Fig 1.
Flow diagram depicting the inclusion and exclusion criteria of the test and validation sets.
Fig 2.
Representative H&E staining and single desmin or CD31 immunolabeling images of different patterns of venous invasion of pancreatobiliary tract cancers.
Left column (A, D, and G), suspicious destructive pattern. Smooth muscle layer is not identified on H&E but observed on desmin immunolabeling; central column; (B, E, and H), intraepithelial (IN)-like pattern, cancer cells within venous space mimic dysplastic cells of intraepithelial neoplasia showing well-circumscribe outer desmin-positive smooth muscle layer (E) and no CD31 expression (H); right column (C, F, and I), conventional pattern, well-circumscribe outer desmin-positive smooth muscle layer and cancer cells within venous space (F), cancer cells within venous space attached to endothelial cells showing partial loss of CD31 expression (I), while cancer cell-non-attached endothelial cells show brown color (A-C, H&E staining; D-F, desmin immunolabeling; G-I, CD31 immunolabeling; all ×200 magnification).
Fig 3.
Representative H&E staining and dual CD31‒desmin immunolabeling (magenta, CD31; desmin, brown) images of different patterns of venous invasion of pancreatobiliary tract cancers.
A and E) Conventional pattern. Cancer cells within venous space partly attach to endothelial cells showing partial loss of CD31 expression (arrows), while cancer cell-non-attached endothelial cells show magenta color (intact CD31 labeling); B and F) IN-like pattern; cancer cells within venous space show spotty CD31 expression (magenta color) with well- circumscribe outer desmin-positive (brown color) smooth muscle layer; C and G) definite destructive pattern of venous invasion on H&E staining (C). Some parts of smooth muscular layer are destroyed by infiltrating cancer cells and remnant smooth muscle layer of muscular vein is identified on the H&E-stained (C) and confirmed by desmin (G; brown color) immunolabeled slide; D and H) suspicious destructive pattern of venous invasion on H&E staining (D). The single artery sign is observed; however, almost all smooth muscular layer are destroyed by infiltrating cancer cells and the equivocally remnant smooth muscle layer of muscular vein cannot be definitely identified on the H&E-stained slide (D) but is confirmed by the desmin (H; brown color) immunolabeling. (A-D, H&E staining; E-H, dual CD31‒desmin immunolabeling, CD31, magenta color; desmin, brown color; ×200 magnification).
Table 1.
Clinicopathologic characteristics of the test and validation sets.
Table 2.
Foci number of VI based on invasion patterns and staining methods in the test set.
Table 3.
Comparisons of vascular invasion patterns based on staining methods in the test set.
Fig 4.
Comparisons of mean foci number of VI of pancreatobiliary tract cancers in the test set.
(A) Total foci of venous invasion regardless of invasion pattern. Invasive foci of the (B) conventional, (C) IN-like, and (D) destructive patterns, as assessed by H&E staining and desmin and CD31 immunolabeling in the test set. (A) The mean total number of foci of VI regardless of invasion pattern detected by CD31 (P = 0.022) and desmin (P = 0.027) immunolabeling is significantly higher than that detected by H&E staining. (B) CD31 immunolabeling detected more invasion foci of the conventional pattern of VI than H&E staining (P < 0.001). (C) There is no significant difference in the detection of the IN-like pattern of VI by H&E staining, desmin, and CD31 immunolabeling. (D) Desmin immunolabeling detected more invasion foci of the destructive pattern (P < 0.001).
Fig 5.
Representative images of mixed patterns of VI.
(A) A focus of VI showing all three patterns—destructive (D), IN-like (I), and conventional (C). This rapid transition from destructive to IN-like and finally to conventional patterns suggests that these three patterns are strongly associated between each pattern. (B) VI exhibiting mixed patterns showing transition from IN-like (I) to conventional (C) patterns (all, ×100 magnification).
Fig 6.
Representative images of CD31 immunolabeling of VI.
(A) Partial loss of CD31 labeling (arrows) of endothelial cells where attachment of cancer cells on the endothelial portion. (B) Dual CD31 and desmin immunolabeling showing total loss of CD31 labeling on endothelial cells where attachment of cancer cells on the entire endothelial cells (arrows) with outer desmin positive smooth muscle of muscular vein. (C) Intact CD31 labeling of endothelial cells (all, ×400 magnification).
Fig 7.
Comparisons of mean number of foci of VI of pancreatobiliary tract cancers on the validation set.
(A) Total foci of VI regardless of invasion pattern and invasive foci exhibiting the (B) conventional, (C) IN-like, and (D) destructive patterns as assessed by H&E staining and dual CD31‒desmin immunolabeling among the validation set. (A) Dual CD31‒desmin immunolabeling detected more invasion foci of VI regardless of the invasion pattern than H&E staining (P = 0.012). (B) No significant difference was observed in the detection of the conventional pattern of VI between H&E staining and dual CD31‒desmin immunolabeling. (C) Dual CD31‒desmin immunolabeling detected more invasion foci exhibiting the IN-like pattern of VI than H&E staining (P = 0.0045). (D) Similarly, dual CD31‒desmin immunolabeling detected more invasion foci exhibiting the destructive pattern than H&E staining (P < 0.001).
Table 4.
Focus-by-focus comparisons of vascular invasion patterns observed by H&E and dual CD31‒desmin immunolabeling of the validation set.
Table 5.
Association between VI and other clinicopathologic features in pancreatobiliary tract cancers.
Fig 8.
Kaplan–Meier survival analyses of disease-free survival of pancreatobiliary tract cancers based on VI status.
(A) Patients with pancreatobiliary tract cancer and venous invasion detected by H&E staining have significantly worse disease-free survival (3-year survival rate, 23%) than those without VI (46%; P = 0.016). (B) Patients with pancreatobiliary tract cancer and venous invasion detected by dual CD31‒desmin immunolabeling have significantly worse disease-free survival (3-year survival rate, 18%) than those without venous invasion (67%; P <0.001). (C) Patients with pancreas cancer and venous invasion have significantly shorter disease-free survival (3-year survival rate, 14%) than those without venous invasion (42%; P = 0.02) by dual CD31‒desmin immunolabeling. Patients with (D) ampulla of Vater and (E) biliary tract cancers and venous invasion had a tendency of shorter disease-free survival than those without venous invasion (ampulla of Vater cancer, 3-year survival rate, 34% vs 86%, P = 0.192; biliary tract cancer, 100% vs 25%, P = 0.250) by dual CD31‒desmin immunolabeling.
Table 6.
Univariate and multivariate analyses of pancreatobiliary tract cancer in the validation set.
Fig 9.
Diagram showing the proposed sequence of VI of pancreatobiliary tract cancers.