Continual pretraining Prostate MRI ViT Encoder for task alignment

Key points

• Building foundation models for MRI has emerged as a promising direction for advancing medical image analysis and improving accessibility of task specific models.1
• Existing approaches primarily focus on effective pretraining strategies and evaluating their performance across diverse downstream tasks.1
• We further perform layer-wise representational analysis using Centered Kernel Alignment (CKA) to assess representational shifts and their correlation with task outcomes.1
• Our Feature analysis reveals that, for segmentation tasks, continually pretrained encoder features exhibit greater similarity to those of the finetuned encoder than the original pretrained encoder, correlating with observed performance gains.1

Sources (1)

• [1]Continual pretraining Prostate MRI ViT Encoder for task alignment
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 14, 01:57 PM

  Specifically, continually pretraining a pretrained ViT encoder using Low-Rank Adaptation (LoRA) on a prostate segmentation task to enhance task alignment and downstream performance.

  Building foundation models for MRI has emerged as a promising direction for advancing medical image analysis and improving accessibility of task specific models.

Extractive summary: sentences quoted from the sources.