Skip to main content
. 2025 Dec 18;4:e76586. doi: 10.2196/76586

Table 2.

Overall performance comparison by sampling rate.

Model and sampling rate (s) Recall (95% CI) IOUa (95% CI) Precision (95% CI) PEb length (s; 95% CI) Predicted length (s; 95% CI)
GPT-4o

1 0.998 (0.994-1.000) 0.784 (0.765-0.803) 0.792 (0.774-0.811) 126 (121-132) 175 (165-187)

2 0.991 (0.985-0.996) 0.809 (0.794-0.824) 0.822 (0.808-0.838) 126 (121-132) 159 (150-167)

3 0.979 (0.970-0.986) 0.796 (0.778-0.813) 0.817 (0.802-0.833) 126 (121-132) 157 (149-166)
GPT-4o-mini

1 0.945 (0.933-0.956) 0.835 (0.820-0.849) 0.878 (0.863-0.894) 126 (121-132) 141 (133-149)

2 0.918 (0.903-0.931) 0.822 (0.805-0.838) 0.885 (0.871-0.900) 126 (121-132) 134 (127-141)

3 0.886 (0.868-0.903) 0.791 (0.771-0.809) 0.880 (0.864-0.894) 126 (121-132) 130 (123-137)
Gemini 2.0 Flash

1 0.999 (0.999-1.000) 0.631 (0.607-0.656) 0.639 (0.616-0.667) 126 (121-132) 262 (245-277)

2 0.996 (0.990-0.999) 0.719 (0.699-0.740) 0.728 (0.708-0.750) 126 (121-132) 202 (188-214)

3 0.993 (0.986-0.998) 0.722 (0.701-0.743) 0.731 (0.711-0.751) 126 (121-132) 197 (185-210)
Gemma 3

1 0.995 (0.990-0.998) 0.531 (0.506-0.558) 0.541 (0.514-0.571) 126 (121-132) 332 (309-353)

2 0.993 (0.987-0.997) 0.621 (0.596-0.647) 0.631 (0.605-0.658) 126 (121-132) 270 (250-288)

3 0.989 (0.983-0.994) 0.634 (0.611-0.656) 0.647 (0.623-0.673) 126 (121-132) 254 (235-270)
Qwen2.5-VL-72B

1 0.919 (0.903-0.934) 0.765 (0.741-0.784) 0.827 (0.806-0.851) 126 (121-132) 166 (155-178)

2 0.879 (0.857-0.899) 0.752 (0.729-0.774) 0.846 (0.826-0.867) 126 (121-132) 152 (140-162)

3 0.847 (0.823-0.869) 0.736 (0.714-0.757) 0.854 (0.837-0.875) 126 (121-132) 139 (130-148)
Qwen-2.5VL-7B

1 0.929 (0.914-0.943) 0.425 (0.401-0.450) 0.462 (0.436-0.493) 126 (121-132) 382 (361-400)

2 0.881 (0.861-0.902) 0.470 (0.445-0.493) 0.531 (0.503-0.559) 126 (121-132) 319 (299-338)

3 0.860 (0.837-0.884) 0.475 (0.449-0.501) 0.547 (0.520-0.580) 126 (121-132) 299 (278-319)

aIOU: intersection over union.

bPE: physical examination.