Visual foundation models with adapter learning for Parkinson’s disease recognition based on OCT images
Year: 2026
Type: article
Abstract: Parkinson’s disease (PD) is a progressive neurodegenerative disorder. Optical coherence tomography (OCT) images have emerged as a promising means for PD recognition. However, constructing a precise DNN-based PD recognition model requires a large number of OCT images and expensive computational resources. This paper aims to transfer pre-trained visual foundation models (VFMs) with the aid of the adapter learning paradigm to recognize PD from OCT images, by adding only a few trainable parameters. We propose a novel VFM-Adapter framework for automatic OCT-based PD recognition based on the adapter learning paradigm. To be specific, we introduce three lightweight adapters: global adapter, local adapter, and token-wise adapter, to effectively leverage the transferability of feature representations from VFMs to OCT images, for equipping VFMs with strong PD recognition capability. Our VFM-Adapter achieves an accuracy of 77.66%, sensitivity of 78.01%, specificity of 77.31%, F1-score of 77.93%, and precision of 79.58%. It outperforms state-of-the-art deep neural networks, such as convolutional neural networks (CNNs), attention-based CNNs, and vision transformers by updating only a small number of parameters. Extensive ablation studies further demonstrated the effectiveness of each adapter. The proposed VFM-Adapter provides an effective yet parameter-efficient transfer learning framework for OCT-based PD recognition. By substantially finetuning the small number of trainable parameters while maintaining competitive PD performance, the proposed framework offers a promising altetnative for automated PD screening in real-world diagnosis scenarios. Not applicable.
Language: English
Topic: Retinal Imaging and Analysis
Subfield: Radiology, Nuclear Medicine and Imaging
Field: Medicine
Domain: Health Sciences
Open Access status: gold
Estimated APC paid: $2,990
OpenAlex ID: W7204101337
