Multiple-instance-learning-based detection of coeliac disease in histological whole-slide images

Advancing Artificial Intelligence Integration Into the Pathology Workflow: Exploring Opportunities in Gastrointestinal Tract Biopsies

This review aims to present a comprehensive overview of the current landscape of artificial intelligence (AI) applications in the analysis of tubular gastrointestinal biopsies. These publications cover a spectrum of conditions, ranging from inflammatory ailments to malignancies. Moving beyond the conventional diagnosis based on hematoxylin and eosin-stained whole-slide images, the review explores additional implications of AI, including its involvement in interpreting immunohistochemical results, molecular subtyping, and the identification of cellular spatial biomarkers. Furthermore, the review examines how AI can contribute to enhancing the quality and control of diagnostic processes, introducing new workflow options, and addressing the limitations and caveats associated with current AI platforms in this context.

CD, or not CD, that is the question: a digital interobserver agreement study in coeliac disease

OBJECTIVE

Coeliac disease (CD) diagnosis generally depends on histological examination of duodenal biopsies. We present the first study analysing the concordance in examination of duodenal biopsies using digitised whole-slide images (WSIs). We further investigate whether the inclusion of immunoglobulin A tissue transglutaminase (IgA tTG) and haemoglobin (Hb) data improves the interobserver agreement of diagnosis.

DESIGN

We undertook a large study of the concordance in histological examination of duodenal biopsies using digitised WSIs in an entirely virtual reporting setting. Our study was organised in two phases: in phase 1, 13 pathologists independently classified 100 duodenal biopsies (40 normal; 40 CD; 20 indeterminate enteropathy) in the absence of any clinical or laboratory data. In phase 2, the same pathologists examined the (re-anonymised) WSIs with the inclusion of IgA tTG and Hb data.

RESULTS

We found the mean probability of two observers agreeing in the absence of additional data to be 0.73 (±0.08) with a corresponding Cohen's kappa of 0.59 (±0.11). We further showed that the inclusion of additional data increased the concordance to 0.80 (±0.06) with a Cohen's kappa coefficient of 0.67 (±0.09).

CONCLUSION

We showed that the addition of serological data significantly improves the quality of CD diagnosis. However, the limited interobserver agreement in CD diagnosis using digitised WSIs, even after the inclusion of IgA tTG and Hb data, indicates the importance of interpreting duodenal biopsy in the appropriate clinical context. It further highlights the unmet need for an objective means of reproducible duodenal biopsy diagnosis, such as the automated analysis of WSIs using artificial intelligence.

Rapid artefact removal and H&E-stained tissue segmentation

We present an innovative method for rapidly segmenting haematoxylin and eosin (H&E)-stained tissue in whole-slide images (WSIs) that eliminates a wide range of undesirable artefacts such as pen marks and scanning artefacts. Our method involves taking a single-channel representation of a low-magnification RGB overview of the WSI in which the pixel values are bimodally distributed such that H&E-stained tissue is easily distinguished from both background and a wide variety of artefacts. We demonstrate our method on 30 WSIs prepared from a wide range of institutions and WSI digital scanners, each containing substantial artefacts, and compare it to segmentations provided by Otsu thresholding and Histolab tissue segmentation and pen filtering tools. We found that our method segmented the tissue and fully removed all artefacts in 29 out of 30 WSIs, whereas Otsu thresholding failed to remove any artefacts, and the Histolab pen filtering tools only partially removed the pen marks. The beauty of our approach lies in its simplicity: manipulating RGB colour space and using Otsu thresholding allows for the segmentation of H&E-stained tissue and the rapid removal of artefacts without the need for machine learning or parameter tuning.

Pathologist-trained machine learning classifiers developed to quantitate celiac disease features differentiate endoscopic biopsies according to modified marsh score and dietary intervention response

BACKGROUND

Histologic evaluation of the mucosal changes associated with celiac disease is important for establishing an accurate diagnosis and monitoring the impact of investigational therapies. While the Marsh-Oberhuber classification has been used to categorize the histologic findings into discrete stages (i.e., Type 0-3c), significant variability has been documented between observers using this ordinal scoring system. Therefore, we evaluated whether pathologist-trained machine learning classifiers can be developed to objectively quantitate the pathological changes of villus blunting, intraepithelial lymphocytosis, and crypt hyperplasia in small intestine endoscopic biopsies.

METHODS

A convolutional neural network (CNN) was trained and combined with a secondary algorithm to quantitate intraepithelial lymphocytes (IEL) with 5 classes on CD3 immunohistochemistry whole slide images (WSI) and used to correlate feature outputs with ground truth modified Marsh scores in a total of 116 small intestine biopsies.

RESULTS

Across all samples, median %CD3 counts (positive cells/enterocytes) from villous epithelium (VE) increased with higher Marsh scores (Type 0%CD3 VE = 13.4; Type 1-3%CD3 VE = 41.9, p < 0.0001). Indicators of villus blunting and crypt hyperplasia were also observed (Type 0-2 villous epithelium/lamina propria area ratio = 0.81; Type 3a-3c villous epithelium/lamina propria area ratio = 0.29, p < 0.0001), and Type 0-1 crypt/villous epithelial area ratio = 0.59; Type 2-3 crypt/villous epithelial area ratio = 1.64, p < 0.0001). Using these individual features, a combined feature machine learning score (MLS) was created to evaluate a set of 28 matched pre- and post-intervention biopsies captured before and after dietary gluten restriction. The disposition of the continuous MLS paired biopsy result aligned with the Marsh score in 96.4% (27/28) of the cohort.

CONCLUSIONS

Machine learning classifiers can be developed to objectively quantify histologic features and capture additional data not achievable with manual scoring. Such approaches should be further investigated to improve biopsy evaluation, especially for clinical trials.

The Comparative Pathology Workbench: Interactive visual analytics for biomedical data

Pathologists need to compare histopathological images of normal and diseased tissues between different samples, cases, and species. We have designed an interactive system, termed Comparative Pathology Workbench (CPW), which allows direct and dynamic comparison of images at a variety of magnifications, selected regions of interest, as well as the results of image analysis or other data analyses such as scRNA-seq. This allows pathologists to indicate key diagnostic features, with a mechanism to allow discussion threads amongst expert groups of pathologists and other disciplines. The data and associated discussions can be accessed online from anywhere in the world. The Comparative Pathology Workbench (CPW) is a web-browser-based visual analytics platform providing shared access to an interactive "spreadsheet" style presentation of image and associated analysis data. The CPW provides a grid layout of rows and columns so that images that correspond to matching data can be organised in the form of an image-enabled "spreadsheet". An individual workbench can be shared with other users with read-only or full edit access as required. In addition, each workbench element or the whole bench itself has an associated discussion thread to allow collaborative analysis and consensual interpretation of the data. The CPW is a Django-based web-application that hosts the workbench data, manages users, and user-preferences. All image data are hosted by other resource applications such as OMERO or the Digital Slide Archive. Further resources can be added as required. The discussion threads are managed using WordPress and include additional graphical and image data. The CPW has been developed to allow integration of image analysis outputs from systems such as QuPath or ImageJ. All software is open-source and available from a GitHub repository.

Stain normalization gives greater generalizability than stain jittering in neural network training for the classification of coeliac disease in duodenal biopsy whole slide images

Around 1% of the population of the UK and North America have a diagnosis of coeliac disease (CD), due to a damaging immune response to the small intestine. Assessing whether a patient has CD relies primarily on the examination of a duodenal biopsy, an unavoidably subjective process with poor inter-observer concordance. Wei et al. [11] developed a neural network-based method for diagnosing CD using a dataset of duodenal biopsy whole slide images (WSIs). As all training and validation data came from one source, there was no guarantee that their results would generalize to WSIs obtained from different scanners and laboratories. In this study, the effects of applying stain normalization and jittering to the training data were compared. We trained a deep neural network on 331 WSIs obtained with a Ventana scanner (WSIs; CD: n = 190 ; normal: n = 141 ) to classify presence of CD. In order to test the effects of stain processing when validating on WSIs scanned on varying scanners and from varying laboratories, the neural network was validated on 4 datasets: WSIs of slides scanned on a Ventana scanner (WSIs; CD: n = 48 ; normal: n = 35 ), WSIs of the same slides rescanned on a Hamamatsu scanner (WSIs; CD: n = 48 ; normal: n = 35 ), WSIs of the same slides rescanned on an Aperio scanner (WSIs; CD: n = 48 ; normal: n = 35 ), and WSIs of different slides scanned on an Aperio scanner (WSIs; CD: n = 38 ; normal: n = 37 ). Without stain processing, the F1 scores of the neural network were 0.947 , 0.619 , 0.746 , and 0.727 when validating on the Ventana validation WSIs, Hamamatsu and Aperio rescans of the Ventana validation WSIs, and Aperio WSIs from a different source respectively. With stain normalization, the performance of the neural network improved significantly with respective F1 scores 0.982 , 0.943 , 0.903 , and 0.847 . Stain jittering resulted in a better performance than stain normalization when validating on data from the same source F1 score 1.000 , but resulted in poorer performance than stain normalization when validating on WSIs from different scanners (F1 scores 0.939 , 0.814 , and 0.747 ). This study shows the importance of stain processing, in particular stain normalization, when training machine learning models on duodenal biopsy WSIs to ensure generalizability between different scanners and laboratories.