Our client has a medical device that collects time series measurements from a patient's skin to provide skin cancer diagnostic predictions in real-time. The client has collected data via clinical trials across healthy and disease populations. The client has a deep-learning model to perform the prediction.
As our client prepares for their pivotal clinical trial, they wanted assurance that the model was optimized before entering the clinical trial, and our advice on whether they should collect more data, or if the model was ready for the pivotal trial.
To improve the accuracy of the model, our team expanded the feature set and implemented a machine learning pipeline to manage the feature search and model hyperparameter search at scale to optimize the performance.
We built an MLOps pipeline using MLflow and Kubernetes to perform and manage feature extraction, model training, and hyperparameter tuning. The pipeline enables experiment tracking of the modeling performance of different combinations of feature extraction strategies, model types, and hyperparameters.
We have increased the model’s performance and simplified the architecture to enable edge deployment. Our client is ready for their pivotal clinical trial.