. 2024 Nov 30;14:29805. doi: 10.1038/s41598-024-81136-0

Table 2.

FLASH-TV 2.0 performance on two independent samples.

Dataset	Gold standard TV time, minutes (SD)	FLASH predicted TV time, minutes (SD)	Kappa, PABAK (SD)	ICC	Accuracy (%)	Sensitivity (%)	Specificity (%)
Pilot test (n = 10)	29.02 (10.78)	29.77 (10.46)	0.68 (0.14)	0.96	83.80 (7.24)	81.71 (19.29)	84.22 (6.85)
Lab validation test, initial sample (n = 15)*	27.75 (10.36)	25.33 (10.81)	0.54 (0.21)	0.46	77.71 (10.41)	69.34 (22.78)	83.61 (8.50)

Dataset

Gold standard TV time, minutes (SD)

FLASH predicted TV time, minutes (SD)

Kappa, PABAK
(SD)

ICC

Accuracy (%)

Sensitivity (%)

Specificity (%)

Pilot test (n = 10)

29.02

(10.78)

29.77

(10.46)

0.68

(0.14)

0.96

83.80 (7.24)

81.71 (19.29)

84.22

(6.85)

Lab validation test, initial sample (n = 15)*

27.75

(10.36)

25.33

(10.81)

0.54

(0.21)

0.46

77.71

(10.41)

69.34

(22.78)

83.61

(8.50)

SD - standard deviation.

PABAK - prevalence and bias-adjusted Kappa, compares the gaze and no-gaze of FLASH-TV and gold standard at 5-second epoch, nested within each participant.

ICC-intraclass correlation, compares the overall duration of TV watching of FLASH-TV and gold standard per participant.

*Here we used the initial n = 15 from our full validation test set of 30 participants.

Comparisons of FLASH-TV and gold-standard do not include video segments labeled as uncertain by the team.