Ground-truth preparation

We developed a ground-truth preparation process for medical imaging projects where we use machine learning algorithms.

The model operation results are conditional on the quality of learning data and their relevant preparation.

The objective data prepared in the appropriate way which will be used to teach the model will guarantee the anticipated results. And excellent results are the success of the whole enterprise.

Why us

  • We have organized the whole pipeline including both data annotation and verification of these annotations’ accuracy.
  • Our process allows us to ensure a high degree of objectivity in the analysis of a given study and to eliminate the risk related to an incorrect assessment resulting from personal experience of a given doctor.
  • We engage experienced radiologists with the expertise necessary to analyze a given study properly to prepare the data.
  • The test used for model validation and the test annotations are verified in three stages.

The process of ground-truth preparation 

  • We obrain learning data – we use data provided by the customer or obtained by us from proven providers to create the algorithm.
  • We create teams of specialists – if necessary, we involve not only experienced radiologists but also other medicine specialists. It is important to ensure the maximum degree of objective analysis of a given study.
  • Preparing the annotations – our task is to ensure a smooth flow of studies and annotations between doctors and to monitor the whole process. The selected team of involved specialists is in charge of preparing the annotations.
  • Evaluation – another team of involved experts (with more than 10 years’ radiology experience) evaluate annotations made by specialists from the annotators’ team. They accept such a study or refer it for improvement. Thus, each study is described by one specialist. This description is then reviewed by the second, more experienced specialist (expert). Thus, the annotation of each study is double-checked, for the first time during its preparation and for the second time during its evaluation.
  • Validating the model and evaluating its performance – after the annotation process had been completed and all studies had been accepted by the experts, we select a test set from the studies that had not been used for teaching. The annotations of the studies from the test set are passed on for evaluation to a second member of the evaluation team (second expert) who had not previously evaluated a particular study.
  • Result – as a result, the test study and its annotations were evaluated three times, the first time was when preparing the annotations, the second time when evaluating them, and the third time by another expert. Only after the study had been accepted by both radiologists could it be used for model validation.

Learn more about our competences

Medical image segmentation as a method for the preparation of learning data.

Mean Opinion Score as a method for the validation of R&D project results in imaging diagnostics.

Medical software development difficult beginnings. How to cooperate?

Let’s work on
your challenges together!

Contact with us:

T: +48 609 995 887
E: purbanski@graylight-imaging.com