Cascaded-Recalibrated Multiple Instance Deep Model for Pathologic-Level Lung Cancer Prediction in CT Images

Table 2

Performance comparison of lung cancer prediction (pmal) by transferring from attribute malignancy in terms of the models of higher-order transfer learning, instance-space MIL, embedding-space MIL, and the proposed recalibrated MIL.

Mal  Pmal	Accuracy	AUC	F1-score
Higher-order transfer [13]	0.702 ± 0.050	0.669 ± 0.045	0.644 ± 0.035
Instance-space MIL	0.820 ± 0.044	0.793 ± 0.059	0.780 ± 0.054
Embedding-space MIL	0.820 ± 0.034	0.800 ± 0.043	0.783 ± 0.031
Recalibrated MIL (ours)	0.837 ± 0.035	0.830 ± 0.021	0.800 ± 0.037

The bold values mean the superiority to others.