Assessment of emerging pretraining strategies in interpretable multimodal deep learning for cancer prognostication

Table 2 Comparison of multimodal models and top unimodal model via their C-indices and 95% confidence interval

Subtype	Top Unimodal	Multi-Self	Multi-Cross	Multi-Transfer	PORPOISE	Cheerla and Geveart^a
BLCA	0.692 ± 0.002	0.551 ± 0.003	0.739 ± 0.002	0.621 ± 0.002	0.631 ± 0.050	0.73
BRCA	0.711 ± 0.002	0.633 ± 0.003	0.606 ± 0.004	0.836 ± 0.002	0.628 ± 0.053	0.79
HNSC	0.680 ± 0.002	0.651 ± 0.002	0.687 ± 0.002	0.605 ± 0.002	0.573 ± 0.010	0.67
KIRC	0.794 ± 0.004	0.864 ± 0.003	0.835 ± 0.003	0.628 ± 0.003	0.659 ± 0.075	0.73
LIHC	0.746 ± 0.002	0.725 ± 0.003	0.742 ± 0.003	0.739 ± 0.003	0.622 ± 0.042	0.77
LUAD	0.669 ± 0.003	0.697 ± 0.003	0.718 ± 0.002	0.776 ± 0.003	0.600 ± 0.046	0.73
PAAD	0.628 ± 0.004	0.666 ± 0.004	0.666 ± 0.003	0.700 ± 0.003	0.653 ± 0.030	0.74
SKCM	0.640 ± 0.003	0.582 ± 0.003	0.626 ± 0.002	0.643 ± 0.002	0.651 ± 0.056	0.72

ISSN: 1756-0381