Fully automatic identification of post-treatment infarct lesions after endovascular therapy based on non-contrast computed tomography

Non-contrast computed tomography (NCCT) of the brain is critical to patients with acute ischemic stroke who receive thrombolysis and thrombectomy. It can help identify reperfusion-related hemorrhage, edema which need intervention. It also can guide the timing and intensity of antithrombotic therapy. Rapid, accurate, and automated detection and segmentation of acute ischemic lesions after endovascular therapy (EVT) are highly needed. In this work, we propose a novel encoder-decoder network for fully automatic segmentation of acute ischemic lesions after EVT on NCCT, which is named ISCT-EDN. NCCT images of AIS (acute ischemic stroke) patients who underwent EVT in a multicenter cohort study were collected in this study. ISCT-EDN takes hierarchical network as backbone. Feature pyramid network (FPN) is designed to aggregate features from multi stages of backbone. Reasonable feature fusion strategy is considered in FPN to enhance multi-level propagation. In addition, to overcome the limitation of fixed geometric structure of convolution for multi-range dependency exploitation, non-local parallel decoder is introduced with deformable convolution and self-attention. The proposed model was compared with 7 segmentation models which are commonly used in the medical domain and the performance was superior to other models in in the segmentation of post-treatment infarct lesions on NCCT images of AIS patients after EVT.

Automated summary

Non-contrast computed tomography (NCCT) is crucial for patients with acute ischemic stroke who undergo thrombolysis and thrombectomy. This study proposes a novel encoder-decoder network, named ISCT-EDN, for fully automatic segmentation of acute ischemic lesions after endovascular therapy (EVT) on NCCT images. The proposed model outperforms other commonly used segmentation models in the segmentation of post-treatment infarct lesions.