The paper discusses a new approach for whole-slide image analysis in computational pathology. The researchers propose using the MLP-Mixer model, a less explored alternative to common vision transformers, for large-scale datasets. The model lacks a self-attention mechanism, resulting in linear computational complexity with the number of input patches. To preprocess the whole-slide image, the researchers suggest a combination of feature embedding and clustering. This process creates a reduced prototype representation that can be used as input to the MLP-Mixer architecture. The proposed method shows comparable performance to current state-of-the-art methods, but with lower computational time and memory load.
Publication date: 19 Oct 2023
Project Page: https://github.com/butkej/ProtoMixer
Paper: https://arxiv.org/pdf/2310.12769
