Preoperative CT-based radiomics combined with tumour spread through air spaces can accurately predict early recurrence of stage I lung adenocarcinoma: a multicentre retrospective cohort study

Wang, Yuhang; Ding, Yun; Liu, Xin; Li, Xin; Jia, Xiaoteng; Li, Jiuzhen; Zhang, Han; Song, Zhenchun; Xu, Meilin; Ren, Jie; Sun, Daqiang

doi:10.1186/s40644-023-00605-3

Table 2 Performance evaluation of the models in the development cohort and validation cohort

From: Preoperative CT-based radiomics combined with tumour spread through air spaces can accurately predict early recurrence of stage I lung adenocarcinoma: a multicentre retrospective cohort study

	AUC (95%CI)	Sensitivity	Specificity	PPV	NPV	Accuracy	Youden Index
Training cohort
Rad-tumoral	0.785 (0.681–0.890)	0.840	0.705	0.339	0.961	0.726	0.545
Rad-peritumoral-3u	0.680 (0.568–0.792)	0.920	0.432	0.225	0.968	0.506	0.352
Rad-peritumoral-6u	0.685 (0.574–0.796)	0.800	0.583	0.256	0.942	0.616	0.383
Rad-peritumoral-12u	0.688 (0.576–0.800)	0.760	0.583	0.247	0.931	0.610	0.343
Rad-DeepL-2d	0.692 (0.579–0.805)	0.840	0.590	0.269	0.953	0.628	0.430
Rad-DeepL-3d	0.629 (0.509–0.750)	0.640	0.633	0.239	0.907	0.634	0.273
STAS	0.727 (0.654–0.799)	0.920	0.511	0.253	0.973	0.573	0.431
RAISm	0.847 (0.762–0.932)	0.800	0.871	0.526	0.960	0.860	0.671
Test cohort
Rad-tumoral	0.727(0.533–0.921)	0.875	0.556	0.226	0.968	0.597	0.431
Rad-peritumoral-3u	0.780(0.560–1.000)	0.750	0.833	0.400	0.957	0.823	0.583
Rad-peritumoral-6u	0.745(0.552–0.939)	0.750	0.722	0.286	0.951	0.726	0.472
Rad-peritumoral-12u	0.829(0.685–0.972)	0.875	0.630	0.259	0.971	0.661	0.505
Rad-DeepL-2d	0.734(0.590–0.878)	0.875	0.574	0.233	0.969	0.613	0.449
Rad-DeepL-3d	0.576(0.333–0.819)	0.500	0.778	0.250	0.913	0.742	0.278
STAS	0.606(0.433–0.780)	0.750	0.463	0.171	0.926	0.500	0.213
RAISm	0.750(0.531–0.969)	0.875	0.667	0.280	0.973	0.694	0.542
Validation cohort
Rad-tumoral	0.768(0.625–0.912)	0.585	0.900	0.960	0.346	0.647	0.485
Rad-peritumoral-3u	0.663(0.509–0.818)	0.439	1.000	1.000	0.303	0.549	0.439
Rad-peritumoral-6u	0.583(0.406–0.760)	0.366	0.900	0.938	0.257	0.471	0.266
Rad-peritumoral-12u	0.593(0.374–0.812)	0.659	0.600	0.871	0.300	0.647	0.259
Rad-DeepL-2d	0.568(0.390–0.746)	0.512	0.800	0.913	0.286	0.569	0.312
Rad-DeepL-3d	0.522(0.332–0.712)	0.220	1.000	1.000	0.238	0.373	0.220
STAS	0.713(0.555–0.872)	0.585	0.800	0.923	0.320	0.627	0.385
RAISm	0.817(0.625–1.000)	0.951	0.700	0.929	0.778	0.902	0.651

AUC Area under the receiver operating characteristics curve, CI Confidence interval, PPV Positive predictive values, NPV Negative predictive values, Rad-tumoral Tumoral radiomics signature, Rad-peritumoral-3u Peritumoral radiomic signature extracted from the 3 voxel units peritumoral area, Rad-peritumoral-6u Peritumoral radiomic signature extracted from the 6 voxel units peritumoral area, Rad-peritumoral-12u Peritumoral radiomic signature extracted from the 12 voxel units peritumoral area, Rad-DeepL-2d Radiomics signature constructed by 2d deep learning features, Rad-DeepL-3d Radiomics signature constructed by 3d deep learning features, STAS Spread through air spaces, RAISm RAdiomcs Integrated with STAS status model

Back to article page

ISSN: 1470-7330

Contact us

Submission enquiries: alyzza.javate@springernature.com
General enquiries: info@biomedcentral.com

Cancer Imaging

Contact us