Overview
Accurate preoperative detection of muscle-invasive bladder cancer remains a clinical challenge. The investigators aimed to develop and validate a knowledge-guided causal diagnostic network for the detection of muscle-invasive bladder cancer with multiparametric magnetic resonance imaging(MRI).
Description
Patients who underwent bladder MRI were retrospectively collected at three centers between January 2013 and September 2023. The investigators first constructed a nnUNet to segment causal region where muscle-invasive bladder cancer may occur. Subsequently, the investigators explored a causal network based on a modified ResNet3d-18 by striking a fine balance between nnUNet awareness and a self-supervised learning (SSL) model, which steered model to emulate diagnostic acumen of expert in staging muscle-invasive bladder cancer at MRI. Model was trained in center 1, and independently tested in center 1, center 2 and center 3. Ablation test was performed among all 13 Ablation-Test models using either single or multi-parametric MRI. Benefit was tested in six radiologists using vesical imaging-reporting and data system (VI-RADS) versus network-adjusted VI-RADS.
Eligibility
Inclusion Criteria:
- Urothelial carcinoma of the bladder confirmed by final histopathology ②Received a standard contrast-enhanced 3.0T mpMRI before surgery ③All tumors within patients included were resected and received pathologic examination separately
Exclusion Criteria:
①Absence of surgical interventions
②With inadequate image quality or with inadequate pathology for analysis