Table 2.
Performance comparison among LPAIDS, model W, and model N
| Accuracy (95% CI) | Sensitivity (95% CI) | Specificity (95% CI) | PPV (95% CI) | NPV (95% CI) | AUC | |
|---|---|---|---|---|---|---|
| Images | ||||||
| LPAIDS (All images) | 0·956 (0·951–0·960) | 0·948 (0·941–0·955) | 0·964 (0·958–0·970) | 0·965 (0·959–0·971) | 0·946 (0·939–0·953) | 0·974 |
| Model W (WLI images) | 0·948 (0·942–0·954) | 0·885 (0·870–0·899) | 0·988 (0·984–0·992) | 0·978 (0·971–0·985) | 0·932 (0·923–0·941) | 0·955 |
| LPAIDS (WLI images) | 0·957 (0·951–0·962) | 0·918 (0·906–0·931) | 0·980 (0·975–0·985) | 0·967 (0·959–0·975) | 0·950 (0·943–0·958) | 0·964 |
| Model N (NBI images) | 0·935 (0·926–0·945) | 0·978 (0·971–0·984) | 0·798 (0·765–0·831) | 0·940 (0·930–0·951) | 0·917 (0·893–0·941) | 0·978 |
| LPAIDS (NBI images) | 0·954 (0·946–0·962) | 0·977 (0·971–0·984) | 0·878 (0·852–0·905) | 0·963 (0·955–0·971) | 0·923 (0·900–0·945) | 0·981 |
| Videos | ||||||
| LPAIDS (All videos) | 0·949 (0·931–0·968) | 0·948 (0·922–0·974) | 0·951 (0·924–0·977) | 0·955 (0·930–0·979) | 0·943 (0·916–0·971) | 0·974 |
| Model W (WLI videos) | 0·945 (0·921–0·968) | 0·886 (0·835–0·937) | 0·986 (0·970–1·000) | 0·978 (0·953–1·000) | 0·925 (0·890–0·959) | 0·948 |
| LPAIDS (WLI videos) | 0·950 (0.928–0.973) | 0·919 (0·876–0·963) | 0·972 (0·949–0·994) | 0·958 (0·925–0·991) | 0·945 (0·915–0·975) | 0·968 |
| Model N (NBI videos) | 0·926 (0·889–0·963) | 0·971 (0·943–0·999) | 0·804 (0·695–0·913) | 0·931 (0·890–0·972) | 0·911 (0·828–0·994) | 0·974 |
| LPAIDS (NBI videos) | 0·947 (0·916–0·979) | 0·978 (0·954–1·000) | 0·863 (0·768–0·957) | 0·951 (0·916–0·986) | 0·936 (0·866–1·000) | 0·974 |
AUC: area under the curve; PPV, positive predictive value;CI: confidence interval; LPAIDS: Laryngopharyngeal Artificial Intelligence Diagnostic System; NBI: narrow-band imaging; NPV, negative predictive value; WLI: white light imaging