Tumor-lymph cross-plane projection excavates features of spatial relations: ResNet-CBAM model for prognostic prediction in esophageal cancer

Jiayang Xu ^1,2 Chen Huang ² Qianshun Chen ² Jieyang Wang ² Yuyu Lin ² Wei Tang ^1,2 Wei Shen ^1,2 Xunyu Xu ^1,2*

• ¹ Shengli Clinical Medical College, Fujian Medical University, Fuzhou, China
• ² Department of Thoracic Surgery, Fujian Provincial Hospital, Fuzhou, Fujian Province, China

Background: Prognostic models for esophageal cancer based on contrast-enhanced chest CT can assist thoracic surgeons in devising personalized treatment plans to optimize patient outcomes. However, the extensive lymphatic drainage and early lymph node metastasis of the esophagus pose significant challenges in extracting and analyzing effective lymph node characteristics. Previous studies have focused on tumor and lymph node features separately, neglecting spatial correlations such as position, direction, and volumetric ratio. Methods: A total of 285 patients undergone radical resection surgery were retrospectively collected at Fujian Provincial Hospital from 2018 to 2022. This study proposed a tumor-lymph node projection plane constructed by projecting lymph node ROI to the tumor ROI plane. A ResNet-CBAM model, combining a residual convolutional neural network with a CBAM attention module, was employed for feature extraction and survival prediction. The PJ-group utilized tumor-lymph node projection planes as training data, while TM-group and ZC-group respectively utilized tumor ROI and concatenated images of tumor and lymph ROI as control. Additional comparisons were conducted with traditional machine learning models (Support Vector Machines, Logistic Regression, and K-Nearest Neighbors). Survival outcomes (median, 批注 [嘉徐1]: Samples were marked in 'abstract' paragraph now.