Detecting Cancer Metastases on Gigapixel Pathology Images
By Yun Liu, Krishna Gadepalli, Mohammad Norouzi, George E. Dahl, Timo Kohlberger, A. Boyko, Subhashini Venugopalan, Aleksei Timofeev, Phil Q. Nelson, Greg S. Corrado and 3 more. 2017
676 citations recorded by the source corpus.
Each year, the treatment decisions for more than 230,000 breast cancer patients in the U.S. hinge on whether the cancer has metastasized away from the breast. Metastasis detection is currently performed by pathologists reviewing large expanses of biological tissues. This process is labor intensive and error-prone. We present a framework to automatically detect and localize tumors as small as 100 x 100 pixels in gigapixel microscopy images sized 100,000 x 100,000 pixels. Our method leverages a convolutional neural network (CNN) architecture and obtains state-of-the-art results on the Camelyon16 dataset in the challenging lesion-level tumor detection task. At 8 false positives per image, we detect 92.4% of the tumors, relative to 82.7% by the previous best automated approach. For comparison, a human pathologist attempting exhaustive search achieved 73.2% sensitivity. We achieve image-level AUC scores above 97% on both the Camelyon16 test set and an independent set of 110 slides. In addition, we discover that two slides in the Camelyon16 training set were erroneously labeled normal. Our approach could considerably reduce false negative rates in metastasis detection.
Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Radiology, Nuclear Medicine and Imaging, Medical Imaging
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
- Nicolas Papernot
- Nicholas Carlini
- Percy Liang
- Yoshua Bengio
- Quoc V. Le
- Florian Tramèr
- Mohit Bansal
- Philip S. Yu
In Computer Vision and Pattern Recognition
- Luc Van Gool
- Andrew Zisserman
- Xiaogang Wang
- Dacheng Tao
- Ming-Hsuan Yang
- Jitendra Malik
- Trevor Darrell
- Bernt Schiele
In Radiology, Nuclear Medicine and Imaging
- P. Heng
- B. van Ginneken
- Qi-Yan Dou
- Hao Chen
- Francesco Ciompi
- Geert J. S. Litjens
- Anne L. Martel
- Ozan Ciga
In Medical Imaging
- Anne L. Martel
- Ozan Ciga
- Chetan L. Srinidhi
- Francesco Ciompi
- J. A. van der Laak
- Geert J. S. Litjens
- Bin Li
- Kevin W. Eliceiri
Citations in the corpus
- A Large-scale Study of Representation Learning with the Visual Task Adaptation Benchmark (2019, neutral): “Further, although the literature reports mixed results (Liu et al., 2017; Raghu et al., 2019; Neumann et al., 2019), we find that they also work for specialized tasks despite the domain-shift (medical or remote sensing).”
- Rotation Equivariant CNNs for Digital Pathology (2018, methodology): “limited receptive eld of the model, with 5 dense-block/transition-block pairs. The model spatially-pools the input by a factor of 25, the output of which resembles the segmentation resolution used in [1]. Full-model group equivariance is achieved by replacing all convolution layers with group-equiva”
- Scalable Private Learning with PATE (2018, background): “For example, models trained on people’s personal messages or detailed medical information can offer invaluable insights into real-world language usage or the diagnoses and treatment of human diseases (McMahan et al., 2017; Liu et al., 2017).”
- A Survey on Explainable Artificial Intelligence (XAI): Toward Medical XAI (2019, background): “ data. The so-called ground truths for medical tasks, provided by professionals, are not always absolutely correct. In fact, news regarding how AI beats human performance in medical imaging diagnosis [168] indicates that human judgment could be brittle. This is true even of trained medical personnel”
- Machine Unlearning (2019, background): “This data is often sensitive in nature and could include information like medical records [1] or personal emails [2].”
- A survey on deep learning in medical image analysis (2017, methodology): “Consequently, AlexNet or other simple models such as GG are still popular for medical data, though recent landmark tudies all use a version of GoogleNet called Inception v3 ( Gulshan t al., 2016; Esteva et al., 2017; Liu et al., 2017 ).”
- WILDS: A Benchmark of in-the-Wild Distribution Shifts (2020, neutral): “These typically fall into one of two categories: data augmentation methods that perturb the colors in the training images (e.g., Liu et al. (2017); Bug et al. (2017); Tellez et al. (2018)) or stain normalization methods that seek to standardize colors across training images (e.g., Macenko et al.…”
- Quantifying the effects of data augmentation and stain color normalization in convolutional neural networks for computational pathology (2019, methodology): “We followed the color deconvolution approach proposed by (Macenko et al. (2009)).”
- Fairness Constraints: A Flexible Approach for Fair Classification (2019, background): “By leveraging vast amounts of (training) data and modern machine learning methods, algorithmic decision systems are able to provide highly accurate predictions, often surpassing even human accuracy (Kleinberg et al., 2017a; Liu et al., 2017).”
- Weakly Supervised Deep Learning for Whole Slide Lung Cancer Image Analysis (2020, background): “Furthermore, a range of magnification levels of WSIs were also considered to improve the performance [17], [20].”
- A Survey of Deep Learning for Scientific Discovery (2020, background): “In semantic segmentation, all the pixels corresponding to balloon are identified, while in instance segmentation, each individual balloon is identified separately. pathology [135, 55], analyzing brain scans (PET, fMRI) [202, 45].”
- Deep neural network models for computational histopathology: A survey (2019, methodology): “One may combat staining variation by augmenting the training data by varying each pixel value per channel within a predefined range on transformed color spaces, such as HSV (hue, saturation and value) or HED (Hematoxylin, Eosin, and Diaminobenzidine) (Liu et al., 2017; Li and Ping, 2018; Tellez et a”
- Explainable Reinforcement Learning: A Survey (2020, neutral): “Especially Machine Learning (ML) as one branch of AI has numerous fields of application, be it transportation [59], advertisement and content recommendation [47], or medicine [39].”
- Near real-time intraoperative brain tumor diagnosis using stimulated Raman histology and deep neural networks (2020, neutral): “6) 36,37 .”