Deep Learning for 3D CT Organ Segmentation: A Systematic Review

A comprehensive systematic literature review of deep learning methods for 3D organ segmentation from computed tomography (CT) scans, following PRISMA 2020 guidelines.

---

Paper

• Manuscript: https://github.com/gameguild-gg/docdo-paper
• Format: IEEE Conference
• Studies Reviewed: 52 peer-reviewed papers (2015–2025)

Key Findings

Metric	Value
Papers reviewed	52
Architectures compared	CNN, Transformer, Hybrid
Datasets analyzed	BTCV, AMOS, KiTS, MSD, TotalSegmentator
Best overall method	nnU-Net (self-configuring)
Highest Dice (liver)	97.2%
Highest Dice (pancreas)	88.1%

Building the Paper

The LaTeX manuscript lives in https://github.com/game-guild/docdo-paper. This repo holds the data, scripts, and governance.

---

Repository Structure

docdo-paper-research/
├── core/                       # Governance, glossary, metadata
│   ├── governance/
│   │   └── research-log/       #   Development log, task board, chat log
│   ├── conceptual-architecture/
│   └── metadata/
│
├── artifacts/                  # Data: search results, PDFs, annotations
│   ├── data/
│   │   └── evidence/
│   │       ├── raw/            #   Original search exports
│   │       ├── interim/        #   Deduplicated records
│   │       ├── processed/      #   Screening batches & results
│   │       ├── pdfs/           #   Downloaded papers
│   │       └── supplementary/  #   PRISMA supplementary materials (S1–S12)
│   └── reports/
│
├── work/                       # Authored outputs
│   └── projects/papers/        #   (LaTeX lives in separate docdo-paper repo)
│
├── operations/                 # Python library + tests
│   ├── src/docdo/              #   9-module pip-installable library
│   └── tests/                  #   150 pytest tests
│
├── scholar/                    # Bibliographic infrastructure
│   └── bib/                    #   references.bib, additional-references.bib, reviewed-papers.bib
│
└── documentation/              # Documentation & reports
    └── reports/                #   Peer review, critical analysis, QA reports

---

Where to Put Things

What	Where
Search result CSVs	artifacts/data/evidence/
Downloaded PDFs	artifacts/data/evidence/pdfs/
PRISMA supplementary data	artifacts/data/evidence/supplementary/
BibTeX files	scholar/bib/
Python code	operations/src/docdo/
Tests	operations/tests/
Reports	documentation/reports/

---

Methodology

This review follows PRISMA 2020 guidelines with:

• Systematic search across PubMed, IEEE Xplore, arXiv
• AI-assisted screening (GPT-4o cascade with human validation)
• Quality assessment using modified QUADAS-2
• Complete audit trail in supplementary materials

Citation

@article{docdo2025segmentation,
  title={Deep Learning for 3D CT Organ Segmentation: A Systematic Review},
  author={[Authors]},
  journal={[Journal]},
  year={2025}
}

---

Key Documents

Document	Purpose
documentation/repository-structure.md	Full structure tree + architecture rationale
documentation/reports/	Peer review, critical analysis, QA reports
core/governance/research-log/	Development history, task board
artifacts/data/evidence/supplementary/	PRISMA supplementary materials

---

License

See LICENSE.md.