My paper, "Confidence in Angle Predictions for Clinical Decision Support”, introduces a novel method for estimating uncertainty in angle measurements derived from predicted anatomical landmarks. We use a UNet++ architecture to generate landmark heatmaps that capture localization uncertainty. From these, we apply a Monte Carlo-like method to estimate an angle distribution. We then introduce a confidence metric derived from this distribution, capturing both the predicted value and the model's certainty. Our results show strong alignment between machine-derived confidence scores and clinician assessments. This is the first work to estimate distributions over clinically relevant angles from predicted landmarks, enabling more robust and trustworthy decision support in clinical workflows.
I am extremely proud and lucky to be a part of the MICCAI community and presenting my work this year. Presenting at MICCAI 2025 is a significant milestone in my research journey. It's a unique opportunity to present my work and receive feedback from experts in the field. This experience will be invaluable to my career as a whole and help me as I continue developing my thesis and refining the direction of my work.
I am especially eager to learn more about confidence and uncertainty estimation methods for medical imaging tasks in computer vision. I am also deeply interested in multimodal learning (image and text integration) and model fairness—areas I hope to explore further as I build on this work.
I'm looking forward to connecting with fellow researchers, exploring new approaches, and learning how others are tackling similar challenges in the field. I'm excited to be inspired by fresh ideas, innovative applications, and potential collaborations that could shape the next steps in my research. I'm also eager to attend the doctoral community events and engage with other students who are at a similar stage in their academic journey.