publishing calendar
click a month to filter the blog list.
- formatting
- images
- links
- math
- code
- blockquotes
- external-services
•
•
•
•
•
•
-
how i structure ablations in small-data medical imaging projects
Why ablations in medical imaging need to be boring, disciplined, and suspicious of lucky splits.
-
when to use cnns, transformers, or foundation models in medical imaging
Why model choice in medical imaging should start with the data, not the trend.
-
neural active contours: why old geometry ideas still matter in deep segmentation
Why contour evolution still has something useful to say in an era of deep segmentation models.
-
how to choose between explicit geometry, implicit fields, and gaussian splats
A practical way to think about 3D representations when you care about sparse clinical reconstruction and not just rendering demos.
-
federated graph learning for pathology: why distribution shift matters more than architecture
Why pathology models fail across sites for reasons that are often bigger than the backbone.
-
open-vocabulary surgical tool tracking: what foundation models help with and what they do not
Why foundation models are useful for surgical tool understanding, but not a replacement for tracking, geometry, and temporal discipline.
-
deformable endoscopic scene reconstruction: where current 3d methods break
Why endoscopy is a difficult place for 3D reconstruction methods that assume the world stays still.
-
sparse-view coronary vessel reconstruction from cta: representation tradeoffs
Why coronary reconstruction depends as much on representation choice as it does on the model.
-
designing a reproducible medical imaging project instead of a one-off notebook
Why medical imaging projects become more valuable when the pipeline can be rerun without guesswork.
-
what i learned benchmarking skin lesion segmentation beyond u-net
Why comparing skin lesion segmenters is less about naming the newest architecture and more about finding the failure modes.