Scholarlink

<p>The development of high-quality medical image compression techniques and efficient diagnostic tools has increased due to the growing volume of medical imaging data and the fulfilment of essential needs in storage and real-time processing. This research introduces a deep learning framework using compressed magnetic resonance images to accurately classify brain pictures into no tumour, glioma, meningioma, and pituitary tumour. An analysis and evaluation of the performance of Convolutional Neural Networks and Visual Geometric Group-16 (VGG16) deep learning models are conducted using a compression method using the discrete cosine transform (DCT) and the discrete wavelet transform (DWT). The experimental results indicate that the DWT-based compression surpasses traditional DCT methods, as evidenced by various metrics such as mean squared error, root mean squared error, structural similarity index, mean average error, peak signal-to-noise ratio, entropy, compression time, and compression factor. We utilize a convolutional neural network and VGG16 architecture trained on compressed and uncompressed brain MRI images to accomplish the classification objective. The VGG16 model achieved a validation accuracy of 97.10% with uncompressed images, whereas compressed photos led to a slightly higher accuracy of 98.23%. The receiver operating characteristic curve (AUROC) exhibited an area under the curve approaching 1.0, verifying the model's discriminatory capability. For all four classes of tumours, the model always gives a reliable high accuracy- an average value of 98.23%. The best result was achieved in the ‘No Tumor’ detection with a perfect F1 score, which equals 1.00, for Glioma, Meningioma, and Pituitary; the model also shows good precision and recall in the sense of having good F1 scores on the level of 0.98 and 0.97, respectively. The suggested frameworks are taught, developed, and implemented on NVIDIA's 128-core Jetson Nano single-board computer, which allows the system to be used in portable and real-time applications in healthcare environments. This work combines compression, classification, and deployment into a single pipeline, representing a novel AI system for efficiency and practicality in clinical medical imaging.</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

An efficient deep learning framework for detecting and classifying brain tumour from DWT compressed MRI images

  • 期刊论文

  • 发表日期：

    2025年8月23日

查看全文
  • K. A. Neena,
  • M. N. Anil Kumar

摘要

The development of high-quality medical image compression techniques and efficient diagnostic tools has increased due to the growing volume of medical imaging data and the fulfilment of essential needs in storage and real-time processing. This research introduces a deep learning framework using compressed magnetic resonance images to accurately classify brain pictures into no tumour, glioma, meningioma, and pituitary tumour. An analysis and evaluation of the performance of Convolutional Neural Networks and Visual Geometric Group-16 (VGG16) deep learning models are conducted using a compression method using the discrete cosine transform (DCT) and the discrete wavelet transform (DWT). The experimental results indicate that the DWT-based compression surpasses traditional DCT methods, as evidenced by various metrics such as mean squared error, root mean squared error, structural similarity index, mean average error, peak signal-to-noise ratio, entropy, compression time, and compression factor. We utilize a convolutional neural network and VGG16 architecture trained on compressed and uncompressed brain MRI images to accomplish the classification objective. The VGG16 model achieved a validation accuracy of 97.10% with uncompressed images, whereas compressed photos led to a slightly higher accuracy of 98.23%. The receiver operating characteristic curve (AUROC) exhibited an area under the curve approaching 1.0, verifying the model's discriminatory capability. For all four classes of tumours, the model always gives a reliable high accuracy- an average value of 98.23%. The best result was achieved in the ‘No Tumor’ detection with a perfect F1 score, which equals 1.00, for Glioma, Meningioma, and Pituitary; the model also shows good precision and recall in the sense of having good F1 scores on the level of 0.98 and 0.97, respectively. The suggested frameworks are taught, developed, and implemented on NVIDIA's 128-core Jetson Nano single-board computer, which allows the system to be used in portable and real-time applications in healthcare environments. This work combines compression, classification, and deployment into a single pipeline, representing a novel AI system for efficiency and practicality in clinical medical imaging.