Scholarlink

Objective <p>In dedicated cardiac SPECT systems, attenuation correction using X-ray CT images is often difficult to perform. This study aimed to improve image quality and achieve quantitative image reconstruction for a dedicated cardiac SPECT system equipped with pinhole collimators.</p> Methods <p>A pinhole geometry simulating the detector configuration of the Discovery NM 530c system was employed. Using this geometry, the attenuation coefficient distribution (µ-map) and the projection data of scattered photons were estimated. Attenuation, scatter, and pinhole aperture blurring were corrected using convolutional neural networks (CNNs). For µ-map estimation, reconstructed images from projection data acquired in two low-energy windows were used as CNN inputs, and the µ-map at 140&#xa0;keV was used as the reference data. For scatter estimation, the CNN was trained using projection data including both attenuation and scatter effects, and data including only attenuation effects. The supervised data were projection data of scattered photons that considered attenuation. For aperture correction, projection data containing pinhole aperture blurring were used as inputs, and projection data calculated by a ray-tracing method served as reference data. A total of 269 image datasets were used for µ-map estimation and 5111 projection datasets were used for scatter estimation and aperture correction. All projection data required for these corrections were generated using Monte Carlo photon transport simulations. The reconstructed SPECT images corrected by the proposed CNN-based method without µ-maps were compared with those obtained by a conventional model-based method using known µ-maps, based on the normalized mean squared error (NMSE), structural similarity index (SSIM), and polar maps.</p> Results <p>Simulation results showed that the NMSE(%) of SPECT images decreased from 9.22 ± 3.36 (conventional) to 4.62 ± 1.37 (proposed) and the SSIM of SPECT images increased from 0.816 ± 0.048 (conventional) to 0.910 ± 0.029 (proposed).</p> Conclusions <p>We developed a CNN-based method for attenuation, scatter, and aperture correction without using µ-maps in a dedicated cardiac pinhole SPECT system. The simulation results demonstrated the feasibility and effectiveness of the proposed method for improving quantitative image reconstruction.</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

Accurate CT-free correction of attenuation, scatter, and aperture effects using deep learning in dedicated cardiac pinhole SPECT

  • 期刊论文

  • 发表日期：

    2026年3月9日

查看全文
  • Shuto Inaba,
  • Yudai Nawano,
  • Koichi Ogawa

摘要

Objective

In dedicated cardiac SPECT systems, attenuation correction using X-ray CT images is often difficult to perform. This study aimed to improve image quality and achieve quantitative image reconstruction for a dedicated cardiac SPECT system equipped with pinhole collimators.

Methods

A pinhole geometry simulating the detector configuration of the Discovery NM 530c system was employed. Using this geometry, the attenuation coefficient distribution (µ-map) and the projection data of scattered photons were estimated. Attenuation, scatter, and pinhole aperture blurring were corrected using convolutional neural networks (CNNs). For µ-map estimation, reconstructed images from projection data acquired in two low-energy windows were used as CNN inputs, and the µ-map at 140 keV was used as the reference data. For scatter estimation, the CNN was trained using projection data including both attenuation and scatter effects, and data including only attenuation effects. The supervised data were projection data of scattered photons that considered attenuation. For aperture correction, projection data containing pinhole aperture blurring were used as inputs, and projection data calculated by a ray-tracing method served as reference data. A total of 269 image datasets were used for µ-map estimation and 5111 projection datasets were used for scatter estimation and aperture correction. All projection data required for these corrections were generated using Monte Carlo photon transport simulations. The reconstructed SPECT images corrected by the proposed CNN-based method without µ-maps were compared with those obtained by a conventional model-based method using known µ-maps, based on the normalized mean squared error (NMSE), structural similarity index (SSIM), and polar maps.

Results

Simulation results showed that the NMSE(%) of SPECT images decreased from 9.22 ± 3.36 (conventional) to 4.62 ± 1.37 (proposed) and the SSIM of SPECT images increased from 0.816 ± 0.048 (conventional) to 0.910 ± 0.029 (proposed).

Conclusions

We developed a CNN-based method for attenuation, scatter, and aperture correction without using µ-maps in a dedicated cardiac pinhole SPECT system. The simulation results demonstrated the feasibility and effectiveness of the proposed method for improving quantitative image reconstruction.