Segmentation for pelvic malignancies in radiation oncology practice: a systematic review and meta-analysis protocol

Syst Rev. 2026 Apr 23;15(1):187. doi: 10.1186/s13643-026-03173-2.

Abstract

Background: Deep learning (DL)-based artificial intelligence (AI) models, the fourth generation in autosegmentation, have been adopted both for commercial and research applications worldwide and have shown great promise as reliable and comprehensive resource strategies for radiotherapy workflows.

Methods: A comprehensive search will be conducted on Medline (PubMed), Scopus, the Cochrane Library, EMBASE, and Web of Science between January 2004 and December 2025. We will conduct a title, abstract, and full-text screening of all studies as per the eligibility criteria. Two reviewers will be involved in screening studies, quality appraisal, and data extraction, and a third reviewer will be consulted to resolve conflicts. Based on data availability, the data will be synthesised via meta-analysis and narrative synthesis. The reporting will be performed using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, ensuring the reliability and validity of the results.

Discussion: The model's performance will be assessed using various quantitative indices, qualitative tools, timesavings, and dosimetry. This review will enable us to determine the accuracy of autosegmentation for targets and various pelvic organs, boosting clinicians' confidence in facilitating the clinical implementation of such tools in routine clinical practice.

Systematic review registration: The protocol is prospectively registered on PROSPERO. The registration ID is CRD42024491066.

Keywords: Autosegmentation; Cancer; Convolutional neural network (CNN); Deep learning; Dice coefficient score; Pelvic malignancy.

MeSH terms

  • Deep Learning*
  • Humans
  • Meta-Analysis as Topic
  • Pelvic Neoplasms* / diagnostic imaging
  • Pelvic Neoplasms* / radiotherapy
  • Radiation Oncology* / methods
  • Systematic Reviews as Topic