Clinical-grade computational pathology using weakly supervised deep learning on whole slide images

By Gabriele Campanella, M. Hanna, Luke Geneslaw, Allen P. Miraflor, Vitor Werneck Krauss Silva, K. Busam, E. Brogi, V. Reuter, D. Klimstra, Thomas J. Fuchs. Nature Medicine, 2019

2,579 citations recorded by the source corpus.

The development of decision support systems for pathology and their deployment in clinical practice have been hindered by the need for large manually annotated datasets. To overcome this problem, we present a multiple instance learning-based deep learning system that uses only the reported diagnoses as labels for training, thereby avoiding expensive and time-consuming pixel-wise manual annotations. We evaluated this framework at scale on a dataset of 44,732 whole slide images from 15,187 patients without any form of data curation. Tests on prostate cancer, basal cell carcinoma and breast cancer metastases to axillary lymph nodes resulted in areas under the curve above 0.98 for all cancer types. Its clinical application would allow pathologists to exclude 65–75% of slides while retaining 100% sensitivity. Our results show that this system has the ability to train accurate classification models at unprecedented scale, laying the foundation for the deployment of computational decision support systems in clinical practice. A deep learning model trained on real-world digital pathology data achieves clinical performance in cancer diagnosis.

Subfields: Artificial Intelligence, Biophysics, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/addae423490bbe82da4fb2fc265237178686b4e8

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Biophysics

In Computer Vision and Pattern Recognition

Citations in the corpus

WILDS: A Benchmark of in-the-Wild Distribution Shifts (2020, neutral): “…the gold standard is to evaluate models on an independent test set collected from a different hospital (e.g.,Beck et al. (2011); Liu et al. (2017); Courtiol et al. (2019); McKinney et al. (2020)) or at least with a different scanner within the same hospital (e.g., Campanella et al. (2019)).”
Multimodal Co-Attention Transformer for Survival Prediction in Gigapixel Whole Slide Images (2021, background): “…a tiny proportion of the total image ( i.e. - a needle-in-a-haystack problem such as differentiating normal tissue vs micro-metastases) [4, 3, 5, 45], or 2) spanning the entire composition of a WSI and dependent on the interactions of its components ( i.e. - a ﬁne-grained visual recognition…”
Self supervised contrastive learning for digital histopathology (2020, methodology): “This is unlike some recent work that shows for various histopathology tasks, models comparable to Resnet 34 in terms of network capacity tend to perform significantly better than both smaller and larger models (Campanella et al., 2019).”
Artificial Intelligence in Cancer Research and Precision Medicine (2021, background): “These model architectures have indeed excelled at classification of images such as determining whether a digitized stained slide contains cancer cells or not (2, 3, 7–13).”
Dual-stream Multiple Instance Learning Network for Whole Slide Image Classification with Self-supervised Contrastive Learning (2020, methodology): “To address this challenge, several recent studies [20, 3, 17] have demonstrated the promise of weakly supervised WSI classiﬁcation, where only slide-level labels are used to train a patch-based classiﬁer.”
A foundation model for clinical-grade computational pathology and rare cancers detection (2024, positive): “Using multiple-instance learning 14,49,55 with labels at the level of groups of slides has enabled clinically relevant diagnostics by scaling to training datasets on the order of 10,000 WSIs 10–14 .”
Artificial intelligence in healthcare: An essential guide for health leaders (2019, background): “Their study shows that their system has the ability to train accurate classification models without annotated data sets, thus overcome a great challenge for deploying computational decisionsupport systems in clinical practice.(15) Massachusetts Institute of Technology (MIT) has recently developed a ”