Artificial Intelligence and Deep Learning in Pathology

Artificial Intelligence and Deep Learning in Pathology
Copyright
Dedication
Contributors
Preface
1. The evolution of machine learning: past, present, and future
	Introduction
	Rules-based versus machine learning: a deeper look
	Varieties of machine learning
	General aspects of machine learning
	Deep learning and neural networks
	The role of AI in pathology
		Limitations of AI
		General aspects of AI
	References
2. The basics of machine learning: strategies and techniques
	Introduction
	Shallow learning
		Geometric (distance-based) models
		The K-Means Algorithm (KM)
		Probabilistic models
		Decision Trees and Random Forests
	The curse of dimensionality and principal component analysis
	Deep learning and the artificial neural network
		Neuroscience 101
		The rise of the machines
		The basic ANN
		The weights in an ANN
		Learning from examples; Backprop and stochastic gradient descent.
		Convolutional Neural Networks
			Overview
			Detailed explanation
	Overfitting and underfitting
	Things to come
	References
3. Overview of advanced neural network architectures
	Introduction
	Network depth and residual connections
	Autoencoders and unsupervised pretraining
	Transfer learning
	Generative models and generative adversarial networks
	Recurrent neural networks
	Reinforcement learning
	Ensembles
	Genetic algorithms
	References
4. Complexity in the use of artificial intelligence in anatomic pathology
	Introduction
	Life before machine learning
	Multilabel classification
		Single object detection
	Multiple objects
	Advances in multilabel classification
	Graphical neural networks
		Capsule networks
	Weakly supervised learning
	Synthetic data
	N-shot learning
	One-class learning
		Risk analysis
	General considerations
	Summary and conclusions
	References
5. Dealing with data: strategies of preprocessing data
	Introduction
	Overview of preprocessing
	Feature selection, extraction, and correction
	Feature transformation, standardization, and normalization
	Feature engineering
	Mathematical approaches to dimensional reduction
	Dimensional reduction in deep learning
	Imperfect class separation in the training set
	Fairness and bias in machine learning
	Summary
	References
6. Digital pathology as a platform for primary diagnosis and augmentation via deep learning
	Introduction
	Digital imaging in pathology
	Telepathology
	Whole slide imaging
	Whole slide image viewers
	Whole slide image data and workflow management
	Selection criteria for a whole slide scanner
	Evolution of whole slide imaging systems
	Infrastructure requirements and checklist for rolling out high-throughput whole slide imaging workflow solution
	Whole slide imaging and primary diagnosis
	Whole slide imaging and image analysis
	Whole slide imaging and deep learning
	Conclusions
	References
7. Applications of artificial intelligence for image enhancement in pathology
	Introduction
	Common machine learning tasks
		Classification
		Segmentation
		Image translation and style transfer
	Commonly used deep learning methodologies
		Convolutional neural networks
		U-nets
		Generative adversarial networks and their variants
	Common training and testing practices
		Dataset preparation and preprocessing
		Loss functions
		Metrics
	Deep learning for microscopy enhancement in histopathology
		Stain color normalization
		Mode switching
		In silico labeling
		Super-resolution, extended depth-of-field, and denoising
	Deep learning for computationally aided diagnosis in histopathology
		A rationale for AI-assisted imaging and interpretation
		Approaches to rapid histology interpretations
	Future prospects
	References
8. Precision medicine in digital pathology via image analysis and machine learning
	Introduction
		Precision medicine
		Digital pathology
	Applications of image analysis and machine learning
		Knowledge-driven image analysis
		Machine learning for image segmentation
		Deep learning for image segmentation
		Spatial resolution
		Machine learning on extracted data
		Beyond augmentation
	Practical concepts and theory of machine learning
		Machine learning and digital pathology
		Common techniques
		Supervised learning
			Naïve Bayes assumption–based methods
			Logistic regression–based methods
			Support vector–based methods
			Nonparametric, k-nearest neighbor–based methods
			Random forests
		Unsupervised learning
	Image-based digital pathology
		Conventional approaches to image analysis
		Deep learning on images
	Regulatory concerns and considerations
	References
9. Artificial intelligence methods for predictive image-based grading of human cancers
	Introduction
	Tissue preparation and staining
	Image acquisition
	Stain normalization
	Unmixing of immunofluorescence spectral images
	Automated detection of tumor regions in whole-slide images
		Localization of diagnostically relevant regions of interest in whole-slide images
		Tumor detection
	Image segmentation
		Nuclear and epithelial segmentation in IF images
		Nuclei detection and segmentation in H&E images
		Epithelial segmentation in H&E images
		Mitotic figure detection
		Ring segmentation
	Protein biomarker features
	Morphological features for cancer grading and prognosis
	Modeling
		Cox proportional hazards model
		Neural networks
		Decision trees and random forests
		SVM-based methods: Survival-SVM, SVCR, and SVRc
		Feature selection tools
	Ground truth data for AI-based features
	Conclusion
	References
10. Artificial intelligence and the interplay between tumor and immunity
	Introduction
	Immune surveillance and immunotherapy
	Identifying TILs with deep learning
	Multiplex immunohistochemistry with digital pathology and deep learning
	Vendor platforms
	Conclusion
	References
11. Overview of the role of artificial intelligence in pathology: the computer as a pathology digital assistant
	Introduction
	Computational pathology: background and philosophy
		The current state of diagnostics in pathology and the evolving computational opportunities: “why now?”
		Digital pathology versus computational pathology
		Data on scale
	Machine learning tools in computational pathology: types of artificial intelligence
	The need for human intelligence–artificial intelligence partnerships
	Human transparent machine learning approaches
		Explainable artificial intelligence
		Cognitive artificial intelligence
		Human in the loop
		One-shot learning
	Image-based computational pathology
		Core premise of image analytics: what is a high-resolution image?
		The targets of image-based calculations
	First fruits of computational pathology: the evolving digital assistant
		The digital assistant for quality control
		The digital assistant for histological object segmentation
			Nuclei
			Mitoses
			Tumor-infiltrating lymphocytes
			Glands and acini
		The digital assistant in immunohistochemistry
		The digital assistant in tissue classification
		The digital assistant in finding metastases
		The digital assistant in predictive modeling and precision medicine
		The digital assistant for anatomical simulation learning
		The digital assistant for image-omics data fusion
	Artificial intelligence and regulatory challenges
	Educating machines–educating us: learning how to learn with machines
	References
Index