Pixel
Results CSV
Identifiers and Case Structure
| Term | Definition |
|---|---|
| Case ID | Unique identifier for cases in our system. |
| Origin | Original file identifier or filepath. |
| Origin Created At | Date and time file was uploaded to our Centaur system. |
| Content ID | Unique identifier for the content (e.g. image, text, audio, etc.) in the case. |
| URL | Click this URL to view the case on the Centaur Portal. |
| Labeling State | Click the information icon next to “Status” to see the definitions of labeling states. |
| Series (if applicable) | Unique identifier for the video or content sequence (e.g. image frames) that the case is associated with. A series can contain multiple cases, and a case can cover part or all of a series. |
| Series Index (if applicable) | Unique identifier for the content as part of its series (e.g. image frames). |
| Patch (if applicable) | Area of interest on an image to be annotated for the case, in WKT format with coordinates rescaled to a 0-100 scale. |
| Study UID (if applicable) | Unique ID of the DICOM study that the annotated image was from. |
| Series UID (if applicable) | Unique ID of the series within the DICOM study that the annotated image was from. |
| Instance UID (if applicable) | Unique ID of the image within the DICOM series that was annotated. |
Reads and Annotations
| Term | Definition |
|---|---|
| Qualified Reads | Number of reads (“opinions”) on the case. Qualified Reads are those which meet a predefined accuracy criterion. |
| Correct Label | Centaur generated mask_uuid that can be used to identify the segmentation in the Masks folder provided in the Results CSV |
| Majority Label | Centaur generated mask_uuid that can be used to identify the segmentation in the Masks folder provided in the Results CSV |
| Top Reads | Qualified Reads used to create the Majority Label. |
Measures of Confidence
| Term | Definition |
|---|---|
| Difficulty | [1 - average IoU] (Intersection over Union) between each of the Top Reads and the Correct Label. |
| Agreement | Average IoU (Intersection over Union) between each of the Top Reads and the Majority Label. |
Other Information
| Term | Definition |
|---|---|
| Internal Notes | Any notes internal to your team or tracked by Centaur relevant to the case. |
| Comments | Comments on the case by our labelers. |
| Flags | Flags on the case by our labelers. |
| Explanation | Notes written by you and/or an admin to help train and educate labelers. The explanation is shown to labelers after they have submitted a read. |
| Pre-answer explainers | Additional text information provided to labelers about a case prior to viewing. |
| Static Highlight | Range and entity highlighted for any static highlights present on the case (e.g., [{'range':[6,15], 'text': "this cell"}]). |
Masks Folder*
A Masks file is provided alongside the results download. This includes a .tiff or .nifti file for each majority and correct label associated with a case. The files are named using the format mask_uuid.tiff(nifti). In the case of NIfTI, each series is contained within a single NIfTI file. For TIFF, a separate mask file is generated for each slice, frame, instance.
You can link these mask files to the results CSV by using the Majority Label and Correct Label columns, where each Case ID in the results CSV has a corresponding mask_uuid.
Reads CSV
| Term | Definition |
|---|---|
| Case ID | Unique identifier for cases in our system. |
| User ID | Unique identifier for users in our system. |
| Read ID | Unique identifier for reads in our system |
| Qscore | The trailing average accuracy for the user when they submitted an answer to the associated case ID. Read more about trailing average accuracy (Qscore) here. |
| Frame ID | Identifier for the frame number in the case. |
| Label Class | The class assigned to the label. |
| User Label | Read provided by the user. |
_Note: Only the top reads used to create the majority label are included.
Updated about 2 months ago
