Stroke is one of the top ten leading causes of death in Taiwan and an important topic in clinical medicine. The diagnosis and treatment of acute ischemic stroke rely heavily on medical imaging technology, particularly Diffusion Weighted Imaging (DWI) in Magnetic Resonance Imaging (MRI), which provides rich lesion information. However, existing artificial intelligence (AI) analysis tools still face issues of insufficient image analysis depth, affecting the efficiency and accuracy of clinical decision-making.

To address this challenge, the team led by Assistant Professor Yi-Chia Wei at Keelung Chang Gung Memorial Hospital has developed the "SGD-Net Acute Ischemic Stroke Lesion Deep Analysis System," which integrates AI deep learning and neuroimaging to achieve more precise lesion analysis.

Breaking Technological Barriers, Pioneering New Standards in Smart Healthcare

SGD-Net is built on a two-stage architecture. In the first stage, a U-Net deep learning network is used for precise lesion segmentation. The second stage is flexible, with multiple applications that can be expanded based on different clinical needs, including:

• Lesion Classification: Differentiates between various types of stroke lesions to improve diagnostic accuracy.
  
  
• Semantic Analysis: Combines multi-modal MRI and brain mapping to provide analysis of the correlation between functional areas and lesions.
  
  
• Imaging Phenotyping Analysis: Extracts image features from the whole brain and lesions to assist clinical decision-making.
  
  
• Prognostic Prediction: Using the LightGBM machine learning model, it accurately predicts patients' short-term prognosis and large vessel stenosis risks with an accuracy of up to 0.85.
  
  

The innovation of SGD-Net lies in its powerful scalability and multi-level analytical capabilities. The system automatically processes DICOM images and seamlessly integrates with existing hospital PACS and HIS systems, enabling real-time analysis and alert functions. For example, when a large area stroke or embolic stroke is detected, the system can automatically trigger an alarm, improving the clinical response efficiency of medical teams. Additionally, through semantic image analysis, SGD-Net integrates neuroscience knowledge, providing in-depth insights into the interactions between stroke lesions and brain function areas.

Compared to existing commercial AI imaging analysis tools (such as RAPID AI, Viz.ai, Olea Medical), SGD-Net is specifically optimized for MRI, offering a more comprehensive stroke lesion assessment. By integrating brain mapping and imaging phenotyping analysis, the system not only enhances diagnostic accuracy but also opens new perspectives for clinical decision-making and stroke research.

Future Outlook: Promoting a New Era of AI-Driven Smart Healthcare

Regarding the future development of SGD-Net, Assistant Professor Yi-Chia Wei stated that SGD-Net is not just a technological breakthrough but a significant milestone in smart healthcare. The goal is for AI to not only assist in medical diagnosis but also help physicians conduct deeper lesion analysis and decision-making. Currently, SGD-Net has completed the core technology development and is in the stage of user interface construction and software optimization. It plans to undergo multi-site testing to ensure clinical applicability. In the future, the system will further integrate clinical education and research, providing strong support for medical education, neuroscience research, and clinical applications.

Resource: AI深度解析病灶 精準診斷急性缺血性腦中風