Deep learning identifies histopathologic changes in bladder cancers associated with smoke exposure status

Area of Science:

• Oncology
• Pathology
• Artificial Intelligence

Background:

• Smoking is a known risk factor for bladder cancer (BC).
• The relationship between BC histology and smoking status is not well understood.

Purpose of the Study:

• To investigate if bladder cancer histology images can predict smoking exposure status.
• To develop and validate a deep learning model for this prediction.

Main Methods:

• A deep learning model was trained on whole-slide histology images from 66 BC cases.
• The model was externally validated on 94 BC cases, assessing prediction accuracy using AUC.
• Multivariate analyses controlled for BC grade, gender, age, and time to diagnosis.

Main Results:

• The model achieved an AUC of 0.67 (95% CI: 0.58-0.76) on external validation, indicating non-random classification.
• The model independently predicted smoking exposure status with an odds ratio of 1.710 (95% CI: 1.148-2.54).
• Histologic patterns predictive of smoking status were identified.

Conclusions:

• Deep learning can identify histopathologic features in bladder cancer that correlate with smoking exposure.
• Histology-based prediction of smoking status may provide valuable clinical information.