Scholarlink

Purpose <p>To evaluate current evidence on machine learning (ML) for the diagnosis of Hirschsprung disease (HSCR) and summarize its diagnostic performance and potential clinical utility.</p> Methods <p>PubMed, Web of Science, Cochrane Library, and Scopus were systematically searched (January 2016–November 2025) for studies applying ML to HSCR diagnosis. Study quality was assessed using QUADAS-2. Findings were narratively synthesized, with exploratory meta-analysis performed where feasible.</p> Results <p>Eleven studies were included, with substantial heterogeneity in design, data modalities, and outcomes. Three barium enema–based studies were eligible for meta-analysis, showing pooled sensitivity of 0.857 (95% CI 0.738–0.936), specificity of 0.880 (95% CI 0.790–0.941), and an area under the curve of 0.927. In rectal biopsy–based studies, ML-assisted approaches appeared to reduce interpretation time, while evidence for improved diagnostic performance remains limited and heterogeneous.</p> Conclusion <p>ML may have potential value in supporting HSCR diagnosis, particularly when combined with imaging and clinical data. In histopathology, ML appears more likely to serve as an assistive tool to improve efficiency and potentially enhance diagnostic performance rather than replace expert interpretation. Further prospective multicenter studies are needed before routine clinical implementation.</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

Machine learning in the diagnosis of Hirschsprung disease: a systematic review and meta-analysis

  • 期刊论文

  • 发表日期：

    2026年6月15日

查看全文
  • Haoyang Liu,
  • Ying Zhou,
  • Xin Wang,
  • Chen Wang,
  • Sijia Guo,
  • Miaomiao Sun,
  • Shuai Li,
  • Yong Wang

摘要

Purpose

To evaluate current evidence on machine learning (ML) for the diagnosis of Hirschsprung disease (HSCR) and summarize its diagnostic performance and potential clinical utility.

Methods

PubMed, Web of Science, Cochrane Library, and Scopus were systematically searched (January 2016–November 2025) for studies applying ML to HSCR diagnosis. Study quality was assessed using QUADAS-2. Findings were narratively synthesized, with exploratory meta-analysis performed where feasible.

Results

Eleven studies were included, with substantial heterogeneity in design, data modalities, and outcomes. Three barium enema–based studies were eligible for meta-analysis, showing pooled sensitivity of 0.857 (95% CI 0.738–0.936), specificity of 0.880 (95% CI 0.790–0.941), and an area under the curve of 0.927. In rectal biopsy–based studies, ML-assisted approaches appeared to reduce interpretation time, while evidence for improved diagnostic performance remains limited and heterogeneous.

Conclusion

ML may have potential value in supporting HSCR diagnosis, particularly when combined with imaging and clinical data. In histopathology, ML appears more likely to serve as an assistive tool to improve efficiency and potentially enhance diagnostic performance rather than replace expert interpretation. Further prospective multicenter studies are needed before routine clinical implementation.