Practical DICOM: Troubleshooting & Best Practices

This intensive 8 hour training provides a practical, hands-on exploration of real-world issues and best practices in working with DICOM medical imaging data. Participants will learn to recognize and troubleshoot common pitfalls in DICOM conformance, metadata integrity, and data conversion while gaining familiarity with widely used tools and libraries for data inspection, validation, and workflow automation. Through a mix of lectures, demonstrations, and case-based exercises, the course emphasizes strategies for building reliable and reproducible imaging pipelines. Designed for developers, researchers, and clinical data professionals, the training equips attendees with both the technical skills and best-practice frameworks needed to confidently manage DICOM data in research and clinical settings.

The target audience for this training is product developers, imaging informatics professionals, researchers, clinical data managers, and AI/ML developers working with medical imaging data (MRI, CT, PET, etc.).

Objectives

• Understand the most frequent and impactful DICOM issues in real-world settings.
• Equip participants to detect and fix metadata problems.
• Introduce and compare tools for inspecting, modifying, and using DICOM data.
• Learn to structure, transform, and quality-check datasets.
• Establish reliable imaging pipelines for clinical or research work.
• Address real-world issues participants have faced.

Prerequisites

• Experience using 3D Slicer
• Familiarity with DICOM

Agenda

This course involves approximately 8 hours of instruction. For onsite courses, we recommend either one 8-hour day of instruction (morning and afternoon sessions with a lunch break) or two 4-hour sessions on consecutive days. When taught as an online course, we can accommodate a variety of schedules including 2-hour sessions on four consecutive days or 1-hour sessions weekly for eight weeks.

DICOM in Practice – What Can Go Wrong and Why

• Non-conformance and vendor-specific deviations
• DICOM object types (images, RTSTRUCT, SEG, SR): which ones break and why
• Hierarchy challenges: Patient > Study > Series > Instance
• Enhanced vs. Legacy DICOM
• UIDs and duplication issues in longitudinal/multi-center data
• Hidden problems with secondary captures and screenshots

DICOM Metadata and Content Validation

• Critical tags: Patient, Study, Series, Image — what matters most
• Private tags and manufacturer extensions
• Anonymization vs. de-identification: pitfalls and compliance (HIPAA/GDPR)
• Using DICOM dictionaries, public tag databases
• Tools for validation
• Hands-on: Reviewing bad metadata examples

DICOM Handling Tools and Ecosystem Overview

• Libraries:
  • pydicom, dicom-numpy, dcm2niix, dcm4che, DCMTK
  • Scripting workflows: parsing, patching, and converting
• DICOM Servers & Interoperability:
  • Orthanc DICOMweb services
• 3D Slicer and similar platforms: using DICOM plugins, import/export issues, visualization pitfalls
• Export issues: When “Save as DICOM” isn’t enough

Advanced Data Handling & Conversion Strategies

• Managing timepoints and multi-series studies (e.g., pre/post contrast, multi-phase CT)
• Batch conversion strategies: DICOM → NIfTI, NRRD, HDF5
• Tools: dcm2niix, plastimatch, slicer.cli, in-house scripts
• Working with SEG/RTSTRUCT: flattening, resampling, interpreting ROI labels
• Spatial orientation consistency checks

Best Practices for Workflow Design and Reproducibility

• Checklist: What makes a “clean” DICOM dataset?
• Dealing with multi-institution data: harmonization strategies
• Version control and provenance for DICOM metadata
• Logging & documentation for regulatory or reproducibility purposes
• DICOM for AI/ML: Label leakage, data leakage, and augmentation dangers

Case Reviews and Open Discussion

• Participant-submitted examples (anonymized)
• Live debugging of corrupted (anonymized) DICOMs