Scholarlink

Purpose <p>Multimodal data are crucial for the detection of different stages of Alzheimer’s disease (AD) compared to relying on a single modality. Combining biomedical modalities, such as magnetic resonance imaging (MRI) and positron emission tomography (PET), provide distinct but complementary diagnostic insights, each of which captures unique data that enhance the general understanding of patient health. However, not every patient requires both MRI and PET scans. The use of radioactive tracers, the high costs and the limited availability of PET imaging often discourage patients from undergoing PET scans. To address the concerns of health, time, cost, and effort, a cross-modality synthesis approach is an efficient solution.</p> Methods <p>In this work, we propose a novel modified pix2pix network to translate MRI images into PET-like images. In this study, we propose specific modifications to the pix2pix architecture by replacing the traditional U-Net with a Dense-UNet to enhance its effectiveness in generating PET-like images from MRI scans. In addition, we present a convolutional neural network (CNN) for diagnosing AD within a binary classification framework, distinguishing between AD and cognitive normal.</p> Results <p>Our model outperforms other models by generating high-quality synthesized images, achieving significantly higher SSIM and PSNR, with a substantial reduction in MSE and FID. Using a combination of MRI, PET, and our synthetic PET images, we have achieved enhanced accuracy in the diagnosis of binary AD, outperforming current approaches.</p> Conclusion <p>The classification performance demonstrates that the synthetic PET images produced by the modified pix2pix model are realistic and of high quality, offering additional information that improves the precision of AD diagnosis.</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

MRI to PET image synthesis using modified pix2pix for Alzheimer’s disease diagnosis

  • 期刊论文

  • 发表日期：

    2025年2月12日

查看全文
  • M. S. Meharban,
  • M. K. Sabu,
  • T. Santhanakrishnan

摘要

Purpose

Multimodal data are crucial for the detection of different stages of Alzheimer’s disease (AD) compared to relying on a single modality. Combining biomedical modalities, such as magnetic resonance imaging (MRI) and positron emission tomography (PET), provide distinct but complementary diagnostic insights, each of which captures unique data that enhance the general understanding of patient health. However, not every patient requires both MRI and PET scans. The use of radioactive tracers, the high costs and the limited availability of PET imaging often discourage patients from undergoing PET scans. To address the concerns of health, time, cost, and effort, a cross-modality synthesis approach is an efficient solution.

Methods

In this work, we propose a novel modified pix2pix network to translate MRI images into PET-like images. In this study, we propose specific modifications to the pix2pix architecture by replacing the traditional U-Net with a Dense-UNet to enhance its effectiveness in generating PET-like images from MRI scans. In addition, we present a convolutional neural network (CNN) for diagnosing AD within a binary classification framework, distinguishing between AD and cognitive normal.

Results

Our model outperforms other models by generating high-quality synthesized images, achieving significantly higher SSIM and PSNR, with a substantial reduction in MSE and FID. Using a combination of MRI, PET, and our synthetic PET images, we have achieved enhanced accuracy in the diagnosis of binary AD, outperforming current approaches.

Conclusion

The classification performance demonstrates that the synthetic PET images produced by the modified pix2pix model are realistic and of high quality, offering additional information that improves the precision of AD diagnosis.