A comparison of sampling strategies for histological image analysis

One of the foremost and challenging tasks in hematoxylin and eosin stained histological image analysis is to reduce color variation present among images, which may significantly affect the performance of computer-aided histological image analysis. In this regard, the paper introduces a new rough-fuzzy circular clustering algorithm for stain color normalization. It judiciously integrates the merits of both fuzzy and rough sets. While the theory of rough sets deals with uncertainty, vagueness, and incompleteness in stain class definition, fuzzy set handles the overlapping nature of histochemical stains. The proposed circular clustering algorithm works on a weighted hue histogram, which considers both saturation and local neighborhood information of the given image. A new dissimilarity measure is introduced to deal with the circular nature of hue values. Some new quantitative measures are also proposed to evaluate the color constancy after normalization. The performance of the proposed method, along with a comparison with other state-of-the-art methods, is demonstrated on several publicly available standard data sets consisting of hematoxylin and eosin stained histological images.

Breast cancer will affect 1 in 8 women in their lifetime. Approximately 20-30% of all breast cancer cases become metastatic. This study was designed to evaluate the effect of tumor met knockdown with individual and combination drug therapy. In this mouse model, with 3 per group, we used the following chemotherapeutic agents: a vehicle control, carboplatin (20mg/kg i.p.), Nutlin-3a (200mg/kg p.o.) and a combination group with the same doses. Nutlin-3 is a small-molecule inhibitor that binds to the p53-binding pocket of MDM2, and activates the p53 pathway. This has been shown to be effective in mouse models of cancer that have a wild-type p53, including prostate, leukemia and multiple myeloma. The lungs were removed and examined for metastatic lung lesions, fixed in 10% neutral buffered formalin (NBF), processed into a paraffin block, microtomed, and slides were stained with H&E. Ki67 immunostain was performed to clearly define the met lesions and the whole slide was imaged with an Aperio Digital Imaging System. The Image Scope positive pixel algorithm was used to quantitate the met lesions in the lungs. The data revealed upon drug treatment the number of mets in the lungs were unchanged in the carboplatin and nutlin-3a individual treatment groups compared to the vehicle group. The lungs in the combination group had the lowest number of tumors with both visual and quantitative image analysis compared to the vehicle control group. The Aperio automated positive pixel count for Ki67 immunostain was consistent with the H&E pathology read. The values decreased with combination treatment (control vs. combination 20mg/kg carboplatin, 200mg/kg nutlin 3a) and were 6% and 1.7% respectively. It was concluded that mice treated with carboplatin and Nutlin-3a combination therapy had fewer metastatic breast cancer mets in the lungs. The individual drug doses were low compared to previous studies and did not have an effect as was expected. This finding suggests that the combination of a traditional chemotherapy with a small protein inhibitor could inhibit the spread of metastatic breast cancer. Citation Format: Margaret Strack, Eva Tonsing-Carter, Tony Sinn, Tiaishia Spragin, Kacie Peterson, Barbara Bailey, Karen Pollok, George Sandusky. Histologic and whole slide quantitative image analysis of lung metastases in a mouse breast cancer model using carboplatin and Nutlin 3A alone and in combination. [abstract]. In: Proceedings of the 106th Annual Meeting of the American Association for Cancer Research; 2015 Apr 18-22; Philadelphia, PA. Philadelphia (PA): AACR; Cancer Res 2015;75(15 Suppl):Abstract nr 5196. doi:10.1158/1538-7445.AM2015-5196

Features extracted from cell networks have become popular tools in histological image analysis. However, existing features do not take sufficient advantage of the cycle structure present within the cell networks. We introduce a new class of network cycle features that take advantage of such structures. We demonstrate the utility of these features for automated prostate cancer scoring using histological images. Prostate cancer is commonly scored by pathologists using the Gleason grading system and our automated system based upon network cycle features serves an important need in making this process less labor-intensive and more reproducible. Our system first extracts the cells from the histological images, computes networks from the cell locations and then computes features based upon statistics for the different cycles present in these networks. Using an SVM (Support Vector Machine) classifier on these features, we demonstrate the efficacy of our system in distinguishing between grade 3 and grade 4 prostate tumors. We also show the superiority of our approach over previously developed systems for this problem based upon texture features, fractal features and alternative network features.

Evaluation of genetic mutations in cancers is important because distinct mutational profiles help determine individualized drug therapy. However, molecular analyses are not routinely performed in all cancers because they are expensive, time-consuming and not universally available. Artificial intelligence (AI) has shown the potential to determine a wide range of genetic mutations on histologic image analysis. Here, we assessed the status of mutation prediction AI models on histologic images by a systematic review. A literature search using the MEDLINE, Embase and Cochrane databases was conducted in August 2021. The articles were shortlisted by titles and abstracts. After a full-text review, publication trends, study characteristic analysis and comparison of performance metrics were performed. Twenty-four studies were found mostly from developed countries, and their number is increasing. The major targets were gastrointestinal, genitourinary, gynecological, lung and head and neck cancers. Most studies used the Cancer Genome Atlas, with a few using an in-house dataset. The area under the curve of some of the cancer driver gene mutations in particular organs was satisfactory, such as 0.92 of BRAF in thyroid cancers and 0.79 of EGFR in lung cancers, whereas the average of all gene mutations was 0.64, which is still suboptimal. AI has the potential to predict gene mutations on histologic images with appropriate caution. Further validation with larger datasets is still required before AI models can be used in clinical practice to predict gene mutations.

As is known, pathologists visually examine the tissue distributions by using microscopes traditionally. The rise in digital image processing and machine learning also allows high-performance computerized analysis of histology images taken with modern imaging systems. In general, histological image segmentation is the first step in the quantitative analysis of histology images. Therefore, a high-accuracy segmentation is essential for histology image analysis in most cases. In this paper, we performed a deep Convolutional Neural Networks (CNNs) based nuclei segmentation study on colon histology images. By using the U-Net biomedical image segmentation model, it is aimed to classify each pixel in colon histology images into one of the following 6 types of nucleus: epithelial, lymphocyte, plasma, eosinophil, neutrophil, connective tissue or classify it as the image background. In comprehensive experimental tests performed on Colon Nuclei Identification and Counting (CoNIC) Challenge dataset, commonly used segmentation and classification metrics were measured, and promising segmentation performances were achieved.

The analysis of histological images is based on visual assessment of tissues by specialists using an optical microscopy. This task can be time-consuming and challenging, mainly due to the complexity of the structures and diseases under investigation. These facts have motivated the development of computational methods to support specialists in research and decision-making. Despite the different computational strategies available in the literature, the solutions based on genetic algorithm have not been fully explored to provide the best combination of features, selection algorithms and classifiers. In this paper, we describe an approach based on genetic algorithm able to evaluate a significant number of features, selection methods and classifiers in order to provide an acceptable association for the diagnosis and pattern recognition of non-Hodgkin lymphomas and colorectal cancer. The chromosomal structure was represented with four genes. The evaluation and selection of individuals, as well as the crossover and mutation processes, were defined to distinguish the groups under investigation, with the highest AUC value and the smallest number of features. The tests were performed considering 1512 features from histological images, different population sizes and number of iterations. An initial population of 50 individuals and 50 iterations provided the best result (AUC value of 0.984) for the colorectal histological images. For non-Hodgkin lymphoma images, the best result (AUC value of 0.947) was obtained with a population of 500 individuals and 50 iterations. The proposed methodology with detailed information regarding the methods, features and best associations are relevant contributions for the community interested in the study of pattern recognition of colorectal cancer and lymphomas.

Iron overload has been measured in 100 hepatic biopsies by three different methods: (i) biochemical assay of the liver iron concentration (LIC), (ii) histological grading (HISTO) and (iii) automated image analysis with a Leitz Texture Analysis System by estimating two parameters, (a) the total iron area (TIA) and (b) the first grey level step (FGLS) at which the iron is detected. Image analysis appears as a specific, sensitive, quick, reproducible and valid method.

Texture analysis (TA) of histological images has recently received attention as an automated method of characterizing liver fibrosis. The colored staining methods used to identify different tissue components reveal various patterns that contribute in different ways to the digital texture of the image. A histological digital image can be represented with various color spaces. The approximation processes of pixel values that are carried out while converting between different color spaces can affect image texture and subsequently could influence the performance of TA. Conventional TA is carried out on grey scale images, which are a luminance approximation to the original RGB (Red, Green, and Blue) space. Currently, grey scale is considered sufficient for characterization of fibrosis but this may not be the case for sophisticated assessment of fibrosis or when resolution conditions vary. This paper investigates the accuracy of TA results on three color spaces, conventional grey scale, RGB, and Hue-Saturation-Intensity (HSI), at different resolutions. The results demonstrate that RGB is the most accurate in texture classification of liver images, producing better results, most notably at low resolution. Furthermore, the green channel, which is dominated by collagen fiber deposition, appears to provide most of the features for characterizing fibrosis images. The HSI space demonstrated a high percentage error for the majority of texture methods at all resolutions, suggesting that this space is insufficient for fibrosis characterization. The grey scale space produced good results at high resolution; however, errors increased as resolution decreased.

The underlying objective of epidemiological investigations is to extrapolate results from a sample to the relevant population. The simplest way of achieving this is to adopt a sampling strategy in which each individual in the population has the same chance of being selected--that is, to employ an 'equal probability of selection method' (epsem). The easiest ways of achieving this are to use simple random sampling or stratified random sampling with a constant sampling fraction. These strategies are often impracticable, however, particularly in large investigations covering a wide geographical area where resource implications dictate a more complex approach such as multi-stage or cluster sampling. Following detailed definitions and appropriate illustrations of these terms, the main purpose of this paper is to provide a working guide of how to achieve epsem using these various random sampling techniques. In brief, for multi-stage sampling with the rare feature of equal-sized first stage units, epsem is achieved by applying the above simple or stratified approaches to the first stage units. Even in the more realistic scenario of unequal first stage units, the same options apply provided that a fixed proportion of second stage units are to be selected (cluster sampling is in fact just one example of this, with 100% sampling of second stage units). If on the other hand a fixed number of second stage units are to be selected then for epsem the first stage units should be selected with each one having a probability proportionate to its size.

OBJECTIVES/GOALS: Our objective is to investigate the antibacterial and regenerative properties of a novel AHA dental coating for the prevention and treatment of deep caries (cavities). Further, we aim to investigate and compare these properties through in vivo murine models and assessment on human saliva samples and human pulpal cells collected from clinical samples. METHODS/STUDY POPULATION: In vitro antibacterial studies were performed by collecting and culturing human salivary bacteria with AHA substrates and quantifying survival of cariogenic Sm (S. mutans). In vivo, C57BL/6 mice were treated with AHA composite fillings, infected with Sm clinical isolates, and fed a high sucrose diet with cavity formation assessment after 6 weeks. To evaluate regeneration, mice were similarly given composite with AHA or MTA (standard of care) upon pulpal exposure with regeneration quantified by microCT and histological analysis of dentin bridge formation, ALP production, and odontoblast migration. In vitro, AHA substrates were cultured with MC3T3-E1 pre-osteoblast cells and dental pulp stem cells obtained from clinical samples over 21 days, with mineralization and ALP assessed indicating osteogenesis. RESULTS/ANTICIPATED RESULTS: In vivo studies have shown the reduction of cavity formation in mice treated with AHA as well as dentin regeneration upon pulpal exposure using microCT and histological image analysis. Coinciding with these findings, AHA substrates eradicated cariogenic Sm in human saliva samples and single species cultures in vitro. Further, preliminary results in vitro have shown increased mineralization of MC3T3-E1 cells when cultured with AHA substrates in comparison to uncoated substrates. We anticipate similar increased mineralization as well as increased ALP production of human pulpal stem cells from clinical samples when cultured with AHA substrates, suggesting osteogenesis. Further, we anticipate increased odontoblast migration and ALP production upon additional analysis of in vivo tissue samples. DISCUSSION/SIGNIFICANCE: This work will elucidate the antibacterial and regenerative properties of AHA dental coatings. These results further support the translation of AHA coatings into the clinic as a novel therapeutic for the prevention and treatment of dental decay, which may overcome the limitations associated with the current treatments.

When multiple instance learning meets foundation models: Advancing histological whole slide image analysis.

Random sampling and reconstruction of concentrated signals in a reproducing kernel space

A systematic assessment of goblet cell sampling of the bulbar conjunctiva by impression cytology

Low spatial resolution satellite sensors provide information over relatively large targets with typical pixel resolutions of hundreds of km2. However, the spatial scales of ground measurements are usually much smaller. Such differences in spatial scales makes the interpretation of comparisons between quantities derived from low resolution sensors and ground measurements particularly difficult. It also highlights the importance of developing appropriate sampling strategies when designing ground campaigns for validation studies of low resolution sensors. We make use of statistical modelling of high resolution surface shortwave radiation budget (SSRB) data to look into this problem. A spatial model that describes the SSRB over a selected region is proposed, and the impact of different sampling schemes in the performance of the model is analysed. Both systematic and random sampling schemes can efficiently represent the full observations set. **Current affiliation: Meteorological Office, Hadley Centre for Climate Prediction and Research, UK

Deep convolutional neural networks for automatic classification of gastric carcinoma using whole slide images in digital histopathology