Raptor: Scalable Train-Free Embeddings for 3D Medical Volumes Leveraging Pretrained 2D Foundation Models

TL;DR

Raptor introduces a train-free, scalable method for creating meaningful embeddings of 3D medical images by leveraging pretrained 2D models and random projections, outperforming existing methods without additional training.

Contribution

It proposes Raptor, a novel train-free approach that uses pretrained 2D models and random projections to efficiently generate embeddings for volumetric medical data.

Findings

Outperforms state-of-the-art medical volume methods (+3% SuPreM, +6% MISFM, +10% Merlin, +13% VoCo, +14% SLIViT)

Reduces computational complexity significantly

Eliminates the need for costly training on volumetric data

Abstract

Current challenges in developing foundational models for volumetric imaging data, such as magnetic resonance imaging (MRI), stem from the computational complexity of training state-of-the-art architectures in high dimensions and curating sufficiently large datasets of volumes. To address these challenges, we introduce Raptor (Random Planar Tensor Reduction), a train-free method for generating semantically rich embeddings for volumetric data. Raptor leverages a frozen 2D foundation model, pretrained on natural images, to extract visual tokens from individual cross-sections of medical volumes. These tokens are then spatially compressed using random projections, significantly reducing computational complexity while retaining semantic information. Extensive experiments on ten diverse medical volume tasks verify the superior performance of Raptor over state-of-the-art methods, including those pretrained exclusively on medical volumes (+3% SuPreM, +6% MISFM, +10% Merlin, +13% VoCo, and +14% SLIViT), while entirely bypassing the need for costly training. Our results highlight the effectiveness and versatility of Raptor as a foundation for advancing deep learning-based methods for medical volumes.