Figure 5
Two example cases with grades between vPatho and the pathology reports. The proportions were estimated by vPatho (AI), and the resulting ISUP grade and tertiary Gleason pattern were compared with pathology reports (clinical routine). The tumor volume as a percentage was calculated by vPatho. Although both the vPatho and pathology results revealed the right Gleason pattern in these patients, differences in the ISUP grade were observed. Although the Gleason pattern 4 detected by vPatho was affected by a greater false negative rate than was the other Gleason patterns 3 and 5, we found that the grade difference between vPatho and pathology reports was associated with a greater proportion of estimations by vPatho for Gleason pattern 4 than by that for Gleason pattern 3 (see Fig. 4). This finding revealed that the proportion of patients with a Gleason pattern marked as a tertiary Gleason pattern by pathologists is twice as high as the 10% threshold (a corrected threshold for determining the secondary Gleason pattern), indirectly highlighting the inaccurate size estimation (50% of the original size) made by human observers (pathologists). This result is in accordance with our previous study in which we showed that human observers (pathologists) significantly underestimate the size proportion (by 50% of the original tumor) compared to computer-assisted size estimation on an independent dataset with 255 prostatectomy specimens4.
