Deep-learning methods for computational pathology require either manual annotation of gigapixel whole-slide images (WSIs) or large datasets of WSIs with slide-level labels and typically suffer from ...poor domain adaptation and interpretability. Here we report an interpretable weakly supervised deep-learning method for data-efficient WSI processing and learning that only requires slide-level labels. The method, which we named clustering-constrained-attention multiple-instance learning (CLAM), uses attention-based learning to identify subregions of high diagnostic value to accurately classify whole slides and instance-level clustering over the identified representative regions to constrain and refine the feature space. By applying CLAM to the subtyping of renal cell carcinoma and non-small-cell lung cancer as well as the detection of lymph node metastasis, we show that it can be used to localize well-known morphological features on WSIs without the need for spatial labels, that it overperforms standard weakly supervised classification algorithms and that it is adaptable to independent test cohorts, smartphone microscopy and varying tissue content.
There is increasing interest in using whole slide imaging (WSI) for diagnostic purposes (primary and/or consultation). An important consideration is whether WSI can safely replace conventional light ...microscopy as the method by which pathologists review histologic sections, cytology slides, and/or hematology slides to render diagnoses. Validation of WSI is crucial to ensure that diagnostic performance based on digitized slides is at least equivalent to that of glass slides and light microscopy. Currently, there are no standard guidelines regarding validation of WSI for diagnostic use.
To recommend validation requirements for WSI systems to be used for diagnostic purposes.
The College of American Pathologists Pathology and Laboratory Quality Center convened a nonvendor panel from North America with expertise in digital pathology to develop these validation recommendations. A literature review was performed in which 767 international publications that met search term requirements were identified. Studies outside the scope of this effort and those related solely to technical elements, education, and image analysis were excluded. A total of 27 publications were graded and underwent data extraction for evidence evaluation. Recommendations were derived from the strength of evidence determined from 23 of these published studies, open comment feedback, and expert panel consensus.
Twelve guideline statements were established to help pathology laboratories validate their own WSI systems intended for clinical use. Validation of the entire WSI system, involving pathologists trained to use the system, should be performed in a manner that emulates the laboratory's actual clinical environment. It is recommended that such a validation study include at least 60 routine cases per application, comparing intraobserver diagnostic concordance between digitized and glass slides viewed at least 2 weeks apart. It is important that the validation process confirm that all material present on a glass slide to be scanned is included in the digital image.
Validation should demonstrate that the WSI system under review produces acceptable digital slides for diagnostic interpretation. The intention of validating WSI systems is to permit the clinical use of this technology in a manner that does not compromise patient care.
The hematoxylin and eosin (H&E) stain is the standard used for microscopic examination of tissues that have been fixed, processed, embedded, and sectioned. It can be performed manually or by ...automation. For economic reasons, the manual technique is generally the method of choice for facilities with a low sample volume. This protocol describes manual H&E staining of fixed, processed, paraffin-embedded, and sectioned mouse tissues. In H&E-stained tissues, the nucleic acids stain dark blue and the proteins stain red to pink or orange. For accurate phenotyping and delineation of tissue detail, the protocol must be adhered to rigorously. This includes frequent reagent changes as well as the use of "in-date" reagents. Appropriate color in a good H&E stain allows for identification of many tissue subtleties that are necessary for accurate diagnosis.
Little is known about differences in rates of fibrosis progression between patients with nonalcoholic fatty liver (NAFL) vs nonalcoholic steatohepatitis (NASH). We conducted a systematic review and ...meta-analysis of all studies that assessed paired liver biopsy specimens to estimate the rates of fibrosis progression in patients with nonalcoholic fatty liver disease (NAFLD) including NAFL and NASH.
Through a systematic search of multiple databases and author contact, up to June 2013, we identified studies of adults with NAFLD that collected paired liver biopsy specimens at least 1 year apart. From these, we calculated a pooled-weighted annual fibrosis progression rate (number of stages changed between the 2 biopsy samples) with 95% confidence intervals (CIs), and identified clinical risk factors associated with progression.
We identified 11 cohort studies including 411 patients with biopsy-proven NAFLD (150 with NAFL and 261 with NASH). At baseline, the distribution of fibrosis for stages 0, 1, 2, 3, and 4 was 35.8%, 32.5%, 16.7%, 9.3%, and 5.7%, respectively. Over 2145.5 person-years of follow-up evaluation, 33.6% had fibrosis progression, 43.1% had stable fibrosis, and 22.3% had an improvement in fibrosis stage. The annual fibrosis progression rate in patients with NAFL who had stage 0 fibrosis at baseline was 0.07 stages (95% CI, 0.02-0.11 stages), compared with 0.14 stages in patients with NASH (95% CI, 0.07-0.21 stages). These findings correspond to 1 stage of progression over 14.3 years for patients with NAFL (95% CI, 9.1-50.0 y) and 7.1 years for patients with NASH (95% CI, 4.8-14.3 y).
Based on a meta-analysis of studies of paired liver biopsy studies, liver fibrosis progresses in patients with NAFL and NASH.
Mass spectrometry (MS) imaging links molecular information and the spatial distribution of analytes within a sample. In contrast to most histochemical techniques, mass spectrometry imaging can ...differentiate molecular modifications and does not require labeling of targeted compounds. We have recently introduced the first mass spectrometry imaging method that provides highly specific molecular information (high resolution and accuracy in mass) at cellular dimensions (high resolution in space). This method is based on a matrix-assisted laser desorption/ionization (MALDI) imaging source working at atmospheric pressure which is coupled to an orbital trapping mass spectrometer. Here, we present a number of application examples and demonstrate the benefit of ‘mass spectrometry imaging with high resolution in mass and space.’ Phospholipids, peptides and drug compounds were imaged in a number of tissue samples at a spatial resolution of 5–10 μm. Proteins were analyzed after on-tissue tryptic digestion at 50-μm resolution. Additional applications include the analysis of single cells and of human lung carcinoma tissue as well as the first MALDI imaging measurement of tissue at 3 μm pixel size. MS image analysis for all these experiments showed excellent correlation with histological staining evaluation. The high mass resolution (
R
= 30,000) and mass accuracy (typically 1 ppm) proved to be essential for specific image generation and reliable identification of analytes in tissue samples. The ability to combine the required high-quality mass analysis with spatial resolution in the range of single cells is a unique feature of our method. With that, it has the potential to supplement classical histochemical protocols and to provide new insights about molecular processes on the cellular level.
Background & Aims A “resect and discard” policy has been proposed for diminutive polyps detected by screening colonoscopy, because hyperplastic and adenomatous polyps can be distinguished, in vivo, ...by using narrow-band imaging (NBI). We modeled the cost-effectiveness of this policy. Methods Markov modeling was used to compare the cost-effectiveness of universal pathology evaluations with a resect and discard policy for colonoscopy screening. In a resect and discard approach, diminutive lesions (≤5 mm), classified by endoscopy with high confidence, were not analyzed by a pathologist. Base case assumptions of an 84% rate of high-confidence classification, with a sensitivity and specificity for adenomas of 94% and 89%, respectively, were used. Census data were used to project outputs of the model onto the US population, assuming 23% as the current rate of adherence to a colonoscopy screening. Results With universal referral of resected polyps to pathology, colonoscopy screening costs an estimated $3222/person, with a gain of 51 days/person. Endoscopic polypectomy accounted for $179/person, of which $46/person was related to pathology examination. Adoption of a resect and discard policy for eligible diminutive polyps resulted in a savings of $25/person, without any meaningful effect on screening efficacy. Projected onto the US population, this approach would result in an undiscounted annual savings of $33 million. In the sensitivity analysis, the rate of high-confidence diagnosis and the accuracy for endoscopic polyp determination were the most meaningful variables. Conclusions In a simulation model, a resect and discard strategy for diminutive polyps detected by screening colonoscopy resulted in a substantial economic benefit without an impact on efficacy.
A Dataset for Breast Cancer Histopathological Image Classification Spanhol, Fabio A.; Oliveira, Luiz S.; Petitjean, Caroline ...
IEEE transactions on biomedical engineering,
2016-July, 2016-07-00, 2016-7-00, 20160701, 2016-07, Letnik:
63, Številka:
7
Journal Article
Recenzirano
Today, medical image analysis papers require solid experiments to prove the usefulness of proposed methods. However, experiments are often performed on data selected by the researchers, which may ...come from different institutions, scanners, and populations. Different evaluation measures may be used, making it difficult to compare the methods. In this paper, we introduce a dataset of 7909 breast cancer histopathology images acquired on 82 patients, which is now publicly available from http://web.inf.ufpr.br/vri/breast-cancer-database. The dataset includes both benign and malignant images. The task associated with this dataset is the automated classification of these images in two classes, which would be a valuable computer-aided diagnosis tool for the clinician. In order to assess the difficulty of this task, we show some preliminary results obtained with state-of-the-art image classification systems. The accuracy ranges from 80% to 85%, showing room for improvement is left. By providing this dataset and a standardized evaluation protocol to the scientific community, we hope to gather researchers in both the medical and the machine learning field to advance toward this clinical application.
The lack of publicly available ground-truth data has been identified as the major challenge for transferring recent developments in deep learning to the biomedical imaging domain. Though ...crowdsourcing has enabled annotation of large scale databases for real world images, its application for biomedical purposes requires a deeper understanding and hence, more precise definition of the actual annotation task. The fact that expert tasks are being outsourced to non-expert users may lead to noisy annotations introducing disagreement between users. Despite being a valuable resource for learning annotation models from crowdsourcing, conventional machine-learning methods may have difficulties dealing with noisy annotations during training. In this manuscript, we present a new concept for learning from crowds that handle data aggregation directly as part of the learning process of the convolutional neural network (CNN) via additional crowdsourcing layer (AggNet). Besides, we present an experimental study on learning from crowds designed to answer the following questions. 1) Can deep CNN be trained with data collected from crowdsourcing? 2) How to adapt the CNN to train on multiple types of annotation datasets (ground truth and crowd-based)? 3) How does the choice of annotation and aggregation affect the accuracy? Our experimental setup involved Annot8, a self-implemented web-platform based on Crowdflower API realizing image annotation tasks for a publicly available biomedical image database. Our results give valuable insights into the functionality of deep CNN learning from crowd annotations and prove the necessity of data aggregation integration.