Scholarlink

<p>Timely and accurate detection of brain tumors is critical for clinical decision-making, yet many deep learning-based diagnostic tools remain impractical for real-time deployment in healthcare settings due to high computational demands. This study introduces NeuroYOLO, an ultra-lightweight, high-accuracy tumor detection framework that integrates the MobileNetV3 feature extractor with YOLOv10’s streamlined detection head. The system was trained and validated on two publicly available MRI datasets comprising 1,008 binary-labeled and 3,064 multi-class-labeled brain images. NeuroYOLO achieved mAP@50 of 97.2% (binary) and 96.0% (multi-class), with sub-10&#xa0;ms inference latency, making it suitable for edge deployment in radiology equipment. Precision reached 92.6% and 100%, respectively, with robust generalization across glioma, meningioma, and pituitary tumors. Comparative benchmarking against YOLOv7, YOLOv8, and baseline YOLOv10 confirms a superior speed–accuracy tradeoff. Ablation studies demonstrate MobileNetV3’s essential role in reducing model complexity while preserving detection fidelity. The system is designed for seamless integration into PACS environments and portable diagnostic devices, advancing the development of AI-augmented imaging tools for neuro-oncology. Future extensions include 3D volumetric inference, support for underrepresented tumor subtypes, and clinical trials in multicenter cohorts.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

NeuroYOLO: A Lightweight YOLOv10-MobileNetV3 Framework for Real-Time Brain Tumor Detection in MRI Scans

  • 期刊论文

  • 发表日期：

    2025年9月9日

查看全文
  • Md. Mehedi Hasan,
  • Bikash Kumar Paul,
  • Mohammad Shorif Uddin,
  • Mehedi Hasan Talukder,
  • Rafid Mostafiz

摘要

Timely and accurate detection of brain tumors is critical for clinical decision-making, yet many deep learning-based diagnostic tools remain impractical for real-time deployment in healthcare settings due to high computational demands. This study introduces NeuroYOLO, an ultra-lightweight, high-accuracy tumor detection framework that integrates the MobileNetV3 feature extractor with YOLOv10’s streamlined detection head. The system was trained and validated on two publicly available MRI datasets comprising 1,008 binary-labeled and 3,064 multi-class-labeled brain images. NeuroYOLO achieved mAP@50 of 97.2% (binary) and 96.0% (multi-class), with sub-10 ms inference latency, making it suitable for edge deployment in radiology equipment. Precision reached 92.6% and 100%, respectively, with robust generalization across glioma, meningioma, and pituitary tumors. Comparative benchmarking against YOLOv7, YOLOv8, and baseline YOLOv10 confirms a superior speed–accuracy tradeoff. Ablation studies demonstrate MobileNetV3’s essential role in reducing model complexity while preserving detection fidelity. The system is designed for seamless integration into PACS environments and portable diagnostic devices, advancing the development of AI-augmented imaging tools for neuro-oncology. Future extensions include 3D volumetric inference, support for underrepresented tumor subtypes, and clinical trials in multicenter cohorts.