.png){kind=icon}
返回
Multilevel Multimodal Framework for Automatic Collateral Scoring in Brain Stroke
.png){kind=icon}
OA
AI
DOI：10.1109/ACCESS.2024.3368504.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
In patients with ischemic brain stroke, collateral circulation plays a crucial role in selecting patients suitable for endovascular therapy. The presence of well-developed collaterals improves the patient's chances of recovery. In clinical practice, the presence of collaterals is diagnosed on a Computed Tomography Angiography scan. The radiologist grades it on the basis of subjective visual assessment, which is prone to interobserver and intraobserver variability. Computer-based methods of collateral assessment face the challenge of non-uniform scan volume, leading to manual selection of slices, meaning that the most imperative slices have to be manually selected by the radiologist. This paper proposes a multilevel multimodal hierarchical framework for automated collateral scoring. Specifically, we propose deploying a Convolutional Neural Network for image selection based on the visibility of collaterals and a multimodal model for comparing the occluded and contralateral sides of the brain for collateral scoring. We also generate a patient-level prediction by integrating automated machine learning in the proposed framework. While the proposed multimodal predictor contributes to Artificial Intelligence, the proposed end-to-end framework is an application in engineering. The proposed framework has been trained and tested on 116 patients, with five-fold cross-validation, achieving an accuracy of 91.17% for multi-class collateral scores and 94.118% for binary class collateral scores. The proposed multimodal predictor achieved a weighted F1 score of 0.86 and 0.95 on multi-class and binary-class collateral scores, respectively. The proposed framework is fast, efficient, and scalable for real-world deployments. Automated evaluation of collaterals with attention maps for explainability would complement radiologists' efforts. Code for the proposed framework is available
Keyword:
Biomedical imaging
Stroke (medical condition)
Deep learning
Feature extraction
Task analysis
Numerical models
Manuals
Medical expert systems
Circulatory system
Patient monitoring
Surgery
Convolutional neural networks
Medical treatment
biomedical image processing
medical expert systems
stroke (medical condition)
期刊
IF:
3.6
论文数:
9.8W
被引数:
29.4W
机构
引用论文
The Reform and Innovation of English Course: A Coherent Whole of MOOC, Flipped Classroom and ESP英语课程改革与创新：MOOC、翻转课堂与ESP的有机融合
OAAI
OAAI分享收藏分享收藏Systematic comparison of perfusion-CT and CT-angiography in acute stroke patients.png)
ANNALS OF NEUROLOGY
ANNALS OF NEUROLOGYIF7.7
PREAI
分享收藏分享收藏
查看更多内容