Table 2: Performance comparison of peritumoral texture augmented patches.

CNN Model	Input patch	Accuracy[%]	Sensitivity[%]	Specificity[%]	PPV[%]	NPV[%]
ResNet 152 Single Model	image patch	82.75	92.3	63.15	83.72	80.0
	peritumoral patch	87.93	89.74	84.21	92.1	80.0
	shape-focused intratumoral patch	72.41	92.3	31.57	73.46	66.66
	coupled intra and peritumoral patch	74.13	71.79	78.94	87.5	57.69
ResNetl52 Ensemble Model (soft-voting)	image patch & peritumoral patch	84.48	94.87	57.89	82.22	84.61
	shape-focused intratumoral patch & peritumoral patch	72.41	76.92	42.1	73.17	47.05
	shape-focused intratumoral patch & coupled intra and peritumoral patch	68.96	79.48	26.31	68.88	38.46
	image patch & peritumoral patch & shape-focused intratumoral patch	75.86	92.3	47.36	78.26	75.0
ResNet 152 Ensemble Model (3 channel input)	image patch & peritumoral patch & shape-focused intratumoral patch	87.93	94.87	73.68	88.09	87.5
EffieientNet-b7 Single Model	image patch	79.31	94.87	47.36	78.72	81.81
	peritumoral patch	81.03	82.05	78.94	88.88	68.18
	shape-focused intratumoral patch	79.31	92.3	52.63	80.0	76.92
	16x16 peritumoral texture augmented patch	71.42	97.36	16.66	71.15	75.0
EfficientNet-b7 Ensemble Model (soft-voting)	image patch & peritumoral patch	84.48	89.74	73.68	87.5	77.77
	shape-focused intratumoral patch & peritumoral patch	81.03	87.17	73.68	87.17	73.68
	image patch & 1×16 peritumoral texture augmented patch	75.0	91.89	10.52	66.66	40.0
EfficientNet-b7 Ensemble Model (3 channel input)	image patch & peritumoral patch & shape-focused intratumoral patch	74.13	71.79	78.94	87.5	57.69