Lung nodule malignancy classification with associated pulmonary fibrosis using 3D attention-gated convolutional network with CT scans

Table 2 Quantitative results

Dataset	LIDC-IDRI	In-house	In-house
Training strategy/data processing	NA	NA	Pretrained on LIDC-IDRI data	Pretrained on LIDC-IDRI data/Nodule surrounding tissues removed	Pretrained on LIDC-IDRI data/semantic fibrosis data added
Micro-environment information	Available	Available	Available	Not available	Available
Accuracy (%)^*	85.11 (3.19)	78.84 (5.88)	79.03 (2.97)^*	75.61 (7.02)^*	80.84 (3.31)^*
Sensitivity (%)^†	77.78 (12.24)	62.00 (13.65)	65.46 (18.64)^†	50.00 (25.46)^†	74.67 (14.78)^†
Specificity (%)^§	88.54 (5.87)	87.29 (5.98)	85.86 (6.29)^§	88.46 (7.88)^§	84.95 (5.43)^§
AUC	0.90 (0.04)	0.83 (0.03)	0.84 (0.06)	0.78 (0.08)	0.89 (0.05)

Data are presented in the format of Mean (Standard deviation), AUC area under the receiver operating characteristic
*One-way ANOVA analysis performed with Bonferroni correction on accuracy: p-value = 0.0011 (statistical significance)
^†One-way ANOVA analysis performed with Bonferroni correction on sensitivity: p-value = 0.0013 (statistical significance)
^§One-way ANOVA analysis performed with Bonferroni correction on specificity p-value = 0.35556 (statistical non-significance)

ISSN: 1479-5876