Efficient Self-Supervised Adaptation for Medical Image Analysis

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Sorkhei, Moein, Konuk, Emir, Guo, Jingyu, Meng, Chanjuan, Matsoukas, Christos, Smith, Kevin
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911180382011392
author Sorkhei, Moein
Konuk, Emir
Guo, Jingyu
Meng, Chanjuan
Matsoukas, Christos
Smith, Kevin
author_facet Sorkhei, Moein
Konuk, Emir
Guo, Jingyu
Meng, Chanjuan
Matsoukas, Christos
Smith, Kevin
contents Self-supervised adaptation (SSA) improves foundation model transfer to medical domains but is computationally prohibitive. Although parameter efficient fine-tuning methods such as LoRA have been explored for supervised adaptation, their effectiveness for SSA remains unknown. In this work, we introduce efficient self-supervised adaptation (ESSA), a framework that applies parameter-efficient fine-tuning techniques to SSA with the aim of reducing computational cost and improving adaptation performance. Among the methods tested, Attention Projection Layer Adaptation (APLA) sets a new state-of-the-art, consistently surpassing full-parameter SSA and supervised fine-tuning across diverse medical tasks, while reducing GPU memory by up to 40.1% and increasing training throughput by 25.2%, all while maintaining inference efficiency.
format Preprint
id arxiv_https___arxiv_org_abs_2503_18873
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Efficient Self-Supervised Adaptation for Medical Image Analysis
Sorkhei, Moein
Konuk, Emir
Guo, Jingyu
Meng, Chanjuan
Matsoukas, Christos
Smith, Kevin
Computer Vision and Pattern Recognition
Self-supervised adaptation (SSA) improves foundation model transfer to medical domains but is computationally prohibitive. Although parameter efficient fine-tuning methods such as LoRA have been explored for supervised adaptation, their effectiveness for SSA remains unknown. In this work, we introduce efficient self-supervised adaptation (ESSA), a framework that applies parameter-efficient fine-tuning techniques to SSA with the aim of reducing computational cost and improving adaptation performance. Among the methods tested, Attention Projection Layer Adaptation (APLA) sets a new state-of-the-art, consistently surpassing full-parameter SSA and supervised fine-tuning across diverse medical tasks, while reducing GPU memory by up to 40.1% and increasing training throughput by 25.2%, all while maintaining inference efficiency.
title Efficient Self-Supervised Adaptation for Medical Image Analysis
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2503.18873