Non-Invasive Tumor Budding Evaluation and Correlation with Treatment Response in Bladder Cancer: A Multi-Center Cohort Study

Abstract

The clinical benefits of neoadjuvant chemoimmunotherapy (NACI) are demonstrated in patients with bladder cancer (BCa); however, more than half fail to achieve a pathological complete response (pCR). This study utilizes multi-center cohorts of 2322 patients with pathologically diagnosed BCa, collected between January 1, 2014, and December 31, 2023, to explore the correlation between tumor budding (TB) status and NACI response and disease prognosis. A deep learning model is developed to noninvasively evaluate TB status based on CT images. The deep learning model accurately predicts the TB status, with area under the curve values of 0.932 (95% confidence interval: 0.898-0.965) in the training cohort, 0.944 (0.897-0.991) in the internal validation cohort, 0.882 (0.832-0.933) in external validation cohort 1, 0.944 (0.908-0.981) in the external validation cohort 2, and 0.854 (0.739-0.970) in the NACI validation cohort. Patients predicted to have a high TB status exhibit a worse prognosis (p < 0.05) and a lower pCR rate of 25.9% (7/20) than those predicted to have a low TB status (pCR rate: 73.9% [17/23]; p < 0.001). Hence, this model may be a reliable, noninvasive tool for predicting TB status, aiding clinicians in prognosis assessment and NACI strategy formulation.

Keywords: bladder cancer; deep learning; multicenter study; neoadjuvant chemoimmunotherapy; tumor budding.

Figure 1

The overall workflow for the discovery and validation of a deep learning model based on CT images to predict the TB status and assess NACI response in patients with BCa. BCa, bladder cancer; CT, computed tomography; H&E, hematoxylin, and eosin; NACI, neoadjuvant chemoimmunotherapy; TB, tumor budding.

Figure 2

Correlation between TB status and NACI response and prognosis. A) Representative H&E‐stained slides showing high TB and low TB statuses (20×). B) The correlation between TB status and response to NACI in the BGB‐A317‐2002 cohort (n = 57; p < 0.001; Pearson's chi‐squared test). C) Changes in the TB status before and after NACI treatment in patients. D–I) Kaplan‐Meier analysis of overall and cancer‐specific survival of patients with BCa stratified by the TB status in the SYMH cohort (n = 514; p < 0.001), external cohort 1 (n = 460; p < 0.001), and external cohort 2 (n = 647; p < 0.001). p‐values were calculated using the Cox proportional hazards model. H&E, hematoxylin and eosin; HR, hazard ratio; K‐M curve, Kaplan‐Meier curve; NACI, neoadjuvant chemoimmunotherapy; pCR, pathological complete response; SYMH, Sun Yat‐sen Memorial Hospital; TB, tumor budding.

Figure 3

Performance of the TB‐based deep learning prediction model across different cohorts. A) Representative examples of the original CT images showing the three input channels: binary tumor mask, arterial‐phase ROI, and venous‐phase ROI. The model prediction is visualized using Guided Grad‐CAM, along with the corresponding pathological images. B–E) AUCs and 95% CIs of results obtained using the deep learning TB prediction model for the training cohort (n = 257), internal validation cohort (n = 110), external validation cohort 1 (n = 342), and external validation cohort 2 (n = 385). F–I) Sensitivity and specificity of results obtained using the TB‐based deep learning prediction model for the training cohort (n = 257), internal validation cohort (n = 110), external validation cohort 1 (n = 342), and external validation cohort 2 (n = 385). AUC, the area under the curve; 95% CI, 95% confidence interval; CT, computed tomography; Guided Grad‐CAM, guided gradient‐weighted class activation mapping; H&E, hematoxylin, and eosin; ROI, region of interest; TB, tumor budding.

Figure 4

Evaluation ability of the model for prognosis and response to NACI. A–D) Kaplan‐Meier analysis of the overall survival of patients with BCa stratified by the predicted TB status in the training cohort (n = 257; p < 0.001), internal validation cohort (n = 110; p < 0.001), external validation cohort 1 (n = 342; p < 0.001), and external validation cohort 2 (n = 385; p < 0.001). P‐values were calculated using the Cox proportional hazards model. E–H) Kaplan‐Meier analysis of the cancer‐specific survival of patients with BCa stratified by the predicted TB status in the training cohort (n = 257; p = 0.002), internal validation cohort (n = 110; p < 0.001), external validation cohort 1 (n = 342; p < 0.001), and external validation cohort 2 (n = 385; p = 0.026). p‐values were calculated using the Cox proportional hazards model. I) H&E‐based TB status and response to NACI in the NACI real‐world cohort (n = 108; p < 0.001; Pearson's chi‐squared test). J) AUCs and 95% CIs of results obtained using the TB‐based deep learning prediction model for the NACI validation cohort (n = 50). K) Sensitivity and specificity of results obtained using the TB‐based deep learning prediction model for the NACI validation cohort (n = 50). L) Predicted TB status and response to NACI in the NACI validation cohort (n = 50; p < 0.001; Pearson's chi‐squared test). AUC, the area under the curve; BCa, bladder cancer; 95% CI, 95% confidence interval; H&E, hematoxylin, and eosin; HR, hazard ratio; NACI, neoadjuvant chemoimmunotherapy; pCR, pathological complete response; TB, tumor budding.