Problem
3D medical image segmentation models are expensive to train and iterate on. At Philips, the research team needed faster iteration cycles for MRI workflow automation without sacrificing model quality.
What I built
Built an automated hyperparameter optimisation and experiment-tracking framework using Optuna and MLflow. The framework systematically searched the hyperparameter space for 3D segmentation models, tracked all runs, and surfaced the best configurations — removing the manual tuning bottleneck from the research workflow.
Also worked on Deep Learning solutions to automate MRI workflow components, focusing on 3D image segmentation and computational geometry.
Impact
60%
reduction in compute time
- 10% improvement in inference pipeline speed
- Automated what was previously a manual, time-intensive tuning process
- Enabled faster research iteration cycles for the Philips team