Self-DANA: A Resource-Efficient Channel-Adaptive Self-Supervised Approach for Foundation Models

Foundation Models (FMs) are large-scale models trained on extensive datasets that can be adapted to a wide range of downstream tasks with minimal fine-tuning. They have recently also gained attention in Electrocardiogram (ECG) signal analysis. One of the key properties of FMs is their transferability to a wide range of downstream scenarios. However, the adaptation of ECG FMs to downstream scenarios with fewer available channels (i.e., wearable and portable devices) still has to be properly investigated. In this work, we propose Self-DANA, an easy-to-integrate solution that enables FMs to be adaptable to a reduced number of input channels, ensuring resource efficiency and high performance. We also introduce Random Lead Selection, a novel augmentation to build more robust and channel-agnostic FMs. Our experiments on three datasets and five reduced-channel configurations demonstrate that Self-DANA significantly enhances resource efficiency while achieving superior or comparable performance to the literature alternative.