Substantial Interobserver Variability Research Articles

Enhancing Accuracy in Breast Density Assessment Using Deep Learning: A Multicentric, Multi-Reader Study.

The evaluation of mammographic breast density, a critical indicator of breast cancer risk, is traditionally performed by radiologists via visual inspection of mammography images, utilizing the Breast Imaging-Reporting and Data System (BI-RADS) breast density categories. However, this method is subject to substantial interobserver variability, leading to inconsistencies and potential inaccuracies in density assessment and subsequent risk estimations. To address this, we present a deep learning-based automatic detection algorithm (DLAD) designed for the automated evaluation of breast density. Our multicentric, multi-reader study leverages a diverse dataset of 122 full-field digital mammography studies (488 images in CC and MLO projections) sourced from three institutions. We invited two experienced radiologists to conduct a retrospective analysis, establishing a ground truth for 72 mammography studies (BI-RADS class A: 18, BI-RADS class B: 43, BI-RADS class C: 7, BI-RADS class D: 4). The efficacy of the DLAD was then compared to the performance of five independent radiologists with varying levels of experience. The DLAD showed robust performance, achieving an accuracy of 0.819 (95% CI: 0.736-0.903), along with an F1 score of 0.798 (0.594-0.905), precision of 0.806 (0.596-0.896), recall of 0.830 (0.650-0.946), and a Cohen's Kappa (κ) of 0.708 (0.562-0.841). The algorithm achieved robust performance that matches and in four cases exceeds that of individual radiologists. The statistical analysis did not reveal a significant difference in accuracy between DLAD and the radiologists, underscoring the model's competitive diagnostic alignment with professional radiologist assessments. These results demonstrate that the deep learning-based automatic detection algorithm can enhance the accuracy and consistency of breast density assessments, offering a reliable tool for improving breast cancer screening outcomes.

Scoring PD-L1 Expression in Urothelial Carcinoma: An International Multi-Institutional Study on Comparison of Manual and Artificial Intelligence Measurement Model (AIM-PD-L1) Pathology Assessments.

Assessing programmed death ligand 1 (PD-L1) expression on tumor cells (TCs) using Food and Drug Administration-approved, validated immunoassays can guide the use of immune checkpoint inhibitor (ICI) therapy in cancer treatment. However, substantial interobserver variability has been reported using these immunoassays. Artificial intelligence (AI) has the potential to accurately measure biomarker expression in tissue samples, but its reliability and comparability to standard manual scoring remain to be evaluated. This multinational study sought to compare the %TC scoring of PD-L1 expression in advanced urothelial carcinoma, assessed by either an AI Measurement Model (AIM-PD-L1) or expert pathologists. The concordance among pathologists and between pathologists and AIM-PD-L1 was determined. The positivity rate of ≥ 1%TC PD-L1 was between 20-30% for 8/10 pathologists, and the degree of agreement and scoring distribution for among pathologists and between pathologists and AIM-PD-L1 was similar both scored as a continuous variable or using the pre-defined cutoff. Numerically higher score variation was observed with the 22C3 assay than with the 28-8 assay. A 2-h training module on the 28-8 assay did not significantly impact manual assessment. Cases exhibiting significantly higher variability in the assessment of PD-L1 expression (mean absolute deviation > 10) were found to have patterns of PD-L1 staining that were more challenging to interpret. An improved understanding of sources of manual scoring variability can be applied to PD-L1 expression analysis in the clinical setting. In the future, the application of AI algorithms could serve as a valuable reference guide for pathologists while scoring PD-L1.

An algorithm for standardization of tumor Infiltrating lymphocyte evaluation in head and neck cancers

PurposeThe prognostic and predictive significance of pathologist-read tumor infiltrating lymphocytes (TILs) in head and neck cancers have been demonstrated through multiple studies over the years. TILs have not been broadly adopted clinically, perhaps due to substantial inter-observer variability. In this study, we developed a machine-based algorithm for TIL evaluation in head and neck cancers and validated its prognostic value in independent cohorts. Experimental designA network classifier called NN3-17 was trained to identify and calculate tumor cells, lymphocytes, fibroblasts and “other” cells on hematoxylin-eosin stained sections using the QuPath software. These measurements were used to construct three predefined TIL variables. A retrospective collection of 154 head and neck squamous cell cancer cases was used as the discovery set to identify optimal association of TIL variables and survival. Two independent cohorts of 234 cases were used for validation. ResultsWe found that electronic TIL variables were associated with favorable prognosis in both the HPV-positive and -negative cases. After adjusting for clinicopathologic factors, Cox regression analysis demonstrated that electronic total TILs% (p = 0.025) in the HPV-positive and electronic stromal TILs% (p < 0.001) in the HPV-negative population were independent markers of disease specific outcomes (disease free survival). ConclusionsNeural network TIL variables demonstrated independent prognostic value in validation cohorts of HPV-positive and HPV-negative head and neck cancers. These objective variables can be calculated by an open-source software and could be considered for testing in a prospective setting to assess potential clinical implications.

A narrative review on the application of artificial intelligence in renal ultrasound.

Kidney disease is a serious public health problem and various kidney diseases could progress to end-stage renal disease. The many complications of end-stage renal disease. have a significant impact on the physical and mental health of patients. Ultrasound can be the test of choice for evaluating the kidney and perirenal tissue as it is real-time, available and non-radioactive. To overcome substantial interobserver variability in renal ultrasound interpretation, artificial intelligence (AI) has the potential to be a new method to help radiologists make clinical decisions. This review introduces the applications of AI in renal ultrasound, including automatic segmentation of the kidney, measurement of the renal volume, prediction of the kidney function, diagnosis of the kidney diseases. The advantages and disadvantages of the applications will also be presented clinicians to conduct research. Additionally, the challenges and future perspectives of AI are discussed.

Deep Learning Algorithms to Detect Murmurs Associated With Structural Heart Disease.

Background The success of cardiac auscultation varies widely among medical professionals, which can lead to missed treatments for structural heart disease. Applying machine learning to cardiac auscultation could address this problem, but despite recent interest, few algorithms have been brought to clinical practice. We evaluated a novel suite of Food and Drug Administration-cleared algorithms trained via deep learning on >15 000 heart sound recordings. Methods and Results We validated the algorithms on a data set of 2375 recordings from 615 unique subjects. This data set was collected in real clinical environments using commercially available digital stethoscopes, annotated by board-certified cardiologists, and paired with echocardiograms as the gold standard. To model the algorithm in clinical practice, we compared its performance against 10 clinicians on a subset of the validation database. Our algorithm reliably detected structural murmurs with a sensitivity of 85.6% and specificity of 84.4%. When limiting the analysis to clearly audible murmurs in adults, performance improved to a sensitivity of 97.9% and specificity of 90.6%. The algorithm also reported timing within the cardiac cycle, differentiating between systolic and diastolic murmurs. Despite optimizing acoustics for the clinicians, the algorithm substantially outperformed the clinicians (average clinician accuracy, 77.9%; algorithm accuracy, 84.7%.) Conclusions The algorithms accurately identified murmurs associated with structural heart disease. Our results illustrate a marked contrast between the consistency of the algorithm and the substantial interobserver variability of clinicians. Our results suggest that adopting machine learning algorithms into clinical practice could improve the detection of structural heart disease to facilitate patient care.

The Tissue Systems Pathology Test Outperforms Pathology Review in Risk Stratifying Patients With Low-Grade Dysplasia

Low-grade dysplasia (LGD) is associated with an increased risk of progression in Barrett's esophagus (BE); however, the diagnosis of LGD is limited by substantial interobserver variability. Multiple studies have shown that an objective tissue systems pathology test (TissueCypher Barrett's Esophagus Test, TSP-9), can effectively predict neoplastic progression in patients with BE. This study aimed to compare the risk stratification performance of the TSP-9 test vs benchmarks of generalist and expert pathology. A blinded cohort study was conducted in the screening cohort of a randomized controlled trial of patients with BE with community-based LGD. Biopsies from the first endoscopy with LGD were assessed by the TSP-9 test and independently reviewed by 30 pathologists from 5 countries per standard practice. The accuracy of the test and the diagnoses in predicting high-grade dysplasia (HGD) and esophageal adenocarcinoma (EAC) were compared. A total of 154 patients with BE (122 men), mean age 60.9 ± 9.8 years were studied. Twenty-four patients progressed to HGD/EAC within 5 years (median time of 1.7 years) and 130 did not progress to HGD/EAC within 5 years (median 7.8 years follow-up). The TSP-9 test demonstrated higher sensitivity (71% vs mean 63%, range 33%-88% across 30 pathologists), than the pathology review in detecting patients who progressed (P=.01186). The TSP-9 test outperformed the pathologists in risk stratifying patients with BE with LGD. Care guided by the test can provide an effective solution to variable pathology review of LGD, improving health outcomes by upstaging care to therapeutic intervention for patients at high risk for progression, while reducing unnecessary interventions in low-risk patients.

Deep learning model improves tumor-infiltrating lymphocyte evaluation and therapeutic response prediction in breast cancer

Tumor-infiltrating lymphocytes (TILs) have been recognized as key players in the tumor microenvironment of breast cancer, but substantial interobserver variability among pathologists has impeded its utility as a biomarker. We developed a deep learning (DL)-based TIL analyzer to evaluate stromal TILs (sTILs) in breast cancer. Three pathologists evaluated 402 whole slide images of breast cancer and interpreted the sTIL scores. A standalone performance of the DL model was evaluated in the 210 cases (52.2%) exhibiting sTIL score differences of less than 10 percentage points, yielding a concordance correlation coefficient of 0.755 (95% confidence interval [CI], 0.693–0.805) in comparison to the pathologists’ scores. For the 226 slides (56.2%) showing a 10 percentage points or greater variance between pathologists and the DL model, revisions were made. The number of discordant cases was reduced to 116 (28.9%) with the DL assistance (p < 0.001). The DL assistance also increased the concordance correlation coefficient of the sTIL score among every two pathologists. In triple-negative and human epidermal growth factor receptor 2 (HER2)-positive breast cancer patients who underwent the neoadjuvant chemotherapy, the DL-assisted revision notably accentuated higher sTIL scores in responders (26.8 ± 19.6 vs. 19.0 ± 16.4, p = 0.003). Furthermore, the DL-assistant revision disclosed the correlation of sTIL-high tumors (sTIL ≥ 50) with the chemotherapeutic response (odd ratio 1.28 [95% confidence interval, 1.01–1.63], p = 0.039). Through enhancing inter-pathologist concordance in sTIL interpretation and predicting neoadjuvant chemotherapy response, here we report the utility of the DL-based tool as a reference for sTIL scoring in breast cancer assessment.

Magnetic Resonance Enterography Reinvented: Exploring the Potential of a New Natural Beverage as an Alternative to Polyethylene Glycol Solution

The aim of the present study was to test a new oral contrast medium composed of natural components for the magnetic resonance (MR) imaging of small bowel diseases. Between January 2018 and June 2019, 35 patients affected by ileocolic Crohn’s disease (CD) were enrolled in the present study. Each patient underwent two sequential MR enterographies, first with the standard polyethylene glycol (PEG) water solution and, after 3 weeks, with the new natural beverage designed by our team. At the end of the administration of each oral contrast, a satisfaction survey was given to the patients to assess the palatability of both beverages. The intestinal distention and the quality of images were evaluated by two expert radiologists for both studies and the interreader agreement was calculated. According to the satisfaction questionnaire, 97.1% of patients expressed positive judgments regarding the natural beverage (71.4% very good and 25.7% good) whereas only 8.6% of them appreciated the PEG water solution (8.6% good) (p = 0.0001). The degree of intestinal distention was excellent and good in 97.1% of patients after the administration of PEG and in 94.3% of the patients after the administration of the natural beverage, without significant differences between the two products and with almost perfect (k = 0.821) and substantial (k = 0.754) inter-observer variability, respectively. No statistical differences were observed between the two expert radiologists regarding the evaluation of the imaging quality; in particular, they were considered good and excellent in 100% of patients after the administration of PEG water solution and in 97.2% of those who took the natural beverage, with substantial (k = 0.618) and almost perfect (k = 0.858) inter-observer variability, respectively. The new natural beverage demonstrated the same intestinal distension and excellent image quality compared to the synthetic standard oral contrast administered during MRE for small bowel diseases, proving to be a valid alternative with better palatability.

An Automated Tissue Systems Pathology Test can Standardize the Management and Improve Health Outcomes for Barrett's Esophagus Patients.

Low-grade dysplasia (LGD) in Barrett's esophagus (BE) is associated with an increased risk for progression to high-grade dysplasia (HGD) or esophageal adenocarcinoma (EAC). However, due to substantial interobserver variability in the diagnosis of LGD, a patient's management plan and health outcome depend largely on which pathologist reviews their case. This study evaluated the ability of a tissue systems pathology test that objectively risk stratifies patients with BE (TissueCypher, TSP-9) to standardize management in a manner consistent with improved health outcomes for BE patients. 154 BE patients with community-based LGD from the prospectively-followed screening cohort of the SURF trial were studied. Management decisions were simulated 500 times with varying generalist (n=16) and expert (n=14) pathology reviewers to determine the most likely care plan with or without use of the TSP-9 test for guidance. The percentage of patients receiving appropriate management based on the known progression/non-progression outcomes was calculated. The percentage of patients with 100% of simulations resulting in appropriate management significantly increased from 9.1% for pathology alone, to 58.4% when TSP-9 results were used with pathology, and further increased to 77.3% of patients receiving appropriate management when only TSP-9 results were used. Use of the test results also significantly increased the consistency of management decisions for patients when their slides were reviewed by different pathologists (P<0.0001). Management guided by the TSP-9 test can standardize care plans by increasing the early detection of progressors who can receive therapeutic interventions, while also increasing the percentage of non-progressors who can avoid unnecessary therapy and be managed by surveillance alone.

Stroke risk study based on deep learning-based magnetic resonance imaging carotid plaque automatic segmentation algorithm.

The primary factor for cardiovascular disease and upcoming cardiovascular events is atherosclerosis. Recently, carotid plaque texture, as observed on ultrasonography, is varied and difficult to classify with the human eye due to substantial inter-observer variability. High-resolution magnetic resonance (MR) plaque imaging offers naturally superior soft tissue contrasts to computed tomography (CT) and ultrasonography, and combining different contrast weightings may provide more useful information. Radiation freeness and operator independence are two additional benefits of M RI. However, other than preliminary research on MR texture analysis of basilar artery plaque, there is currently no information addressing MR radiomics on the carotid plaque. For the automatic segmentation of MRI scans to detect carotid plaque for stroke risk assessment, there is a need for a computer-aided autonomous framework to classify MRI scans automatically. We used to detect carotid plaque from MRI scans for stroke risk assessment pre-trained models, fine-tuned them, and adjusted hyperparameters according to our problem. Our trained YOLO V3 model achieved 94.81% accuracy, RCNN achieved 92.53% accuracy, and MobileNet achieved 90.23% in identifying carotid plaque from MRI scans for stroke risk assessment. Our approach will prevent incorrect diagnoses brought on by poor image quality and personal experience. The evaluations in this work have demonstrated that this methodology produces acceptable results for classifying magnetic resonance imaging (MRI) data.

Clinical consequences of diagnostic variability in the histopathological evaluation of early rectal cancer

IntroductionIn early rectal cancer, organ sparing treatment strategies such as local excision have gained popularity. The necessity of radical surgery is based on the histopathological evaluation of the local excision specimen. This study aimed to describe diagnostic variability between pathologists, and its impact on treatment allocation in patients with locally excised early rectal cancer. Materials and methodsPatients with locally excised pT1-2 rectal cancer were included in this prospective cohort study. Both quantitative measures and histopathological risk factors (i.e. poor differentiation, deep submucosal invasion, and lymphatic- or venous invasion) were evaluated. Interobserver variability was reported by both percentages and Fleiss’ Kappa- (ĸ) or intra-class correlation coefficients. ResultsA total of 126 patients were included. Ninety-four percent of the original histopathological reports contained all required parameters. In 73 of the 126 (57.9%) patients, at least one discordant parameter was observed, which regarded histopathological risk factors for lymph node metastases in 36 patients (28.6%). Interobserver agreement among different variables varied between 74% and 95% or ĸ 0.530–0.962. The assessment of lymphovascular invasion showed discordances in 26% (ĸ = 0.530, 95% CI 0.375–0.684) of the cases. In fourteen (11%) patients, discordances led to a change in treatment strategy. ConclusionThis study demonstrated that there is substantial interobserver variability between pathologists, especially in the assessment of lymphovascular invasion. Pathologists play a key role in treatment allocation after local excision of early rectal cancer, therefore interobserver variability needs to be reduced to decrease the number of patients that are over- or undertreated.

Considerable interobserver variation calls for unambiguous definitions of thyroid nodule ultrasound characteristics.

Thyroid nodule ultrasound characteristics are used as an indication for fine-needle aspiration cytology, usually as the basis for Thyroid Imaging Reporting and Data System (TIRADS) score calculation. Few studies on interobserver variation are available, all of which are based on analysis of preselected still ultrasound images and often lack surgical confirmation. After the blinded online evaluation of video recordings of the ultrasound examinations of 47 consecutive malignant and 76 consecutive benign thyroid lesions, 7 experts from 7 thyroid centers answered 17 TIRADS-related questions. Surgical histology was the reference standard. Interobserver variations of each ultrasound characteristic were compared using Gwet's AC1 inter-rater coefficients; higher values mean better concordance, the maximum being 1.0. On a scale from 0.0 to 1.0, the Gwet's AC1 values were 0.34, 0.53, 0.72, and 0.79 for the four most important features in decision-making, i.e. irregular margins, microcalcifications, echogenicity, and extrathyroidal extension, respectively. The concordance in the discrimination between mildly/moderately and very hypoechogenic nodules was 0.17. The smaller the nodule size the better the agreement in echogenicity, and the larger the nodule size the better the agreement on the presence of microcalcifications. Extrathyroidal extension was correctly identified in just 45.8% of the cases. Examination of video recordings, closely simulating the real-world situation, revealed substantial interobserver variation in the interpretation of each of the four most important ultrasound characteristics. In view of the importance for the management of thyroid nodules, unambiguous and widely accepted definitions of each nodule characteristic are warranted, although it remains to be investigated whether this diminishes observer variation.

Non-invasive biomarkers for liver inflammation in non-alcoholic fatty liver disease: present and future.

Inflammation is the key driver of liver fibrosis progression in non-alcoholic fatty liver disease (NAFLD). Unfortunately, it is often challenging to assess inflammation in NAFLD due to its dynamic nature and poor correlation with liver biochemical markers. Liver histology keeps its role as the standard tool, yet it is well-known for substantial sampling, intraobserver, and interobserver variability. Serum proinflammatory cytokines and apoptotic markers, namely cytokeratin-18, are well-studied with reasonable accuracy, whereas serum metabolomics and lipidomics have been adopted in some commercially available diagnostic models. Ultrasound and computed tomography imaging techniques are attractive due to their wide availability; yet their accuracies may not be comparable with magnetic resonance imaging-based tools. Machine learning and deep learning models, be they supervised or unsupervised learning, are promising tools to identify various subtypes of NAFLD, including those with dominating liver inflammation, contributing to sustainable care pathways for NAFLD.

Automatic diagnosis and grading of Prostate Cancer with weakly supervised learning on whole slide images

Background:The workflow of prostate cancer diagnosis and grading is cumbersome and the results suffer from substantial inter-observer variability. Recent trials have shown potential in using machine learning to develop automated systems to address this challenge. Most automated deep learning systems for prostate cancer Gleason grading focused on supervised learning requiring demanding fine-grained pixel-level annotations. Methods:A weakly-supervised deep learning model with slide-level labels is presented in this study for the diagnosis and grading of prostate cancer with whole slide image (WSI). WSIs are first cropped into small patches and then processed with a deep learning model to extract patch-level features. A graph convolution network (GCN) is used to aggregate the features for classifications. Throughout the training process, the noisy labels are progressively filtered out to reduce inter-observer variations in clinical reports. Finally, multi-center independent test cohorts with 6,174 slides are collected to evaluate the prostate cancer diagnosis and grading performance of our model. Results:The cancer diagnosis (2-level classification) results on two external test sets (n= 4,675, n= 844) show an area under the receiver operating characteristic curve (AUC) of 0.985 and 0.986. The Gleason grading (6-level classification) results reach 0.931 quadratic weighted kappa on the internal test set (n= 531). It generalizes well on the external test dataset (n= 844) with 0.801 quadratic weighted kappa with the reference standard set independently. The model enables pathological meaningful interpretability by visualizing the most attended lesions which are highly consistent with expert annotations. Conclusion:The proposed model incorporates a graph network in weakly supervised learning with only slide-level reports. A robust learning strategy is also employed to correct the label noise. It is highly accurate (>0.985 AUC for diagnosis) and also interpretable with intuitive heatmap visualization. It can be unified with a digital pathology pipeline to deliver prostate cancer metrics for a pathology report.

Association of artificial intelligence-powered and manual quantification of programmed death-ligand 1 (PD-L1) expression with outcomes in patients treated with nivolumab ± ipilimumab

AbstractAssessment of programmed death ligand 1 (PD-L1) expression by immunohistochemistry (IHC) has emerged as an important predictive biomarker across multiple tumor types. However, manual quantitation of PD-L1 positivity can be difficult and leads to substantial inter-observer variability. Although the development of artificial intelligence (AI) algorithms may mitigate some of the challenges associated with manual assessment and improve the accuracy of PD-L1 expression scoring, use of AI-based approaches to oncology biomarker scoring and drug development has been sparse, primarily due to the lack of large-scale clinical validation studies across multiple cohorts and tumor types. We developed AI-powered algorithms to evaluate PD-L1 expression on tumor cells by IHC and compared it with manual IHC scoring in urothelial carcinoma, non-small cell lung cancer, melanoma, and squamous cell carcinoma of the head and neck (prospectively determined during the phase II and III CheckMate clinical trials). 1,746 slides were retrospectively analyzed, the largest investigation of digital pathology algorithms on clinical trial datasets performed to date. AI-powered quantification of PD-L1 expression on tumor cells identified more PD-L1–positive samples compared with manual scoring at cutoffs of ≥1% and ≥5% in most tumor types. Additionally, similar improvements in response and survival were observed in patients identified as PD-L1–positive compared with PD-L1–negative using both AI-powered and manual methods, while improved associations with survival were observed in patients with certain tumor types identified as PD-L1–positive using AI-powered scoring only. Our study demonstrates the potential for implementation of digital pathology-based methods in future clinical practice to identify more patients who would benefit from treatment with immuno-oncology therapy compared with current guidelines using manual assessment.

A deep learning-based algorithm for tall cell detection in papillary thyroid carcinoma.

According to the World Health Organization, the tall cell variant (TCV) is an aggressive subtype of papillary thyroid carcinoma (PTC) comprising at least 30% epithelial cells two to three times as tall as they are wide. In practice, applying this definition is difficult causing substantial interobserver variability. We aimed to train a deep learning algorithm to detect and quantify the proportion of tall cells (TCs) in PTC. We trained the deep learning algorithm using supervised learning, testing it on an independent dataset, and further validating it on an independent set of 90 PTC samples from patients treated at the Hospital District of Helsinki and Uusimaa between 2003 and 2013. We compared the algorithm-based TC percentage to the independent scoring by a human investigator and how those scorings associated with disease outcomes. Additionally, we assessed the TC score in 71 local and distant tumor relapse samples from patients with aggressive disease. In the test set, the deep learning algorithm detected TCs with a sensitivity of 93.7% and a specificity of 94.5%, whereas the sensitivity fell to 90.9% and specificity to 94.1% for non-TC areas. In the validation set, the deep learning algorithm TC scores correlated with a diminished relapse-free survival using cutoff points of 10% (p = 0.044), 20% (p < 0.01), and 30% (p = 0.036). The visually assessed TC score did not statistically significantly predict survival at any of the analyzed cutoff points. We observed no statistically significant difference in the TC score between primary tumors and relapse tumors determined by the deep learning algorithm or visually. We present a novel deep learning-based algorithm to detect tall cells, showing that a high deep learning-based TC score represents a statistically significant predictor of less favorable relapse-free survival in PTC.

Comparison of immunohistochemistry and RT-qPCR for assessing ER, PR, HER2, and Ki67 and evaluating subtypes in patients with breast cancer.

Currently, the most commonly applied method for the determination of breast cancer subtypes is to test estrogen receptor (ER), progesterone receptor (PR), human epidermal growth factor receptor 2 (HER2), and Ki67 by immunohistochemistry (IHC). However, the IHC method has substantial intraobserver and interobserver variability. ESR1, PGR, ERBB2, and MKi67 mRNA tests by reverse transcription-quantitative polymerase chain reaction (RT-qPCR) assay may improve the diagnostic objectivity and efficiency. Here, we compared the concordance between RT-qPCR and IHC for assessment of the same biomarkers and evaluated the subtypes. A total of 265 eligible cases were divided into a training cohort and a validation cohort, and the expressions of ER/ESR1, PR/PGR, HER2/ERBB2, and Ki67/MKI67 were tested by IHC and RT-qPCR. Then, the appropriate cutoff of RT-qPCR was calculated in the training cohort. The concordance between RT-qPCR and IHC was calculated for individual marker. In addition, we investigated the subtypes based on the RT-qPCR results. The Spearman correlation coefficients between ER/ESR1, PR/PGR, HER2/ERBB2, and Ki67/MKI67 by IHC and RT-qPCR were 0.768, 0.699, 0.762, and 0.387, respectively. The cutoff values for the RT-qPCR assay of ESR1 (1%), PGR (1%), ERBB2, and MKi67 (14%) were 35.539, 32.139, 36.398, and 29.176, respectively. The overall percent agreement (OPA) between ER/ESR1, PR/PGR, HER2/ERBB2, and Ki67/MKI67 by IHC and RT-qPCR was 92.48%, 73.68%, 92.80%, and 74.44%, respectively. A total of 224 (84.53%) specimens were concordant for the breast cancer subtypes (IHC-based type) by RT-qPCR. Evaluation of breast cancer biomarker status by RT-qPCR was highly concordant with IHC. RT-qPCR can be used as a supplementary method to detect molecular markers of breast cancer.

OTO-Net: An Automated MRA Image Segmentation Network for Intracranial Aneurysms.

Intracranial aneurysms are local dilations of the cerebral blood vessels; people with intracranial aneurysms have a high risk to cause bleeding in the brain, which is related to high mortality and morbidity rates. Accurate detection and segmentation of intracranial aneurysms from Magnetic Resonance Angiography (MRA) images are essential in the clinical routine. Manual annotations used to assess the intracranial aneurysms on MRA images are substantial interobserver variability for both aneurysm detection and assessment of aneurysm size and growth. Many prior automated segmentation works have focused their efforts on tackling the problem, but there is still room for performance improvement due to the significant variability of lesions in the location, size, structure, and morphological appearance. To address these challenges, we propose a novel One-Two-One Fully Convolutional Networks (OTO-Net) for intracranial aneurysms automated segmentation in MRA images. The OTO-Net uses full convolution to achieve intracranial aneurysms automated segmentation through the combination of downsampling, upsampling, and skip connection. In addition, loss ensemble is used as the objective function to steadily improve the backpropagation efficiency of the network structure during the training process. We evaluated the proposed OTO-Net on one public benchmark dataset and one private dataset. Our proposed model can achieve the automated segmentation accuracy with 98.37% and 97.86%, average surface distances with 1.081 and 0.753, dice similarity coefficients with 0.9721 and 0.9813, and Hausdorff distance with 0.578 and 0.642 on these two datasets, respectively.

Unsupervised machine learning for identifying important visual features through bag-of-words using histopathology data from chronic kidney disease

Pathologists use visual classification to assess patient kidney biopsy samples when diagnosing the underlying cause of kidney disease. However, the assessment is qualitative, or semi-quantitative at best, and reproducibility is challenging. To discover previously unknown features which predict patient outcomes and overcome substantial interobserver variability, we developed an unsupervised bag-of-words model. Our study applied to the C-PROBE cohort of patients with chronic kidney disease (CKD). 107,471 histopathology images were obtained from 161 biopsy cores and identified important morphological features in biopsy tissue that are highly predictive of the presence of CKD both at the time of biopsy and in one year. To evaluate the performance of our model, we estimated the AUC and its 95% confidence interval. We show that this method is reliable and reproducible and can achieve 0.93 AUC at predicting glomerular filtration rate at the time of biopsy as well as predicting a loss of function at one year. Additionally, with this method, we ranked the identified morphological features according to their importance as diagnostic markers for chronic kidney disease. In this study, we have demonstrated the feasibility of using an unsupervised machine learning method without human input in order to predict the level of kidney function in CKD. The results from our study indicate that the visual dictionary, or visual image pattern, obtained from unsupervised machine learning can predict outcomes using machine-derived values that correspond to both known and unknown clinically relevant features.

Deep learning-enabled assessment of cardiac allograft rejection from endomyocardial biopsies

Endomyocardial biopsy (EMB) screening represents the standard of care for detecting allograft rejections after heart transplant. Manual interpretation of EMBs is affected by substantial interobserver and intraobserver variability, which often leads to inappropriate treatment with immunosuppressive drugs, unnecessary follow-up biopsies and poor transplant outcomes. Here we present a deep learning-based artificial intelligence (AI) system for automated assessment of gigapixel whole-slide images obtained from EMBs, which simultaneously addresses detection, subtyping and grading of allograft rejection. To assess model performance, we curated a large dataset from the United States, as well as independent test cohorts from Turkey and Switzerland, which includes large-scale variability across populations, sample preparations and slide scanning instrumentation. The model detects allograft rejection with an area under the receiver operating characteristic curve (AUC) of 0.962; assesses the cellular and antibody-mediated rejection type with AUCs of 0.958 and 0.874, respectively; detects Quilty B lesions, benign mimics of rejection, with an AUC of 0.939; and differentiates between low-grade and high-grade rejections with an AUC of 0.833. In a human reader study, the AI system showed non-inferior performance to conventional assessment and reduced interobserver variability and assessment time. This robust evaluation of cardiac allograft rejection paves the way for clinical trials to establish the efficacy of AI-assisted EMB assessment and its potential for improving heart transplant outcomes.