Biology subjects

Leh, S.

Publications and source records attributed to Leh, S..

5 recordsLinked to original sources

bioRxiv2025

Morphometry-based detection of deep learning faults in glomerular segmentation

Deep learning-based segmentation has evolved to a powerful strategy for automatically annotating glomeruli in kidney biopsy images. However, since any artificial intelligence can make mistakes, strategies for identifying and correcting faulty annotations are often indispensable. Yet, how can such a validation be achieved without the laborious task of a pathologist manually checking every single image? To address this issue, the current project performed an extensive study on the use of shape analysis to automatically evaluate the glomerular annotations produced by deep-learning segmentation. Examining a large repertoire of shape descriptors on over 168000 glomerular predictions, the study found that morphometry could successfully highlight and distinguish between three different types of segmentation inconsistencies. In addition, using shape descriptors to rank segmentation annotations, it was possible to obtain a distinct enrichment of errors on the leading edge of the ranking, implying that pathologists would only have to inspect and correct the most suspicious fraction of all annotations. Ultimately, the study suggested a panel of three shape descriptors that enabled an efficient enrichment of all errors, respective or irrespective of error type. In summary, the work demonstrates the methodological aspects and benefits of shape analysis for evaluating glomerular segmentation results. We are convinced that, by applying such a strategy for detecting segmentation errors, it will be possible to approach a more time-efficient correction of deep learning-derived glomerular annotations.

pathology↗

bioRxiv2025

Image Analysis for Non-Neoplastic Kidney Disease: Utilizing Morphological Segmentation to Improve Quantification of Interstitial Fibrosis

Interstitial fibrosis (IF) is a hallmark of chronic kidney disease (CKD) and a strong predictor of progression to end-stage kidney disease (ESKD). Current biopsy-based IF assessments rely on subjective visual estimations, limiting reproducibility. Sirius Red staining is widely used for visualizing fibrotic tissue, yet its application in digital pathology is limited by non-specific staining. This study investigates the impact of cortical structure segmentation on fibrosis quantification in Sirius Red-stained, non-neoplastic kidney biopsies. Fibrosis measurements before and after segmentation were compared using two image analysis methods (stain deconvolution and red-green), with ground truth fibrosis measured by point counting and expert pathologist grading. Excluding non-interstitial structures led to a significant reduction of quantified fibrosis and improved correlation with pathology grading and point counting for both the stain deconvolution and the red green method. Bland-Altman analysis showed reduced bias after segmentation: for the deconvolution method, mean difference decreased from +2.5% (95% LoA: -8% to +13%) to +1% (-7% to +9%); for the red-green method, from +3% (-10% to +16%) to +1% (-8% to +10%). Correlation with pathology grading also improved (Spearmans {rho} rose from 0.55 to 0.58 for deconvolution and from 0.53 to 0.59 for red-green). These findings confirm that targeted segmentation enhances the accuracy and consistency of automated fibrosis assessment, supporting its integration into digital pathology workflows as a critical step toward reliable quantification of fibrosis in kidney disease.

pathology↗

bioRxiv2025

GlomExtractor: a versatile tool for extracting glomerular patches from whole slide kidney biopsy images

The extraction of glomerular image patches is a key step prior to the training and application of glomerular classification models. Numerous algorithms for detecting and segmenting glomeruli have already been described in the literature, but standards for how to extract glomeruli from such an initial annotation remain lacking. Furthermore, the impact of different choices in extracting and preprocessing, such as cropping and scaling, are poorly understood and researched. To address this gap, the current paper introduces the GlomExtractor, a versatile tool implementing the key steps for extracting glomerular patches from whole slide images, including optional filtering of segmentations, patch extraction and scaling, and postprocessing such as background removal. By utilizing the GlomExtractor in combination with glomerular clustering experiments, the current study demonstrated how different processing strategies can impact the performance of downstream machine learning applications. We believe that the GlomExtractor and the experiments demonstrated in the current paper will help researchers in (i) the development of glomerular classification models, in (ii) advancing our understanding of the impact of patch processing on model performance, and in (iii) establishing community standards for how glomeruli should be extracted depending on downstream use.

pathology↗

bioRxiv2025

StainStyleSampler: Clustering-based sampling of whole slide image appearances

The appearance of whole slide biopsy images is greatly affected by various factors such as laboratory procedures or the choice of digital slide scanners. The resulting variations in image styles within and across batches of histological images represent one of the major obstacles to the development of generalizable machine learning algorithms. To overcome this challenge, a lot of research has focused on stain normalization and stain augmentation techniques. While such approaches provide effective strategies to reduce stain variation or increase stain invariance, respectively, they typically involve only limited modeling or sampling of the underlying stain style distribution. Tools for a streamlined sampling of different aspects of such a distribution, which would be crucial e.g. for explicitly evaluating machine learning robustness across or with respect to major stain styles, remain largely missing. Here, we present the StainStyleSampler, a toolkit for (i) the exploration and modeling of stain style variations, and (ii) the automated sampling of images or styles capturing the core components of this variation. The tool enables the extraction of various color features and deconvolved stain components, visualization of such features directly or after dimensionality reduction, modeling of style distributions using binning, clustering, and density mapping, and automated sampling of the most representative reference images. We believe that this software will equip pathologists and computer-scientists with a more versatile set of tools that can aid substantially in both the exploration and sampling of stain variation across whole slide images.

pathology↗

bioRxiv2024

Unsupervised learning for labeling global glomerulosclerosis

Current deep learning models for classifying glomeruli in nephropathology are trained almost exclusively in a supervised manner, requiring expert-labeled images. Very little is known about the potential for unsupervised learning to overcome this bottleneck. To address this open question in a proof-of-concept, the project focused on the most fundamental classification task: globally sclerosed versus non-globally sclerosed glomeruli. The performance of clustering between the two classes was extensively studied across a variety of labeled datasets with diverse compositions and histological stains, and across the feature embeddings produced by 34 different pre-trained CNN models. As demonstrated by the study, clustering of globally and non-globally sclerosed glomeruli is generally highly feasible, yielding accuracies of over 95% in most datasets. Further work will be required to expand these experiments towards the clustering of additional glomerular lesion categories. We are convinced that these efforts (i) will open up opportunities for semi-automatic labeling approaches, thus alleviating the need for labor-intensive manual labeling, and (ii) illustrate that glomerular classification models can potentially be trained even in the absence of expert-derived class labels.

pathology↗