Oral leukoplakia (OL) evaluation through photographs can be performed with the aid of Artificial Intelligence (AI). Supervised Machine Learning (SML) processes, which are based on labeling, are indicated to ensure a reliable computational mechanism of lesion identification. Thus, OL classification and demarcation within a photograph are crucial for SML.

Objective:

To label OL lesions in homogeneous and non-homogeneous using photographs, and to test a segmentation procedure, aiming for its use in a trustworthy dataset.

Methods:

Fifty-five OL photographs were inserted into Fiji/ImageJ, and a region of interest (ROI) was defined to obtain a three-dimensional plot of pixel color clustering. Then, the photography and the plot were used for OL classification by a panel of 5 experts in Oral Medicine. The segmentation process was performed by two operators which created a second ROI for evaluation of the lesion by area, perimeter, centroid, and circularity. The intraclass correlation coefficient was calculated and a comparative analysis was performed (Mann Whitney and Unpaired t-test). Then, segmentation was accomplished by creating a computer code including the precise information of the lesional site, in an automated binarization fashion.

Results:

The experts agreed in 53% of the cases regarding OL classification. An excellent level of operator agreement related to the size and site of the lesion was found. Although, differences were found comparing the lesion’s area, perimeter, and centroid (p<0.05). The code was effective for the segmentation separating the lesion from the background.

Conclusion:

The agreement on OL classification among experts accounted for half of the cases. The lesion segmentation was possible using a computer code based on interactive drawing. With an excellent agreement between operators, the manual delimitation of lesional sites can be used for SML, but the differences regarding lesional perimeter and its classification should be considered before labeling and creating a good dataset. (AU)