Deep learning on MIP FDG-PET predicts lymph node spread in esophageal squamous cell carcinoma

A deep learning model on rotational MIP FDG-PET achieved AUC 0.82 (95% CI 0.57–0.94) and 92% sensitivity for lymph node metastasis in esophageal squamous cell carcinoma, vs radiologists’ 58% sensitivity.

• At the optimal threshold, the CNN model accuracy was 86%, compared to 67% for SUVmax and 69% for radiologists, though differences were not statistically significant (p>0.05).
• The model’s sensitivity of 92% outperformed radiologists (58%) and SUVmax (63%), but specificity was lower (75% vs. 92% for radiologists), acting as a diagnostic safety net.
• This retrospective proof-of-concept suggests potential for preoperative PET-based AI to reduce understaging in ESCC.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.