NA-MIC Project Weeks

Back to Projects List

Creating DICOM-compatible cancer annotations for NLST

Key Investigators

• Deepa Krishnaswamy (Brigham and Women's Hospital, USA)
• Suraj Pai (Brigham and Women's Hospital, USA)
• Leonard Nürnberg (Brigham and Women's Hospital, USA)
• Andrey Fedorov (Brigham and Women's Hospital, USA)
• David Clunie (PixelMed Publishing, USA)

Project Description

The National Lung Screening Trial (NLST) is one of the largest lung cancer collections, with over 25K patients. In Imaging Data Commons (IDC), we have segmentations of anatomical regions using the TotalSegmentator model, but, we are missing any annotations of cancer.

There were several initiatives to add cancer nodule annotations to NLST data in IDC. One set of nodule segmentations was created from an AI model from this initiative, but only a percentage of them have been verified by an expert.

However, there is one initiative from MIT (https://github.com/reginabarzilaygroup/Sybil) that had experts annotate center points and bounding boxes for nodules in NLST patients. Our plan is to convert these json annotations to DICOM Structured Reports, which can then be ingested into IDC and displayed.

Objective

We will first convert the json point annotations to DICOM Structured Reports. Then we will ingest them into a DICOM datastore, and deploy our own OHIF application to display the points overlaid on the image data.

Approach and Plan

1. We will understand the format of the json files by plotting them in Slicer.
2. We will create a DICOM SR for a patient, starting with one point annotation per patient.
3. We will store these DICOM SR objets in a DICOM data store.
4. We will deploy OHIF, and display our point annotations along with the image.
5. If that works, we will add the ability for the DICOM SR to store multiple annotations.

Progress and Next Steps

1. We have started to understand the format of the json files by plotting them in Slicer.

Illustrations

No response

Background and References

No response