気管支動脈塞栓術におけるアダムケビッツ動脈および前脊髄動脈検出のための深層学習モデル：多施設共同後ろ向き研究

Spinal cord infarction is the most serious complication of bronchial artery embolization (BAE) for massive hemoptysis. The prevention of this complication requires precise identification of the spinal artery during angiography. However, artificial intelligence-assisted radiographic image analysis, particularly for spinal artery recognition, is still underdeveloped.

To evaluate the efficacy of a deep learning model for recognizing the artery of Adamkiewicz and anterior spinal artery (ASA) to avoid spinal artery embolization during BAE.

A novel deep learning-based framework was proposed for spinal artery identification, comprising region of interest (ROI) perception and target spinal artery identification. The ROI perception extracts vessel-related regions to improve the conspicuity of the spinal artery. The target spinal artery identification utilizes a progressive refinement learning network, localizing the artery from the global view and progressively refine the identification results through cross-scale information interaction. Therefore, the proposed framework can identify the spinal artery.

This multi-center, retrospective study was conducted from January 2019 to December 2023, involving 2,036 patients with hemoptysis who underwent de novo BAE. Among these, 78 (3.8%) patients had identifiable artery of Adamkiewicz and ASA on right intercostal-bronchial artery angiography. Experimental results demonstrated the effectiveness of the proposed method, achieving a sensitivity of 92.10% and specificity of 84.62%, with no statistically significant difference from the radiologist fellow interpretation.

The study presents a novel deep learning system for spinal artery detection during BAE. It shows high sensitivity and performance comparable to radiology fellows but has false negatives, rendering it a negative study.