The Accuracy of Analyzing Oral Lesions with Artificial Intelligence Devices

Abstract

Artificial intelligence is a tool to predict the probability of oral precancerous lesions. Photography techniques and the direction for importing photos should be used properly. The results of this research can be used as part of establishing a development program for community health worker (CHWs) to screen oral precancerous lesions and obtain the most accuracy in predicting lesions. This research was a cross-sectional descriptive study with the objective of comparing the accuracy in predicting lesions using different brightness of photos employing mobile light and natural light, the distance and direction of taking photos.

A purposive sampling was used to recruit the samples who came to visit at the dental department of Thalang Hospital, Phuket during October - December 2023. All samples who had oral precancerous or cancerous lesion were diagnosed by oral surgeon. The percent probability of precancerous lesions was analyzed by artificial intelligence technology.

The results showed that the medians of percent probability of oral precancerous lesions from 4 groups of techniques; 1) the natural light, non close-up group 2) the mobile flash light, non close-up group 3) the natural light, close up group and 4) the mobile flash light, close-up group were equal to 57.18, 63.35, 79.38, and 80.28 respectively. The medians of percent probability of all 4 groups were significantly difference (p-value <0.001), using Kruskal - Wallis test. Man-Whitney U test was used to identify the difference of pairs. For close-up flash light group, the percent probabilities of oral precancerous lesions with the different directions of photographs rotating 90, 180, 270 degrees from normal images while importing the data in artificial intelligence devices were significantly difference. (p-value = 0.037)

Conclusion close-up photography combined with mobile flash light from a mobile camera with a magnification power of more than 2 times showed the highest percent probability of oral precancerous lesion's prediction.

Percentage of artificial intelligence's probability of precancerous lesions in the oral cavity from horizontal direction (the normal photo group and the 180 degree rotation group) was higher than the vertical direction. (the 90 degree and the 270 degree rotation groups)