Automatic Non-rigid Histological Image Registration (ANHIR) challenge was organized to compare the performance of image registration algorithms on several kinds of microscopy histology images in a fair and independent manner. We have assembled 8 datasets, containing 355 images with 18 different stains, resulting in 481 image pairs to be registered. Registration accuracy was evaluated using manually placed landmarks. In total, 256 teams registered for the challenge, 10 submitted the results, and 6 participated in the workshop. Here, we present the results of 7 well-performing methods from the challenge together with 6 well-known existing methods. The best methods used coarse but robust initial alignment, followed by non-rigid registration, used multiresolution, and were carefully tuned for the data at hand. They outperformed off-the-shelf methods, mostly by being more robust. The best methods could successfully register over 98% of all landmarks and their mean landmark registration accuracy (TRE) was 0.44% of the image diagonal. The challenge remains open to submissions and all images are available for download.

ANHIR: Automatic Non-Rigid Histological Image Registration Challenge

Automated segmentation of cell membranes to evaluate HER2 status in whole slide images using a modified deep learning network.

Image segmentation is one of the most basic tasks in computer vision and remains an initial step of many applications. In this paper, we focus on interactive image segmentation (IIS), often referred to as foreground-background separation or object extraction, guided by user interaction. We provide an overview of the IIS literature by covering more than 150 publications, especially recent works that have not been surveyed before. Moreover, we try to give a comprehensive classification of them according to different viewpoints and present a general and concise comparison of the most recent published works. Furthermore, we survey widely used datasets, evaluation metrics, and available resources in the field of IIS.

/pdf/a-survey-of-recent-interactive-image-segmentation-methods-n00uq3mgmb.pdf

A survey of recent interactive image segmentation methods

Image segmentation is widely used as an initial phase of many image analysis tasks. It is often advantageous to first group pixels into compact, edge-respecting superpixels, because these reduce the size of the segmentation problem and thus the segmentation time by an order of magnitudes. In addition, features calculated from superpixel regions are more robust than features calculated from fixed pixel neighborhoods. We present a fast and general multiclass image segmentation method consisting of the following steps: (i) computation of superpixels; (ii) extraction of superpixel-based descriptors; (iii) calculating image-based class probabilities in a supervised or unsupervised manner; and (iv) regularized superpixel classification using graph cut. We apply this segmentation pipeline to five real-world medical imaging applications and compare the results with three baseline methods: pixelwise graph cut segmentation, supertexton-based segmentation, and classical superpixel-based segmentation. On all datasets, we outperform the baseline results. We also show that unsupervised segmentation is surprisingly efficient in many situations. Unsupervised segmentation provides similar results to the supervised method but does not require manually annotated training data, which is often expensive to obtain.

Supervised and unsupervised segmentation using superpixels, model estimation, and graph cut

We present a novel approach to image manipulation and understanding by simultaneously learning to segment object masks, paste objects to another background image, and remove them from original images. For this purpose, we develop a novel generative model for compositional image generation, SEIGAN (Segment-Enhance-Inpaint Generative Adversarial Network), which learns these three operations together in an adversarial architecture with additional cycle consistency losses. To train, SEIGAN needs only bounding box supervision and does not require pairing or ground truth masks. SEIGAN produces better generated images (evaluated by human assessors) than other approaches and produces high-quality segmentation masks, improving over other adversarially trained approaches and getting closer to the results of fully supervised training.

SEIGAN: Towards Compositional Image Generation by Simultaneously Learning to Segment, Enhance, and Inpaint.

A main problem with reproducing machine learning publications is the variance of metric implementations across papers. A lack of standardization leads to different behavior in mechanisms such as checkpointing, learning rate schedulers or early stopping, that will influence the reported results. For example, a complex metric such as Fréchet inception distance (FID) for synthetic image quality evaluation (Heusel et al., 2017) will differ based on the specific interpolation method used.

/pdf/torchmetrics-measuring-reproducibility-in-pytorch-3ptqjujv.pdf

TorchMetrics - Measuring Reproducibility in PyTorch

Region growing is a classical image segmentation method based on hierarchical region aggregation using local similarity rules. Our proposed method differs from classical region growing in three important aspects. First, it works on the level of superpixels instead of pixels, which leads to a substantial speed-up. Second, our method uses learned statistical shape properties that encourage plausible shapes. In particular, we use ray features to describe the object boundary. Third, our method can segment multiple objects and ensure that the segmentations do not overlap. The problem is represented as an energy minimization and is solved either greedily or iteratively using graph cuts. We demonstrate the performance of the proposed method and compare it with alternative approaches on the task of segmenting individual eggs in microscopy images of Drosophila ovaries.

Region growing using superpixels with learned shape prior

We present an image processing pipeline which accepts a large number of images, containing spatial expression information for thousands of genes in Drosophila imaginal discs. We assume that the gene activations are binary and can be expressed as a union of a small set of non-overlapping spatial patterns, yielding a compact representation of the spatial activation of each gene. This lends itself well to further automatic analysis, with the hope of discovering new biological relationships. Traditionally, the images were labeled manually, which was very time consuming. The key part of our work is a binary pattern dictionary learning algorithm, that takes a set of binary images and determines a set of patterns, which can be used to represent the input images with a small error. We also describe the preprocessing phase, where input images are segmented to recover the activation images and spatially aligned to a common reference. We compare binary pattern dictionary learning to existing alternative methods on synthetic data and also show results of the algorithm on real microscopy images of the Drosophila imaginal discs.

Binary Pattern Dictionary Learning for Gene Expression Representation in Drosophila Imaginal Discs

Drosophila melanogaster is a well-known model organism that can be used for studying oogenesis (egg chamber development) including gene expression patterns. Standard analysis methods require manual segmentation of individual egg chambers, which is a difficult and time-consuming task. We present an image processing pipeline to detect and localize Drosophila egg chambers that consists of the following steps: (i) superpixel-based image segmentation into relevant tissue classes; (ii) detection of egg center candidates using label histograms and ray features; (iii) clustering of center candidates and; (iv) area-based maximum likelihood ellipse model fitting. Our proposal is able to detect 96% of human-expert annotated egg chambers at relevant developmental stages with less than 1% false-positive rate, which is adequate for the further analysis.

Jiří Borovec

Papers

TorchMetrics - Measuring Reproducibility in PyTorch

Supervised and unsupervised segmentation using superpixels, model estimation, and graph cut

Region growing using superpixels with learned shape prior

Binary Pattern Dictionary Learning for Gene Expression Representation in Drosophila Imaginal Discs

Detection and Localization of Drosophila Egg Chambers in Microscopy Images