The deep learning method enhances the segmentation of the boat and the plaque in the diagnosis of stroke

Stroke is the second leading cause of death worldwide. The ischemic stroke, which is strongly linked to atherosclerotic plates, requires accurate division and quantification and quantification of the plaque and boat for a definitive diagnosis. However, conventional manual segmentation remains time consuming and dependent on the operator, while current computer -supported tools are lagging behind to achieve the accuracy required for clinical applications. These technological congestion points seriously prevent the precise diagnosis and treatment of ischemic stroke.

In a study published in European radiologyA research team led by Dr. Zhang NA from the Institutes of Advanced Technologies Shenzhen (SIAT) of the Chinese Academy of Sciences, along with its partners, has developed a fully apprentice of multiple -work segmenting modeling and a two -way structure by the MR -based methods. This approach allows automated and accurate division and quantitative analysis of carotid arterial vessels, walls of the vessels and plates, offering a reliable diagnostic tool for clinical risk assessment of the ischemic stroke.

In this study, the proposed method consists of two main steps. The first step involves the construction of a purely learning ordinary neuronal network (CNN), called the-sgnet container, to separate the wall of the lumen and vessel. The second step utilizes the walls of the boat-especially manual premiums and automatic backed-based permits based on the tversky-loss-loss-loss, using the morphological resemblance between the wall of the vessel and the atherosclerotic plaque.

This study included data from 193 patients with atherosclerotic plaque in five centers, all of them underwent magnetic resonance scan (MRI) (MRI). The data set was divided into three subsets: 107 patients for training and validation, 39 for internal tests and 47 for external tests.

Experimental results have shown that most dice similarity (DSC) rates for wall and wall fragmentation exceeded 90%. The incorporation of vessel wall priors improved the DSC to segregate the plaque by more than 10%, reaching 88.45%. In addition, compared to dice-based priors, the Tversky-Loss-based priors further reinforced the DSC by almost 3%, reaching 82.84%.

In contrast to manual methods, the proposed technique provides accurate, automated plaque segmentation and completes the quantitative characteristic plaque evaluation for a single patient in less than 3 seconds.

The purpose of our research is to use AI models to produce expensive, reproducible and clinically related quantitative results, which can help health professionals in the diagnosis of stroke and therapeutic decision -making. “

Dr. Zhang Na, Shenzhen Institutes of Advanced Technology

Dr. Zhang added: “In the future, we should conduct additional studies using other equipment, populations and anatomical analyzes to further validate the credibility of research results.”