Deep learning-based quantification of total bleeding volume and its association with complications, disability, and death in patients with aneurysmal subarachnoid hemorrhage

Ping Hu,Yanze Wu,Tengfeng Yan,Lei Shu,Feng Liu,Bing Xiao,Minhua Ye,Miaojing Wu,Shigang Lv,Xingen Zhu
DOI: https://doi.org/10.3171/2024.1.JNS232280
2024-03-29
Abstract:Objective: The relationships between immediate bleeding severity, postoperative complications, and long-term functional outcomes in patients with aneurysmal subarachnoid hemorrhage (aSAH) remain uncertain. Here, the authors apply their recently developed automated deep learning technique to quantify total bleeding volume (TBV) in patients with aSAH and investigate associations between quantitative TBV and secondary complications, adverse long-term functional outcomes, and death. Methods: Electronic health record data were extracted for adult patients admitted to a single institution within 72 hours of aSAH onset between 2018 and 2021. An automatic deep learning model was used to fully segment and quantify TBV on admission noncontrast head CT images. Patients were subgrouped by TBV quartile, and multivariable logistic regression, restricted cubic splines, and subgroup analysis were used to explore the relationships between TBV and each clinical outcome. Results: A total of 819 patients were included in the study. Sixty-six (8.1%) patients developed hydrocephalus, while 43 (5.3%) experienced rebleeding, 141 (17.2%) had delayed cerebral ischemia, 88 (10.7%) died in the 12 months after discharge, and 208 (25.7%) had a modified Rankin Scale score ≥ 3 12 months after discharge. On multivariable analysis, patients in the highest TBV quartile (> 37.94 ml) had an increased risk of hydrocephalus (adjusted OR [aOR] 4.38, 95% CI 1.61-11.87; p = 0.004), rebleeding (aOR 3.26, 95% CI 1.03-10.33; p = 0.045), death (aOR 6.92, 95% CI 1.89-25.37; p = 0.004), and 12-month disability (aOR 3.30, 95% CI 1.62-6.72; p = 0.001) compared with the lowest TBV quantile (< 8.34 ml). The risks of hydrocephalus (nonlinear, p = 0.025), rebleeding, death, and disability (linear, p > 0.05) were positively associated with TBV by restricted cubic splines. In subgroup analysis, TBV had a stronger effect on 12-month outcome in female than male patients (p for interaction = 0.0499) and on rebleeding prevalence in patients with endovascular coiling than those with surgical clipping (p for interaction = 0.008). Conclusions: Elevated TBV is associated with a greater risk of hydrocephalus, rebleeding, death, and poor prognosis.
What problem does this paper attempt to address?