Researchers from China have demonstrated the ability of a deep learning radiomics signature (DLRS)— constructed with computed tomography images—to accurately predict the muscle invasion status of bladder cancer.
The preoperative assessment of muscular invasion in bladder cancer is crucial for appropriate and optimal treatment. However, preoperative diagnosis of muscle invasion status in bladder cancer can be challenging.
In a retrospective review of 173 patients, researchers noted 43 patients with pathologically proven muscle-invasive disease and 130 patients with non-muscle-invasive disease. Of these patients, 129 were randomly assigned to a training cohort and the remaining 44 were assigned to a test cohort.
Then, researchers constructed 6 machine learning classifiers based on deep learning radiomics features, which were adapted to predict muscle invasion. To evaluate the performance of their models, researchers used area under the curve (AUC), accuracy, sensitivity, and specificity.
Results of the analysis showed DLRS-based models performed best in predicting muscle invasion status. In particular, the multilayer perceptron (MLP) model outperformed the other models (training cohort AUC: 0.97% [95% CI, 0.95-0.99] and test cohort AUC: 0.88 [95% CI, 0.78-0.98]).
Additionally, the sensitivity, specificity, and accuracy of the MLP model in the test cohort were 0.91 (95% CI, 0.55-0.87), 0.78 (95% CI, 0.59-0.86), and 0.58 (95% CI, 0.73-0.83), respectively.
In their concluding remarks, researchers wrote that these results indicated “DLRS-based MLP provided great clinical utility in distinguishing non-muscle-invasive bladder cancer from muscle-invasive bladder cancer.” This model, they added, was superior to models constructed using radiomics features only.