Ultrasound labels with a review record.
Nitsor connects labelling and expert review on uncompressed single-frame DICOM ultrasound images, with each decision tied to the image it describes.
Real 2D breast ultrasound, Gómez-Flores et al., BUS-BRA, CC BY 4.0 (opens in a new tab), modified: enlarged 2x without interpolation and outlined by Nitsor. Outline from the published lesion mask. Model proposal from SAM 2.1, outside Nitsor. Not a Nitsor import. Dataset notice. Sources
Ultrasound research
Keep the image behind each decision.
A finding needs its source image, its label and the reader who checked it. Keep them together as the research dataset changes.
Connect
Start with one image.
Register an uncompressed single-frame DICOM ultrasound image. Keep images without spatial geometry as individual images.
Label
Mark the structure you study.
Draw labels against the image and keep its calibration with the source. Use physical measurements only when the image supplies valid calibration.
Review
Keep the expert decision.
A second reader examines the label. The record names the author, the reviewer and the revision reviewed.
Fix the dataset
Name the images behind the result.
Give your research or training job a fixed dataset version while the team continues editing. Training runs in your own workflow.
What each frame shows
- Frame 1: the published pubic symphysis and fetal head masks as white outlines, and a grey box fitted to the published head mask.
- SAM 2.1 proposes the head from that box, drawn dashed.
- Frames 2 to 24: the image steps through the sampled frames; the white outlines are the masks published for each frame, and the dashed line is the SAM 2.1 proposal carried forward from the first box.
Real intrapartum transperineal ultrasound, Lu, Bai, Zhou and Zhou, JNU-IFM, CC BY 4.0 (opens in a new tab), modified: cropped and outlined by Nitsor. Outlines from the published masks. Model proposal from SAM 2.1, outside Nitsor. Sources
Public research data
Published masks and a model proposal, frame by frame.
Twenty-four frames sampled from one video, with the pubic symphysis and fetal head masks its authors published in white, and a SAM 2.1 proposal for the head, carried forward from one box on the first frame, dashed.
Source frames from a video. Nitsor imports uncompressed single-frame DICOM images.
Uncompressed single-frame DICOM ultrasound images
Uncompressed single-frame DICOM ultrasound is handled as a 2D image. Compressed transfer syntaxes, multi-frame, cine and native 3D or 4D acquisitions are outside this workflow.
Image and measurement scope
Keep calibration with the image.
DICOM PixelSpacing or valid Ultrasound Regions calibration supplies physical scale. An uncalibrated image carries no physical scale.
Keep separate images separate.
Images without spatial geometry remain individual images. A collection of images does not establish a native 3D or 4D acquisition.
One uncompressed single-frame DICOM ultrasound image, as recorded. Its Ultrasound Region, 852 × 565 pixels, gives 0.2284 mm per pixel on both axes (PhysicalDeltaX and PhysicalDeltaY, 0.02284 cm, read from the header).
Real liver ultrasound, Eisenbrey, Lyshchik and Wessner, B-mode-and-CEUS-Liver, TCIA, CC BY 4.0 (opens in a new tab), modified: cropped, the scanner model name blacked out. Sources
- Can we use uncalibrated images?
- Keep the image and its labels together, without physical measurements. A visible ruler or a familiar structure does not establish a calibrated pixel size.
- Can a model suggest labels for these images?
- Model suggestions require a model compatible with your images. Opening a DICOM image does not establish that a model supports it. The SAM 2.1 proposals on this page were run by us on public images, outside Nitsor.
- Can Nitsor make a clinical diagnosis?
- Nitsor is for research and AI development. It does not diagnose, approve a clinical decision or replace the reader's judgement.
Show us what you inspect.
Describe your uncompressed single-frame DICOM ultrasound images and the review handoff you want to improve. We can discuss the workflow using public research data.