融合脑电与核磁的智能数字孪生系统，实现脑瘤动态监测与预测。
A Scalable AI Driven, IoT Integrated Cognitive Digital Twin for Multi-Modal Neuro-Oncological Prognostics and Tumor Kinetics Prediction using Enhanced Vision Transformer and XAI
- 用增强视觉变压器融合脑电与核磁数据，提升肿瘤定位精度。
- 94.6%准确率、93.2%召回率，Dice分数达0.91，表现优异。
- 支持实时可视化与可解释性分析，适合临床神经病学研究者使用。
神经肿瘤预后在现代神经科学中至关重要，因脑瘤检测与管理极具挑战。本文提出一种认知数字孪生框架，结合可穿戴头帽采集的实时脑电（EEG）信号与结构磁共振（MRI）数据，实现动态个性化肿瘤监测。核心为增强型视觉变压器（ViT++），引入块级注意力正则化（PLAR）与自适应阈值机制，提升肿瘤定位与理解能力。基于双向LSTM的神经分类器分析脑电时序模式，区分癫痫、发作间期与健康状态。通过Grad-CAM热力图与three.js驱动的3D可视化模块，提供交互式解剖洞察。此外，肿瘤动力学引擎基于MRI趋势与脑电异常预测体积增长。在多项指标上表现卓越：94.6%精确率、93.2%召回率，Dice分数达0.91，为实时可解释神经诊断树立新标准，推动智能脑健康监测发展。
原文摘要 · Abstract (English)
Neuro-oncological prognostics are now vital in modern clinical neuroscience because brain tumors pose significant challenges in detection and management. To tackle this issue, we propose a cognitive digital twin framework that combines real-time EEG signals from a wearable skullcap with structural MRI data for dynamic and personalized tumor monitoring. At the heart of this framework is an Enhanced Vision Transformer (ViT++) that includes innovative components like Patch-Level Attention Regularization (PLAR) and an Adaptive Threshold Mechanism to improve tumor localization and understanding. A Bidirectional LSTM-based neural classifier analyzes EEG patterns over time to classify brain states such as seizure, interictal, and healthy. Grad-CAM-based heatmaps and a three.js-powered 3D visualization module provide interactive anatomical insights. Furthermore, a tumor kinetics engine predicts volumetric growth by looking at changes in MRI trends and anomalies from EEG data. With impressive accuracy metrics of 94.6% precision, 93.2% recall, and a Dice score of 0.91, this framework sets a new standard for real-time, interpretable neurodiagnostics. It paves the way for future advancements in intelligent brain health monitoring.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。