Time of onset should make zero difference, the absolute requirement is still 100% recovery. You just have to create protocols for every contingency, at least that is what competent stroke researchers would be doing and competent stroke leadership would be demanding.
Weakly supervised multitask learning models to identify symptom onset time of unclear-onset intracerebral hemorrhage
Chang Jianbo1, Pei Hanqi2, Chen Yihao1, Jiang Cheng2, Shang Hong2, Wang Yuxiang3, Wang Xiaoning2, Ye Zeju4, Wang Xingong5, Tian Fengxuan6, Chai Jianjun7, Xu Jijun8, Li Zhaojian3, Ma Wenbin1, Wei Junji1, Jianhua Yao2, Feng Ming1, and Wang Renzhihttps://orcid.org/0000-0003-2080-54741
Background
Approximately one-third of spontaneous intracerebral hemorrhage patients did not know the onset time and were excluded from studies about time-dependent treatments for hyperacute spontaneous intracerebral hemorrhage.
Aims
To help clinicians explore the benefit of time-dependent treatments for unclear-onset patients, we presented artificial intelligence models to identify onset time using non-contrast computed tomography (NCCT) based on weakly supervised multitask learning (WS-MTL) structure.
Methods
The patients with reliable symptom onset time (strong label) or repeat CT (weak label) were included and split into training set and test set (internal and external). The WS-MTL structure utilized strong and weak labels simultaneously to improve performance. The models included three binary classification models for classifying whether NCCT acquired within 6, 8 or 12 h for different treatments measured by area under curve, and a regression model for determining the exact onset time measured by mean absolute error. The generalizability of models was also explored in comprehensive analysis.
Results
A total of 4004 patients with 10,780 NCCT scans were included. The performance of WS-MTL classification model showed high accuracy, and that of regression model was satisfactory in ≤6 h subgroup. In comprehensive analysis, the WS-MTL showed better performance for larger hematomas and thinner scans. And the performance improved effectively as training amounts increasing and could be improved steadily through retraining.
Conclusions
The WS-MTL models showed good performance and generalizability. Considering the large number of unclear-onset spontaneous intracerebral hemorrhage patients, it may be worth to integrate the WS-MTL model into clinical practice to identify the onset time.
Keywords
Artificial intelligence, intracerebral hemorrhage, unclear onset time, stroke, deep learning; non-contrast CT
1Neurosurgery, Peking Union Medical College Hospital, Peking Union Medical College, Chinese Academy of Medical Sciences, Beijing, China
2Tencent AI Lab, Shen Zhen, China
3Neurosurgery, The Affiliated Hospital of Qingdao University, Qingdao University, China
4Neurosurgery, Dongguan People’s Hospital, Dongguan, China
5Neurosurgery, Linyi People Hospital, Linyi, China
6Neurosurgery, Qinghai Provincial People's Hospital, Qinghai, China
7Neurosurgery, Zhangqiu People’s Hospital, Jinan, China
8Neurosurgery, Tengzhou Central People's Hospital, Zaozhuang, China
Corresponding author(s):
Jianhua Yao, Tencent AI Lab, Shenzhen 518000, China. Email: Jianhua.yao@gmail.com Feng Ming, Peking Union Medical College Hospital, Beijing 100730, China. Email: jackietz@163.com Wang Renzhi, Neurosurgery, Peking Union Medical College Hospital, Peking Union Medical College, Chinese Academy of Medical Sciences, Beijing 100730, China. Email: wangrz@126.com
Introduction
Accounting for approximately 15%–20% of all stroke cases, spontaneous intracerebral hemorrhage (sICH) was the most serious type of stroke and caused a significantly high healthcare economic burden.1 Approximately one-third of patients did not know the symptom onset time and had a higher mortality than patients who knew onset time.2–4
Although the impact of hyperacute therapy for ICH was far behind that for ischemic stroke. In recent years, there were many time-dependent treatments which could be potential breakthroughs in the management of hyperacute ICH.5 For example, the rapid blood pressure reduction within 6 h improved functional outcome in sICH patients6; tranexamic acid treatment within 8 h reduced the risk of hematoma expansion,7 and the early surgery within 8–12 h improved patient's prognosis.8–10 However, unclear-onset sICH patients were excluded from these studies and it leaded to a high risk of selection bias. Developing models to identify the symptom onset time for unclear-onset sICH patients4 could help clinicians investigate the benefit of time-dependent treatments.
Aims
The present study described novel models to determine the symptom onset time for unclear-onset sICH patients based on non-contrast computed tomography (NCCT) by the same algorithm structure. The models included three binary classification models for classifying whether NCCT acquired within 6, 8 or 12 h for different treatments and a regression model for determining the exact onset time.
Methods
Participants
All data were retrospectively obtained from Chinese Intracranial Hemorrhage Image Database (CICHID).11 All medical records and CT scans were exported from hospital information system (HIS) and recoded for anonymizing. The characteristics of patients and parameters of scans were collected by independent research assistants. The retrospective study was approved by Institutional Review Board of Peking Union Medical College Hospital (Ethics code: S-K1175).
We included the NCCT with reliable symptom onset time, and the patients with both initial CT and repeat CT. There were two kinds of labels in our study, strong label and weak label, as detailed in supplement and Figure 1. NCCT scans with reliable symptom onset time were marked with strong label and were the truth label in our study. The temporal ordering of initial and repeat CT pairs was marked with weak label, which was accurately recorded by scanner machine.
Figure 1. Flow diagram of case selection. The strong label was the truth label in our study, defined as the CT scans with reliable symptom onset time, showed as blue boxes in the figure. The weak label was temporal ordering of CT pairs from one patient, showed as purple boxes in the figure. A total of 909 cases in the training set had repeat CT and transferred into weak label as supplement data to auxiliary training set. Of them, 650 cases met the inclusion criteria and were finally included in auxiliary training set, with 498 cases in Cohort 1 and 252 cases in Cohort 2. There was no overlap between training set and test set.
NCCT: non-contrast computed tomography.
Keywords
Artificial intelligence, intracerebral hemorrhage, unclear onset time, stroke, deep learning; non-contrast CT
1Neurosurgery, Peking Union Medical College Hospital, Peking Union Medical College, Chinese Academy of Medical Sciences, Beijing, China
2Tencent AI Lab, Shen Zhen, China
3Neurosurgery, The Affiliated Hospital of Qingdao University, Qingdao University, China
4Neurosurgery, Dongguan People’s Hospital, Dongguan, China
5Neurosurgery, Linyi People Hospital, Linyi, China
6Neurosurgery, Qinghai Provincial People's Hospital, Qinghai, China
7Neurosurgery, Zhangqiu People’s Hospital, Jinan, China
8Neurosurgery, Tengzhou Central People's Hospital, Zaozhuang, China
Corresponding author(s):
Jianhua Yao, Tencent AI Lab, Shenzhen 518000, China. Email: Jianhua.yao@gmail.com Feng Ming, Peking Union Medical College Hospital, Beijing 100730, China. Email: jackietz@163.com Wang Renzhi, Neurosurgery, Peking Union Medical College Hospital, Peking Union Medical College, Chinese Academy of Medical Sciences, Beijing 100730, China. Email: wangrz@126.com
Introduction
Accounting for approximately 15%–20% of all stroke cases, spontaneous intracerebral hemorrhage (sICH) was the most serious type of stroke and caused a significantly high healthcare economic burden.1 Approximately one-third of patients did not know the symptom onset time and had a higher mortality than patients who knew onset time.2–4
Although the impact of hyperacute therapy for ICH was far behind that for ischemic stroke. In recent years, there were many time-dependent treatments which could be potential breakthroughs in the management of hyperacute ICH.5 For example, the rapid blood pressure reduction within 6 h improved functional outcome in sICH patients6; tranexamic acid treatment within 8 h reduced the risk of hematoma expansion,7 and the early surgery within 8–12 h improved patient's prognosis.8–10 However, unclear-onset sICH patients were excluded from these studies and it leaded to a high risk of selection bias. Developing models to identify the symptom onset time for unclear-onset sICH patients4 could help clinicians investigate the benefit of time-dependent treatments.
Aims
The present study described novel models to determine the symptom onset time for unclear-onset sICH patients based on non-contrast computed tomography (NCCT) by the same algorithm structure. The models included three binary classification models for classifying whether NCCT acquired within 6, 8 or 12 h for different treatments and a regression model for determining the exact onset time.
Methods
Participants
All data were retrospectively obtained from Chinese Intracranial Hemorrhage Image Database (CICHID).11 All medical records and CT scans were exported from hospital information system (HIS) and recoded for anonymizing. The characteristics of patients and parameters of scans were collected by independent research assistants. The retrospective study was approved by Institutional Review Board of Peking Union Medical College Hospital (Ethics code: S-K1175).
We included the NCCT with reliable symptom onset time, and the patients with both initial CT and repeat CT. There were two kinds of labels in our study, strong label and weak label, as detailed in supplement and Figure 1. NCCT scans with reliable symptom onset time were marked with strong label and were the truth label in our study. The temporal ordering of initial and repeat CT pairs was marked with weak label, which was accurately recorded by scanner machine.
Figure 1. Flow diagram of case selection. The strong label was the truth label in our study, defined as the CT scans with reliable symptom onset time, showed as blue boxes in the figure. The weak label was temporal ordering of CT pairs from one patient, showed as purple boxes in the figure. A total of 909 cases in the training set had repeat CT and transferred into weak label as supplement data to auxiliary training set. Of them, 650 cases met the inclusion criteria and were finally included in auxiliary training set, with 498 cases in Cohort 1 and 252 cases in Cohort 2. There was no overlap between training set and test set.
NCCT: non-contrast computed tomography.
More at link.
No comments:
Post a Comment