Project: Project: DICOM metadata databases

[nolden]

Draft Status

Ready - team will start page creating immediately

Category

DICOM

Key Investigators

• Marco Nolden (German Cancer Research Center, Germany)
• Andrey Fedorov (BWH, USA)
• Steve Pieper (Isomics, Inc., USA)

Project Description

Medical imaging applications and systems which manage large collections of DICOM images usually need some kind of database to allow for browsing and selecting images or image collections, to support curation and control of ML training tasks, batch analysis etc.
Goal of the project is to investigate existing and new approaches to handle the metadata of large image collections for different purposes, create experimental setups, and report on results.

• DICOM objects contains rich metadata

• depending on the use case, record linkage to non-imaging data might be an additional requirement

• extracted metadata can be represented in different JSON styles, stored in document databases like CouchDB, Apache OpenSearch etc..

• there is a FHIR imaging study (https://www.hl7.org/fhir/imagingstudy.html), FHIR data could be stored in FHIR stores, or regular SQL databases …

• custom approaches, like the CTK DICOM database, or IDC's representation in BigQuery; one has also observed flattened FHIR in SQL databases, combined with object stores etc.

• DICOM to JSON could be done according to the DICOM JSON model (https://dicom.nema.org/medical/dicom/current/output/chtml/part18/chapter_F.html) , e.g. using DCMTK, or custom approaches, but also generic metadata extractors like Apache Tika could be an option

Objective

1. Objective A. A report on the experiments and their results.

Approach and Plan

1. put DICOM JSON in JSON columns of sqlite or postgres, test jsonpath and similar
2. create FHIR imaging studies, put to FHIR endpoint or other databases
3. Connect with out-of-the-box visualization solutions for e.g. json documents

Progress and Next Steps

1. Describe specific steps you have actually done.

Illustrations

No response

Background and References

• Opensearch dashboards
• https://www.metabase.com/
• https://superset.apache.org/
• https://python.langchain.com/v0.1/docs/modules/model_io/output_parsers/types/json/
• https://echarts.apache.org/examples/en/index.html#chart-type-line
• https://projectweek.na-mic.org/PW41_2024_MIT/Projects/Dcm2Parquet/
• [2407.09064] Multi-Modal Dataset Creation for Federated Learning with DICOM Structured Reports
• https://learn.microsoft.com/en-us/industry/healthcare/healthcare-data-solutions/dicom-data-transformation-mapping
• Adding private tag names pydicom/pydicom#2187
• Potential example queries for experiiments: https://docs.google.com/document/d/1qC5_qUFBQ2HmEjfYQa9WaH1Y-erMlfis00bbU8UPnRs/edit?tab%3Dt.0%23heading%3Dh.ti11twc8h457&sa=D
• https://learn.canceridc.dev/cookbook/bigquery