MedImgAnnotator: batch wise medical image annotation and preprocessing tool

Ewunate Assaye Kassaw, Amit Mehndiratta, Peush Sahni, Anup SingharXiv 2026 · arXiv

TL;DR

MedImgAnnotator is a new semi-automated tool designed to solve the "bottleneck" of medical image labeling by supporting high-throughput batch processing and intelligent image-mask matching. It reduces active annotation time by nearly 70%—dropping from over 200 seconds to just 59 seconds per case—while integrating essential preprocessing and AI-assisted plugins into a single, lightweight workflow.

What Problem Does This Paper Address?

The development of robust deep learning models for medical imaging requires massive, high-quality datasets. However, the "ground truth" curation process is notoriously slow. Existing open-source tools like 3D Slicer, ITK-SNAP, and ImageJ were primarily designed for clinical visualization or single-volume exploration.

When researchers need to annotate thousands of volumes, these tools become "tedious" because they require manual file loading, lack automated case queuing, and often struggle to correctly pair raw images with AI-predicted masks. The authors highlight a pilot study where annotating just 50 liver CT volumes took 24 hours due to these navigational hurdles. MedImgAnnotator was built to bridge this gap between clinical viewing and high-throughput research needs.

Methods

MedImgAnnotator is a Python-based standalone application that prioritizes a "batch-first" architecture. Key technical features include:

  • Intelligent Data Management: It uses the Ratcliff-Obershelp pattern-matching algorithm to automatically pair masks with images, even when file naming conventions are inconsistent.
  • Hybrid Slice Interpolation: To reduce manual labor, users can annotate a few "key slices." The tool then uses a combination of Fast Symmetric Forces Demons (deformable registration) and Signed Distance Transforms (SDT) to mathematically guess and fill in the intermediate masks, accounting for anatomical shifts and deformations.
  • ImPrep Module: An integrated suite for DICOM-to-NIfTI conversion, resampling, intensity normalization (like CT Hounsfield Unit windowing), and spatial harmonization.
  • Plugin System: A modular framework that allows users to drop in external AI models (like nnU-Net) for "human-in-the-loop" workflows, where the AI predicts a mask and the human quickly refines it.
Figure 1: MedImgAnnotation tool development procedure

Datasets

The tool was validated across several diverse medical imaging datasets:

  • MICCAI 2019 Chronic Heart Disease: Chest CT scans.
  • AMOS 2022: Abdominal Multi-Organ Segmentation (MR and CT).
  • Pancreas MR Dataset: For testing organ-specific annotation.
  • MICCAI 2026 CARE Challenge: The tool was used to refine liver masks for a massive cohort of 6,412 MRI volumes, helping the user secure first place in the challenge.

Key Insights and Ideas

  • The "Queue" vs. "File" Paradigm: By treating a dataset as a queue rather than a collection of individual files, the tool eliminates the "open/save/close" overhead that consumes a significant portion of a researcher's day.
  • Fuzzy Matching is a UX Necessity: In research environments, file names are often messy. Using string-matching algorithms to pair masks and images is a small but critical improvement that prevents manual alignment errors.
  • Deformable Interpolation Matters: Simple linear interpolation often fails in medical imaging because organs curve and shift between slices. Using deformable registration for interpolation allows for much smoother, "biologically organic" curves with less manual input.
  • Tool-Hopping is a Productivity Killer: By embedding preprocessing (ImPrep) and AI inference (Plugins) directly into the annotator, the authors prevent the need to switch between three or four different software packages.

Key Results

  • Speedup: MedImgAnnotator reduced the active annotation time for liver segmentation to 59 seconds, compared to an average of 206.5 seconds for other openly available tools (a 2.55x to 3.95x improvement).
  • User Preference: In a trial with 12 independent researchers, the tool received a 4.5/5 ease-of-use rating, significantly higher than the 3.67 rating of their routinely used tools.
  • Statistical Significance: Wilcoxon signed-rank tests confirmed that the reduction in time (p=0.032) and the increase in ease-of-use (p=0.008) were statistically significant.
  • Versatility: The tool maintained high efficiency across different organs, with mean active times ranging from 25.1s (pancreas) to 315.2s (complex brain tumors).

Limitations and Future Work

While highly efficient, the authors acknowledge a few limitations:

  • OS Constraints: The executable is currently restricted to Windows.
  • Visualization: It lacks a 3D volume rendering view and the ability to show multiparametric/multiphasic scans side-by-side.
  • Evaluation Scale: The tool has only been tested by 12 users so far; the authors suggest that a larger, multi-center cohort study would better establish its generalizability.

Future work may involve expanding the tool to support more complex multi-modal visualization and cross-platform compatibility.

Summary updated

Abstract

The curation of large cohort datasets for developing deep learning models demands annotating or editing over the AI-predicted labels. However, openly available annotation tools are time-consuming, as they involve complex navigational steps and do not support batch processing. This study aimed to develop a simple, fast, semiautomated batch-annotation tool named MedImgAnnotator. The tool was created by integrating editing tools, image mask pair matching logic, preprocessing functions, object detection tool, and a plugin management system. Its performance was evaluated in terms of active annotation time and ease of use by the developer and 12 users using open-source MR and CT scans. Additionally, the tool had been used by the MICCAI 2026 CARE Challenge liver segmentation best performance winner to annotate liver masks from 6412 MRI volumes. The evaluation results showed that MedImgAnnotator reduced active annotation time from 206.5+/-32.87 (average active annotation time of openly available tools) seconds to 59 seconds. The Wilcoxon signed-rank test of 12 users evaluation results recorded P values of 0.032 and 0.008 for the active annotation time and higher ease of use rating metrics, respectively, indicating that the current tool significantly reduces annotation time and enhances ease of use. Moreover, the tool utilization for annotating different regions of interest from different modality scans and body regions shows its usability across modalities and areas of interest. MedImgAnnotator, which significantly reduces active annotation time for large imaging cohorts, is available as an executable file and as a Python script. Access link: https://ewunateassaye.github.io/developed-tools/medimgannotator.html
BibTeX
@misc{kassaw2026medimgannotator,
  author = {Ewunate Assaye Kassaw and Amit Mehndiratta and Peush Sahni and Anup Singh},
  title = {{MedImgAnnotator: batch wise medical image annotation and preprocessing tool}},
  year = {2026},
  eprint = {2611.22450},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.22450},
}
Figure