- Collagen in the tumor microenvironment predicts the likelihood that patients with early gastric cancer will have lymph node metastases (LNM).
- Collagen features were extracted from multiphoton images using MATLAB 2015b (MathWorks), including 12 morphologic features and 134 textural features.
Why this matters
- Endoscopic ultrasonography and CT are of limited use in determining LNM, which is a key factor in determining surgical strategy.
- The arrangement and orientation of extracellular matrix collagen are known to be associated with metastasis in other cancers.
- Retrospective analysis of a primary cohort (n=232) at 2 medical centers in China who underwent radical gastrectomy and were diagnosed with T1 gastric cancer. The derived model was tested against a second cohort (n=143) from a separate hospital.
- Funding: Wide range of nonindustry sources in China.
- The collagen signature was associated with LNM (OR, 5.470; P<.001>
- Primary cohort area under the receiver operating characteristic curve for metastasis (AUROC, 0.955; 95% CI, 0.919-0.991). Validation cohort AUROC, 0.938 (95% CI, 0.897-0.981).
- At the maximum Youden index of 0.301, the model had good sensitivity (90.0%), specificity (90.3%), accuracy (90.2%), positive predictive value (71.1%), and negative predictive value (97.1%) in the validation cohort.
- Retrospective analysis.