Highlight
– Integration of CT radiomic features with clinical data enhances prediction of recurrence-free survival (RFS) in high-grade serous ovarian carcinoma (HGSOC).
– Multi-time point imaging, including baseline and post-neoadjuvant chemotherapy (NACT), provides incremental prognostic insight.
– Residual disease, patient age, and radiomic feature kurtosis are significant individual predictors of shorter RFS.
– Incorporation of radiomics changes between time points after surgery did not significantly improve predictive model performance.
Study Background
High-grade serous ovarian carcinoma (HGSOC) is the most lethal subtype of epithelial ovarian cancer, characterized by high rates of recurrence despite aggressive multimodal treatment. Neoadjuvant chemotherapy (NACT) followed by interval debulking surgery is a common treatment strategy for selected patients. Predicting recurrence-free survival (RFS) in HGSOC remains challenging yet crucial for individualized patient management and therapy optimization. Conventional prognostic models rely on clinical and pathological data, which often lack precision. Recent advances in radiomics—the extraction of quantitative imaging features—hold promise for noninvasive tumor characterization. Integrating radiomics data from computed tomography (CT) imaging with clinical and genomic parameters may refine prognostication and foster precision medicine in ovarian cancer.
Study Design
This retrospective single-center study enrolled 91 patients diagnosed with HGSOC who were treated with neoadjuvant chemotherapy followed by surgical debulking. All patients underwent portal venous phase contrast-enhanced CT imaging at two critical time points: baseline (pre-NACT) and after completion of NACT but prior to surgery. Using standardized 2D segmentation, first-order texture radiomic features—parameters describing the distribution and intensity of pixel values within the tumor region—were extracted from CT images of selected disease sites with commercially available texture analysis software. The primary endpoint was recurrence-free survival.
Multivariate Cox proportional hazards models were constructed to evaluate the prognostic value of clinical, radiomic, and genomic features at baseline, post-NACT, and post-surgery time points. Models were compared by their concordance (C-) statistics to assess discriminatory accuracy in predicting RFS. The study also explored whether changes in radiomic features between imaging time points could further improve prognostic modeling.
Key Findings
Clinical variables alone at baseline yielded a poor predictive performance (C-statistic 0.53), demonstrating limited ability to predict RFS. When baseline radiomic features were added to clinical data, model performance improved moderately (C-statistic 0.63), indicating additive prognostic value of CT texture features.
Post-NACT models that incorporated all baseline clinical data along with the delta (change) in radiomic features between baseline and post-NACT imaging maintained a similar predictive accuracy (C-statistic 0.63), suggesting the importance of early imaging-based tumor response characterization but no major gain from dynamic radiomic changes at this stage.
In the post-surgery setting, integrating baseline data plus surgical outcome variables, most notably residual disease burden, further improved model performance (C-statistic 0.69). Inclusion of changes in radiomic features across all time points did not significantly enhance the post-surgical model (C-statistic 0.7), indicating diminishing returns of radiomic temporal changes after definitive surgery.
Among individual predictors, older age, presence of residual disease after surgery, and the radiomic feature kurtosis (a measure of image texture peakedness) were significantly associated with shorter recurrence-free survival. These findings underscore the multifactorial nature of recurrence risk and highlight kurtosis as a novel imaging biomarker with potential biological underpinnings.
Expert Commentary
This study exemplifies the incremental value of radiomics in complementing traditional clinical prognostic factors in HGSOC. CT-based radiomic features, which capture intra-tumoral heterogeneity noninvasively, may reflect underlying tumor biology such as cellularity, necrosis, or stromal composition. The finding that radiomics improves risk stratification supports growing evidence that quantitative imaging biomarkers can be integrated into clinical decision-making frameworks.
However, the moderate C-statistic values suggest room for improvement and the need for larger, multicenter validation cohorts. The lack of substantial benefit from longitudinal changes in radiomic features post-surgery may reflect that surgical removal of tumor mass alters tumor characteristics that are not fully captured by imaging texture metrics. Moreover, the retrospective single-center nature limits generalizability. Incorporation of genomic data, beyond clinical and radiomic features, as well as advanced machine learning approaches, might further enhance predictive accuracy.
Current guidelines do not yet include radiomics-based prognostic tools, making this research an important step toward their future clinical adoption. Further mechanistic studies linking radiomic signatures to molecular pathways could elucidate biological plausibility and therapeutic implications.
Conclusion
The integration of computed tomography radiomic features with clinical variables offers a promising approach to refine prognostication for recurrence-free survival in patients with high-grade serous ovarian cancer treated with neoadjuvant chemotherapy and surgery. Radiomic kurtosis and residual disease after surgery emerged as key predictors, emphasizing the complementary value of imaging biomarkers. While incremental gains from multi-time point radiomic changes were modest, the results support further prospective validation and incorporation of radiomics into personalized management strategies that aim to improve patient outcomes in HGSOC.
Funding and Clinical Trials
Details regarding funding sources or clinical trial registration were not provided within the original publication.
References
- Roller L, Zhou N, Atre I, et al. Toward precision prognosis: Predicting recurrence-free survival in high-grade serous ovarian cancer patients using multi-time point clinical and computed tomography radiomics data. Gynecol Oncol. 2026 Jul 2;211:98-107. PMID: 42391853.
- Huang YQ, Liang CH, He L, et al. Development and validation of radiomics nomogram for preoperative prediction of lymph node metastasis in colorectal cancer. J Clin Oncol. 2016;34(18):2150-2157.
- Lambin P, Rios-Velazquez E, Leijenaar R, et al. Radiomics: extracting more information from medical images using advanced feature analysis. Eur J Cancer. 2012 Mar;48(4):441-6.
- Bookman MA, Brady MF, McGuire WP, et al. Evaluation of new platinum-based treatment regimens in advanced-stage ovarian cancer: a Gynecologic Oncology Group Study. J Clin Oncol. 2009;27(9):1419-1425.