Background: For patients with acute ischemic stroke, haemorrhagic transformation (HT) is a feared complication following reperfusion therapies. Automatic segmentation of haemorrhage provides an objective assessment of outcome and improves audit quality. This study aims to automatically segment haemorrhagic regions on follow-up non-contrast head CT (NCCT) for stroke patients treated with endovascular thrombectomy (EVT).
Methods: We propose a semi-automated approach with adaptive thresholding methods, eliminating the requirement for large training data and reducing computational burden. We compare the proposed method with a state-of-the-art segmentation model SAMIHS to evaluate the effectiveness using Dice Similarity Coefficient (DSC).
Results: A total of 51 patients from 10 stroke centres were included, with 28 Type 2 haemorrhagic infarction (HI2) cases and 23 parenchymal hematoma (PH) cases. Our method achieved DSC scores of 0.66 ± 0.17, 0.73 ± 0.14, and 0.61 ± 0.18 for total, PH, and HI2 cases respectively, which is comparable to SAMIHS (DSC scores of 0.66 ± 0.02, 0.69 ± 0.03 and 0.64 ± 0.07 for total, PH, and HI2 cases respectively). In addition, the algorithm demonstrated excellent processing time, with an average of 2.7 seconds for each patient case.
Conclusions: To our knowledge, this is the first study to perform automated segmentation of post-treatment haemorrhage for acute stroke patients and evaluate the performance based on the radiological severity of HT. Our approach offers a balance between accuracy and practicality, presenting a promising tool for aiding prognosis prediction in stroke patients with HT after EVT, without the computational overhead of novel methods.