A DEEP LEARNING model identified all lymph node metastases in a bladder cancer test set, achieving 100% sensitivity and 60% specificity, according to a new study. The findings suggest that artificial intelligence (AI) could support the assessment of histopathological slides and improve efficiency in routine bladder cancer pathology.
Researchers developed a supervised deep learning model to identify lymph node metastases in patients with bladder cancer. Accurate pathological assessment of dissected lymph nodes is important for prognosis and treatment decisions, although conventional histopathological examination can be labour intensive and time consuming.
The study included 100 histopathological slides of lymph nodes collected between 2015 and 2024 from Sahlgrenska University Hospital. The slides were digitised into whole slide images and divided into training and validation or test sets.
The model was trained using 3,437 pixelwise annotations. For validation, 50 regions containing normal tissue and metastases were assessed using the AI model and two specialist pathologists.
In the test set, the deep learning model achieved 100% sensitivity and 60% specificity for detecting lymph node metastases.
The researchers also assessed whether AI assistance could reduce the time required for pathologists to review complete slides. For the first pathologist, median review time fell from 9 seconds without AI assistance to 4.2 seconds with AI assistance (p<0.001).
For the second pathologist, median review time decreased from 12 seconds to 4 seconds following the introduction of AI assistance (p<0.001).
These reductions indicate that the model could help pathologists identify relevant areas of histopathological slides more rapidly.
The findings suggest that deep learning could provide a practical tool for supporting the detection of lymph node metastases in bladder cancer, particularly where pathology workloads make comprehensive slide assessment challenging.
Importantly, the model was developed using a relatively limited dataset. The researchers suggested that institutions with lower case volumes could potentially develop locally trained AI models to support routine histopathological assessment.
Further evaluation will be important to establish how such systems perform across broader clinical settings and datasets. The study highlights the potential for AI-assisted pathology to complement, rather than replace, specialist assessment of bladder cancer lymph nodes.
Reference
Gorczynski A et al. Deep learning-assisted detection of lymph node metastases in bladder cancer. Diagn Pathol. 2026;21:58.
Featured image: natali_mis on Adobe Stock
Each article is made available under the terms of the Creative Commons Attribution-Non Commercial 4.0 License.
