recompute_results.md

Recomputed results

We have recomputed the classification results with argmax prediction rule, where the threshold for binary classification is 0.5, and they are different from the results in our paper, which are computed based on the cut-off value in ROC.

The performance in prediction of ALN status (N0 vs. N(+))

Methods		AUC	Accuracy (%)	Precision (%)	Recall (%)	F1-score (%)
Clinical data only	T	0.661	64.13	52.46	62.08	56.87
	V	0.709	68.10	56.52	65.82	60.82
	I-T	0.613	57.80	45.92	53.57	49.45
DL-CNB model	T	0.901	81.43	72.53	82.50	77.19
	V	0.808	71.43	66.10	49.37	56.52
	I-T	0.816	70.18	67.27	44.05	53.24
DL-CNB+C model	T	0.878	78.89	70.82	75.83	73.24
	V	0.823	74.76	64.13	74.68	69.01
	I-T	0.831	76.15	70.51	65.48	67.90

The performance in prediction of ALN status (N0 vs. N + (1-2))

Methods		AUC	Accuracy (%)	Precision (%)	Recall (%)	F1-score (%)
Clinical data only	T	0.638	61.58	34.08	59.38	43.30
	V	0.677	64.57	36.76	56.82	44.64
	I-T	0.627	61.05	29.58	55.26	38.53
DL-CNB model	T	0.912	81.66	66.67	51.56	58.15
	V	0.756	70.86	42.86	47.73	45.16
	I-T	0.845	80.23	54.55	63.16	58.54
DL-CNB+C model	T	0.936	85.71	75.00	63.28	68.64
	V	0.789	77.14	61.11	25.00	35.48
	I-T	0.878	84.88	77.27	44.74	56.67

The performance in prediction of ALN status (N0 vs. N + (>2))

Methods		AUC	Accuracy (%)	Precision (%)	Recall (%)	F1-score (%)
Clinical data only	T	0.680	66.86	38.05	65.00	48.00
	V	0.748	68.48	37.14	76.47	50.00
	I-T	0.629	60.12	30.26	58.97	40.00
DL-CNB model	T	0.906	82.94	64.10	62.50	63.29
	V	0.755	76.97	38.89	20.59	26.92
	I-T	0.837	80.92	71.43	25.64	37.74
DL-CNB+C model	T	0.918	81.96	68.92	42.50	52.58
	V	0.761	79.39	50.00	29.41	37.04
	I-T	0.838	80.35	66.67	25.64	37.04