Quality control stress test for deep learning-based diagnostic model in digital pathology.
Ontology highlight
ABSTRACT:
SUBMITTER: Schomig-Markiefka B
PROVIDER: S-EPMC8592835 | biostudies-literature |
REPOSITORIES: biostudies-literature
Similar Datasets
Project description:For a method to be widely adopted in medical research or clinical practice, it needs to be reproducible so that clinicians and regulators can have confidence in its use. Machine learning and deep learning have a particular set of challenges around reproducibility. Small differences in the settings or the data used for training a model can lead to large differences in the outcomes of experiments. In this work, three top-performing algorithms from the Camelyon grand challenges are reproduced using only information presented in the associated papers and the results are then compared to those reported. Seemingly minor details were found to be critical to performance and yet their importance is difficult to appreciate until the actual reproduction is attempted. We observed that authors generally describe the key technical aspects of their models well but fail to maintain the same reporting standards when it comes to data preprocessing which is essential to reproducibility. As an important contribution of the present study and its findings, we introduce a reproducibility checklist that tabulates information that needs to be reported in histopathology ML-based work in order to make it reproducible. Author summary For a method to be used a lot in medical research or clinical practice, it needs to be able to be reproduced so that people can trust it. Machine learning and deep learning have some challenges around this. For example, small changes in how a model is trained can lead to significant changes in the results of experiments. This makes it essential that researchers report how they do things in enough detail for the results of their experiments to be reproducible. In this work, we looked at three different algorithms used for digital pathology image analysis. We tried to reproduce them using only the information reported in their papers. We confirmed that even minor details could be essential. Authors often do not report all the details needed to reproduce their work. We also created a checklist of things that need to be reported to help other researchers make their work reproducible.
| S-EPMC9931349 | biostudies-literature
Project description:BackgroundAn increasing number of studies within digital pathology show the potential of artificial intelligence (AI) to diagnose cancer using histological whole slide images, which requires large and diverse data sets. While diversification may result in more generalizable AI-based systems, it can also introduce hidden variables. If neural networks are able to distinguish/learn hidden variables, these variables can introduce batch effects that compromise the accuracy of classification systems.ObjectiveThe objective of the study was to analyze the learnability of an exemplary selection of hidden variables (patient age, slide preparation date, slide origin, and scanner type) that are commonly found in whole slide image data sets in digital pathology and could create batch effects.MethodsWe trained four separate convolutional neural networks (CNNs) to learn four variables using a data set of digitized whole slide melanoma images from five different institutes. For robustness, each CNN training and evaluation run was repeated multiple times, and a variable was only considered learnable if the lower bound of the 95% confidence interval of its mean balanced accuracy was above 50.0%.ResultsA mean balanced accuracy above 50.0% was achieved for all four tasks, even when considering the lower bound of the 95% confidence interval. Performance between tasks showed wide variation, ranging from 56.1% (slide preparation date) to 100% (slide origin).ConclusionsBecause all of the analyzed hidden variables are learnable, they have the potential to create batch effects in dermatopathology data sets, which negatively affect AI-based classification systems. Practitioners should be aware of these and similar pitfalls when developing and evaluating such systems and address these and potentially other batch effect variables in their data sets through sufficient data set stratification.
| S-EPMC7886613 | biostudies-literature
Project description:BackgroundEvaluation of tumor-tissue images stained with hematoxylin and eosin (H&E) is pivotal in diagnosis, yet only a fraction of the rich phenotypic information is considered for clinical care. Here, we propose a survival deep learning (SDL) framework to extract this information to predict glioma survival.MethodsDigitized whole slide images were downloaded from The Cancer Genome Atlas (TCGA) for 766 diffuse glioma patients, including isocitrate dehydrogenase (IDH)-mutant/1p19q-codeleted oligodendroglioma, IDH-mutant/1p19q-intact astrocytoma, and IDH-wildtype astrocytoma/glioblastoma. Our SDL framework employs a residual convolutional neural network with a survival model to predict patient risk from H&E-stained whole-slide images. We used statistical sampling techniques and randomized the transformation of images to address challenges in learning from histology images. The SDL risk score was evaluated in traditional and recursive partitioning (RPA) survival models.ResultsThe SDL risk score demonstrated substantial univariate prognostic power (median concordance index of 0.79 [se: 0.01]). After adjusting for age and World Health Organization 2016 subtype, the SDL risk score was significantly associated with overall survival (OS; hazard ratio = 2.45; 95% CI: 2.01 to 3.00). Four distinct survival risk groups were characterized by RPA based on SDL risk score, IDH status, and age with markedly different median OS ranging from 1.03 years to 14.14 years.ConclusionsThe present study highlights the independent prognostic power of the SDL risk score for objective and accurate prediction of glioma outcomes. Further, we show that the RPA delineation of patient-specific risk scores and clinical prognostic factors can successfully demarcate the OS of glioma patients.
| S-EPMC9389424 | biostudies-literature
Project description:PURPOSE:Digital pathology (DP), referring to the digitization of tissue slides, is beginning to change the landscape of clinical diagnostic workflows and has engendered active research within the area of computational pathology. One of the challenges in DP is the presence of artefacts and batch effects, unintentionally introduced during both routine slide preparation (eg, staining, tissue folding) and digitization (eg, blurriness, variations in contrast and hue). Manual review of glass and digital slides is laborious, qualitative, and subject to intra- and inter-reader variability. Therefore, there is a critical need for a reproducible automated approach of precisely localizing artefacts to identify slides that need to be reproduced or regions that should be avoided during computational analysis. METHODS:Here we present HistoQC, a tool for rapidly performing quality control to not only identify and delineate artefacts but also discover cohort-level outliers (eg, slides stained darker or lighter than others in the cohort). This open-source tool employs a combination of image metrics (eg, color histograms, brightness, contrast), features (eg, edge detectors), and supervised classifiers (eg, pen detection) to identify artefact-free regions on digitized slides. These regions and metrics are presented to the user via an interactive graphical user interface, facilitating artefact detection through real-time visualization and filtering. These same metrics afford users the opportunity to explicitly define acceptable tolerances for their workflows. RESULTS:The output of HistoQC on 450 slides from The Cancer Genome Atlas was reviewed by two pathologists and found to be suitable for computational analysis more than 95% of the time. CONCLUSION:These results suggest that HistoQC could provide an automated, quantifiable, quality control process for identifying artefacts and measuring slide quality, in turn helping to improve both the repeatability and robustness of DP workflows.
| S-EPMC6552675 | biostudies-literature
Project description:Colonoscopy is an effective tool to detect colorectal lesions and needs the support of pathological diagnosis. This study aimed to develop and validate deep learning models that automatically classify digital pathology images of colon lesions obtained from colonoscopy-related specimen. Histopathological slides of colonoscopic biopsy or resection specimens were collected and grouped into six classes by disease category: adenocarcinoma, tubular adenoma (TA), traditional serrated adenoma (TSA), sessile serrated adenoma (SSA), hyperplastic polyp (HP), and non-specific lesions. Digital photographs were taken of each pathological slide to fine-tune two pre-trained convolutional neural networks, and the model performances were evaluated. A total of 1865 images were included from 703 patients, of which 10% were used as a test dataset. For six-class classification, the mean diagnostic accuracy was 97.3% (95% confidence interval [CI], 96.0-98.6%) by DenseNet-161 and 95.9% (95% CI 94.1-97.7%) by EfficientNet-B7. The per-class area under the receiver operating characteristic curve (AUC) was highest for adenocarcinoma (1.000; 95% CI 0.999-1.000) by DenseNet-161 and TSA (1.000; 95% CI 1.000-1.000) by EfficientNet-B7. The lowest per-class AUCs were still excellent: 0.991 (95% CI 0.983-0.999) for HP by DenseNet-161 and 0.995 for SSA (95% CI 0.992-0.998) by EfficientNet-B7. Deep learning models achieved excellent performances for discriminating adenocarcinoma from non-adenocarcinoma lesions with an AUC of 0.995 or 0.998. The pathognomonic area for each class was appropriately highlighted in digital images by saliency map, particularly focusing epithelial lesions. Deep learning models might be a useful tool to help the diagnosis for pathologic slides of colonoscopy-related specimens.
| S-EPMC9329279 | biostudies-literature
Project description:Poor generalizability is a major barrier to clinical implementation of artificial intelligence in digital pathology. The aim of this study was to test the generalizability of a pretrained deep learning model to a new diagnostic setting and to a small change in surgical indication. A deep learning model for breast cancer metastases detection in sentinel lymph nodes, trained on CAMELYON multicenter data, was used as a base model, and achieved an AUC of 0.969 (95% CI 0.926-0.998) and FROC of 0.838 (95% CI 0.757-0.913) on CAMELYON16 test data. On local sentinel node data, the base model performance dropped to AUC 0.929 (95% CI 0.800-0.998) and FROC 0.744 (95% CI 0.566-0.912). On data with a change in surgical indication (axillary dissections) the base model performance indicated an even larger drop with a FROC of 0.503 (95%CI 0.201-0.911). The model was retrained with addition of local data, resulting in about a 4% increase for both AUC and FROC for sentinel nodes, and an increase of 11% in AUC and 49% in FROC for axillary nodes. Pathologist qualitative evaluation of the retrained model´s output showed no missed positive slides. False positives, false negatives and one previously undetected micro-metastasis were observed. The study highlights the generalization challenge even when using a multicenter trained model, and that a small change in indication can considerably impact the model´s performance.
| S-EPMC9659028 | biostudies-literature
Project description: Not available
| S-EPMC10728828 | biostudies-literature
Project description:IntroductionDigital pathology solutions are increasingly implemented for primary diagnostics in departments of pathology around the world. This has sparked a growing engagement on validation studies to evaluate the diagnostic performance of whole slide imaging (WSI) regarding safety, reliability, and accuracy. The aim of this review was to evaluate the performance of digital pathology for diagnostic purposes compared to light microscopy (LM) in human pathology, based on validation studies designed to assess such technologies.MethodsIn this systematic review based on PRISMA guidelines, we analyzed validation studies of WSI compared with LM. We included studies of diagnostic performance of WSI regarding diagnostic test accuracy (DTA) indicators, degree of overdiagnosis, diagnostic concordance, and observer variability as a secondary outcome. Overdiagnosis is (for example) detecting a pathological condition that will either not progress or progress very slowly. Thus, the patient will never get symptoms from this condition and the pathological condition will never be the cause of death. From a search comprising four databases: PubMed, EMBASE, Cochrane Library, and Web of Science, encompassing the period 2010-2021, we selected and screened 12 peer-reviewed articles that fulfilled our selection criteria. Risk of bias was conducted through QUADAS-2 tool, and data analysis and synthesis were performed in a qualitative format.ResultsWe found that diagnostic performance of WSI was not inferior to LM for DTA indicators, concordance, and observer variability. The degree of overdiagnosis was not explicitly reported in any of the studies, while the term itself was used in one study and could be implicitly calculated in another.ConclusionWSI had an overall high diagnostic accuracy based on traditional accuracy measurements; however, the degree of overdiagnosis is unknown.
| S-EPMC9577136 | biostudies-literature
Project description:Biomedical imaging Is an important source of information in cancer research. Characterizations of cancer morphology at onset, progression, and in response to treatment provide complementary information to that gleaned from genomics and clinical data. Accurate extraction and classification of both visual and latent image features Is an increasingly complex challenge due to the increased complexity and resolution of biomedical image data. In this paper, we present four deep learning-based image analysis methods from the Computational Precision Medicine (CPM) satellite event of the 21st International Medical Image Computing and Computer Assisted Intervention (MICCAI 2018) conference. One method Is a segmentation method designed to segment nuclei in whole slide tissue images (WSIs) of adult diffuse glioma cases. It achieved a Dice similarity coefficient of 0.868 with the CPM challenge datasets. Three methods are classification methods developed to categorize adult diffuse glioma cases into oligodendroglioma and astrocytoma classes using radiographic and histologic image data. These methods achieved accuracy values of 0.75, 0.80, and 0.90, measured as the ratio of the number of correct classifications to the number of total cases, with the challenge datasets. The evaluations of the four methods indicate that (1) carefully constructed deep learning algorithms are able to produce high accuracy in the analysis of biomedical image data and (2) the combination of radiographic with histologic image information improves classification performance.
| S-EPMC7046596 | biostudies-literature
Project description:The diagnosis of disease often requires analysis of a biopsy. Many diagnoses depend not only on the presence of certain features but on their location within the tissue. Recently, a number of deep learning diagnostic aids have been developed to classify digitized biopsy slides. Clinical workflows often involve processing of more than 500 slides per day. But, clinical use of deep learning diagnostic aids would require a preprocessing workflow that is cost-effective, flexible, scalable, rapid, interpretable, and transparent. Here, we present such a workflow, optimized using Dask and mixed precision training via APEX, capable of handling any patch-level or slide level classification and prediction problem. The workflow uses a flexible and fast preprocessing and deep learning analytics pipeline, incorporates model interpretation and has a highly storage-efficient audit trail. We demonstrate the utility of this package on the analysis of a prototypical anatomic pathology specimen, liver biopsies for evaluation of hepatitis from a prospective cohort. The preliminary data indicate that PathFlowAI may become a cost-effective and time-efficient tool for clinical use of Artificial Intelligence (AI) algorithms.
| S-EPMC6919317 | biostudies-literature