Table 2 Results on the test set of the external databases

From: Deep learning for obstructive sleep apnea diagnosis based on single channel oximetry

 

ODI

OBM

OxiNet

 

ICC

F1,M

ICC

F1,M

ICC

F1,M

SHHS1

0.89

0.69

0.93

0.74

0.96

0.84

 

(0.89–0.93)

(0.68–0.72)

(0.90–0.95)

(0.70–0.78)

(0.95–0.97)

(0.82–0.86)

SHHS2

0.89

0.69

0.93

0.74

0.95

0.83

 

(0.87–0.91)

(0.65–0.72)

(0.92–0.93)

(0.74–0.75)

(0.94–0.98)

(0.83–0.85)

UHV

0.75

0.61

0.86

0.67

0.92

0.77

 

(0.74–0.78)

(0.58–0.62)

(0.88–0.92)

(0.60–0.74)

(0.91–0.96)

(0.77–0.79)

CFS

0.70

0.56

0.75

0.60

0.92

0.78

 

(0.66–0.78)

(0.50–0.70)

(0.68–0.80)

(0.51–0.69)

(0.90–0.96)

(0.74–0.82)

MROS

0.70

0.52

0.81

0.65

0.94

0.80

 

(0.68–0.72)

(0.50–0.58)

(0.79–0.83)

(0.62–0.68)

(0.95–0.99)

(0.78–0.84)

MESA

0.75

0.60

0.75

0.65

0.94

0.75

 

(0.71–0.77)

(0.54–0.64)

(0.70–0.80)

(0.62–0.68)

(0.92–0.94)

(0.72–0.76)

  1. ODI, OBM, and OxiNet are trained on 90% of SHHS1. Confidence intervals are computed via bootstrapping on each test set separately.