. 2022 Jun 23;12:10634. doi: 10.1038/s41598-022-13928-1

Table 3.

3-fold averaged confusion matrix value results after applying optimal thresholds and the single tile removal post-processing to the DenseNet-161 and ensemble models, for the hold-out test sets.

Model	TN		TP		FN		FP
DenseNet-161	30880	$+$ 672	1081	$+$ 3	70	$+$ 5	3171	$+$ 651
DenseNet-161	30880	$-$ 651	1081	$-$ 5	70	$-$ 3	3171	$-$ 672
DenseNet-161 (threshold per fold + post-processed)	33505	$+$ 27	929	$+$ 32	222	$+$ 37	546	$+$ 19
DenseNet-161 (threshold per fold + post-processed)	33505	$-$ 19	929	$-$ 37	222	$-$ 32	546	$-$ 27
DenseNet-161 (mean 3-fold threshold + post-processed)	33479	$+$ 143	929	$+$ 12	222	$+$ 12	572	$+$ 126
DenseNet-161 (mean 3-fold threshold + post-processed)	33479	$-$ 126	929	$-$ 12	222	$-$ 12	572	$-$ 143
Ensemble	32119	$+$ 1053	1063	$+$ 26	88	$+$ 44	1932	$+$ 1048
Ensemble	32119	$-$ 1048	1063	$-$ 44	88	$-$ 26	1932	$-$ 1053
Ensemble (threshold per fold + post-processed)	33221	$+$ 101	983	$+$ 57	168	$+$ 69	830	$+$ 184
Ensemble (threshold per fold + post-processed)	33221	$-$ 184	983	$-$ 69	168	$-$ 57	830	$-$ 101
Ensemble (mean 3-fold threshold + post-processed)	33180	$+$ 504	1002	$+$ 37	149	$+$ 68	871	$+$ 460
Ensemble (mean 3-fold threshold + post-processed)	33180	$-$ 460	1002	$-$ 68	149	$-$ 37	871	$-$ 504

Presented in this table are the true negative (TN), true positive (TP), false negative (FN) and false positive (FP) values averaged across all three folds, rounded to the nearest whole figure. Plus/minus values shows the mean subtracted from the highest and lowest results from the 3-fold experiments, for each model.