A deep learning model for breast ductal carcinoma in situ classification in whole slide images

The pathological differential diagnosis between breast ductal carcinoma in situ (DCIS) and invasive ductal carcinoma (IDC) is of pivotal importance for determining optimum cancer treatment(s) and clinical outcomes. Since conventional diagnosis by pathologists using microscopes is limited in terms of human resources, it is necessary to develop new techniques that can rapidly and accurately diagnose large numbers of histopathological specimens. Computational pathology tools which can assist pathologists in detecting and classifying DCIS and IDC from whole slide images (WSIs) would be of great benefit for routine pathological diagnosis. In this paper, we trained deep learning models capable of classifying biopsy and surgical histopathological WSIs into DCIS, IDC, and benign. We evaluated the models on two independent test sets (n=1,382, n=548), achieving ROC areas under the curves (AUCs) up to 0.960 and 0.977 for DCIS and IDC, respectively.

An Open-Source Whole Slide Image Registration Workflow at Cellular Precision Using Fiji, QuPath and Elastix

Image analysis workflows for Histology increasingly require the correlation and combination of measurements across several whole slide images. Indeed, for multiplexing, as well as multimodal imaging, it is indispensable that the same sample is imaged multiple times, either through various systems for multimodal imaging, or using the same system but throughout rounds of sample manipulation (e.g. multiple staining sessions). In both cases slight deformations from one image to another are unavoidable, leading to an imperfect superimposition Redundant and thus a loss of accuracy making it difficult to link measurements, in particular at the cellular level. Using pre-existing software components and developing missing ones, we propose a user-friendly workflow which facilitates the nonlinear registration of whole slide images in order to reach sub-cellular resolution level. The set of whole slide images to register and analyze is at first defined as a QuPath project. Fiji is then used to open the QuPath project and perform the registrations. Each registration is automated by using an elastix backend, or semi-automated by using BigWarp in order to interactively correct the results of the automated registration. These transformations can then be retrieved in QuPath to transfer any regions of interest from an image to the corresponding registered images. In addition, the transformations can be applied in QuPath to produce on-the-fly transformed images that can be displayed on top of the reference image. Thus, relevant data can be combined and analyzed throughout all registered slides, facilitating the analysis of correlative results for multiplexed and multimodal imaging.

TIAToolbox: An End-to-End Toolbox for Advanced Tissue Image Analytics

Computational Pathology (CPath) has seen rapid growth in recent years, driven by advanced deep learning (DL) algorithms. These algorithms typically share the same sequence of steps. However, due to the sheer size and complexity of handling large multi-gigapixel whole-slide images, there is no open-source software library that provides a generic end-to-end API for pathology image analysis using best practices for CPath. Most researchers have designed custom pipelines from the bottom-up, restricting the development of advanced CPath algorithms to specialist users. To help overcome this bottleneck, we present TIAToolbox, a Python toolbox designed to make CPath more accessible to new and advanced CPath scientists and pathologists alike. We provide a usable and adaptable library with efficient, cutting-edge and unit-tested tools for data loading, pre-processing, model inference, post-processing and visualization. This enables all kinds of users to easily build upon recent DL developments in the CPath literature. TIAToolbox provides a user-friendly modular API to enable seamless integration of advanced DL algorithms. We show with the help of examples how state-of-the-art DL algorithms can be streamlined using TIAToolbox.

ProxyFL: Decentralized Federated Learning through Proxy Model Sharing

Institutions in highly regulated domains such as finance and healthcare often have restrictive rules around data sharing. Federated learning is a distributed learning framework that enables multi-institutional collaborations on decentralized data with improved protection for each collaborator’s data privacy. In this paper, we propose a communication-efficient scheme for decentralized federated learning called ProxyFL, or proxy-based federated learning. Each participant in ProxyFL maintains two models, a private model, and a publicly shared proxy model designed to protect the participant’s privacy. Proxy models allow efficient information exchange among participants using the PushSum method without the need of a centralized server. The proposed method eliminates a significant limitation of canonical federated learning by allowing model heterogeneity; each participant can have a private model with any architecture. Furthermore, our protocol for communication by proxy leads to stronger privacy guarantees using differential privacy analysis. Experiments on popular image datasets, and a pan-cancer diagnostic problem using over 30,000 high-quality gigapixel histology whole slide images, show that ProxyFL can outperform existing alternatives with much less communication overhead and stronger privacy.

Transferability of Deep Learning Models for Focus Quality Assessment in Digital Pathology

Out-of-focus sections of whole slide images are a significant source of false positives and other systematic errors in clinical diagnoses. As a result, focus quality assessment (FQA) methods must be able to quickly and accurately differentiate between focus levels in a scan. Recently, deep learning methods using convolutional neural networks (CNNs) have been adopted for FQA. However, the biggest obstacles impeding their wide usage in clinical workflows are their generalizability across different test conditions and their potentially high computational cost. In this study, we focus on the transferability and scalability of CNN-based FQA approaches. We carry out an investigation on ten architecturally diverse networks using five datasets with stain and tissue diversity. We evaluate the computational complexity of each network and scale this to realistic applications involving hundreds of whole slide images. We assess how well each full model transfers to a separate, unseen dataset without fine-tuning. We show that shallower networks transfer well when used on small input patch sizes, while deeper networks work more effectively on larger inputs. Furthermore, we introduce neural architecture search (NAS) to the field and learn an automatically designed low-complexity CNN architecture using differentiable architecture search which achieved competitive performance relative to established CNNs.