Multi-scale Attention-Based Multiple Instance Learning for Classification of Multi-gigapixel Histology Images

Histology images with multi-gigapixel of resolution yield rich information for cancer diagnosis and prognosis. Most the time, only slide-level label is available because pixel-wise annotation labour intensive task. In this paper, we propose a deep learning pipeline classification in histology images. Using multiple instance learning, attempt to predict latent membrane protein 1 (LMP1) status nasopharyngeal carcinoma (NPC) based on haematoxylin eosin-stain (H &E) We utilised attention mechanism residual connection our aggregation layers. 3-fold cross-validation experiment, achieved average accuracy, AUC F1-score 0.936, 0.995 0.862, respectively. This method also allows us examine model interpretability by visualising scores. To best knowledge, first LMP1 NPC using learning.