Purpose:
The Guidelines for Data Annotation detail how datasets should be annotated by medical experts to apply the labels necessary for training a machine learning model. They establish a standardized annotation style, defining both the accuracy and format of annotations. These guidelines should be precise enough to allow any third party to assess the quality of the annotations. Additionally, they should include the rationale behind label selection and the methodology used to derive the ground truth for model development.
Related Documentation:
- SOP Machine Learning Model Development
- Intended Use
- Guidelines for Data Acquisition
Note: The content here serves as an example tailored to a specific use case. All details should be replaced and customized to suit your specific product application. Consider presenting this information as a slide deck for easier incorporation of helpful images.
General Instructions
Annotation experts are required to segment the following findings:
- Tumors
- Other suspicious findings
- (…)
It is crucial to segment all visible findings, as missing any could lead to misclassification as non-suspicious and negatively impact model performance.
Annotation experts should NOT segment the following findings:
- Benign finding XYZ
- (…)
Tumors
Annotation Guidelines:
– Ensure the segmentation encompasses the entire tumor
– (…)
Record the following details:
– Tumor classification
– Biopsy information
– Confidence level
– (…)
Other Suspicious Findings
Annotation Guidelines:
– Ensure the segmentation fully captures the finding
– (…)
Record the following details:
– Type of suspicious finding
– Biopsy information
– Confidence level
– (…)