Radiomics and deep learning for myocardial scar screening in hypertrophic cardiomyopathy

Table 3 Performance of the scar prediction models trained using 5 different splits of the development dataset

Cross-validation		1	2	3	4	5	Mean ± SD
Radiomics	Sensitivity	0.92	0.91	0.92	0.91	0.91	0.91 ± 0.01
	Specificity	0.42	0.36	0.37	0.37	0.29	0.36 ± 0.05
	Recall	0.85	0.81	0.84	0.82	0.77	0.82 ± 0.03
	Precision	0.58	0.56	0.57	0.56	0.53	0.56 ± 0.02
	Accuracy	0.65	0.62	0.63	0.62	0.58	0.62 ± 0.03
	AUC	0.77	0.76	0.77	0.72	0.71	0.75 ± 0.03
Deep learning	Sensitivity	0.92	0.92	0.91	0.92	0.91	0.92 ± 0.01
	Specificity	0.45	0.42	0.42	0.42	0.30	0.40 ± 0.06
	Recall	0.86	0.85	0.83	0.85	0.78	0.83 ± 0.03
	Precision	0.60	0.58	0.58	0.58	0.54	0.58 ± 0.02
	Accuracy	0.67	0.65	0.65	0.65	0.58	0.64 ± 0.03
	AUC	0.76	0.76	0.75	0.78	0.77	0.76 ± 0.01
Deep learning–radiomics	Sensitivity	0.91	0.91	0.91	0.91	0.91	0.91 ± 0.00
	Specificity	0.43	0.45	0.43	0.49	0.32	0.42 ± 0.06
	Recall	0.84	0.84	0.84	0.85	0.79	0.83 ± 0.02
	Precision	0.59	0.60	0.59	0.61	0.54	0.59 ± 0.03
	Accuracy	0.65	0.67	0.65	0.69	0.60	0.65 ± 0.03
	AUC	0.79	0.82^*†	0.83^*†	0.82^*	0.79^*	0.81 ± 0.02

Performance is evaluated using the internal testing dataset
^*Statistical significance vs Radiomics model.
^†Statistical significance vs DL model. All metrics were computed at an operating point corresponding to a sensitivity of at least 90%

ISSN: 1532-429X