Smielewski P, Beqiri E, Mataczynski C, Placek M, Kazimierska A, Hutchinson P J, Czosnyka M, Kasprowicz M
Brain Physics Laboratory, Division of Neurosurgery, Department of Clinical Neurosciences, University of Cambridge, Cambridge, UK.
Department of Computer Engineering, Faculty of Information and Communication Technology, Wroclaw University of Science and Technology, Wroclaw, Poland.
Brain Spine. 2024 May 19;4:102835. doi: 10.1016/j.bas.2024.102835. eCollection 2024.
Adoption of the ICM+® brain monitoring software by clinical research centres worldwide has been continuously growing over the past 20 years. This has necessitated ongoing updates to accommodate evolving neuromonitoring research needs, including recent explosion of artificial intelligence (AI).
We sought to provide an update on the current features of the software. In particular, we aimed to highlight the new options of integrating AI models.
We reviewed all currently available ICM+ analytical areas and discussed potential AI based extensions in each. We tested a proof-of-concept integration of an AI model and evaluated its performance for real-time data processing.
ICM+ current analytical tools serve both real-time (bed-side) and offline (file based) analysis, including the calculation engine, Signal Calculator, Custom Statistics, Batch tools, ScriptLab and charting. The ICM+ Python plugin engine allows to execute custom Python scripts and take advantage of complex AI frameworks. For the proof-of-concept, we used a neural network convolutional model with 207,000 trainable parameters that classifies morphology of intracranial pressure (ICP) pulse waveform into 5 pulse categories (normal to pathological plus artefactual). When evaluated within ICM+ plugin script on a Windows 10 laptop the classification of a 5 min ICP waveform segment took only 0.19s with a 2.3s of initial, one-off, model loading time required.
Modernised ICM+ analytical tools, reviewed in this manuscript, include integration of custom AI models allowing them to be shared and run in real-time, facilitating rapid prototyping and validating of new AI ideas at the bed-side.
在过去20年里，全球临床研究中心对ICM+®脑监测软件的采用率一直在持续增长。这就需要不断更新以适应不断发展的神经监测研究需求，包括近期人工智能（AI）的迅猛发展。
我们试图提供该软件当前功能的最新情况。特别是，我们旨在突出整合AI模型的新选项。
我们回顾了所有当前可用的ICM+分析领域，并讨论了每个领域基于AI的潜在扩展。我们测试了一个AI模型的概念验证集成，并评估了其对实时数据处理的性能。
ICM+当前的分析工具可用于实时（床边）和离线（基于文件）分析，包括计算引擎、信号计算器、自定义统计、批处理工具、ScriptLab和图表绘制。ICM+ Python插件引擎允许执行自定义Python脚本并利用复杂的AI框架。对于概念验证，我们使用了一个具有207,000个可训练参数的神经网络卷积模型，该模型将颅内压（ICP）脉冲波形的形态分为5种脉冲类别（从正常到病理性加伪迹）。在一台Windows 10笔记本电脑上的ICM+插件脚本中进行评估时，对一段5分钟的ICP波形段进行分类仅需0.19秒，初始一次性模型加载时间为2.3秒。
本手稿中回顾的现代化ICM+分析工具包括自定义AI模型的集成，使其能够实时共享和运行，便于在床边快速进行新AI想法的原型设计和验证。
