Deep Interpretable Classification and Weakly-Supervised Segmentation of Histology Images via Max-Min Uncertainty (2020-11-14T00:00:00.000000Z)

TL;DR

High uncertainty is introduced as a criterion to localize non-discriminative regions that do not affect classifier decision, and is described with original Kullback-Leibler (KL) divergence losses evaluating the deviation of posterior predictions from the uniform distribution.

Abstract

Weakly-supervised learning (WSL) has recently triggered substantial interest as it mitigates the lack of pixel-wise annotations. Given global image labels, WSL methods yield pixel-level predictions (segmentations), which enable to interpret class predictions. Despite their recent success, mostly with natural images, such methods can face important challenges when the foreground and background regions have similar visual cues, yielding high false-positive rates in segmentations, as is the case in challenging histology images. WSL training is commonly driven by standard classification losses, which implicitly maximize model confidence, and locate the discriminative regions linked to classification decisions. Therefore, they lack mechanisms for modeling explicitly non-discriminative regions and reducing false-positive rates. We propose novel regularization terms, which enable the model to seek both non-discriminative and discriminative regions, while discouraging unbalanced segmentations. We introduce high uncertainty as a criterion to localize non-discriminative regions that do not affect classifier decision, and describe it with original Kullback-Leibler (KL) divergence losses evaluating the deviation of posterior predictions from the uniform distribution. Our KL terms encourage high uncertainty of the model when the latter inputs the latent non-discriminative regions. Our loss integrates: (i) a cross-entropy seeking a foreground, where model confidence about class prediction is high; (ii) a KL regularizer seeking a background, where model uncertainty is high; and (iii) log-barrier terms discouraging unbalanced segmentations. Comprehensive experiments and ablation studies over the public GlaS colon cancer data and a Camelyon16 patch-based benchmark for breast cancer show substantial improvements over state-of-the-art WSL methods, and confirm the effect of our new regularizers (our code is publicly available at https://github.com/sbelharbi/deep-wsl-histo-min-max-uncertainty).

References92 items

Deep Active Learning for Joint Classification & Segmentation with Weak Annotator

Colorectal Cancer Detection Based on Deep Learning

Black-box Explanation of Object Detectors via Saliency Maps

Automated histologic diagnosis of CNS tumors with machine learning

Evaluating and Aggregating Feature-based Model Explanations

Understanding Integrated Gradients with SmoothTaylor for Deep Neural Network Attribution

Toward Interpretable Machine Learning: Transparent Deep Neural Networks and Beyond

Towards Ground Truth Evaluation of Visual Explanations

Bounding boxes for weakly supervised segmentation: Global constraints get close to full supervision

Evaluating Weakly Supervised Object Localization Methods Right

Deep Ordinal Classification with Inequality Constraints

Iterative Augmentation of Visual Evidence for Weakly-Supervised Lesion Localization in Deep Interpretability Frameworks: Application to Color Fundus Images

Understanding Deep Networks via Extremal Perturbations and Smooth Masks

Deep weakly-supervised learning methods for classification and localization in histology images: a survey

Constrained domain adaptation for segmentation

Automated brain histology classification using machine learning

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

MixMatch: A Holistic Approach to Semi-Supervised Learning

Curriculum semi-supervised segmentation

Constrained deep networks: Lagrangian optimization via Log-barrier extensions

Using a Deep Learning Algorithm and Integrated Gradients Explanation to Assist Grading for Diabetic Retinopathy.

Deep Learning With Sampling in Colon Cancer Histology

Visualizing Deep Learning Models for the Detection of Referable Diabetic Retinopathy and Glaucoma

Definitions, methods, and applications in interpretable machine learning

Interpretable Deep Learning under Fire

Deep Clustering: On the Link Between Discriminative Models and K-Means

RISE: Randomized Input Sampling for Explanation of Black-box Models

Constrained‐CNN losses for weakly supervised segmentation☆

Adversarial Complementary Learning for Weakly Supervised Object Localization

Normalized Cut Loss for Weakly-Supervised CNN Segmentation

Tell Me Where to Look: Guided Attention Inference Network

Attention-based Deep Multiple Instance Learning

Visual interpretability for deep learning: a survey

A Deep Learning Interpretable Classifier for Diabetic Retinopathy Disease Grading

Diagnostic Assessment of Deep Learning Algorithms for Detection of Lymph Node Metastases in Women With Breast Cancer

Grad-CAM++: Generalized Gradient-Based Visual Explanations for Deep Convolutional Networks

Explainable and Interpretable Models in Computer Vision and Machine Learning

Machine Learning Methods for Histopathological Image Analysis

Explainable Artificial Intelligence: Understanding, Visualizing and Interpreting Deep Learning Models

Two-Phase Learning for Weakly Supervised Object Localization

WILDCAT: Weakly Supervised Learning of Deep ConvNets for Image Classification, Pointwise Localization and Segmentation

Weakly-supervised localization of diabetic retinopathy lesions in retinal fundus images

Zoom-in-Net: Deep Mining Lesions for Diabetic Retinopathy Detection

Large scale tissue histopathology image classification, segmentation, and visualization via deep convolutional activation features

Real Time Image Saliency for Black Box Classifiers

Network Dissection: Quantifying Interpretability of Deep Visual Representations

Virtual Adversarial Training: A Regularization Method for Supervised and Semi-Supervised Learning

Hide-and-Seek: Forcing a Network to be Meticulous for Weakly-Supervised Object and Action Localization

Interpretable Explanations of Black Boxes by Meaningful Perturbation

Object Region Mining with Adversarial Erasing: A Simple Classification to Semantic Segmentation Approach

A survey on deep learning in medical image analysis

Constrained Deep Weak Supervision for Histopathology Image Segmentation

3D fully convolutional networks for subcortical segmentation in MRI: A large-scale study

Deep image mining for diabetic retinopathy screening

Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization

Top-Down Neural Attention by Excitation Backprop

Breast cancer histopathological image classification using Convolutional Neural Networks

A Dataset for Breast Cancer Histopathological Image Classification

ScribbleSup: Scribble-Supervised Convolutional Networks for Semantic Segmentation

Simple Does It: Weakly Supervised Instance and Semantic Segmentation

Gland segmentation in colon histology images: The glas challenge contest

“Why Should I Trust You?”: Explaining the Predictions of Any Classifier

Learning Deep Features for Discriminative Localization

Deep Residual Learning for Image Recognition

Look and Think Twice: Capturing Top-Down Visual Attention with Feedback Convolutional Neural Networks

ProNet: Learning to Propose Object-Specific Boxes for Cascaded Neural Networks

On Pixel-Wise Explanations for Non-Linear Classifier Decisions by Layer-Wise Relevance Propagation

Constrained Convolutional Neural Networks for Weakly Supervised Segmentation

Is object localization for free? - Weakly-supervised learning with convolutional neural networks

What's the Point: Semantic Segmentation with Point Supervision

U-Net: Convolutional Networks for Biomedical Image Segmentation

A Stochastic Polygons Model for Glandular Structures in Colon Histology Images

BoxSup: Exploiting Bounding Boxes to Supervise Convolutional Networks for Semantic Segmentation

Striving for Simplicity: The All Convolutional Net

From image-level to pixel-level labeling with Convolutional Networks

Fully convolutional networks for semantic segmentation

Breast cancer histopathology image analysis

Breast Cancer Histopathology Image Analysis: A Review

Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps

Visualizing and Understanding Convolutional Networks

ImageNet classification with deep convolutional neural networks

Histology image analysis for carcinoma detection and grading

Convex optimization

Histopathological Image Analysis: A Review

Semi-supervised Learning by Entropy Minimization

F-CAM: Full Resolution CAM via Guided Parametric Upscaling

A brief introduction to weakly supervised learning

Attention Networks for Weakly Supervised Object Localization

Grad-CAM: Why did you say that? Visual Explanations from Deep Networks via Gradient-based Localization

Deep Learning

Belharbi

for histopathological breast cancer image classiﬁcation