MRG'ye Dayalı Radyomik Tahmin Edilen TRG Heterojenliği, Neoadjuvan Kemoterapi Sonrası KRKC Hastalarında Hayatta Kalmayı Öngörüyor

Özet (abstract)

To develop and validate a radiomics model for predicting tumor regression grade (TRG) of liver metastases in patients with colorectal cancer liver metastases (CRLM). To further explore the prognostic value of the best, the worst and the heterogeneity of radiomics-predicted TRG (R min , R max and R H ) for predicting survival. The radiomics model was constructed to predict TRG using the latest preoperative MRI, with tumor residual rate as the label. Radiomics-predicted TRG heterogeneity (R H ) was defined as the presence of both good (predicted-TRG 1–3, predicted-tumor residual rate ≤ 50%) and poor (predicted-TRG 4–5, predicted-tumor residual rate > 50%) predicted responses across different liver lesions within the same patient. The Kaplan-Meier method was used to draw survival curves of R H+ (with R H ) and R H− (without R H ) patients. Cox regression analysis was used to establish a predictive model for local tumor disease-free survival (LTDFS) with R H , R min , R max and clinical characteristics. A total of 295 liver lesions from 83 patients were included in this single center retrospective study (199 in training group, 96 in validation group). Twenty-five radiomic features were screened to construct the radiomics model. The correlation between tumor residual rate and radiomics score was moderate in both groups (r 2 = 0.360, r 2 = 0.440). The median LTDFS was 3.0 months (95% CI: 1.0–5.0) in 13 R H+ patients and 10.2 months (95% CI: 8.4–12.0) in 70 R H− patients ( p = 0.002). R H , R min and intraoperative RFA were independent predictors for LTDFS ( p = 0.012, p = 0.020, p = 0.010, respectively). A gadoxetic acid-enhanced MR based radiomics model achieved modest predictive performance for TRG. Radiomics-predicted TRG heterogeneity had strong predictive effect on LTDFS. R H and R min were independent prognostic factors for LTDFS.

Yazarların özeti; kaynağından alınmıştır. BMC Medical Imaging, 2026 · DOI ↗