A machine learning-based clinical tool for diagnosing myopathy using multi-cohort microarray expression profiles

Table 1 Disease-specific 95% confidence intervals for the AUC using the top 500 genes

Diseases	AUC (95% CI)	Specificity	Sensitivity	Precision	F1 score
Control	0.861 (0.826–0.895)	0.814	0.747	0.883	0.810
Chronic	0.872 (0.824–0.920)	0.708	0.851	0.957	0.901
Congenital	0.848 (0.805–0.892)	0.805	0.776	0.900	0.833
IM	0.794 (0.713–0.876)	0.585	0.883	0.951	0.916
ICUAW	0.777 (0.668–0.887)	0.812	0.614	0.988	0.758
Immobility	0.789 (0.716–0.861)	0.850	0.598	0.974	0.741

IM inflammatory myositis, ICUAW intensive care unit acquired weakness, CI confidence interval. Classes were balanced using an augmentation strategy of sampling to twice the mean class size (N = 2520). Confidence intervals were generated using 2000 stratified bootstrapping replications. Optimal thresholds were determined using a Youden’s J statistic. Specificity, sensitivity, precision, and F1 score were calculated at the optimal threshold

ISSN: 1479-5876