Clinical Scorecard: Automated Deep Learning Segmentation and CT Elastography Radiomics for Predicting Lymph Node Metastasis Near the Right Recurrent Laryngeal Nerve in Esophageal Cancer Preoperatively

At a Glance

Key Highlights

• Automatic segmentation model achieved high accuracy (Dice coefficient 0.898 ± 0.024) for lymph nodes near right RLN.
• Five DEM-derived radiomic features, especially first-order entropy, showed strong diagnostic performance (AUC 0.814, sensitivity 0.895, specificity 0.709).
• The combined approach provides a non-invasive, reproducible, and clinically useful tool for preoperative prediction of right RLN lymph node metastasis in ESCC.

Guideline-Based Recommendations

Diagnosis

• Use automated nnU-Net segmentation to delineate lymph nodes near the right RLN on CT images.
• Incorporate CT elastography-derived radiomic features, including entropy and fractal dimensions, to improve detection of metastatic lymph nodes beyond conventional size-based criteria.

Management

• Integrate imaging-based prediction of right RLN lymph node metastasis into preoperative treatment planning to personalize therapeutic strategies.

Monitoring & Follow-up

• Apply probability calibration and decision curve analysis to assess and monitor diagnostic model performance across clinical thresholds.

Risks

• Conventional CT criteria may miss micrometastases (<5 mm), leading to underdiagnosis.
• Manual or semi-automatic segmentation methods are time-consuming and prone to operator bias, reducing reproducibility.

Patient & Prescribing Data

415 patients with esophageal squamous cell carcinoma included in retrospective study

Automated segmentation combined with CT elastography radiomics enables accurate preoperative prediction of lymph node metastasis, facilitating personalized treatment decisions and potentially improving prognosis.

Clinical Best Practices

• Employ fully automated deep learning segmentation (nnU-Net) to reduce operator variability and improve reproducibility in lymph node delineation.
• Utilize CT elastography-derived radiomic features that reflect tissue biomechanical heterogeneity for enhanced metastatic detection.
• Incorporate advanced imaging biomarkers such as entropy and fractal dimension features alongside clinical data for comprehensive risk assessment.

References

• Global incidence and mortality of esophageal cancer
• Radiomics in esophageal cancer lymph node metastasis prediction
• nnU-Net framework for automated medical image segmentation
• CT elastography and biomechanical tissue properties