Fig. 3

Bland–Altman plot—assessing the repeatability of AVE scores obtained from 2 different J8 images captured from the same patient at the same screening visit. It displays the repeatability of AVE scores obtained from 2 different J8 images captured from the same patient at the same visit by Bland–Altman plot. The x-axis shows the average of 2 AVE scores obtained from 2 different J8 images while the y-axis shows the difference of these scores. Continuous AVE score is obtained as the summation of class label (0, 1, 2) multiplied by its corresponding class AVE prediction. Each point in the plot is colored according to its ground truth. Blue points represent ground truth normal patients, yellows are indeterminate cases, and reds are confirmed CIN2+ cases. Under perfect repeatability, score differences are expected to be zero; therefore, in an ideal situation, all of the points on the graph are expected to be lying on the y = 0 line (horizontal line passing through 0). However, in our situation points vary around this horizontal line, and the variability is highest at the middle (where x = 1). This means that the variability in score differences is dependent on score averages. The variability is smallest at each end 0 (corresponding to normal) and 2 (corresponding to precancer/cc), and is highest at the middle which means there is low repeatability at indeterminate class compared to definite normal and precancer/cancer classes