Initial situation
A startup with its own development of neural network analysis of CT scans applied for registration of a SaMD solution. At the time of work in Russia there was no established practice of registering systems based on deep learning. It was necessary to justify the risk class, develop a methodology for validating the algorithm, and undergo clinical evaluation using Russian data.
What we did
- Product classification and justification for risk class 2a
- Preparation of an algorithm validation protocol with AUC, sensitivity and specificity metrics
- Organization of clinical evaluation on a dataset of 4,800 studies from 6 medical centers
- Development of a plan for post-registration monitoring of the quality of the algorithm
- Support of extended examination in Roszdravnadzor and presentation at the specialized commission
Issued Registration Certificate
The registration certificate has been received. Based on our methodology, the customer subsequently registered 2 more models from his line in half the time.