癌患者预后更差             [22] ,但随着临床治疗的
                                                                  发展,HER⁃2 也可做为乳腺癌靶向治疗的一个靶
                3  讨 论
                                                                  点,因此,早期对乳腺癌患者的 HER⁃2 状态进行准
                    HER⁃2是一种原癌基因,与细胞的生长、分化、                       确的评估至关重要。目前,HER⁃2状态主要通过免
                转移过程调控作用有关           [21] ;20%~25%的乳腺癌会出          疫组织化学(IHC)或荧光原位杂交(FISH)检测,两
   80   81   82   83   84   85   86   87   88   89   90