Quality and normalisation for diagnostic algorithms
About the client
This global player is active in a variety of different sectors ranging from diagnostic imaging to home care. They owe their leading position in many fields to their ability to deliver fully integrated solutions through advanced technology and clever clinical and consumer insights.
The objective of this project was two-fold. The first objective aimed at identifying quality-related markers to devise a methodology for excluding “bad” samples. Moreover, this approach should be applicable to samples from different batches, potentially from distinct protocols. The second objective was to devise and implement normalisation approaches that enable cross-batch/cross-platform application of their diagnostic algorithm. The outcome of both objectives was required to be readily applicable for the diagnostic prediction algorithm at hand, enabling the use of this prediction algorithm across different batches and technology platforms.
“Data-driven quality control and normalization methods can significantly broaden the scope of a diagnostic prediction algorithm.” – The client
• Implemented a wide array of quality control metrics in order to determine quality features deterministic for the performance of the diagnostic algorithm.
• State-of-the art normalisation algorithms were implemented and investigated in order to extend the scope of the diagnostic algorithm.
• To ensure statistical integrity, BioLizard collaborated with an external subject matter expert to verify their approach.
• Next to standard reporting, BioLizard motivated their solution through step by step visualisation in order to not only provide a solution but also insights on why this approach was deemed best while other approaches failed.
A data driven quality control approach enables cross-platform quality assessment.
An efficient normalisation approach enables the use of prediction algorithms cross-batch/platform.
The application of the diagnostic prediction algorithm can be increased in scope.