Automated lung ultrasound image assessment using artificial intelligence to identify fluid overload in dialysis patients

Table 2 Diagnostic ability of fluid overload (LUS B lines ≥5) using different cutoffs of overhydration

		Sensitivity	Specificity	Positive predictive value	Negative predictive value	Accuracy
Physician count	OH > = 4.5	0.639	0.657	0.479	0.787	0.651
AI count	OH > = 4.5	0.636	0.662	0.457	0.803	0.654
Physician count	OH > = 4.75	0.639	0.699	0.511	0.797	0.679
AI count	OH > = 4.75	0.658	0.707	0.511	0.815	0.691
Physician count	OH > = 5	0.583	0.712	0.5	0.776	0.67
AI count	OH > = 5	0.576	0.716	0.475	0.791	0.673
Physician count	OH > =4.5 + clinical	0.76	0.655	0.396	0.902	0.679
AI count	OH > =4.5 + clinical	0.833	0.621	0.217	0.967	0.645
Physician count	OH > =4.75 + clinical	0.76	0.69	0.422	0.906	0.706
AI count	OH > =4.75 + clinical	0.833	0.653	0.233	0.969	0.673
Physician count	OH > =5 + clinical	0.72	0.714	0.429	0.896	0.716
AI count	OH > =5 + clinical	0.75	0.674	0.255	0.955	0.682

ISSN: 1471-2369