Scholarlink

Background <p>To investigate the diagnostic value of PET/MR fusion images for the noninvasive prediction of IDH status using various fusion strategies. </p> Methods <p>This retrospective study involved 45 patients with histopathologically confirmed gliomas who underwent preoperative integrated 18F-FET PET/MR from May 2023 to August 2024. Image fusion was performed by combining PET and the IVIM-derived α map using three commonly used fusion algorithms: discrete wavelet transform (DWT), Laplacian pyramid (LP), and ratio-of-low-pass pyramid (RP). PET parameters and first-order histogram features were extracted from 18F-FET PET, IVIM α map, and fused images. A support vector machine (SVM) classifier with a radial basis function (RBF) kernel was used for classification. Model performance was evaluated through stratified cross-validation, with the area under the receiver operating characteristic curve (AUC) as the main metric.</p> Results <p>All three fusion-based models outperformed single-modality models and the combined model baseline. Specifically, the LP-based fusion model showed an improved AUC of 0.790 (95% CI: 0.783–0.798); the RP-based fusion model achieved the highest AUC of 0.820 (95% CI: 0.813–0.828); and the DWT-based fusion model reached an AUC of 0.814 (95% CI: 0.806–0.821), with the highest sensitivity at 0.747. Additionally, the parameter analyses demonstrated that the fusion performance was highly sensitive to the weighting strategy across frequency bands. Optimized fusion weights are critical to realize the full benefit of PET/MR image fusion.</p> Conclusion <p>Integrated ¹⁸F-FET PET/MR image fusion offers a promising alternative to traditional multimodal analysis. Its application in individualized treatment and precision medicine is highly promising.</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

Preoperative assessment of IDH status in adult glioma by integrated 18F-FET PET/MR fusion: a comparative analysis of fusion algorithms

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月20日

查看全文
  • Kun Zheng,
  • Qingqing Han,
  • Tuo Li,
  • Zihang Liu,
  • Enhui Li,
  • Jiahui Liu,
  • Wenbin Ma,
  • Hui You,
  • Yu Wang,
  • Li Huo

摘要

Background

To investigate the diagnostic value of PET/MR fusion images for the noninvasive prediction of IDH status using various fusion strategies.

Methods

This retrospective study involved 45 patients with histopathologically confirmed gliomas who underwent preoperative integrated 18F-FET PET/MR from May 2023 to August 2024. Image fusion was performed by combining PET and the IVIM-derived α map using three commonly used fusion algorithms: discrete wavelet transform (DWT), Laplacian pyramid (LP), and ratio-of-low-pass pyramid (RP). PET parameters and first-order histogram features were extracted from 18F-FET PET, IVIM α map, and fused images. A support vector machine (SVM) classifier with a radial basis function (RBF) kernel was used for classification. Model performance was evaluated through stratified cross-validation, with the area under the receiver operating characteristic curve (AUC) as the main metric.

Results

All three fusion-based models outperformed single-modality models and the combined model baseline. Specifically, the LP-based fusion model showed an improved AUC of 0.790 (95% CI: 0.783–0.798); the RP-based fusion model achieved the highest AUC of 0.820 (95% CI: 0.813–0.828); and the DWT-based fusion model reached an AUC of 0.814 (95% CI: 0.806–0.821), with the highest sensitivity at 0.747. Additionally, the parameter analyses demonstrated that the fusion performance was highly sensitive to the weighting strategy across frequency bands. Optimized fusion weights are critical to realize the full benefit of PET/MR image fusion.

Conclusion

Integrated ¹⁸F-FET PET/MR image fusion offers a promising alternative to traditional multimodal analysis. Its application in individualized treatment and precision medicine is highly promising.