Table 2.

Deep learning feature-based detection efficiency and prognosis prediction

	Training set with WSDL (n = 64)	Test set with WSDL (n = 20)	Training set with CDL (n = 64)	Test set with CDL (n = 20)
Sensitivity	86.67%	87.50%	73.33%	62.5%
Specificity	100%	83.33%	100%	83.33%
Accuracy	93.75%	85.00%	87.50%	75.00%
2-year PFS (PSI > 1)	34.6% ± 9.3%	33.3% ± 15.7%	36.4% ± 10.3%	28.6% ± 17.1%
2-year PFS (PSI < 1)	92.1% ± 4.4%	90.9% ± 8.7%	85.7% ± 5.4%	84.6% ± 10.0%
5-year PFS (PSI > 1)	3.8% ± 3.8%	22.2% ± 13.9%	4.5% ± 4.4%	28.6% ± 17.1%
5-year PFS (PSI < 1)	92.1% ± 4.4%	90.9% ± 8.7%	77.1% ± 9.5%	74.0% ± 13.2%

Abbreviations: PFS, progression-free survival; PSI, prediction similarity index; WSDL, weakly supervised deep learning; CDL, conventional deep learning