A NEW AI model has consistently outperformed existing methods in distinguishing between non-muscle-invasive bladder cancer (NMIBC) and muscle-invasive bladder cancer (MIBC), a 2026 study has found.
The Domain-Adaptive Deep Contrastive Network (DADCNet) achieved an accuracy of 95.5% in MRI-based bladder cancer classification.
Bladder Cancer
Bladder cancer is one of the most common malignancies of the urinary system, with high incidence and mortality rates.
It can either be classified as NMIBC or MIBC.
These carcinomas vary significantly in disease progression, prognosis, and treatment strategies.
Thus, distinguishing between the two is paramount to developing personalised treatment interventions.
Further, accurate analysis of the bladder wall and tumour regions on MRI scans ultimately facilitate better clinical decision-making.
DADCNet
Researchers used a bladder cancer MRI dataset from four medical centres, comprising a total of nearly 300 patients.
They then developed a novel DADCNet model for MRI-based grading and classification.
The model received imaging samples from both training and testing datasets and was assessed 10 times.
DADCNet exhibited superior classification performance, improved class balance, and stronger discriminative ability than a wide range of mainstream methods.
Researchers also found that DADCNet predominantly highlights tumour regions and the adjacent muscular layer when distinguishing between NMIBC and MIBC.
They highlighted that this closely aligns the model with diagnostic criteria for muscle invasion in bladder cancer used by radiologists.
Implications for Clinical Practice
Researchers reported that DADCNet holds significant promise as a clinically useful tool for assisting the preoperative diagnosis of bladder cancer and planning individualised care.
However, they also noted the use of a relatively limited dataset size and potential variations in MRI acquisition protocols across different centres included in the study.
A key concern was the computational cost associated with DADCNet training.
It follows that future work will analyse larger datasets and integrate follow-up assessments to established whether DADCNet can be used for prognosis prediction in cases of bladder cancer.
Reference
Huang J et al. A domain-adaptive deep contrastive network for magnetic resonance imaging-driven bladder cancer classification.
Featured image: ATRPhoto on Adobe Stock
