Performance of a Chest Radiograph AI Diagnostic Tool for COVID-19: A Prospective Observational Study.

Purpose: To conduct a prospective observational study across 12 U.S. hospitals to evaluate real-time performance of an interpretable artificial intelligence (AI) model to detect COVID-19 on chest radiographs. Materials and Methods: A total of 95 363 chest radiographs were included in model training, external validation, and real-time validation. The model was deployed as a clinical decision support system, and performance was prospectively evaluated. There were 5335 total real-time predictions and a COVID-19 prevalence of 4.8% (258 of 5335). Model performance was assessed with use of receiver operating characteristic analysis, precision-recall curves, and F1 score. Logistic regression was used to evaluate the association of race and sex with AI model diagnostic accuracy. To compare model accuracy with the performance of board-certified radiologists, a third dataset of 1638 images was read independently by two radiologists. Results: Participants positive for COVID-19 had higher COVID-19 diagnostic scores than participants negative for COVID-19 (median, 0.1 [IQR, 0.0-0.8] vs 0.0 [IQR, 0.0-0.1], respectively; P < .001). Real-time model performance was unchanged over 19 weeks of implementation (area under the receiver operating characteristic curve, 0.70; 95% CI: 0.66, 0.73). Model sensitivity was higher in men than women (P = .01), whereas model specificity was higher in women (P = .001). Sensitivity was higher for Asian (P = .002) and Black (P = .046) participants compared with White participants. The COVID-19 AI diagnostic system had worse accuracy (63.5% correct) compared with radiologist predictions (radiologist 1 = 67.8% correct, radiologist 2 = 68.6% correct; McNemar P < .001 for both). Conclusion: AI-based tools have not yet reached full diagnostic potential for COVID-19 and underperform compared with radiologist prediction.Keywords: Diagnosis, Classification, Application Domain, Infection, Lung Supplemental material is available for this article.. © RSNA, 2022.

Nonlinear multi-objective flux balance analysis of the Warburg Effect.

Due to its implication in cancer treatment, the Warburg Effect has received extensive in silico investigation. Flux Balance Analysis (FBA), based on constrained optimization, was successfully applied in the Warburg Effect modelling. Yet, the assumption that cell types have one invariant cellular objective severely limits the applicability of the previous FBA models. Meanwhile, we note that cell types with different objectives show different extents of the Warburg Effect. To extend the applicability of the previous model and model the disparate cellular pathway preferences in different cell types, we built a Nonlinear Multi-Objective FBA (NLMOFBA) model by including three key objective terms (ATP production rate, lactate generation rate and ATP yield) into one objective function through linear scalarization. By constructing a cellular objective map and iteratively varying the objective weights, we showed disparate cellular pathway preferences manifested by different cell types driven by their unique cellular objectives, and we gained insights about the causal relationship between cellular objectives and the Warburg Effect. In addition, we obtained other biologically consistent results by using our NLMOFBA model. For example, augmented with the constraint associated with inefficient mitochondria function, low oxygen availability, or limited substrate, NLMOFBA predicts cellular pathways supported by the biology literature. Collectively, our NLMOFBA model can help build a complete understanding towards the Warburg Effect in different cell types. Finally, we investigated the impact of glutaminolysis, an important pathway related to glycolysis, on the occurrence of the Warburg Effect by using linear programming.

Potential and limitations of radiomics in neuro-oncology.

Radiomics seeks to apply classical methods of image processing to obtain quantitative parameters from imaging. Derived features are subsequently fed into algorithmic models to aid clinical decision making. The application of radiomics and machine learning techniques to clinical medicine remains in its infancy. The great potential of radiomics lies in its objective, granular approach to investigating clinical imaging. In neuro-oncology, advanced machine learning techniques, particularly deep learning, are at the forefront of new discoveries in the field. However, despite the great promise of machine learning aided radiomic approaches, the current use remains confined to scholarly research, without real-world deployment in neuro-oncology. The paucity of data, inconsistencies in preprocessing, radiomic feature instability, and the rarity of the events of interest are critical barriers to clinical translation. In this article, we will outline the major steps in the process of radiomics, as well as review advances and challenges in the field as they pertain to neuro-oncology.

A Prospective Observational Study to Investigate Performance of a Chest X-ray Artificial Intelligence Diagnostic Support Tool Across 12 U.S. Hospitals.

Importance: An artificial intelligence (AI)-based model to predict COVID-19 likelihood from chest x-ray (CXR) findings can serve as an important adjunct to accelerate immediate clinical decision making and improve clinical decision making. Despite significant efforts, many limitations and biases exist in previously developed AI diagnostic models for COVID-19. Utilizing a large set of local and international CXR images, we developed an AI model with high performance on temporal and external validation. Objective: Investigate real-time performance of an AI-enabled COVID-19 diagnostic support system across a 12-hospital system. Design: Prospective observational study. Setting: Labeled frontal CXR images (samples of COVID-19 and non-COVID-19) from the M Health Fairview (Minnesota, USA), Valencian Region Medical ImageBank (Spain), MIMIC-CXR, Open-I 2013 Chest X-ray Collection, GitHub COVID-19 Image Data Collection (International), Indiana University (Indiana, USA), and Emory University (Georgia, USA). Participants: Internal (training, temporal, and real-time validation): 51,592 CXRs; Public: 27,424 CXRs; External (Indiana University): 10,002 CXRs; External (Emory University): 2002 CXRs. Main Outcome and Measure: Model performance assessed via receiver operating characteristic (ROC), Precision-Recall curves, and F1 score. Results: Patients that were COVID-19 positive had significantly higher COVID-19 Diagnostic Scores (median .1 [IQR: 0.0-0.8] vs median 0.0 [IQR: 0.0-0.1], p < 0.001) than patients that were COVID-19 negative. Pre-implementation the AI-model performed well on temporal validation (AUROC 0.8) and external validation (AUROC 0.76 at Indiana U, AUROC 0.72 at Emory U). The model was noted to have unrealistic performance (AUROC > 0.95) using publicly available databases. Real-time model performance was unchanged over 19 weeks of implementation (AUROC 0.70). On subgroup analysis, the model had improved discrimination for patients with "severe" as compared to "mild or moderate" disease, p < 0.001. Model performance was highest in Asians and lowest in whites and similar between males and females. Conclusions and Relevance: AI-based diagnostic tools may serve as an adjunct, but not replacement, for clinical decision support of COVID-19 diagnosis, which largely hinges on exposure history, signs, and symptoms. While AI-based tools have not yet reached full diagnostic potential in COVID-19, they may still offer valuable information to clinicians taken into consideration along with clinical signs and symptoms.

Detection of Isocitrate Dehydrogenase Mutated Glioblastomas Through Anomaly Detection Analytics.

BACKGROUND: The rarity of Isocitrate Dehydrogenase mutated (mIDH) glioblastomas relative to wild-type IDH glioblastomas, as well as their distinct tumor physiology, effectively render them "outliers". Specialized tools are needed to identify these outliers. OBJECTIVE: To carefully craft and apply anomaly detection methods to identify mIDH glioblastoma based on radiomic features derived from magnetic resonance imaging. METHODS: T1-post gadolinium images for 188 patients and 138 patients were downloaded from The Cancer Imaging Archive's (TCIA) The Cancer Genome Atlas (TCGA) glioblastoma collection, and from the University of Minnesota Medical Center (UMMC), respectively. Anomaly detection methods were tested on glioblastoma image features for the precision of mIDH detection and compared to standard classification methods. RESULTS: Using anomaly detection training methods, we were able to detect IDH mutations from features in noncontrast-enhancing regions in glioblastoma with an average precision of 75.0%, 69.9%, and 69.8% using three different models. Anomaly detection methods consistently outperformed traditional two-class classification methods from 2 unique learning models (67.9%, 67.6%). The disparity in performances could not be overcome through newer, popular models such as neural networks (67.4%). CONCLUSION: We employed an anomaly detection strategy in the detection of IDH mutation in glioblastoma using preoperative T1 postcontrast imaging. We show these methods outperform traditional two-class classification in the setting of dataset imbalances inherent to IDH mutation prevalence in glioblastoma. We validate our results using an external dataset and highlight new possible avenues for radiogenomic rare event prediction in glioblastoma and beyond.

State of Radiomics in Glioblastoma.

Radiomics is an emerging discipline that aims to make intelligent predictions and derive medical insights based on quantitative features extracted from medical images as a means to improve clinical diagnosis or outcome. Pertaining to glioblastoma, radiomics has provided powerful, noninvasive tools for gaining insights into pathogenesis and therapeutic responses. Radiomic studies have yielded meaningful biological understandings of imaging features that are often taken for granted in clinical medicine, including contrast enhancement on glioblastoma magnetic resonance imaging, the distance of a tumor from the subventricular zone, and the extent of mass effect. They have also laid the groundwork for noninvasive detection of mutations and epigenetic events that influence clinical outcomes such as isocitrate dehydrogenase (IDH) and O6-methylguanine-DNA methyltransferase (MGMT). In this article, we review advances in the field of glioblastoma radiomics as they pertain to prediction of IDH mutation status and MGMT promoter methylation status, as well as the development of novel, higher order radiomic parameters.

Prokaryote autoimmunity in the context of self-targeting by CRISPR-Cas systems.

Prokaryote adaptive immunity (CRISPR-Cas systems) can be a threat to its carriers. We analyze the risks of autoimmune reactions related to adaptive immunity in prokaryotes by computational methods. We found important differences between bacteria and archaea with respect to autoimmunity potential. According to the results of our analysis, CRISPR-Cas systems in bacteria are more prone to self-targeting even though they possess fewer spacers per organism on average than archaea. The results of our study provide opportunities to use self-targeting in prokaryotes for biological and medical applications.

Linear programming model can explain respiration of fermentation products.

Many differentiated cells rely primarily on mitochondrial oxidative phosphorylation for generating energy in the form of ATP needed for cellular metabolism. In contrast most tumor cells instead rely on aerobic glycolysis leading to lactate to about the same extent as on respiration. Warburg found that cancer cells to support oxidative phosphorylation, tend to ferment glucose or other energy source into lactate even in the presence of sufficient oxygen, which is an inefficient way to generate ATP. This effect also occurs in striated muscle cells, activated lymphocytes and microglia, endothelial cells and several mammalian cell types, a phenomenon termed the "Warburg effect". The effect is paradoxical at first glance because the ATP production rate of aerobic glycolysis is much slower than that of respiration and the energy demands are better to be met by pure oxidative phosphorylation. We tackle this question by building a minimal model including three combined reactions. The new aspect in extension to earlier models is that we take into account the possible uptake and oxidation of the fermentation products. We examine the case where the cell can allocate protein on several enzymes in a varying distribution and model this by a linear programming problem in which the objective is to maximize the ATP production rate under different combinations of constraints on enzymes. Depending on the cost of reactions and limitation of the substrates, this leads to pure respiration, pure fermentation, and a mixture of respiration and fermentation. The model predicts that fermentation products are only oxidized when glucose is scarce or its uptake is severely limited.

Evolution and Vaccination of Influenza Virus.

In this study, we present an application paradigm in which an unsupervised machine learning approach is applied to the high-dimensional influenza genetic sequences to investigate whether vaccine is a driving force to the evolution of influenza virus. We first used a visualization approach to visualize the evolutionary paths of vaccine-controlled and non-vaccine-controlled influenza viruses in a low-dimensional space. We then quantified the evolutionary differences between their evolutionary trajectories through the use of within- and between-scatter matrices computation to provide the statistical confidence to support the visualization results. We used the influenza surface Hemagglutinin (HA) gene for this study as the HA gene is the major target of the immune system. The visualization is achieved without using any clustering methods or prior information about the influenza sequences. Our results clearly showed that the evolutionary trajectories between vaccine-controlled and non-vaccine-controlled influenza viruses are different and vaccine as an evolution driving force cannot be completely eliminated.

Mathematical models for explaining the Warburg effect: a review focussed on ATP and biomass production.

For producing ATP, tumour cells rely on glycolysis leading to lactate to about the same extent as on respiration. Thus, the ATP synthesis flux from glycolysis is considerably higher than in the corresponding healthy cells. This is known as the Warburg effect (named after German biochemist Otto H. Warburg) and also applies to striated muscle cells, activated lymphocytes, microglia, endothelial cells and several other cell types. For similar phenomena in several yeasts and many bacteria, the terms Crabtree effect and overflow metabolism respectively, are used. The Warburg effect is paradoxical at first sight because the molar ATP yield of glycolysis is much lower than that of respiration. Although a straightforward explanation is that glycolysis allows a higher ATP production rate, the question arises why cells do not re-allocate protein to the high-yield pathway of respiration. Mathematical modelling can help explain this phenomenon. Here, we review several models at various scales proposed in the literature for explaining the Warburg effect. These models support the hypothesis that glycolysis allows for a higher proliferation rate due to increased ATP production and precursor supply rates.

Tensor Dictionary Learning for Positive Definite Matrices.

Sparse models have proven to be extremely successful in image processing and computer vision. However, a majority of the effort has been focused on sparse representation of vectors and low-rank models for general matrices. The success of sparse modeling, along with popularity of region covariances, has inspired the development of sparse coding approaches for these positive definite descriptors. While in earlier work, the dictionary was formed from all, or a random subset of, the training signals, it is clearly advantageous to learn a concise dictionary from the entire training set. In this paper, we propose a novel approach for dictionary learning over positive definite matrices. The dictionary is learned by alternating minimization between sparse coding and dictionary update stages, and different atom update methods are described. A discriminative version of the dictionary learning approach is also proposed, which simultaneously learns dictionaries for different classes in classification or clustering. Experimental results demonstrate the advantage of learning dictionaries from data both from reconstruction and classification viewpoints. Finally, a software library is presented comprising C++ binaries for all the positive definite sparse coding and dictionary learning approaches presented here.

Tensor sparse coding for positive definite matrices.

In recent years, there has been extensive research on sparse representation of vector-valued signals. In the matrix case, the data points are merely vectorized and treated as vectors thereafter (for example, image patches). However, this approach cannot be used for all matrices, as it may destroy the inherent structure of the data. Symmetric positive definite (SPD) matrices constitute one such class of signals, where their implicit structure of positive eigenvalues is lost upon vectorization. This paper proposes a novel sparse coding technique for positive definite matrices, which respects the structure of the Riemannian manifold and preserves the positivity of their eigenvalues, without resorting to vectorization. Synthetic and real-world computer vision experiments with region covariance descriptors demonstrate the need for and the applicability of the new sparse coding model. This work serves to bridge the gap between the sparse modeling paradigm and the space of positive definite matrices.

Finding minimal generating set for metabolic network with reversible pathways.

Elementary flux modes give a mathematical representation of metabolic pathways in metabolic networks satisfying the constraint of non-decomposability. The large cost of their computation shifts attention to computing a minimal generating set which is a conically independent subset of elementary flux modes. When a metabolic network has reversible reactions and also admits a reversible pathway, the minimal generating set is not unique. A theoretical development and computational framework is provided which outline how to compute the minimal generating set in this case. The method is based on combining existing software to compute the minimal generating set for a "pointed cone" together with standard software to compute the Reduced Row Echelon Form.

Parallelization of Nullspace Algorithm for the computation of metabolic pathways.

Elementary mode analysis is a useful metabolic pathway analysis tool in understanding and analyzing cellular metabolism, since elementary modes can represent metabolic pathways with unique and minimal sets of enzyme-catalyzed reactions of a metabolic network under steady state conditions. However, computation of the elementary modes of a genome- scale metabolic network with 100-1000 reactions is very expensive and sometimes not feasible with the commonly used serial Nullspace Algorithm. In this work, we develop a distributed memory parallelization of the Nullspace Algorithm to handle efficiently the computation of the elementary modes of a large metabolic network. We give an implementation in C++ language with the support of MPI library functions for the parallel communication. Our proposed algorithm is accompanied with an analysis of the complexity and identification of major bottlenecks during computation of all possible pathways of a large metabolic network. The algorithm includes methods to achieve load balancing among the compute-nodes and specific communication patterns to reduce the communication overhead and improve efficiency.

Effects of age and dementia on temporal cycles in spontaneous speech fluency.

Spontaneous speech of healthy adults consists of alternating periods of fluent and hesitant segments, forming temporal cycles in speech fluency. The regularity of these cycles may be related to the functioning of brain networks during speech planning and execution. This paper investigates the theoretical link between human cognitive functioning and temporal cycles in speech production using a quantitative time series analysis to characterize the regularity and frequency of temporal cycles in adults with differing levels and etiology of cognitive decline. We compare spontaneous speech of adults without a neurological diagnosis, both older and younger, to that of adults with frontotemporal lobar degeneration (FTLD). Two measures of temporal cycle frequency (mean and mode) calculated from the power spectrum of speech fluency represented as a time series were found to be associated with subjects' age, regardless of diagnosis of dementia. Two measures of periodicity (g-statistic and rhythmicity-index), as well as mean frequency, differentiated between adults with and without dementia. Our study confirms the presence of regular temporal cycles in spontaneous speech and suggests that temporal cycle characteristics are affected in different ways by declines in cognitive functioning due to dementia and aging.

On algebraic properties of extreme pathways in metabolic networks.

We give a concise development of some of the major algebraic properties of extreme pathways (pathways that cannot be the result of combining other pathways) of metabolic networks, contrasting them to those of elementary flux modes (pathways involving a minimal set of reactions). In particular, we show that an extreme pathway can be recognized by a rank test as simple as the existing rank test for elementary flux modes, without computing all the modes. We make the observation that, unlike elementary flux modes, the property of being an extreme pathway depends on the presence or absence of reactions beyond those involved in the pathway itself. Hence, the property of being an extreme pathway is not a local property. As a consequence, we find that the set of all elementary flux modes for a network includes all the elementary flux modes for all its subnetworks, but that this property does not hold for the set of all extreme pathways.

Transcriptional response of Pasteurella multocida to defined iron sources.

Pasteurella multocida was grown in iron-free chemically defined medium supplemented with hemoglobin, transferrin, ferritin, and ferric citrate as iron sources. Whole-genome DNA microarrays were used to monitor global gene expression over seven time points after the addition of the defined iron source to the medium. This resulted in a set of data containing over 338,000 gene expression observations. On average, 12% of P. multocida genes were differentially expressed under any single condition. A majority of these genes encoded P. multocida proteins that were involved in either transport and binding or were annotated as hypothetical proteins. Several trends are evident when the data from different iron sources are compared. In general, only two genes (ptsN and sapD) were expressed at elevated levels under all of the conditions tested. The results also show that genes with increased expression in the presence of hemoglobin did not respond to transferrin or ferritin as an iron source. Correspondingly, genes with increased expression in the transferrin and ferritin experiments were expressed at reduced levels when hemoglobin was supplied as the sole iron source. Finally, the data show that genes that were most responsive to the presence of ferric citrate did not follow a trend similar to that of the other iron sources, suggesting that different pathways respond to inorganic or organic sources of iron in P. multocida. Taken together, our results demonstrate that unique subsets of P. multocida genes are expressed in response to different iron sources and that many of these genes have yet to be functionally characterized.