Human Aligned Reward Modeling for Automated Transfer Function Generation of 3D Rendering of Medical Image Data

David Melenberg, Nele Blum, Cem Adiyaman, Thorsten Buzug, Maik Stille

Keywords: Deep Learning, Human Feedback, Transfer Function, Volume Rendering

Abstract: We propose a reinforcement learning framework to automate the design of 2D transfer functions for direct volume rendering of medical images. By training a reward model based on human feedback, our approach enables an agent to extract transfer functions from joint histograms without manual fine-tuning. Preliminary results demonstrate that the developed method effectively captures human preferences, marking a significant step towards automated, user-aligned 3D renderings for improved patient communication, diagnosis, and treatment planning.

Submission Number: 47