Fig. 3: Overview of the MAGPIE framework. | npj Digital Medicine

Fig. 3: Overview of the MAGPIE framework.

From: Masked autoencoding, generalizable pretraining, and integrated experts for enhanced glioma segmentation

Fig. 3: Overview of the MAGPIE framework.

The framework consists of four main components: (1) a channel-agnostic encoder that processes multi-modal inputs, (2) a Mixture of Experts (MoE) mechanism for dynamic resource allocation, (3) specialized attention mechanisms (MS-DA in encoder, D-LKA in decoder), and (4) a two-stage training strategy combining self-supervised pretraining with supervised fine-tuning.

Back to article page