Recently, Rundo et al. published a study in Clinical Genitourinary Cancer investigating a three-dimensional deep radiomics pipeline to predict disease control in patients with metastatic urothelial cancer (UC) treated with ICIs. They analyzed the computed tomography (CT) images of target lesions from 43 patients with metastatic UC. The imaging dataset was used for initial deep learning (70%) then testing and validation (30%). The deep learning pipeline was based on a 3-dimensional densely connected convolutional neural network (3D-DCNN) with separable convolutions and nonlocal blocks (NLBs). It was used to analyze target lesions before the initiation of ICIs.
Input data from segmented CT scan lesions was used to classify patients into two categories, patients with disease response to ICIs and patients with disease progression. The predictive accuracy of the 3D-DCNN architecture was 82.5%, with a sensitivity of 96% and a specificity of 60%. The addition of baseline clinical factors increased the predictive accuracy of the 3D-DCNN classifier to 92.5%, owing to an improvement of the specificity that increased to 87%.
This study shows the potential of radiomic approaches promising noninvasive biomarkers for predicting disease control on ICIs.
Written by: Bishoy M. Faltas, MD, Director of Bladder Cancer Research, Englander Institute for Precision Medicine, Weill Cornell Medicine, New York City, New York
- Rundo F, Bersanelli M, Urzia V, Friedlaender A, Cantale O, Calcara G, et al. Three-Dimensional Deep Noninvasive Radiomics for the Prediction of Disease Control in Patients With Metastatic Urothelial Carcinoma treated With Immunotherapy. Clin Genitourin Cancer. 2021;doi.10.1016/j.clgc.2021.03.012. PMID: 33849811
Read the Abstract