Journal: Annals of surgical oncology
Study type and population
- Retrospective, single-center study of 84 patients with bladder cancer who underwent pelvic lymphadenectomy.
- Objective: assess whether CT-based “tumor habitat” radiomics can better predict occult nodal metastasis (ONM) than conventional whole-tumor radiomics, and whether combining habitat features with clinical factors in a nomogram improves prediction.
Methods
- Imaging and segmentation: preoperative CT tumor volumes were segmented and then partitioned into sub-regions (“habitats”) using a Fuzzy C-Means clustering algorithm based on Hounsfield unit values.
- Feature extraction: radiomics features were extracted from:
- the whole tumor, and
- each sub-region (habitat).
- Clinical factors: univariate logistic regression identified clinical factors associated with ONM.
- Predictive models: multiple models were built:
- whole-tumor radiomics model,
- single-habitat models,
- multi-habitat combined model (notably regions 1, 2, and 3 together).
- Composite nomogram: constructed by integrating:
- habitat-based radiomics score (radscore) and
- selected clinical factors.
- Model evaluation: performance assessed by AUC in training and validation sets. Clinical utility was evaluated with 1000 bootstrap iterations and decision curve analysis.
Key results
- Whole-tumor radiomics model:
- AUC 0.735 (training), 0.727 (validation) – moderate performance.
- Best single-habitat model (region 3):
- AUC 0.859 (training), 0.788 (validation).
- Combined multi-habitat model (regions 1, 2, and 3):
- AUC 0.920 (training), 0.864 (validation) – highest discrimination among radiomics-only models.
- Integrated clinical–radiomics nomogram:
- AUC 0.952 (training), 0.742 (validation).
- Decision curve analysis: suggested potential clinical benefit for using the nomogram in ONM risk stratification.
Oncologic/clinical implications
- Habitat-based radiomics: which captures intratumoral heterogeneity rather than treating the tumor as a uniform volume, improves prediction of occult nodal metastasis compared with whole-tumor radiomics alone.
- Combined nomogram: integrating habitat radscore with clinical factors shows promising discrimination and net benefit, but the drop in validation AUC and retrospective, single-cohort design highlight the need for robust external validation before clinical adoption.