Weakly supervised histopathology image segmentation with self-attention

Seeing the random forest through the decision trees. Supporting learning health systems from histopathology with machine learning models: Challenges and opportunities

“…Self-attention methods and other innovative approaches 44,[46][47][48][49][50][51][52] as well as future improvements in computational power, [38][39][40][41][42][43]51,52 may help in this regard.…”

Computational Pathology: A Survey Review and The Way Forward

“… 437 MIL models can be improved by considering multi-scale information: one work notably used embeddings from different magnification levels and self-supervised contrastive learning to learn WSI classifiers. 268 Some works explicitly encode the patient-slide-patch hierarchy in the attention mechanism, 457 , 458 with one work using a cellular graph for top-down attention. 459 Graph Neural Networks (GNNs) have been explored to leverage intra- and inter-cell relationships, enabling cancer grading, 460 classification, 461 and survival prediction.…”

“…Whereas patch classification 267 , 287 , 288 , 289 , 290 or pixel segmentation 237 , 329 , 330 , 331 , 332 , 333 was formerly mainstream, these problems appear to have been largely solved, and now there is far more research into higher-level problems dominate, such as multiple-instance learning. 62 , 437 , 457 , 458 , 459 As computational methods continue to improve, it is natural that they are applied not merely as attention aids for pathologists (i.e. at the pixel or patch level), but furthermore are used to make intelligent slide- and patient-level decisions on their own.…”

Unleashing the potential of AI for pathology: challenges and recommendations

“…However, obtaining patch‐level labels can be very time‐consuming and typically only WSI‐level labels are available for training, making a compelling case for the use of weak supervision techniques . Weakly supervised CPath algorithms [ 50 , 89 , 90 , 91 , 92 , 93 , 94 , 95 ] aggregate patch‐level prediction scores by different mechanisms, such as majority voting, average pooling, or multiple instance learning. The success of these approaches depends on the nature of the ML task and the validity of assumptions underlying these approaches.…”