Rotation Equivariant CNNs for Digital Pathology

By Bastiaan S. Veeling, Jasper Linmans, Jim Winkens, Taco Cohen, Max Welling. 2018

735 citations recorded by the source corpus.

We propose a new model for digital pathology segmentation, based on the observation that histopathology images are inherently symmetric under rotation and reflection. Utilizing recent findings on rotation equivariant CNNs, the proposed model leverages these symmetries in a principled manner. We present a visual analysis showing improved stability on predictions, and demonstrate that exploiting rotation equivariance significantly improves tumor detection performance on a challenging lymph node metastases dataset. We further present a novel derived dataset to enable principled comparison of machine learning models, in combination with an initial benchmark. Through this dataset, the task of histopathology diagnosis becomes accessible as a challenging benchmark for fundamental machine learning research.

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

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

On Robustness and Transferability of Convolutional Neural Networks (2020, background): “The specialized group contains remote sensing datasets EuroSAT [25] and Resisc45 [8], and medical image datasets Patch Camelyon [67] and Diabetic Retinopathy [34].”
A Large-scale Study of Representation Learning with the Visual Task Adaptation Benchmark (2019, background): “Patch Camelyon (Veeling et al., 2018) Clevr/count (Johnson et al., 2017) CLEVR is a visual question and answer dataset designed to evaluate algorithmic visual reasoning.”
Combined Scaling for Zero-shot Transfer Learning (2021, methodology): “From Section 9.2, we see that BASIC models have particularly low performance on EuroSat (Helber et al., 2018), MNIST (LeCun et al., 2010), and Patch Camelyon (Veeling et al., 2018).”
ELEVATER: A Benchmark and Toolkit for Evaluating Language-Augmented Visual Models (2022, background): “Visual recognition has become ubiquitous in our society [60], with applications in geolocalization [51], action recognition [57], street number transcription [44], satellite remote sensing [22], medical imaging [62], self-driving cars [19], etc.”
Group Equivariant Stand-Alone Self-Attention For Vision (2020, methodology): “The translation equivariance of CNNs has been extended to additional symmetries ranging from planar rotations (Dieleman et al., 2016; Marcos et al., 2017; Worrall et al., 2017; Weiler et al., 2018b; Li et al., 2018; Cheng et al., 2018; Hoogeboom et al., 2018; Bekkers et al., 2018; Veeling et al., 20”
General E(2)-Equivariant Steerable CNNs (2019, neutral): “22 D4 [6, 1, 38] 1.”
Gauge Equivariant Convolutional Networks and the Icosahedral CNN (2019, neutral): “, 2019), translations and rotations of the plane and 3D space (Oyallon & Mallat, 2015; Cohen & Welling, 2016; 2017; Marcos et al., 2017; Weiler et al., 2018b;a; Worrall et al., 2017; Worrall & Brostow, 2018; Winkels & Cohen, 2018; Veeling et al., 2018; Thomas et al., 2018; Bekkers et al., 2018; Hoog”
First Session Adaptation: A Strong Replay-Free Baseline for Class-Incremental Learning (2023, methodology): “Specifically, by having access to the running terms { A, b , count } , the sample covariance matrix is given by The time complexity of LDA scales as Algorithm 1 Update of sample covariance running terms Require: g ( · ) ≡ a feature extractor backbone Require: end if”
WILDS: A Benchmark of in-the-Wild Distribution Shifts (2020, positive): “To improve accessibility, we pre-process the slides into patches and balance the dataset so that each split has a 50/50 class balance, making average accuracy is a reasonable measure of performance (Veeling et al., 2018; Tellez et al., 2019).”
Self supervised contrastive learning for digital histopathology (2020, methodology): “The Patch Camelyon dataset contains 327680 image patches with size 96 × 96 taken from whole-slide scans of lymph node sections.”
Non-Parametric Calibration for Classification (2019, background): “• PCam [44]: Detection of metastatic tissue in histopathologic scans of lymph node sections conv.”
AUC Maximization in the Era of Big Data and AI: A Survey (2022, methodology): “They use the PathCamelyon dataset for this task, which consists of 294,912 color images (96×96) extracted from histopathologic scans of lymph node section for training and 32,768 images for testing with balanced class ratio [8, 160].”
Deep Ensembles for Low-Data Transfer Learning (2020, background): “We suspect this is because the default augmentations specified by the BiT HyperRule are not well suited for such datasets; in particular, we found PatchCamelyon benefited the most from different augmentations.”
Vision Models Are More Robust And Fair When Pretrained On Uncurated Images Without Supervision (2022, neutral): “(i) standard datasets such as ImageNet-1K, Places205 and VOC07 which haven been widely used for testing features quality in many previous works [17, 18, 50, 128, 134]; (ii) medical and satellite images such as in RESISC45 [26], EuroSAT [59], PatchCamelyon [121]; (iii) structured datasets containing ”
FiT: Parameter Efficient Few-shot Transfer Learning for Personalized and Federated Image Classification (2022, methodology): “The specialized datasets are: EuroSAT (Helber et al., 2019), Resics45 (Cheng et al., 2017),Patch Camelyon (Veeling et al., 2018), and Retinopathy (Kaggle & EyePacs, 2015).”