Treatment Effect Identification in Multimodal N-of-1 Trials #1
● mergedACNE04 fine-tuneN-of-1 scoringGrad-CAMLMM · AR(1)
- Author
- Bashar Arous · M.Sc. Data Science, University of Potsdam
- Advisor
- Thomas Gärtner (HPI)
- Reviewer
- Prof. Dr. Alexandra Carpentier
- Grade
- 1.7
- Raters needed
5 blinded humans → 1 automated pipeline- Result
- VGG16 recovers 1 of 2 expert-confirmed treatment effects (P2, p = 0.0001), with 0 false positives
- Data
- ACNE04 · 1,457 dermatologist-graded images. HIAlab N-of-1 acne trials, HPI · 255 smartphone photos, 5 participants
- Stack
- Python · TensorFlow 2.14 / Keras · scikit-learn · R / nlme · Colab T4