Tissue Contamination Challenges the Credibility of Machine Learning Models in Real World Digital Pathology.

Machine learning (ML) models are poised to transform surgical pathology practice. The most successful use attention mechanisms to examine whole slides, identify which areas of tissue are diagnostic, and use them to guide diagnosis. Tissue contaminants, such as floaters, represent unexpected tissue. Although human pathologists are extensively trained to consider and detect tissue contaminants, we examined their impact on ML models. We trained 4 whole-slide models. Three operate in placenta for the following functions: (1) detection of decidual arteriopathy, (2) estimation of gestational age, and (3) classification of macroscopic placental lesions. We also developed a model to detect prostate cancer in needle biopsies. We designed experiments wherein patches of contaminant tissue are randomly sampled from known slides and digitally added to patient slides and measured model performance. We measured the proportion of attention given to contaminants and examined the impact of contaminants in the t-distributed stochastic neighbor embedding feature space. Every model showed performance degradation in response to one or more tissue contaminants. Decidual arteriopathy detection--balanced accuracy decreased from 0.74 to 0.69 ± 0.01 with addition of 1 patch of prostate tissue for every 100 patches of placenta (1% contaminant). Bladder, added at 10% contaminant, raised the mean absolute error in estimating gestational age from 1.626 weeks to 2.371 ± 0.003 weeks. Blood, incorporated into placental sections, induced false-negative diagnoses of intervillous thrombi. Addition of bladder to prostate cancer needle biopsies induced false positives, a selection of high-attention patches, representing 0.033 mm2, and resulted in a 97% false-positive rate when added to needle biopsies. Contaminant patches received attention at or above the rate of the average patch of patient tissue. Tissue contaminants induce errors in modern ML models. The high level of attention given to contaminants indicates a failure to encode biological phenomena. Practitioners should move to quantify and ameliorate this problem.

Maternal Antibody Response and Transplacental Transfer Following Severe Acute Respiratory Syndrome Coronavirus 2 Infection or Vaccination in Pregnancy.

BACKGROUND: Pregnant persons are at increased risk of severe coronavirus disease 2019 (COVID-19) and adverse obstetric outcomes. Understanding maternal antibody response, duration, and transplacental transfer after severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection and COVID-19 vaccination is important to inform public health recommendations. METHODS: This prospective observational cohort study included 351 pregnant people who had SARS-CoV-2 infection or COVID-19 vaccination during pregnancy. Immunoglobulin (Ig) G and IgM to SARS-CoV-2 S1 receptor binding domain were measured in maternal and cord blood. Antibody levels and transplacental transfer ratios were compared across (1) disease severity for those with SARS-CoV-2 infection and (2) infection versus vaccination. RESULTS: There were 252 individuals with SARS-CoV-2 infection and 99 who received COVID-19 vaccination during pregnancy. Birthing people with more severe SARS-CoV-2 infection had higher maternal and cord blood IgG levels (P = .0001, P = .0001). Median IgG transfer ratio was 0.87-1.2. Maternal and cord blood IgG were higher after vaccination than infection (P = .001, P = .001). Transfer ratio was higher after 90 days in the vaccinated group (P < .001). Modeling showed higher amplitude and half-life of maternal IgG following vaccination (P < .0001). There were no significant differences by fetal sex. CONCLUSIONS: COVID-19 vaccination in pregnancy leads to higher and longer lasting maternal IgG levels, higher cord blood IgG, and higher transfer ratio after 90 days compared with SARS-CoV-2 infection. Greater infection severity leads to higher maternal and cord blood antibodies. Maternal IgG decreases over time following both vaccination and infection, reinforcing the importance of vaccination, even after infection, and vaccine boosters for pregnant patients.

Bronchopulmonary dysplasia is associated with polyhydramnios in a scan for novel perinatal risk factors.

BACKGROUND: The pathogenesis of bronchopulmonary dysplasia (BPD) is multifactorial, and there are limited data about prenatal exposures and risk of BPD. STUDY DESIGN: Our study performed parallel analyses using a logistic regression model in a cohort of 4527 infants with data from a curated registry and using a phenome wide association study (PheWAS) based on ICD9/10-based phecodes. We examined 20 prenatal exposures from a neonatal intensive care unit (NICU) curated registry database related to pregnancy and maternal health as well as 94 maternal diagnosis phecodes with a PheWAS analysis. RESULT: In both the curated registry and PheWAS analyses, polyhydramnios was associated with an increased risk of BPD (OR 5.70, 95% CI 2.78-11.44, p = 1.37 × 10-6). CONCLUSION: Our data suggest that polyhydramnios may be a clinical indicator of premature infants at increased risk for bronchopulmonary dysplasia. Combining curated registry data with PheWAS analysis creates a valuable tool to generate hypotheses. IMPACT: Polyhydramnios was significantly associated with bronchopulmonary dysplasia in both a curated registry and by ICD coding analysis with a phenome wide association study (PheWAS). Preterm polyhydramnios may be a clinical indicator of infants at increased risk for developing bronchopulmonary dysplasia after preterm birth. Combining curated registry with PheWAS analysis creates a valuable tool to generate hypotheses about perinatal risk factors and morbidities associated with preterm birth.

LabWAS: Novel findings and study design recommendations from a meta-analysis of clinical labs in two independent biobanks.

Phenotypes extracted from Electronic Health Records (EHRs) are increasingly prevalent in genetic studies. EHRs contain hundreds of distinct clinical laboratory test results, providing a trove of health data beyond diagnoses. Such lab data is complex and lacks a ubiquitous coding scheme, making it more challenging than diagnosis data. Here we describe the first large-scale cross-health system genome-wide association study (GWAS) of EHR-based quantitative laboratory-derived phenotypes. We meta-analyzed 70 lab traits matched between the BioVU cohort from the Vanderbilt University Health System and the Michigan Genomics Initiative (MGI) cohort from Michigan Medicine. We show high replication of known association for these traits, validating EHR-based measurements as high-quality phenotypes for genetic analysis. Notably, our analysis provides the first replication for 699 previous GWAS associations across 46 different traits. We discovered 31 novel associations at genome-wide significance for 22 distinct traits, including the first reported associations for two lab-based traits. We replicated 22 of these novel associations in an independent tranche of BioVU samples. The summary statistics for all association tests are freely available to benefit other researchers. Finally, we performed mirrored analyses in BioVU and MGI to assess competing analytic practices for EHR lab traits. We find that using the mean of all available lab measurements provides a robust summary value, but alternate summarizations can improve power in certain circumstances. This study provides a proof-of-principle for cross health system GWAS and is a framework for future studies of quantitative EHR lab traits.

GestAltNet: aggregation and attention to improve deep learning of gestational age from placental whole-slide images.

The placenta is the first organ to form and performs the functions of the lung, gut, kidney, and endocrine systems. Abnormalities in the placenta cause or reflect most abnormalities in gestation and can have life-long consequences for the mother and infant. Placental villi undergo a complex but reproducible sequence of maturation across the third-trimester. Abnormalities of villous maturation are a feature of gestational diabetes and preeclampsia, among others, but there is significant interobserver variability in their diagnosis. Machine learning has emerged as a powerful tool for research in pathology. To capture the volume of data and manage heterogeneity within the placenta, we developed GestaltNet, which emulates human attention to high-yield areas and aggregation across regions. We used this network to estimate the gestational age (GA) of scanned placental slides and compared it to a baseline model lacking the attention and aggregation functions. In the test set, GestaltNet showed a higher r2 (0.9444 vs. 0.9220) than the baseline model. The mean absolute error (MAE) between the estimated and actual GA was also better in the GestaltNet (1.0847 weeks vs. 1.4505 weeks). On whole-slide images, we found the attention sub-network discriminates areas of terminal villi from other placental structures. Using this behavior, we estimated GA for 36 whole slides not previously seen by the model. In this task, similar to that faced by human pathologists, the model showed an r2 of 0.8859 with an MAE of 1.3671 weeks. We show that villous maturation is machine-recognizable. Machine-estimated GA could be useful when GA is unknown or to study abnormalities of villous maturation, including those in gestational diabetes or preeclampsia. GestaltNet points toward a future of genuinely whole-slide digital pathology by incorporating human-like behaviors of attention and aggregation.

Seasonal Variation of Chronic Villitis of Unknown Etiology.

INTRODUCTION: Chronic villitis of unknown etiology (VUE) is a chronic inflammatory lesion of the placenta. VUE is hypothesized to result from an alloimmune response or as response to an unidentified infection. Lack of a seasonal trend is thought to support VUE as an alloimmune response, though data on seasonal VUE trends are limited. METHODS: Data were obtained from a hospital in Chicago, Illinois, from 2011-2016. Placentas sent to pathology were reviewed using a standardized protocol, and VUE cases were identified based on an automated text search of pathology records. We used monthly VUE prevalence estimates to investigate the annual trend, and we used Poisson regression to evaluate seasonal variation in the number of VUE cases. RESULTS: There were 79 825 deliveries within the study period. Pathologists evaluated 12 074 placentas and identified 2873 cases of VUE. Regression results indicate that the risk of VUE is 16% to 17% higher in the fall and winter as compared to the summer (fall relative risk [RR]: 1.17, 95% confidence interval [CI]: 1.06-1.29; winter RR: 1.16, 95% CI: 1.05-1.29). DISCUSSION: Our results suggest that there may be seasonal variation in VUE prevalence, particularly for low-grade VUE. Future studies should evaluate seasonal variation in a representative sample rather than relying on pathology reports to estimate prevalence.

Multi-region saliency-aware learning for cross-domain placenta image segmentation.

We propose a multi-region saliency-aware learning (MSL) method for cross-domain placenta image segmentation. Unlike most existing image-level transfer learning methods that fail to preserve the semantics of paired regions, our MSL incorporates the attention mechanism and a saliency constraint into the adversarial translation process, which can realize multi-region mappings in the semantic level. Specifically, the built-in attention module serves to detect the most discriminative semantic regions that the generator should focus on. Then we use the attention consistency as another guidance for retaining semantics after translation. Furthermore, we exploit the specially designed saliency-consistent constraint to enforce the semantic consistency by requiring the saliency regions unchanged. We conduct experiments using two real-world placenta datasets we have collected. We examine the efficacy of this approach in (1) segmentation and (2) prediction of the placental diagnoses of fetal and maternal inflammatory response (FIR, MIR). Experimental results show the superiority of the proposed approach over the state of the art.

Calcium channel blockers as drug repurposing candidates for gestational diabetes: Mining large scale genomic and electronic health records data to repurpose medications.

New therapeutic approaches are needed for gestational diabetes mellitus (GDM), but must show safety and efficacy in a historically understudied population. We studied associations between electronic medical record (EMR) phenotypes and genetic variants to uncover drugs currently considered safe in pregnancy that could treat or prevent GDM. We identified 129 systemically active drugs considered safe in pregnancy targeting the proteins produced from 196 genes. We tested for associations between GDM and/or type 2 diabetes (DM2) and 306 SNPs in 130 genes represented on the Illumina Infinium Human Exome Bead Chip (DM2 was included due to shared pathophysiological features with GDM). In parallel, we tested the association between drugs and glucose tolerance during pregnancy as measured by the glucose recorded during a routine 50-g glucose tolerance test (GTT). We found an association between GDM/DM2 and the genes targeted by 11 drug classes. In the EMR analysis, 6 drug classes were associated with changes in GTT. Two classes were identified in both analyses. L-type calcium channel blocking antihypertensives (CCBs), were associated with a 3.18 mg/dL (95% CI -6.18 to -0.18) decrease in glucose during GTT, and serotonin receptor type 3 (5HT-3) antagonist antinausea medications were associated with a 3.54 mg/dL (95% CI 1.86-5.23) increase in glucose during GTT. CCBs were identified as a class of drugs considered safe in pregnancy could have efficacy in treating or preventing GDM. 5HT-3 antagonists may be associated with worse glucose tolerance.

Excess SMAD signaling contributes to heart and muscle dysfunction in muscular dystrophy.

Disruption of the dystrophin complex causes muscle injury, dysfunction, cell death and fibrosis. Excess transforming growth factor (TGF) ß signaling has been described in human muscular dystrophy and animal models, where it is thought to relate to the progressive fibrosis that characterizes dystrophic muscle. We now found that canonical TGFß signaling acutely increases when dystrophic muscle is stimulated to contract. Muscle lacking the dystrophin-associated protein γ-sarcoglycan (Sgcg null) was subjected to a lengthening protocol to produce maximal muscle injury, which produced rapid accumulation of nuclear phosphorylated SMAD2/3. To test whether reducing SMAD signaling improves muscular dystrophy in mice, we introduced a heterozygous mutation of SMAD4 (S4) into Sgcg mice to reduce but not ablate SMAD4. Sgcg/S4 mice had improved body mass compared with Sgcg mice, which normally show a wasting phenotype similar to human muscular dystrophy patients. Sgcg/S4 mice had improved cardiac function as well as improved twitch and tetanic force in skeletal muscle. Functional enhancement in Sgcg/S4 muscle occurred without a reduction in fibrosis, suggesting that intracellular SMAD4 targets may be important. An assessment of genes differentially expressed in Sgcg muscle focused on those encoding calcium-handling proteins and responsive to TGFß since this pathway is a target for mediating improvement in muscular dystrophy. These data demonstrate that excessive TGFß signaling alters cardiac and muscle performance through the intracellular SMAD pathway.

Differential diagnostic considerations of desmoid-type fibromatosis.

Fibrous and myofibroblastic tumors of soft tissue often present the surgical pathologist with a difficult differential diagnosis because of the number of diagnostic possibilities and morphologic similarities among cytologically bland spindle-cell tumors. Prototypical in this regard is desmoid-type fibromatosis. In a review of 320 surgical specimens diagnosed as desmoid tumor, 94 (29%) were discovered to be misclassified as such. The most common lesions in this series were Gardner fibroma, scar tissue, superficial fibromatosis, nodular fasciitis, myofibroma, and collagenous fibroma. Four sarcomas were also misinterpreted as desmoid-type fibromatosis (3 low-grade fibromyxoid sarcomas and 1 unclassified spindle-cell sarcoma). We take this opportunity to compare and contrast desmoid tumor and several of the soft tissue tumors that should be considered in the differential diagnosis thereof.

Interplay between heart and skeletal muscle disease in heart failure: the 2011 George E. Brown Memorial Lecture.

The study of single gene disorders often provides insight for more complex human disease. Mutations in the genes encoding the dystrophin protein complex cause muscular dystrophy and cardiomyopathy by destabilizing the plasma membrane of skeletal myofibers and cardiomyocytes. In these diseases, progressive skeletal muscle degeneration and weakness contribute to cardiac dysfunction. Moreover, the pace and pattern of muscle weakness, along with onset of cardiomyopathy, is highly variable even when associated with the same identical mutation. Using a mouse model of muscular dystrophy and cardiomyopathy, we identified genetic loci that modify muscle pathology and cardiac fibrosis. Distinct genetic modifiers were identified for diaphragm and abdominal musculature, and these genetic intervals differ from those that regulate pathology in the skeletal muscle of the limbs and the heart. One modifier gene was identified and highlights the importance of the transforming growth factor-ß pathway in the pathogenesis of muscular dystrophy and cardiomyopathy. We determined that canonical transforming growth factor-ß signaling contributes to heart and muscle dysfunction using a Drosophila model. Together, these studies demonstrate the value of using a genetically sensitized model to uncover pathways that regulate heart failure and muscle weakness.

A population-level digital histologic biomarker for enhanced prognosis of invasive breast cancer.

Breast cancer is a heterogeneous disease with variable survival outcomes. Pathologists grade the microscopic appearance of breast tissue using the Nottingham criteria, which are qualitative and do not account for noncancerous elements within the tumor microenvironment. Here we present the Histomic Prognostic Signature (HiPS), a comprehensive, interpretable scoring of the survival risk incurred by breast tumor microenvironment morphology. HiPS uses deep learning to accurately map cellular and tissue structures to measure epithelial, stromal, immune, and spatial interaction features. It was developed using a population-level cohort from the Cancer Prevention Study-II and validated using data from three independent cohorts, including the Prostate, Lung, Colorectal, and Ovarian Cancer trial, Cancer Prevention Study-3, and The Cancer Genome Atlas. HiPS consistently outperformed pathologists in predicting survival outcomes, independent of tumor-node-metastasis stage and pertinent variables. This was largely driven by stromal and immune features. In conclusion, HiPS is a robustly validated biomarker to support pathologists and improve patient prognosis.

SMAD signaling drives heart and muscle dysfunction in a Drosophila model of muscular dystrophy.

Loss-of-function mutations in the genes encoding dystrophin and the associated membrane proteins, the sarcoglycans, produce muscular dystrophy and cardiomyopathy. The dystrophin complex provides stability to the plasma membrane of striated muscle during muscle contraction. Increased SMAD signaling due to activation of the transforming growth factor-ß (TGFß) pathway has been described in muscular dystrophy; however, it is not known whether this canonical TGFß signaling is pathogenic in the muscle itself. Drosophila deleted for the γ/δ-sarcoglycan gene (Sgcd) develop progressive muscle and heart dysfunction and serve as a model for the human disorder. We used dad-lacZ flies to demonstrate the signature of TGFß activation in response to exercise-induced injury in Sgcd null flies, finding that those muscle nuclei immediately adjacent to muscle injury demonstrate high-level TGFß signaling. To determine the pathogenic nature of this signaling, we found that partial reduction of the co-SMAD Medea, homologous to SMAD4, or the r-SMAD, Smox, corrected both heart and muscle dysfunction in Sgcd mutants. Reduction in the r-SMAD, MAD, restored muscle function but interestingly not heart function in Sgcd mutants, consistent with a role for activin but not bone morphogenic protein signaling in cardiac dysfunction. Mammalian sarcoglycan null muscle was also found to exhibit exercise-induced SMAD signaling. These data demonstrate that hyperactivation of SMAD signaling occurs in response to repetitive injury in muscle and heart. Reduction of this pathway is sufficient to restore cardiac and muscle function and is therefore a target for therapeutic reduction.

Machine learning classification of placental villous infarction, perivillous fibrin deposition, and intervillous thrombus.

INTRODUCTION: Placental parenchymal lesions are commonly encountered and carry significant clinical associations. However, they are frequently missed or misclassified by general practice pathologists. Interpretation of pathology slides has emerged as one of the most successful applications of machine learning (ML) in medicine with applications ranging from cancer detection and prognostication to transplant medicine. The goal of this study was to use a whole-slide learning model to identify and classify placental parenchymal lesions including villous infarctions, intervillous thrombi (IVT), and perivillous fibrin deposition (PVFD). METHODS: We generated whole slide images from placental discs examined at our institution with infarct, IVT, PVFD, or no macroscopic lesion. Slides were analyzed as a set of overlapping patches. We extracted feature vectors from each patch using a pretrained convolutional neural network (EfficientNetV2L). We trained a model to assign attention to each vector and used the attentions as weights to produce a pooled feature vector. The pooled vector was classified as normal or 1 of 3 lesions using a fully connected network. Patch attention was plotted to highlight informative areas of the slide. RESULTS: Overall balanced accuracy in a test set of held-out slides was 0.86 with receiver-operator characteristic areas under the curve of 0.917-0.993. Cases of PVFD were frequently miscalled as normal or infarcts, the latter possibly due to the perivillous fibrin found at the periphery of infarctions. We used attention maps to further understand some errors, including one most likely due to poor tissue fixation and processing. DISCUSSION: We used a whole-slide learning paradigm to train models to recognize three of the most common placental parenchymal lesions. We used attention maps to gain insight into model function, which differed from intuitive explanations.

Placental Pathology After SARS-CoV-2 Infection in the Pre-Variant of Concern, Alpha / Gamma, Delta, or Omicron Eras.

Objectives. The goal of this study is to describe placental pathology after infection with SARS-CoV-2 before the predominance of variants of concern (pre-VOC) and during eras of predominant transmission of the Alpha & Gamma (co-circulating), Delta, and Omicron variants. Methods. We used county-level variant data to establish population-level variant proportions, SARS-CoV-2 PCR to identify cases, and IgG serology to exclude latent infections from controls and histopathologic examination to identify placental pathology. Results. We report findings in 870 placentas from pregnancies complicated by SARS-CoV-2 including 90 with infection in the Alpha/Gamma era, 60 from the Delta era and 56 from the Omicron era. Features of maternal vascular malperfusion (MVM), including decidual arteriopathy, were significantly more frequent after SARS-CoV-2 infection. The risk of these findings varied over time, with the highest rates in the Delta era. Increased COVID-19 severity and the presence of comorbidities strengthened these associations. Conclusion. MVM is a feature of SARS-CoV-2 infection in pregnancy. Lesion frequency changed with the predominant circulating virus and should be considered with new variants.

Tissue contamination challenges the credibility of machine learning models in real world digital pathology.

Machine learning (ML) models are poised to transform surgical pathology practice. The most successful use attention mechanisms to examine whole slides, identify which areas of tissue are diagnostic, and use them to guide diagnosis. Tissue contaminants, such as floaters, represent unexpected tissue. While human pathologists are extensively trained to consider and detect tissue contaminants, we examined their impact on ML models. We trained 4 whole slide models. Three operate in placenta for 1) detection of decidual arteriopathy (DA), 2) estimation of gestational age (GA), and 3) classification of macroscopic placental lesions. We also developed a model to detect prostate cancer in needle biopsies. We designed experiments wherein patches of contaminant tissue are randomly sampled from known slides and digitally added to patient slides and measured model performance. We measured the proportion of attention given to contaminants and examined the impact of contaminants in T-distributed Stochastic Neighbor Embedding (tSNE) feature space. Every model showed performance degradation in response to one or more tissue contaminants. DA detection balanced accuracy decreased from 0.74 to 0.69 +/- 0.01 with addition of 1 patch of prostate tissue for every 100 patches of placenta (1% contaminant). Bladder, added at 10% contaminant raised the mean absolute error in estimating gestation age from 1.626 weeks to 2.371 +/ 0.003 weeks. Blood, incorporated into placental sections, induced false negative diagnoses of intervillous thrombi. Addition of bladder to prostate cancer needle biopsies induced false positives, a selection of high-attention patches, representing 0.033mm2, resulted in a 97% false positive rate when added to needle biopsies. Contaminant patches received attention at or above the rate of the average patch of patient tissue. Tissue contaminants induce errors in modern ML models. The high level of attention given to contaminants indicates a failure to encode biological phenomena. Practitioners should move to quantify and ameliorate this problem.

Enhancing Automatic Placenta Analysis through Distributional Feature Recomposition in Vision-Language Contrastive Learning.

The placenta is a valuable organ that can aid in understanding adverse events during pregnancy and predicting issues post-birth. Manual pathological examination and report generation, however, are laborious and resource-intensive. Limitations in diagnostic accuracy and model efficiency have impeded previous attempts to automate placenta analysis. This study presents a novel framework for the automatic analysis of placenta images that aims to improve accuracy and efficiency. Building on previous vision-language contrastive learning (VLC) methods, we propose two enhancements, namely Pathology Report Feature Recomposition and Distributional Feature Recomposition, which increase representation robustness and mitigate feature suppression. In addition, we employ efficient neural networks as image encoders to achieve model compression and inference acceleration. Experiments validate that the proposed approach outperforms prior work in both performance and efficiency by significant margins. The benefits of our method, including enhanced efficacy and deployability, may have significant implications for reproductive healthcare, particularly in rural areas or low- and middle-income countries.