Scholarlink

Background <p>Amide Proton Transfer (APT) imaging, a molecular MRI technique under the chemical exchange saturation transfer (CEST) framework, offers a non-invasive means of detecting malignant tumors by assessing biochemical alterations without contrast agents. Integration with machine learning (ML) and deep learning (DL) enhances diagnostic precision and tumor characterization.</p> Objective <p>To systematically review and synthesize current evidence on the combined use of APT imaging with ML/DL for the diagnosis and characterization of malignant tumors.</p> Methods <p>A systematic review was conducted following PRISMA guidelines. Five databases and gray literature sources were searched. Studies were included if they involved malignant tumors, utilized APT imaging combined with ML or DL, and reported diagnostic or predictive outcomes. Quality assessment was performed using QUADAS-2 and Radiomics Quality Score (RQS) 2.0.</p> Results <p>Seven studies met inclusion criteria, primarily focusing on gliomas. ML/DL algorithms employed included radiomics-based classifiers, CNNs, and neural networks. APT imaging combined with ML/DL achieved high diagnostic accuracy (AUC: 0.82–0.97) for predicting molecular markers (IDH, 1p/19q, H3K27M), treatment response, and differentiating progression from pseudoprogression. Risk of bias was high in patient selection across studies, with limited generalizability due to retrospective, single-center designs and small sample sizes.</p> Conclusions <p>APT imaging, when integrated with ML/DL, shows promise as a non-invasive, contrast-free method for diagnosing and characterizing gliomas. The combined approach improves accuracy in molecular profiling and treatment assessment. However, broader validation, prospective studies, and application to other tumor types are necessary for clinical translation.</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

Amide proton transfer imaging with machine learning and deep learning for malignant tumor diagnosis: a systematic review

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年10月17日

查看全文
  • Mohammadreza Elhaie,
  • Abolfazl Koozari,
  • Hossein Koohi,
  • Salman Mohammed Albeshan

摘要

Background

Amide Proton Transfer (APT) imaging, a molecular MRI technique under the chemical exchange saturation transfer (CEST) framework, offers a non-invasive means of detecting malignant tumors by assessing biochemical alterations without contrast agents. Integration with machine learning (ML) and deep learning (DL) enhances diagnostic precision and tumor characterization.

Objective

To systematically review and synthesize current evidence on the combined use of APT imaging with ML/DL for the diagnosis and characterization of malignant tumors.

Methods

A systematic review was conducted following PRISMA guidelines. Five databases and gray literature sources were searched. Studies were included if they involved malignant tumors, utilized APT imaging combined with ML or DL, and reported diagnostic or predictive outcomes. Quality assessment was performed using QUADAS-2 and Radiomics Quality Score (RQS) 2.0.

Results

Seven studies met inclusion criteria, primarily focusing on gliomas. ML/DL algorithms employed included radiomics-based classifiers, CNNs, and neural networks. APT imaging combined with ML/DL achieved high diagnostic accuracy (AUC: 0.82–0.97) for predicting molecular markers (IDH, 1p/19q, H3K27M), treatment response, and differentiating progression from pseudoprogression. Risk of bias was high in patient selection across studies, with limited generalizability due to retrospective, single-center designs and small sample sizes.

Conclusions

APT imaging, when integrated with ML/DL, shows promise as a non-invasive, contrast-free method for diagnosing and characterizing gliomas. The combined approach improves accuracy in molecular profiling and treatment assessment. However, broader validation, prospective studies, and application to other tumor types are necessary for clinical translation.