Average Average Area Under The Curve Research Articles

Severity Stratification of Coronary Artery Disease Using Novel Inner Ellipse-Based Foveal Avascular Zone Biomarkers.

Given the similarities between the retinal and coronary microvasculature, the retina holds promising potential to serve as a non-invasive screening tool for coronary artery disease (CAD). We aimed to develop novel inner ellipse-based metrics and discern whether foveal avascular zone (FAZ) alterations can serve as indicators for CAD presence and severity. Patients admitted to the Department of Cardiology who underwent coronary angiography were included. This resulted in an inclusion of 212 patients, of which 73 had no CAD. During the same visit, 6 × 6-mm (nominal size) fovea-centered optical coherence tomography angiography images of both eyes were acquired. The Gensini score (GS) was utilized to quantify CAD severity. Six known FAZ shape metrics were assessed and three novel biomarkers based on the inner ellipse were defined: absolute inner ellipse difference, Hausdorff distance, and Chamfer distance. Eight out of nine metrics showed significant associations with the GS in the left eye. However, significant differences across three CAD severity groups were only demonstrated by the novel metrics. Utilizing the Chamfer distance, age, and sex, patients with and without CAD could be distinguished with an average area under the curve (AUC) of 0.89 (95% confidence interval [CI], 0.84-0.95). Moreover, three CAD severity groups could be discerned with a macro average AUC of 0.77 (95% CI, 0.72-0.84). A comprehensive assessment of FAZ shape descriptors was performed, and a strong association with CAD was found. The inner ellipse-based biomarkers especially demonstrated high predictive abilities for CAD presence and severity.

Effect of patient specificity on predicting knee cartilage degeneration in obese adults: Musculoskeletal finite-element modeling of data from the CAROT trial.

Obesity is a known risk factor for development of osteoarthritis (OA). Numerical tools like finite-element (FE) models combined with degenerative algorithms have been developed to understand the interplay between OA and obesity. Inthisstudy,we aimed to predict knee cartilage degeneration in a cohort of obese adults to investigate the importance of patient-specific information on degeneration predictions. We used a validated FE modeling approach and three different age-dependent functions (step-wise, exponential, and linear) to simulate cartilage degradation under overloading in the knee joint. Gait motion analysis and magnetic resonance imaging data from 115 obese individuals with knee OA were used for musculoskeletal and FE modeling. Cartilage degeneration predictions were contrasted with Kellgren-Lawrence (KL)and Boston-Leeds Osteoarthritis Knee Score (BLOKS) grades. The findings show that overall, the similarities between numerical predictions and clinical measures were better for the medial (average area under the curve(AUC) = 0.62) compared to the lateral compartment (average AUC = 0.52) of the knee. Classification results for KL grades, full patient-specific models and patient-specific geometry with generic gait data showed higher AUC values (AUC = 0.71 and AUC = 0.68, respectively) compared to generic geometry and patient-specific gait (AUC = 0.48). For BLOKS grades, AUC values for both full patient-specific models and for patient-specific geometry with generic gait locomotion were higher (AUC = 0.66 and AUC = 0.64, respectively) compared to when the generic geometry and patient-specific gait were used (AUC = 0.53). In summary, our study highlights the importance of considering individual information in knee OA prediction. Nevertheless, our findings suggest that personalized gait play a smaller role in the OA prediction and classification capacity than personalized joint geometry.

Thermal and visual comforts of occupants for a naturally ventilated educational building in low-income economies: A machine learning approach

Thermal and visual comfort (TVC) significantly impact the health and productivity of occupants as well as the indoor environmental quality in educational buildings. Machine learning approaches have currently gained popularity for TVC prediction; however, accurately predicting TVC faces challenges due to instrumental errors, environmental noise, and imbalanced data. Moreover, studies have revealed the unreliability of TVC predictions, particularly in naturally ventilated buildings. Therefore, this study develops three machine learning models, random forest (RF), decision tree (DT), and XGBoost, to accurately predict TVC using 1310 sets of field survey data collected from a naturally ventilated educational building (NVEB) in Dhaka, Bangladesh. Recursive feature elimination with cross-validation is employed to identify optimal features, while two resampling methods are applied to manage imbalanced classification. The RF model achieves the highest accuracy of 96 % for thermal comfort prediction, while the XGBoost model attains the highest accuracy of 95 % for visual comfort prediction. Meanwhile, the DT model achieves an accuracy of 94 % and 93 % for TVC prediction, respectively. Moreover, both RF and XGBoost outperform DT in class prediction performance, with an average area under the curve (AUC) scores above 0.95, while DT achieves an average AUC score above 0.80. In terms of two resampling methods, the results show that the models developed using the synthetic minority over-sampling technique (SMOTE) + edited nearest neighbour approach leads to less misclassification compared to the SMOTE + Tomek link. These findings demonstrate the capability of machine learning models in accurately predicting TVC in NVEBs and handling imbalanced datasets.

Lung Cancer Diagnosis on Virtual Histologically Stained Tissue Using Weakly Supervised Learning

Lung adenocarcinoma (LUAD) is the most common primary lung cancer and accounts for 40% of all lung cancer cases. The current gold standard for lung cancer analysis is based on the pathologists’ interpretation of hematoxylin and eosin (H&E)-stained tissue slices viewed under a brightfield microscope or a digital slide scanner. Computational pathology using deep learning has been proposed to detect lung cancer on histology images. However, the histological staining workflow to acquire the H&E-stained images and the subsequent cancer diagnosis procedures are labor-intensive and time-consuming with tedious sample preparation steps and repetitive manual interpretation, respectively. In this work, we propose a weakly supervised learning method for LUAD classification on label-free tissue slices with virtual histological staining. The autofluorescence images of label-free tissue with histopathological information can be converted into virtual H&E-stained images by a weakly supervised deep generative model. For the downstream LUAD classification task, we trained the attention-based multiple-instance learning model with different settings on the open-source LUAD H&E-stained whole-slide images (WSIs) dataset from the Cancer Genome Atlas (TCGA). The model was validated on the 150 H&E-stained WSIs collected from patients in Queen Mary Hospital and Prince of Wales Hospital with an average area under the curve (AUC) of 0.961. The model also achieved an average AUC of 0.973 on 58 virtual H&E-stained WSIs, comparable to the results on 58 standard H&E-stained WSIs with an average AUC of 0.977. The attention heatmaps of virtual H&E-stained WSIs and ground-truth H&E-stained WSIs can indicate tumor regions of LUAD tissue slices. In conclusion, the proposed diagnostic workflow on virtual H&E-stained WSIs of label-free tissue is a rapid, cost effective, and interpretable approach to assist clinicians in postoperative pathological examinations. The method could serve as a blueprint for other label-free imaging modalities and disease contexts.

A robust microbiome signature for autism spectrum disorder across different studies using machine learning

Autism spectrum disorder (ASD) is a highly complex neurodevelopmental disorder characterized by deficits in sociability and repetitive behaviour, however there is a great heterogeneity within other comorbidities that accompany ASD. Recently, gut microbiome has been pointed out as a plausible contributing factor for ASD development as individuals diagnosed with ASD often suffer from intestinal problems and show a differentiated intestinal microbial composition. Nevertheless, gut microbiome studies in ASD rarely agree on the specific bacterial taxa involved in this disorder. Regarding the potential role of gut microbiome in ASD pathophysiology, our aim is to investigate whether there is a set of bacterial taxa relevant for ASD classification by using a sibling-controlled dataset. Additionally, we aim to validate these results across two independent cohorts as several confounding factors, such as lifestyle, influence both ASD and gut microbiome studies. A machine learning approach, recursive ensemble feature selection (REFS), was applied to 16S rRNA gene sequencing data from 117 subjects (60 ASD cases and 57 siblings) identifying 26 bacterial taxa that discriminate ASD cases from controls. The average area under the curve (AUC) of this specific set of bacteria in the sibling-controlled dataset was 81.6%. Moreover, we applied the selected bacterial taxa in a tenfold cross-validation scheme using two independent cohorts (a total of 223 samples—125 ASD cases and 98 controls). We obtained average AUCs of 74.8% and 74%, respectively. Analysis of the gut microbiome using REFS identified a set of bacterial taxa that can be used to predict the ASD status of children in three distinct cohorts with AUC over 80% for the best-performing classifiers. Our results indicate that the gut microbiome has a strong association with ASD and should not be disregarded as a potential target for therapeutic interventions. Furthermore, our work can contribute to use the proposed approach for identifying microbiome signatures across other 16S rRNA gene sequencing datasets.

Multi-kingdom gut microbiota analyses define bacterial-fungal interplay and microbial markers of pan-cancer immunotherapy across cohorts

The effect of gut bacteria on the response to immune checkpoint inhibitors (ICIs) has been studied, but the relationship between fungi and ICI responses is not fully understood. Herein, 862 fecal metagenomes from 9 different cohorts were integrated for the identification of differentially abundant fungi and subsequent construction of random forest (RF) models to predict ICI responses. Fungal markers demonstrate excellent performance, with an average area under the curve (AUC) of 0.87. Their performance improves even further, reaching an average AUC of 0.89 when combined with bacterial markers. Higher enrichment of exhausted Tcells is detected in responders, as predicted by fungal markers. Multi-kingdom network and functional analysis reveal that the fungus Schizosaccharomyces octosporus may ferment starch into short-chain fatty acids in responders. This study provides a fungal profile of the ICI response and the identification of multi-kingdom microbial markers with good performance that may improve the overall applicability of ICI therapy.

Urinary neopterin and biopterin indicate that inflammation has a role in autism spectrum disorder.

Inflammation is thought to be involved in the pathogenesis of autism spectrum disorder (ASD). Pteridine metabolites are biomarkers of inflammation that increase on immune system activation. In this study, we investigated the urinary pteridine metabolites in ASD patients as a possible biomarker for immune activation and inflammation. This observational, cross-sectional, prospective study collected urine samples from 212 patients with ASD and 68 age- and sex-matched healthy individuals. Urine neopterin (NE) and biopterin (BIO) levels were measured. Patients who had chronic disorders, active infection at the time of sampling, or high C-reactive protein levels were excluded. The urine NE and BIO concentrations were determined by high-performance liquid chromatography. The ratios of both NE and BIO to creatinine (CRE) were used to standardise the measurements. The NE/CRE and NE/BIO levels were significantly higher in ASD patients than controls. Univariate and multivariate models revealed a significant increase in NE/CRE and NE/BIO in ASD patients. There was a significant relationship between the NE/BIO [average area under the curve (AUC) = 0.717; range: 0.637-0.797] and NE/CRE (average AUC = 0.756; range: 0.684-0.828) ratios, which distinguished individuals with ASD from controls. The elevated NE/CRE and NE/BIO ratios suggest that inflammation and T cell-mediated immunity are involved in the pathophysiology of autism. NE/BIO could serve as a diagnostic inflammatory marker in the pathogenesis of ASD.

FedScore: A privacy-preserving framework for federated scoring system development

ObjectiveWe propose FedScore, a privacy-preserving federated learning framework for scoring system generation across multiple sites to facilitate cross-institutional collaborations. Materials and methodsThe FedScore framework includes five modules: federated variable ranking, federated variable transformation, federated score derivation, federated model selection and federated model evaluation. To illustrate usage and assess FedScore’s performance, we built a hypothetical global scoring system for mortality prediction within 30 days after a visit to an emergency department using 10 simulated sites divided from a tertiary hospital in Singapore. We employed a pre-existing score generator to construct 10 local scoring systems independently at each site and we also developed a scoring system using centralized data for comparison. ResultsWe compared the acquired FedScore model’s performance with that of other scoring models using the receiver operating characteristic (ROC) analysis. The FedScore model achieved an average area under the curve (AUC) value of 0.763 across all sites, with a standard deviation (SD) of 0.020. We also calculated the average AUC values and SDs for each local model, and the FedScore model showed promising accuracy and stability with a high average AUC value which was closest to the one of the pooled model and SD which was lower than that of most local models. ConclusionThis study demonstrates that FedScore is a privacy-preserving scoring system generator with potentially good generalizability.

Deep Modelling Strategies for Human Confidence Classification using Audio-visual Data.

Human behavior expressions such as of confidence are time-varying entities. Both vocal and facial cues that convey the human confidence expressions keep varying throughout the duration of analysis. Although, the cues from these two modalities are not always in synchrony, they impact each other and the fused outcome as well. In this paper, we present a deep fusion technique to combine the two modalities and derive a single outcome to infer human confidence. Fused outcome improves the classification performance by capturing the temporal information from both the modalities. The analysis of time-varying nature of expressions in the conversations captured in an interview setup is also presented. We collected data from 51 speakers who participated in interview sessions. The average area under the curve (AUC) of uni-modal models using speech and facial expressions is 70.6% and 69.4%, respectively, for classifying confident videos from non-confident ones in 5-fold cross-validation analysis. Our deep fusion model improves the performance giving an average AUC of 76.8%.

Attention feature fusion methodology with additional constraint for ovarian lesion diagnosis on magnetic resonance images.

It is challenging for radiologists and gynecologists to identify the type of ovarian lesions by reading magnetic resonance (MR) images. Recently developed convolutional neural networks (CNNs) have made great progress in computer vision, but their architectures still need modification if they are used in processing medical images. This study aims to improve the feature extraction capability of CNNs, thus promoting the diagnostic performance in discriminating between benign and malignant ovarianlesions. We introduce a feature fusion architecture and insert the attention models in the neural network. The features extracted from different middle layers are integrated with reoptimized spatial and channel weights. We add a loss function to constrain the additional probability vector generated from the integrated features, thus guiding the middle layers to emphasize useful information. We analyzed 159 lesions imaged by dynamic contrast-enhanced MR imaging (DCE-MRI), including 73 benign lesions and 86 malignant lesions. Senior radiologists selected and labeled the tumor regions based on the pathology reports. Then, the tumor regions were cropped into 7494 nonoverlapping image patches for training and testing. The type of a single tumor was determined by the average probability scores of the image patches belonging toit. We implemented fivefold cross-validation to characterize our proposed method, and the distribution of performance matrics was reported. For all the test image patches, the average accuracy of our method is 70.5% with an average area under the curve (AUC) of 0.785, while the baseline is 69.4% and 0.773, and for the diagnosis of single tumors, our model achieved an average accuracy of 82.4% and average AUC of 0.916, which were better than the baseline (81.8% and 0.899). Moreover, we evaluated the performance of our proposed method utilizing different CNN backbones and different attention mechanisms. The texture features extracted from different middle layers are crucial for ovarian lesion diagnosis. Our proposed method can enhance the feature extraction capabilities of different layers of the network, thereby improving diagnosticperformance.

Extreme gradient boosting model to assess risk of central cervical lymph node metastasis in patients with papillary thyroid carcinoma: Individual prediction using SHapley Additive exPlanations

Background and ObjectivesCentral cervical lymph node metastasis (CLNM) is considered a risk factor for recurrence in patients with papillary thyroid carcinoma (PTC). Traditional machine learning models suffered from “black-box” problems, which could not exactly explain the interactive effects of the risk factors. We aimed to develop an eXtreme Gradient Boosting (XGBoost) model to assess CLNM, including positive and negative effects. Methods1,122 patients with PTC admitted at Tianjin First Central Hospital from 2016 to 2020 were retrospectively selected. They were randomly divided into the training and test datasets with an 8:2 ratio. 108 patients with PTC admitted at Binzhou Medical University Hospital in 2020 served as the validation dataset. The XGBoost model was used to assess CLNM. The 10-fold cross-validation was utilized for model selection, and the metric used to evaluate classification performance was the average area under the curve (AUC) of 10-fold cross-validation. Interpretation and transparency of the “black-box” problem were performed. SHapley Additive exPlanations (SHAP) and local interpretable model-agnostic explanation (LIME) were used to ensure the stability and reliability of the model. ResultsThe XGBoost model based on ultrasound and dual-energy computed tomography images of the solitary primary lesion had an excellent performance for assessing CLNM, with average AUCs of 0.918, 0.903, and 0.881 in the training, test, and validation datasets, respectively. SHAP plots showed the influence of each parameter on the XGBoost model, including positive (i.e., capsular invasion, diameter, iodine concentration in the venous phase, and calcification) and negative (i.e., sex and age) impacts. For all cases, the capsular invasion prediction weight was the highest; for individual cases, different predictors were assigned different weights. Moreover, the performance of the XGBoost model was better than classical machine-learning models. ConclusionsThis study developed and validated an XGBoost model for assessing CLNM in patients with PTC. The ability to visually interpret the positive and negative effects made the XGBoost model an effective tool for guiding clinical treatment.

Machine Learning Based on MRI DWI Radiomics Features for Prognostic Prediction in Nasopharyngeal Carcinoma.

Simple SummaryIn the past, radiomics studies of nasopharyngeal carcinoma (NPC) were only based on basic MR sequences. Previous studies have shown that radiomics methods based on T2-weighted imaging and contrast-enhanced T1-weighted imaging have been successfully used to improve the prognosis of patients with nasopharyngeal carcinoma. The purpose of this study was to explore the predictive efficacy of radiomics analyses based on readout-segmented echo-planar diffusion-weighted imaging (RESOLVE-DWI) which quantitatively reflects the diffusion motion of water molecules for prognosis evaluation in nasopharyngeal carcinoma. Several prognostic radiomics models were established by using diffusion-weighted imaging, apparent diffusion coefficient maps, T2-weighted and contrast-enhanced T1-weighted imaging to predict the risk of recurrence or metastasis of nasopharyngeal carcinoma, and the predictive effects of different models were compared. The results show that the model based on MRI DWI can successfully predict the prognosis of patients with nasopharyngeal carcinoma and has higher predictive efficiency than the model based on the conventional sequence, which suggests MRI DWI-radiomics can provide a useful and alternative approach for survival estimation.Purpose: This study aimed to explore the predictive efficacy of radiomics analyses based on readout-segmented echo-planar diffusion-weighted imaging (RESOLVE-DWI) for prognosis evaluation in nasopharyngeal carcinoma in order to provide further information for clinical decision making and intervention. Methods: A total of 154 patients with untreated NPC confirmed by pathological examination were enrolled, and the pretreatment magnetic resonance image (MRI)—including diffusion-weighted imaging (DWI), apparent diffusion coefficient (ADC) maps, T2-weighted imaging (T2WI), and contrast-enhanced T1-weighted imaging (CE-T1WI)—was collected. The Random Forest (RF) algorithm selected radiomics features and established the machine-learning models. Five models, namely model 1 (DWI + ADC), model 2 (T2WI + CE-T1WI), model 3 (DWI + ADC + T2WI), model 4 (DWI + ADC + CE-T1WI), and model 5 (DWI + ADC + T2WI + CE-T1WI), were constructed. The average area under the curve (AUC) of the validation set was determined in order to compare the predictive efficacy for prognosis evaluation. Results: After adjusting the parameters, the RF machine learning models based on extracted imaging features from different sequence combinations were obtained. The invalidation sets of model 1 (DWI + ADC) yielded the highest average AUC of 0.80 (95% CI: 0.79–0.81). The average AUCs of the model 2, 3, 4, and 5 invalidation sets were 0.72 (95% CI: 0.71–0.74), 0.66 (95% CI: 0.64–0.68), 0.74 (95% CI: 0.73–0.75), and 0.75 (95% CI: 0.74–0.76), respectively. Conclusion: A radiomics model derived from the MRI DWI of patients with nasopharyngeal carcinoma was generated in order to evaluate the risk of recurrence and metastasis. The model based on MRI DWI can provide an alternative approach for survival estimation, and can reveal more information for clinical decision-making and intervention.

Improving Spatiotemporal Transferability of Real-Time Crash Likelihood Prediction Models Using Transfer-Learning Approaches

A real-time crash likelihood prediction model is an important component of the proactive traffic safety management system. Over the past decades, numerous models have been proposed for predicting real-time crash likelihood and achieved promising results. However, most studies have ignored the model transferability, especially for deep-learning models. The transferability of a model could be defined as applying the model to new data which obtain from different locations or periods. The purpose of this study is to improve the spatiotemporal transferability of a deep-learning crash likelihood prediction model by using transfer-learning approaches. Trajectory and crash data from five arterials in Florida were used in this study. A two-layer long short-term memory (LSTM) model was developed to predict real-time crash likelihood. Two scenarios were created for investigating spatial and temporal transferability of the developed model. Experimental results suggested that the model could be accurately transferred to new data by using the fine-tuning transfer-learning approach. In the framework of this research, the transferred models achieved higher predictive accuracy than models which were directly developed on the new data. Moreover, the spatial transfer learning outperformed the temporal transfer learning in predictive accuracy. For example, when the percentage of test data was 70%, the spatial transferred model had an average area under the curve (AUC) of 0.74. while the temporal transferred model had an average AUC of 0.65. Results from this study could be applied to transfer pretrained crash likelihood prediction models to new locations when few crashes are available.

Toward Suicidal Ideation Detection with Lexical Network Features and Machine Learning

In this study, we introduce a new network feature for detecting suicidal ideation from clinical texts and conduct various additional experiments to enrich the state of knowledge. We evaluate statistical features with and without stopwords, use lexical networks for feature extraction and classification, and compare the results with standard machine learning methods using a logistic classifier, a neural network, and a deep learning method. We utilize three text collections. The first two contain transcriptions of interviews conducted by experts with suicidal (n=161 patients that experienced severe ideation) and control subjects (n=153). The third collection consists of interviews conducted by experts with epilepsy patients, with a few of them admitting to experiencing suicidal ideation in the past (32 suicidal and 77 control). The selected methods detect suicidal ideation with an average area under the curve (AUC) score of 95% on the merged collection with high suicidal ideation, and the trained models generalize over the third collection with an average AUC score of 69%. Results reveal that lexical networks are promising for classification and feature extraction as successful as the deep learning model. We also observe that a logistic classifier’s performance was comparable with the deep learning method while promising explainability.

Differentiation of pancreatic ductal adenocarcinoma and chronic pancreatitis using graph neural networks on histopathology and collagen fiber features

Pancreatic ductal adenocarcinoma (PDAC) is one of the most lethal human cancers. However, the symptoms and radiographic appearance of chronic pancreatitis (CP) mimics that of PDAC, and sometimes the 2 entities can also be difficult to differentiate microscopically. The need for accurate differentiation of PDAC and CP has become a major topic in pancreatic pathology. These 2 diseases can present similar histomorphological features, such as excessive deposition of fibrotic stroma in the tissue microenvironment and inflammatory cell infiltration. In this paper, we present a quantitative analysis pipeline empowered by graph neural networks (GNN) capable of automatic detection and differentiation of PDAC and CP in human histological specimens. Modeling histological images as graphs and deploying graph convolutions can enable the capture of histomorphological features at different scales, ranging from nuclear size to the organization of ducts. The analysis pipeline combines image features computed from co-registered hematoxylin and eosin (H&E) images and Second-Harmonic Generation (SHG) microscopy images, with the SHG images enabling the extraction of collagen fiber morphological features. Evaluating the analysis pipeline on a human tissue micro-array dataset consisting of 786 cores and a tissue region dataset consisting of 268 images, it attained 86.4% accuracy with an average area under the curve (AUC) of 0.954 and 88.9% accuracy with an average AUC of 0.957, respectively. Moreover, incorporating topological features of collagen fibers computed from SHG images into the model further increases the classification accuracy on the tissue region dataset to 91.3% with an average AUC of 0.962, suggesting that collagen characteristics are diagnostic features in PDAC and CP detection and differentiation.

Huanglongbing and Citrus Variegated Chlorosis Integrated Management Based on Favorable Periods for Vector Population Increase and Symptom Expression.

Huanglongbing (HLB, associated with Candidatus Liberibacter asiaticus and transmitted by the Asian citrus psyllid Diaphorina citri) and citrus variegated chlorosis (CVC, caused by Xylella fastidiosa subsp. pauca and transmitted by sharpshooter species) have been managed by vector control and removal of symptomatic trees. Although vectors and new symptomatic trees can be detected year round, peaks of vector populations are higher in spring and summer, and the most symptomatic trees are found in autumn and winter. This work aimed to compare the management of both diseases during these favorable periods. The experiment was conducted during 5 years in a commercial orchard and had a three-by-two factorial design. The factor "vector control" had three levels: monthly vector control year round (VCYR), monthly vector control in spring and summer (VCSS), and vector control when a threshold level of 10% occupancy was detected (VCOT). The factor "inoculum removal" had two levels: monthly eradication year round (TEYR) and monthly eradication in autumn and winter (TEAW). Host flush, both HLB and CVC vector populations, and the number of symptomatic citrus plants were visually assessed. The level of vectors over the seasons, as measured using the average area under the curve (AUC), was similar for all treatments with the exception of psyllid abundance, which was approximately 4.5 times higher for VCSS than for other treatments. For both diseases, no difference in the average AUC of disease progress and disease final incidence was observed. VCOT or adjusted VCSS associated with TEAW could be integrated for sustainable citrus production.

Federated learning for predicting clinical outcomes in patients with COVID-19.

Federated learning (FL) is a method used for training artificial intelligence models with data from multiple sources while maintaining data anonymity, thus removing many barriers to data sharing. Here we used data from 20 institutes across the globe to train a FL model, called EXAM (electronic medical record (EMR) chest X-ray AI model), that predicts the future oxygen requirements of symptomatic patients with COVID-19 using inputs of vital signs, laboratory data and chest X-rays. EXAM achieved an average area under the curve (AUC) >0.92 for predicting outcomes at 24 and 72 h from the time of initial presentation to the emergency room, and it provided 16% improvement in average AUC measured across all participating sites and an average increase in generalizability of 38% when compared with models trained at a single site using that site's data. For prediction of mechanical ventilation treatment or death at 24 h at the largest independent test site, EXAM achieved a sensitivity of 0.950 and specificity of 0.882. In this study, FL facilitated rapid data science collaboration without data exchange and generated a model that generalized across heterogeneous, unharmonized datasets for prediction of clinical outcomes in patients with COVID-19, setting the stage for the broader use of FL in healthcare.

An AI-based auxiliary empirical antibiotic therapy model for children with bacterial pneumonia using low-dose chest CT images

PurposeTo construct an auxiliary empirical antibiotic therapy (EAT) multi-class classification model for children with bacterial pneumonia using radiomics features based on artificial intelligence and low-dose chest CT images.Materials and methodsData were retrospectively collected from children with pathogen-confirmed bacterial pneumonia including Gram-positive bacterial pneumonia (122/389, 31%), Gram-negative bacterial pneumonia (159/389, 41%) and atypical bacterial pneumonia (108/389, 28%) from January 1 to June 30, 2019. Nine machine-learning models were separately evaluated based on radiomics features extracted from CT images; three optimal submodels were constructed and integrated to form a multi-class classification model.ResultsWe selected five features to develop three radiomics submodels: a Gram-positive model, a Gram-negative model and an atypical model. The comprehensive radiomics model using support vector machine method yielded an average area under the curve (AUC) of 0.75 [95% confidence interval (CI), 0.65–0.83] and accuracy (ACC) of 0.58 [sensitivity (SEN), 0.57; specificity (SPE), 0.78] in the training set, and an average AUC of 0.73 (95% CI 0.61–0.79) and ACC of 0.54 (SEN, 0.52; SPE, 0.75) in the test set.ConclusionThis auxiliary EAT radiomics multi-class classification model was deserved to be researched in differential diagnosing bacterial pneumonias in children.

CT Radiomics Model for Discriminating the Risk Stratification of Gastrointestinal Stromal Tumors: A Multi-Class Classification and Multi-Center Study.

ObjectiveTo establish and verify a computed tomography (CT)-based multi-class prediction model for discriminating the risk stratification of gastrointestinal stromal tumors (GISTs).Materials and MethodsA total of 381 patients with GISTs were confirmed by surgery and pathology. Information on 213 patients were obtained from one hospital and used as training cohort, whereas the details of 168 patients were collected from two other hospitals and used as independent validation cohort. Regions of interest on CT images of arterial and venous phases were drawn, radiomics features were extracted, and dimensionality reduction processing was performed. Using a one-vs-rest method, a Random Forest-based GISTs risk three-class prediction model was established, and the receiver operating characteristic curve (ROC) was used to evaluate the performance of the multi-class classification model, and the generalization ability was verified using external data.ResultsThe training cohort included 96 very low-risk and low-risk, 60 intermediate-risk and 57 high-risk patients. External validation cohort included 82 very low-risk and low-risk, 48 intermediate-risk and 38 high-risk patients. The GISTs risk three-class radiomics model had a macro/micro average area under the curve (AUC) of 0.84 and an accuracy of 0.78 in the training cohort. It had a stable performance in the external validation cohort, with a macro/micro average AUC of 0.83 and an accuracy of 0.80.ConclusionCT radiomics can discriminate GISTs risk stratification. The performance of the three-class radiomics prediction model is good, and its generalization ability has also been verified in the external validation cohort, indicating its potential to assist stratified and accurate treatment of GISTs in the clinic.

Ensemble machine learning classification of daily living abilities among older people with HIV.

Backgroundclinically relevant methods to identify individuals at risk for impaired daily living abilities secondary to neurocognitive impairment (ADLs) remain elusive. This is especially true for complex clinical conditions such as HIV-Associated Neurocognitive Disorders (HAND). The aim of this study was to identify novel and modifiable factors that have potential to improve diagnostic accuracy of ADL risk, with the long-term goal of guiding future interventions to minimize ADL disruption.Methodsstudy participants included 79 people with HIV (PWH; mean age = 63; range = 55–80) enrolled in neuroHIV studies at University California San Francisco (UCSF) between 2016 and 2019. All participants were virally suppressed and exhibited objective evidence of neurocognitive impairment. ADL status was defined as either normative (n = 39) or at risk (n = 40) based on a task-based protocol. Gradient boosted multivariate regression (GBM) was employed to identify the combination of variables that differentiated ADL subgroup classification. Predictor variables included demographic factors, HIV disease severity indices, brain white matter integrity quantified using diffusion tensor imaging, cognitive test performance, and health co-morbidities. Model performance was examined using average Area Under the Curve (AUC) with repeated five-fold cross validation.Findingsthe univariate GBM yielded an average AUC of 83% using Wide Range Achievement test 4 (WRAT-4) reading score, self-reported thought confusion and difficulty reading, radial diffusivity (RD) in the left external capsule, fractional anisotropy (FA) in the left cingulate gyrus, and Stroop performance. The model allowing for two-way interactions modestly improved classification performance (AUC of 88%) and revealed synergies between race, reading ability, cognitive performance, and neuroimaging metrics in the genu and uncinate fasciculus. Conversion of Neuropsychological Assessment Battery Daily Living Module (NAB-DLM) performance from raw scores into T scores amplified differences between White and non-White study participants.Interpretationdemographic and sociocultural factors are critical determinants of ADL risk status among older PWH who meet diagnostic criteria for neurocognitive impairment. Task-based ADL assessment that relies heavily on reading proficiency may artificially inflate the frequency/severity of ADL impairment among diverse clinical populations. Culturally relevant measures of ADL status are needed for individuals with acquired neurocognitive disorders, including HAND.