3D-Deep Non-Invasive Radiomics for the Prediction of Disease Control in Patients with Metastatic Urothelial Carcinoma treated with Immunotherapy

ABSTRACT Introduction Immunotherapy is effective in a small percentage of patients with cancer and no reliable predictive biomarkers are currently available. Artificial Intelligence (AI) algorithms may automatically quantify radiological characteristics associated with disease response to medical treatments. Methods We investigated an innovative approach based on the use of a 3D Deep Radiomics pipeline to classify visual features of chest-abdomen CT with the aim of distinguishing disease control (complete or partial response and stable disease) from progressive disease (PD) to immune checkpoint inhibitors (ICIs). The analysis was performed in 42 consecutive patients with metastatic urothelial cancer (mUC) who had progressed on first-line platinum-based chemotherapy and had baseline CT scan images at the time of immunotherapy initiation. The 3D-pipeline included self-learned visual features and a deep self-attention mechanism. According to the outcome to the ICIs, a 3D Deep Classifier semi-automatically categorised the most discriminative Region of Interest of the CT scan images. Results With a median follow-up of 13.3 months (95% CI, 11.1-15.6), the median overall survival (OS) for all patients was 8.5 months (95% confidence interval [CI], 3.1-13.8). According to the disease response to immunotherapy, the median OS was 3.6 months (95% CI, 2.0-5.2) for patients with PD, whilst it had not yet been reached for those with disease control. The predictive accuracy of the 3D-pipeline was 82.5%, with a sensitivity and specificity of 96% and 60%, respectively. The addition of baseline clinical factors increased the accuracy to 92.5% by improving specificity to 87%, while the accuracy of other architectures ranged from 72.5% to 90%. Conclusion AI by 3D Deep Radiomics is a potential non-invasive biomarker for the prediction of disease control to ICIs in mUC and deserves validation in larger series. MICROABSTRACT Artificial Intelligence algorithms may automatically quantify radiological characteristics associated with disease response to medical treatments. We investigated a 3D Deep Radiomics pipeline to predict disease control to immunotherapy in 42 consecutive patients with metastatic urothelial cancer. The predictive accuracy of the pipeline was 82.5% and increased to 92.5% by the addition to the imaging of other baseline clinical factors.