Table 2 Performance Metrics of OMMT-PredNet for Multi-Task Analysis in Oral Leukoplakia

From: Next-generation AI framework for comprehensive oral leukoplakia evaluation and management

Parameters

AUC

Brier score

Youden’s index

F1 score

Precision

Balanced accuracy

Sensitivity

Specificity

PPV

NPV

Multi-Task 1: Oral epithelial dysplasia identification

Training & Internal Validation (Hong Kong Cohort = 598)

0.9219 (95% CI = 0.9088, 0.9349)

0.1080 (95% CI = 0.1018, 0.1142)

0.7113 (95% CI = 0.6794, 0.7432)

0.8550 (95% CI = 0.8456, 0.8644)

0.9303 (95% CI = 0.8786, 0.9820)

0.8471 (95% CI = 0.8362, 0.8580)

0.7931 (95% CI = 0.7503, 0.8359)

0.9182 (95% CI = 0.8502, 0.9862)

0.9303 (95% CI = 0.8786, 0.9820)

0.7722 (95% CI = 0.7474, 0.7970)

External Validation (ZheJiang Cohort = 51)

0.8786 (95% CI = 0.8471, 0.9101)

0.1379 (95% CI = 0.1189, 0.1569)

0.6571 (95% CI = 0.5829, 0.7313)

0.8089 (95% CI = 0.7646, 0.8532)

0.9218 (95% CI = 0.8185, 1.0251)

0.8286 (95% CI = 0.7915, 0.8657)

0.7286 (95% CI = 0.6314, 0.8258)

0.9286 (95% CI = 0.8200, 1.0372)

0.9218 (95% CI = 0.8185, 1.0251)

0.7777 (95% CI = 0.7224, 0.8330)

Multi-Task 2: Time-to-event cancer risk prediction

Training & internal validation (Hong Kong cohort = 598)

0.9592 (95% CI = 0.9491, 0.9693)

0.0911 (95% CI = 0.0757, 0.1065)

0.8039 (95% CI = 0.7584, 0.8494)

0.8972 (95% CI = 0.8741, 0.9203)

0.9442 (95% CI = 0.9094, 0.9790)

0.9020 (95% CI = 0.8792, 0.9248)

0.8549 (95% CI = 0.8331, 0.8767)

0.9490 (95% CI = 0.9163, 0.9817)

0.9442 (95% CI = 0.9094, 0.9790)

0.8674 (95% CI = 0.8481, 0.8867)

External validation (ZheJiang cohort = 51)

0.9469 (95% CI = 0.9145, 0.9793)

0.1230 (95% CI = 0.0565, 0.1895)

0.6234 (95% CI = 0.2934, 0.9534)

0.7455 (95% CI = 0.7329, 0.7581)

0.7333 (95% CI = 0.5482, 0.9184)

0.8643 (95% CI = 0.8445, 0.8841)

0.8000 (95% CI = 0.6413, 0.9587)

0.8857 (95% CI = 0.8064, 0.9650)

0.7333 (95% CI = 0.5482, 0.9184)

0.9329 (95% CI = 0.8927, 0.9731)

  1. OMMT-PredNet C-Index = 0.7947 (95% CI = 0.7789, 0.8105).