Crowdsourcing scoring of immunohistochemistry images: Evaluating Performance of the Crowd and an Automated Computational Method

Crowdsourcing with the drift diffusion model of decision making

Crowdsourcing involves the use of annotated labels with unknown reliability to estimate ground truth labels in datasets. A common task in crowdsourcing involves estimating reliabilities of annotators (such as through the sensitivities and specificities of annotators in the binary label setting). In the literature, beta or dirichlet distributions are typically imposed as priors on annotator reliability. In this study, we investigated the use of a neuroscientifically validated model of decision making, known as the drift-diffusion model, as a prior on the annotator labeling process. Two experiments were conducted on synthetically generated data with non-linear (sinusoidal) decision boundaries. Variational inference was used to predict ground truth labels and annotator related parameters. Our method performed similarly to a state-of-the-art technique (SVGPCR) in prediction of crowdsourced data labels and prediction through a crowdsourced-generated Gaussian process classifier. By relying on a neuroscientifically validated model of decision making to model annotator behavior, our technique opens the avenue of predicting neuroscientific biomarkers of annotators, expanding the scope of what may be learnt about annotators in crowdsourcing tasks.

Synergizing Deep Learning-Enabled Preprocessing and Human-AI Integration for Efficient Automatic Ground Truth Generation

The progress of incorporating deep learning in the field of medical image interpretation has been greatly hindered due to the tremendous cost and time associated with generating ground truth for supervised machine learning, alongside concerns about the inconsistent quality of images acquired. Active learning offers a potential solution to these problems of expanding dataset ground truth by algorithmically choosing the most informative samples for ground truth labeling. Still, this effort incurs the costs of human labeling, which needs minimization. Furthermore, automatic labeling approaches employing active learning often exhibit overfitting tendencies while selecting samples closely aligned with the training set distribution and excluding out-of-distribution samples, which could potentially improve the model's effectiveness. We propose that the majority of out-of-distribution instances can be attributed to inconsistent cross images. Since the FDA approved the first whole-slide image system for medical diagnosis in 2017, whole-slide images have provided enriched critical information to advance the field of automated histopathology. Here, we exemplify the benefits of a novel deep learning strategy that utilizes high-resolution whole-slide microscopic images. We quantitatively assess and visually highlight the inconsistencies within the whole-slide image dataset employed in this study. Accordingly, we introduce a deep learning-based preprocessing algorithm designed to normalize unknown samples to the training set distribution, effectively mitigating the overfitting issue. Consequently, our approach significantly increases the amount of automatic region-of-interest ground truth labeling on high-resolution whole-slide images using active deep learning. We accept 92% of the automatic labels generated for our unlabeled data cohort, expanding the labeled dataset by 845%. Additionally, we demonstrate expert time savings of 96% relative to manual expert ground-truth labeling.

Task design for crowdsourced glioma cell annotation in microscopy images

Crowdsourcing has been used in computational pathology to generate cell and cell nuclei annotations for machine learning. Herein, we broaden its scope to the previously unsolved challenging task of glioma cell detection. This requires multiplexed immunofluorescence microscopy due to diffuse invasiveness and exceptional similarity between glioma cells and reactive astrocytes. In four pilot experiments, we iteratively developed a task design enabling high-quality annotations by crowdworkers on Amazon Mechanical Turk. We applied majority or weighted vote and validated them against ground truth in the final setting. On the base of a YOLO convolutional neural network architecture, we used these consensus labels for training with different image representations regarding colors, intensities, and immmunohistochemical marker combinations. A crowd of 712 workers defined aggregated point annotations in 235 images with an average [Formula: see text] score of 0.627 for majority vote. The networks resulted in acceptable [Formula: see text] scores up to 0.69 for YOLOv8 on average and indicated first evidence for transferability to images lacking tumor markers, especially in IDH-wildtype glioblastoma. Our work confirms feasibility of crowdsourcing to generate labels suitable for training of machine learning tools in the challenging and clinically relevant use case of glioma microenvironment.

Toward a generalizable machine learning workflow for neurodegenerative disease staging with focus on neurofibrillary tangles

Machine learning (ML) has increasingly been used to assist and expand current practices in neuropathology. However, generating large imaging datasets with quality labels is challenging in fields which demand high levels of expertise. Further complicating matters is the often seen disagreement between experts in neuropathology-related tasks, both at the case level and at a more granular level. Neurofibrillary tangles (NFTs) are a hallmark pathological feature of Alzheimer disease, and are associated with disease progression which warrants further investigation and granular quantification at a scale not currently accessible in routine human assessment. In this work, we first provide a baseline of annotator/rater agreement for the tasks of Braak NFT staging between experts and NFT detection using both experts and novices in neuropathology. We use a whole-slide-image (WSI) cohort of neuropathology cases from Emory University Hospital immunohistochemically stained for Tau. We develop a workflow for gathering annotations of the early stage formation of NFTs (Pre-NFTs) and mature intracellular (iNFTs) and show ML models can be trained to learn annotator nuances for the task of NFT detection in WSIs. We utilize a model-assisted-labeling approach and demonstrate ML models can be used to aid in labeling large datasets efficiently. We also show these models can be used to extract case-level features, which predict Braak NFT stages comparable to expert human raters, and do so at scale. This study provides a generalizable workflow for various pathology and related fields, and also provides a technique for accomplishing a high-level neuropathology task with limited human annotations.

Prediction of Five-Year Survival Rate for Rectal Cancer Using Markov Models of Convolutional Features of RhoB Expression on Tissue Microarray

The ability to predict survival in cancer is clinically important because the finding can help patients and physicians make optimal treatment decisions. Artificial intelligence in the context of deep learning has been increasingly realized by the informatics-oriented medical community as a powerful machine-learning technology for cancer research, diagnosis, prediction, and treatment. This paper presents the combination of deep learning, data coding, and probabilistic modeling for predicting five-year survival in a cohort of patients with rectal cancer using images of RhoB expression on biopsies. Using about one-third of the patients' data for testing, the proposed approach achieved 90% prediction accuracy, which is much higher than the direct use of the best pretrained convolutional neural network (70%) and the best coupling of a pretrained model and support vector machines (70%).

Interactions between carbon nanotubes and external structures of SARS-CoV-2 using molecular docking and molecular dynamics

Molecular modeling techniques are used to describe the process of interaction between nanotubes and the main structures of the Covid-19 virus: the envelope protein, the main protease, and the Spike glycoprotein. Molecular docking studies show that the ligands have interaction characteristics capable of adsorbing the structures. Molecular dynamics simulations provide information on the mean squared deviation of atomic positions ​​between 0.5 and 3.0 Å. The Gibbs free energy model and solvent accessible surface area approaches are used. Through the results obtained through molecular dynamics simulations, it is noted that the zig-zag nanotube prefers to interact with E-pro, M-pro, and S-gly, respectively. Molecular couplings and free energy showed that the S-gly active site residues strongly interact with zigzag, chiral, and armchair nanotubes, in this order. The interactions demonstrated in this manuscript may predict some promising candidates for virus antagonists, which may be confirmed through experimental approaches.

Improving Crowdsourcing-Based Image Classification Through Expanded Input Elicitation and Machine Learning

This work investigates how different forms of input elicitation obtained from crowdsourcing can be utilized to improve the quality of inferred labels for image classification tasks, where an image must be labeled as either positive or negative depending on the presence/absence of a specified object. Five types of input elicitation methods are tested: binary classification (positive or negative); the (x, y)-coordinate of the position participants believe a target object is located; level of confidence in binary response (on a scale from 0 to 100%); what participants believe the majority of the other participants' binary classification is; and participant's perceived difficulty level of the task (on a discrete scale). We design two crowdsourcing studies to test the performance of a variety of input elicitation methods and utilize data from over 300 participants. Various existing voting and machine learning (ML) methods are applied to make the best use of these inputs. In an effort to assess their performance on classification tasks of varying difficulty, a systematic synthetic image generation process is developed. Each generated image combines items from the MPEG-7 Core Experiment CE-Shape-1 Test Set into a single image using multiple parameters (e.g., density, transparency, etc.) and may or may not contain a target object. The difficulty of these images is validated by the performance of an automated image classification method. Experiment results suggest that more accurate results can be achieved with smaller training datasets when both the crowdsourced binary classification labels and the average of the self-reported confidence values in these labels are used as features for the ML classifiers. Moreover, when a relatively larger properly annotated dataset is available, in some cases augmenting these ML algorithms with the results (i.e., probability of outcome) from an automated classifier can achieve even higher performance than what can be obtained by using any one of the individual classifiers. Lastly, supplementary analysis of the collected data demonstrates that other performance metrics of interest, namely reduced false-negative rates, can be prioritized through special modifications of the proposed aggregation methods.

How the variability between computer-assisted analysis procedures evaluating immune markers can influence patients' outcome prediction

Differences between computer-assisted image analysis (CAI) algorithms may cause discrepancies in the identification of immunohistochemically stained immune biomarkers in biopsies of breast cancer patients. These discrepancies have implications for their association with disease outcome. This study aims to compare three CAI procedures (A, B and C) to measure positive marker areas in post-neoadjuvant chemotherapy biopsies of patients with triple-negative breast cancer (TNBC) and to explore the differences in their performance in determining the potential association with relapse in these patients. A total of 3304 digital images of biopsy tissue obtained from 118 TNBC patients were stained for seven immune markers using immunohistochemistry (CD4, CD8, FOXP3, CD21, CD1a, CD83, HLA-DR) and were analyzed with procedures A, B and C. The three methods measure the positive pixel markers in the total tissue areas. The extent of agreement between paired CAI procedures, a principal component analysis (PCA) and Cox multivariate analysis was assessed. Comparisons of paired procedures showed close agreement for most of the immune markers at low concentration. The probability of differences between the paired procedures B/C and B/A was generally higher than those observed in C/A. The principal component analysis, largely based on data from CD8, CD1a and HLA-DR, identified two groups of patients with a significantly lower probability of relapse than the others. The multivariate regression models showed similarities in the factors associated with relapse for procedures A and C, as opposed to those obtained with procedure B. General agreement among the results of CAI procedures would not guarantee that the same predictive breast cancer markers were consistently identified. These results highlight the importance of developing additional strategies to improve the sensitivity of CAI procedures.

Deep neural network models for computational histopathology: A survey

Histopathological images contain rich phenotypic information that can be used to monitor underlying mechanisms contributing to disease progression and patient survival outcomes. Recently, deep learning has become the mainstream methodological choice for analyzing and interpreting histology images. In this paper, we present a comprehensive review of state-of-the-art deep learning approaches that have been used in the context of histopathological image analysis. From the survey of over 130 papers, we review the field's progress based on the methodological aspect of different machine learning strategies such as supervised, weakly supervised, unsupervised, transfer learning and various other sub-variants of these methods. We also provide an overview of deep learning based survival models that are applicable for disease-specific prognosis tasks. Finally, we summarize several existing open datasets and highlight critical challenges and limitations with current deep learning approaches, along with possible avenues for future research.

Development of an Automated Image Analyzer for Microvessel Density Measurement in Bone Marrow Biopsies

Angiogenesis is important for the proliferation and survival of multiple myeloma (MM) cells. Bone marrow (BM) microvessel density (MVD) is a useful marker of angiogenesis and an increase in MVD can be used as a marker of poor prognosis in MM patients. We developed an automated image analyzer to assess MVD from images of BM biopsies stained with anti-CD34 antibodies using two color models. MVD was calculated by merging images from the red and hue channels after eliminating non-microvessels. The analyzer results were compared with those obtained by two experienced hematopathologists in a blinded manner using the 84 BM samples of MM patients. Manual assessment of the MVD by two hematopathologists yielded mean±SD values of 19.4±11.8 and 20.0±11.8. The analyzer generated a mean±SD of 19.5±11.2. The intraclass correlation coefficient (ICC) and Bland-Altman plot of the MVD results demonstrated very good agreement between the automated image analyzer and both hematopathologists (ICC=0.893 [0.840–0.929] and ICC=0.906 [0.859–0.938]). This automated analyzer can provide time- and labor-saving benefits with more objective results in hematology laboratories.

Crawling the German Health Web: Exploratory Study and Graph Analysis

BACKGROUND

The internet has become an increasingly important resource for health information. However, with a growing amount of web pages, it is nearly impossible for humans to manually keep track of evolving and continuously changing content in the health domain. To better understand the nature of all web-based health information as given in a specific language, it is important to identify (1) information hubs for the health domain, (2) content providers of high prestige, and (3) important topics and trends in the health-related web. In this context, an automatic web crawling approach can provide the necessary data for a computational and statistical analysis to answer (1) to (3).

OBJECTIVE

This study demonstrates the suitability of a focused crawler for the acquisition of the German Health Web (GHW) which includes all health-related web content of the three mostly German speaking countries Germany, Austria and Switzerland. Based on the gathered data, we provide a preliminary analysis of the GHW's graph structure covering its size, most important content providers and a ratio of public to private stakeholders. In addition, we provide our experiences in building and operating such a highly scalable crawler.

METHODS

A support vector machine classifier was trained on a large data set acquired from various German content providers to distinguish between health-related and non-health-related web pages. The classifier was evaluated using accuracy, recall and precision on an 80/20 training/test split (TD1) and against a crowd-validated data set (TD2). To implement the crawler, we extended the open-source framework StormCrawler. The actual crawl was conducted for 227 days. The crawler was evaluated by using harvest rate and its recall was estimated using a seed-target approach.

RESULTS

In total, n=22,405 seed URLs with country-code top level domains .de: 85.36% (19,126/22,405), .at: 6.83% (1530/22,405), .ch: 7.81% (1749/22,405), were collected from Curlie and a previous crawl. The text classifier achieved an accuracy on TD1 of 0.937 (TD2=0.966), a precision on TD1 of 0.934 (TD2=0.954) and a recall on TD1 of 0.944 (TD2=0.989). The crawl yields 13.5 million presumably relevant and 119.5 million nonrelevant web pages. The average harvest rate was 19.76%; recall was 0.821 (4105/5000 targets found). The resulting host-aggregated graph contains 215,372 nodes and 403,175 edges (network diameter=25; average path length=6.466; average degree=1.872; average in-degree=1.892; average out-degree=1.845; modularity=0.723). Among the 25 top-ranked pages for each country (according to PageRank), 40% (30/75) were web sites published by public institutions. 25% (19/75) were published by nonprofit organizations and 35% (26/75) by private organizations or individuals.

CONCLUSIONS

The results indicate, that the presented crawler is a suitable method for acquiring a large fraction of the GHW. As desired, the computed statistical data allows for determining major information hubs and important content providers on the GHW. In the future, the acquired data may be used to assess important topics and trends but also to build health-specific search engines.

Cognitive and Non-Cognitive Outcomes Associated with Student Engagement in a Novel Brain Chemoarchitecture Mapping Course-Based Undergraduate Research Experience

Course-based undergraduate research experiences (CUREs) engage emerging scholars in the authentic process of scientific discovery, and foster their development of content knowledge, motivation, and persistence in the science, technology, engineering, and mathematics (STEM) disciplines. Importantly, authentic research courses simultaneously offer investigators unique access to an extended population of students who receive education and mentoring in conducting scientifically relevant investigations and who are thus able to contribute effort toward big-data projects. While this paradigm benefits fields in neuroscience, such as atlas-based brain mapping of nerve cells at the tissue level, there are few documented cases of such laboratory courses offered in the domain. Here, we describe a curriculum designed to address this deficit, evaluate the scientific merit of novel student-produced brain atlas maps of immunohistochemically-identified nerve cell populations for the rat brain, and assess shifts in science identity, attitudes, and science communication skills of students engaged in the introductory-level Brain Mapping and Connectomics (BM&C) CURE. BM&C students reported gains in research and science process skills following participation in the course. Furthermore, BM&C students experienced a greater sense of science identity, including a greater likelihood to discuss course activities with non-class members compared to their non-CURE counterparts. Importantly, evaluation of student-generated brain atlas maps indicated that the course enabled students to produce scientifically valid products and make new discoveries to advance the field of neuroanatomy. Together, these findings support the efficacy of the BM&C course in addressing the relatively esoteric demands of chemoarchitectural brain mapping.

Structured crowdsourcing enables convolutional segmentation of histology images

MOTIVATION

While deep-learning algorithms have demonstrated outstanding performance in semantic image segmentation tasks, large annotation datasets are needed to create accurate models. Annotation of histology images is challenging due to the effort and experience required to carefully delineate tissue structures, and difficulties related to sharing and markup of whole-slide images.

RESULTS

We recruited 25 participants, ranging in experience from senior pathologists to medical students, to delineate tissue regions in 151 breast cancer slides using the Digital Slide Archive. Inter-participant discordance was systematically evaluated, revealing low discordance for tumor and stroma, and higher discordance for more subjectively defined or rare tissue classes. Feedback provided by senior participants enabled the generation and curation of 20 000+ annotated tissue regions. Fully convolutional networks trained using these annotations were highly accurate (mean AUC=0.945), and the scale of annotation data provided notable improvements in image classification accuracy.

AVAILABILITY AND IMPLEMENTATION

Dataset is freely available at: https://goo.gl/cNM4EL.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Scientific Discovery Games for Biomedical Research

Over the past decade, scientific discovery games (SDGs) have emerged as a viable approach for biomedical research, engaging hundreds of thousands of volunteer players and resulting in numerous scientific publications. After describing the origins of this novel research approach, we review the scientific output of SDGs across molecular modeling, sequence alignment, neuroscience, pathology, cellular biology, genomics, and human cognition. We find compelling results and technical innovations arising in problem-oriented games such as Foldit and Eterna and in data-oriented games such as EyeWire and Project Discovery. We discuss emergent properties of player communities shared across different projects, including the diversity of communities and the extraordinary contributions of some volunteers, such as paper writing. Finally, we highlight connections to artificial intelligence, biological cloud laboratories, new game genres, science education, and open science that may drive the next generation of SDGs.

Crowdsourcing of Histological Image Labeling and Object Delineation by Medical Students

Crowdsourcing in pathology has been performed on tasks that are assumed to be manageable by nonexperts. Demand remains high for annotations of more complex elements in digital microscopic images, such as anatomical structures. Therefore, this paper investigates conditions to enable crowdsourced annotations of high-level image objects, a complex task considered to require expert knowledge. Seventy six medical students without specific domain knowledge who voluntarily participated in three experiments solved two relevant annotation tasks on histopathological images: 1) labeling of images showing tissue regions and 2) delineation of morphologically defined image objects. We focus on methods to ensure sufficient annotation quality including several tests on the required number of participants and on the correlation of participants' performance between tasks. In a set up simulating annotation of images with limited ground truth, we validated the feasibility of a confidence score using full ground truth. For this, we computed a majority vote using weighting factors based on individual assessment of contributors against scattered gold standard annotated by pathologists. In conclusion, we provide guidance for task design and quality control to enable a crowdsourced approach to obtain accurate annotations required in the era of digital pathology.

The diagnosis conundrum: Comparison of crowdsourced and expert assessments of toddlers with high and low risk of autism spectrum disorder

As the diagnosis and treatment of Autism Spectrum Disorder (ASD) becomes a greater public health concern, access to professionals and expert assessment becomes increasingly more challenging. Novel techniques to improve efficiency of assessment of young children with social communication concerns are desperately needed to address significant barriers to accessing diagnostic evaluations. Utilizing crowdsourcing technology, we compared ratings by experts and crowdworkers of social communication behaviors in videos of 18-month-old children. Of note, 563 crowdworkers and 24 expert responses were collected in approximately 7 hr and 57 days, respectively. Summed scores of social communication behaviors observed by crowdworkers showed high correlation to those of experts. These data introduce a novel and perhaps efficient way in which to examine of social communication impairments in toddlers. Autism Research 2018, 11: 1629-1634. © 2018 International Society for Autism Research, Wiley Periodicals, Inc. LAY SUMMARY: This research compared crowdsourced assessments of social communication behaviors to assessments by experts, for toddlers whose behaviors ranged in severity from typical to atypical. Results showed that crowdsourced rankings of social communication behavior significantly correlate with those of experts.

Large-scale medical image annotation with crowd-powered algorithms

Accurate segmentations in medical images are the foundations for various clinical applications. Advances in machine learning-based techniques show great potential for automatic image segmentation, but these techniques usually require a huge amount of accurately annotated reference segmentations for training. The guiding hypothesis of this paper was that crowd-algorithm collaboration could evolve as a key technique in large-scale medical data annotation. As an initial step toward this goal, we evaluated the performance of untrained individuals to detect and correct errors made by three-dimensional (3-D) medical segmentation algorithms. To this end, we developed a multistage segmentation pipeline incorporating a hybrid crowd-algorithm 3-D segmentation algorithm integrated into a medical imaging platform. In a pilot study of liver segmentation using a publicly available dataset of computed tomography scans, we show that the crowd is able to detect and refine inaccurate organ contours with a quality similar to that of experts (engineers with domain knowledge, medical students, and radiologists). Although the crowds need significantly more time for the annotation of a slice, the annotation rate is extremely high. This could render crowdsourcing a key tool for cost-effective large-scale medical image annotation.

Automated analysis of images for molecular quantification in immunohistochemistry

The quantification of the expression of different molecules is a key question in both basic and applied sciences. While protein quantification through molecular techniques leads to the loss of spatial information and resolution, immunohistochemistry is usually associated with time-consuming image analysis and human bias. In addition, the scarce automatic software analysis is often proprietary and expensive and relies on a fixed threshold binarization. Here we describe and share a set of macros ready for automated fluorescence analysis of large batches of fixed tissue samples using FIJI/ImageJ. The quantification of the molecules of interest are based on an automatic threshold analysis of immunofluorescence images to automatically identify the top brightest structures of each image. These macros measure several parameters commonly quantified in basic neuroscience research, such as neuropil density and fluorescence intensity of synaptic puncta, perisomatic innervation and col-localization of different molecules and analysis of the neurochemical phenotype of neuronal subpopulations. In addition, these same macro functions can be easily modified to improve similar analysis of fluorescent probes in human biopsies for diagnostic purposes based on the expression patterns of several molecules.

A Review on the Applications of Crowdsourcing in Human Pathology

The advent of the digital pathology has introduced new avenues of diagnostic medicine. Among them, crowdsourcing has attracted researchers' attention in the recent years, allowing them to engage thousands of untrained individuals in research and diagnosis. While there exist several articles in this regard, prior works have not collectively documented them. We, therefore, aim to review the applications of crowdsourcing in human pathology in a semi-systematic manner. We first, introduce a novel method to do a systematic search of the literature. Utilizing this method, we, then, collect hundreds of articles and screen them against a predefined set of criteria. Furthermore, we crowdsource part of the screening process, to examine another potential application of crowdsourcing. Finally, we review the selected articles and characterize the prior uses of crowdsourcing in pathology.

Deep Learning and Its Applications in Biomedicine

Advances in biological and medical technologies have been providing us explosive volumes of biological and physiological data, such as medical images, electroencephalography, genomic and protein sequences. Learning from these data facilitates the understanding of human health and disease. Developed from artificial neural networks, deep learning-based algorithms show great promise in extracting features and learning patterns from complex data. The aim of this paper is to provide an overview of deep learning techniques and some of the state-of-the-art applications in the biomedical field. We first introduce the development of artificial neural network and deep learning. We then describe two main components of deep learning, i.e., deep learning architectures and model optimization. Subsequently, some examples are demonstrated for deep learning applications, including medical image classification, genomic sequence analysis, as well as protein structure classification and prediction. Finally, we offer our perspectives for the future directions in the field of deep learning.