Figure 4: Comparison with pathologists.

a. 100 slides from 22 studies were randomly selected from TG-4k for comparison with pathologists. Ten veterinary pathologists were asked to report the presence of each lesion and assign a score from normal, minimal, mild, moderate, or severe. After the independent evaluation, a consensus was reached between three pathologists to derive the gold standard. In parallel, each slide was processed by TRACE (FT) to derive patch predictions with 80% patch overlap. Class-wise patch predictions were then thresholded to retain solely high-confidence predictions, from which the quantitative scores describing the percentage of the slide highlighting each lesion were derived. b. Comparison between TRACE (FT) and consensus (black triangles), and pathologists and consensus (pink box plot). Evaluation reporting Quadratic Cohen’s kappa score. Vertical lines indicate the average TRACE (FT) and pathologist agreement with the consensus. Boxes indicate quartile values of Quadratic Cohen’s Kappa score, with the red center line indicating the 50th percentile. Whiskers extend to data points within 1.5x the interquartile range.