Effectiveness, Explainability and Reliability of Machine Meta-Learning Methods for Predicting Mortality in Patients with COVID-19: Results of the Brazilian COVID-19 Registry

The majority prognostic scores proposed for early assessment of coronavirus disease 19 (COVID-19) patients are bounded by methodological flaws. Our group recently developed a new risk score - ABC2SPH - using traditional statistical methods (least absolute shrinkage and selection operator logistic regression - LASSO). In this article, we provide a thorough comparative study between modern machine learning (ML) methods and state-of-the-art statistical methods, represented by ABC2SPH, in the task of predicting in-hospital mortality in COVID-19 patients using data upon hospital admission. We overcome methodological and technological issues found in previous similar studies, while exploring a large sample (5,032 patients). Additionally, we take advantage of a large and diverse set of methods and investigate the effectiveness of applying meta-learning, more specifically Stacking, in order to combine the methods' strengths and overcome their limitations. In our experiments, our Stacking solutions improved over previous state-of-the-art by more than 26% in predicting death, achieving 87.1% of AUROC and MacroF1 of 73.9%. We also investigated issues related to the interpretability and reliability of the predictions produced by the most effective ML methods. Finally, we discuss the adequacy of AUROC as an evaluation metric for highly imbalanced and skewed datasets commonly found in health-related problems.

Differentiating nontuberculous mycobacterium pulmonary disease from pulmonary tuberculosis through the analysis of the cavity features in CT images using radiomics

Objective To differentiate nontuberculous mycobacteria (NTM) pulmonary diseases from pulmonary tuberculosis (PTB) by analyzing the CT radiomics features of their cavity. Methods 73 patients of NTM pulmonary diseases and 69 patients of PTB with the cavity in Shandong Province Chest Hospital and Qilu Hospital of Shandong University were retrospectively analyzed. 20 patients of NTM pulmonary diseases and 20 patients of PTB with the cavity in Jinan Infectious Disease Hospitall were collected for external validation of the model. 379 cavities as the region of interesting (ROI) from chest CT images were performed by 2 experienced radiologists. 80% of cavities were allocated to the training set and 20% to the validation set using a random number generated by a computer. 1409 radiomics features extracted from the Huiying Radcloud platform were used to analyze the two kinds of diseases' CT cavity characteristics. Feature selection was performed using analysis of variance (ANOVA) and least absolute shrinkage and selection operator (LASSO) methods, and six supervised learning classifiers (KNN, SVM, XGBoost, RF, LR, and DT models) were used to analyze the features. Results 29 optimal features were selected by the variance threshold method, K best method, and Lasso algorithm.and the ROC curve values are obtained. In the training set, the AUC values of the six models were all greater than 0.97, 95% CI were 0.95–1.00, the sensitivity was greater than 0.92, and the specificity was greater than 0.92. In the validation set, the AUC values of the six models were all greater than 0.84, 95% CI were 0.76–1.00, the sensitivity was greater than 0.79, and the specificity was greater than 0.79. In the external validation set, The AUC values of the six models were all greater than 0.84, LR classifier has the highest precision, recall and F1-score, which were 0.92, 0.94, 0.93. Conclusion The radiomics features extracted from cavity on CT images can provide effective proof in distinguishing the NTM pulmonary disease from PTB, and the radiomics analysis shows a more accurate diagnosis than the radiologists. Among the six classifiers, LR classifier has the best performance in identifying two diseases.

A Nomogram for Predicting Intraoperative Hemodynamic Instability in Patients With Pheochromocytoma

PurposeSurgical removal of pheochromocytoma (PCC), including open, laparoscopic, and robot-assisted adrenalectomy, is the cornerstone of therapy, which is associated with high risk of intraoperative and postoperative life-threatening complications due to intraoperative hemodynamic instability (IHD). This study aims to develop and validate a nomogram based on clinical characteristics as well as computed tomography (CT) features for the prediction of IHD in pheochromocytoma surgery.MethodsThe data from 112 patients with pheochromocytoma were collected at a single center between January 1, 2010, and December 31, 2019. Clinical and radiological features were selected with the least absolute shrinkage and selection operator regression analysis to predict IHD then constitute a nomogram. The performance of the nomogram was assessed in terms of discrimination, calibration, and clinical utility.ResultsAge, tumor shape, Mayo Adhesive Probability score, laterality, necrosis, body mass index, and surgical technique were identified as risk predictors of the presence of IHD. The nomogram was then developed using these seven variables. The model showed good discrimination with a C-index of 0.773 (95% CI, 0.683–0.862) and an area under the receiver operating characteristic curve (AUC) of 0.739 (95% CI, 0.642–0.837). The calibration plot suggested good agreement between predicted and actual probabilities. Besides, calibration was tested with the Hosmer–Lemeshow test (P = 0.961). The decision curve showed the clinical effectiveness of the nomogram.ConclusionsOur nomogram based on clinical and CT parameters could facilitate the treatment strategy according to assessment of the risk of IHD in patients with pheochromocytoma.

A Preoperative MRI-Based Radiomics-Clinicopathological Classifier to Predict the Recurrence of Pituitary Macroadenoma Within 5 Years

Objective: To investigate the ability of a MRI-based radiomics-clinicopathological model to predict pituitary macroadenoma (PMA) recurrence within 5 years.Materials and Methods: We recruited 74 recurrent and 94 non-recurrent subjects, following first surgery with 5-year follow-up data. Univariate and multivariate analyses were conducted to identify independent clinicopathological risk factors. Two independent and blinded neuroradiologists used 3D-Slicer software to manually delineate whole tumors using preoperative axial contrast-enhanced T1WI (CE-T1WI) images. 3D-Slicer was then used to extract radiomics features from segmented tumors. Dimensionality reduction was carried out by the least absolute shrinkage and selection operator (LASSO). Two multilayer perceptron (MLP) models were established, including independent clinicopathological risk factors (Model 1) and a combination of screened radiomics features and independent clinicopathological markers (Model 2). The predictive performance of these models was evaluated by receiver operator characteristic (ROC) curve analysis.Results: In total, 1,130 features were identified, and 4 of these were selected by LASSO. In the test set, the area under the curve (AUC) of Model 2 was superior to Model 1 {0.783, [95% confidence interval (CI): 0.718—.860] vs. 0.739, (95% CI: 0.665–0.818)}. Model 2 also yielded the higher accuracy (0.808 vs. 0.692), sensitivity (0.826 vs. 0.652), and specificity (0.793 vs. 0.724) than Model 1.Conclusions: The integrated classifier was superior to a clinical classifier and may facilitate the prediction of individualized prognosis and therapy.

A predictive model for the aggregation of polycyclic aromatic compounds

The physical aggregation of polycyclic aromatic compounds (PACs) is a key step in soot inception. In this work, we set out to elucidate which molecular properties influence the physical growth process and use machine learning to quantitatively relate these features to the propensity of these molecules to physically dimerize with other PACs. To this end, we first develop a dataset of PAC monomers along with their calculated free energies of dimerization emphasizing a set of PACs with a diverse range of properties. First, we augment existing calculations of dimerization energies with our own molecular dynamics simulations enhanced by well-tempered Metadyanmics. We then demonstrate that a machine learning model based on the least absolute shrinkage and selection operator (Lasso) is able to quantitatively learn how molecular features contribute to physical aggregation and predict the free energy of dimerization for new pairs of molecules. The model is able to accurately determine the stability for both homodimerization and heterodimerization cases. Our approach also provides a data driven method to determine the molecular features most important to predicting the dimer stability. From this, we determine that the PAC properties most influential to physical dimerization are size, shape, oxygenation, and presence of rotatable bonds. This work highlights the molecular complexity of the PAC monomers that must be accounted for in order to accurately represent physical aggregation. We anticipate that this approach will allow for more effective modeling of the PAC dimerization process as it facilitates the efficient prediction of dimerization propensity from easily calculable molecular features.

A deep learning approach based on stochastic gradient descent and least absolute shrinkage and selection operator for identifying diabetic retinopathy

More than eighty-five to ninety percentage of the diabetic patients are affected with diabetic retinopathy (DR) which is an eye disorder that leads to blindness. The computational techniques can support to detect the DR by using the retinal images. However, it is hard to measure the DR with the raw retinal image. This paper proposes an effective method for identification of DR from the retinal images. In this research work, initially the Weiner filter is used for preprocessing the raw retinal image. Then the preprocessed image is segmented using fuzzy c-mean technique. Then from the segmented image, the features are extracted using grey level co-occurrence matrix (GLCM). After extracting the fundus image, the feature selection is performed stochastic gradient descent, and least absolute shrinkage and selection operator (LASSO) for accurate identification during the classification process. Then the inception v3-convolutional neural network (IV3-CNN) model is used in the classification process to classify the image as DR image or non-DR image. By applying the proposed method, the classification performance of IV3-CNN model in identifying DR is studied. Using the proposed method, the DR is identified with the accuracy of about 95%, and the processed retinal image is identified as mild DR.

Machine Learning-Based Prediction of Myocardial Recovery in Patients With Left Ventricular Assist Device Support

Background: Prospective studies demonstrate that aggressive pharmacological therapy combined with pump speed optimization may result in myocardial recovery in larger numbers of patients supported with left ventricular assist device (LVAD). This study sought to determine whether the use of machine learning (ML) based models predict LVAD patients with myocardial recovery resulting in pump explant. Methods: A total of 20 270 adult patients with a durable continuous-flow LVAD in the INTERMACS registry (Interagency Registry for Mechanically Assisted Circulatory Support) were included in the study. Ninety-eight raw clinical variables were screened using the least absolute shrinkage and selection operator for selection of features associated with LVAD-induced myocardial recovery. ML models were developed in the training data set (70%) and were assessed in the validation data set (30%) by receiver operating curve and Kaplan-Meier analysis. Results: Least absolute shrinkage and selection operator identified 28 unique clinical features associated with LVAD-induced myocardial recovery, including age, cause of heart failure, psychosocial risk factors, laboratory values, cardiac rate and rhythm, and echocardiographic indices. ML models achieved area under the receiver operating curve of 0.813 to 0.824 in the validation data set outperforming logistic regression-based new INTERMACS recovery risk score (area under the receiver operating curve of 0.796) and previously established LVAD recovery risk scores (INTERMACS Cardiac Recovery Score and INTERMACS Recovery Score by Topkara et al.) with area under the receiver operating curve of 0.744 and 0.748 ( P <0.05). Patients who were predicted to recover by ML models demonstrated a significantly higher incidence of myocardial recovery resulting in LVAD explant in the validation cohort compared with those who were not predicted to recover (18.8% versus 2.6% at 4 years of pump support). Conclusions: ML can be a valuable tool to identify subsets of LVAD patients who may be more likely to respond to myocardial recovery protocols.

The Role of Genetic Algorithm Selection Operators in Extending WSN Stability Period: A Comparative Study

A genetic algorithm (GA) contains a number of genetic operators that can be tweaked to improve the performance of specific implementations. Parent selection, crossover, and mutation are examples of these operators. One of the most important operations in GA is selection. The performance of GA in addressing the single-objective wireless sensor network stability period extension problem using various parent selection methods is evaluated and compared. In this paper, six GA selection operators are used: roulette wheel, linear rank, exponential rank, stochastic universal sampling, tournament, and truncation. According to the simulation results, the truncation selection operator is the most efficient operator in terms of extending the network stability period and improving reliability. The truncation operator outperforms other selection operators, most notably the well-known roulette wheel operator, by increasing the stability period by 25.8% and data throughput by 26.86%. Furthermore, the truncation selection operator outperforms other selection operators in terms of the network residual energy after each protocol round.

MRI-based radiomics analysis to evaluate the clinicopathological characteristics of cervical carcinoma: a multicenter study

Background Preoperative prediction of clinical pathological indicators of cervical cancer (CC) is of great significance to the formulation of personalized treatment plans for CC. Purpose To investigate magnetic resonance imaging (MRI) radiomics analysis for the evaluation of pathological types, tumor grade, FIGO stage, and lymph node metastasis (LNM) of CC. Material and Methods A total of 235 patients with CC from three institutes were enrolled in the study. All patients underwent T2/SPAIR and contrast-enhanced T1-weighted (CE-T1WI) imaging scans before radical hysterectomy by pelvic lymph node dissection surgery. Radiomics features extracted from T2/SPAIR and CE-T1WI imaging were selected by the least absolute shrinkage and selection operator (LASSO) methods for further radiomics signature calculation. These radiomic features were used to construct regression and decision tree models to evaluate the performance of radiomic features in distinguishing clinicopathological indicators. Results The area under the curve (AUC) of T2/SPAIR and CE-T1WI imaging were 0.777 and 0.750, respectively, for differentiating between adenocarcinoma and squamous cell carcinoma. From the two sequences, the AUC of the verification group that distinguished low FIGO stage from high FIGO stage was 0.716 and 0.676, respectively. The AUC for moderately well and poorly differentiated tumors were 0.729 on T2/SPAIR and 0.749 on CE-T1WI imaging. The AUC of the verification groups for LNM was 0.730 and 0.618 on T2/SPAIR and CE-T1WI imaging, respectively. Conclusion MRI radiomics features can be used as a non-invasive method to evaluate the clinicopathological indexes of CC and provide an important auxiliary examination method for patients to determine individualized treatment plans before operation.

MRI-Based Radiomics for Differentiating Orbital Cavernous Hemangioma and Orbital Schwannoma

Aim: The purpose of this work was to develop and evaluate magnetic resonance imaging (MRI)-based radiomics for differentiation of orbital cavernous hemangioma (OCH) and orbital schwannoma (OSC).Methods: Fifty-eight patients (40 OCH and 18 OSC, confirmed pathohistologically) screened out from 216 consecutive patients who presented between 2015 and 2020 were divided into a training group (28 OCH and 12 OSC) and a validation group (12 OCH and 6 OSC). Radiomics features were extracted from T1-weighted imaging (T1WI) and T2-weighted imaging (T2WI). T-tests, the least absolute shrinkage and selection operator (LASSO), and principal components analysis (PCA) were used to select features for use in the classification models. A logistic regression (LR) model, support vector machine (SVM) model, decision tree (DT) model, and random forest (RF) model were constructed to differentiate OCH from OSC. The models were evaluated according to their accuracy and the area under the receiver operator characteristic (ROC) curve (AUC).Results: Six features from T1WI, five features from T2WI, and eight features from combined T1WI and T2WI were finally selected for building the classification models. The models using T2WI features showed superior performance on the validation data than those using T1WI features, especially the LR model and SVM model, which showed accuracy of 93% (85–100%) and 92%, respectively, The SVM model showed high accuracy of 93% (91–96%) on the combined feature group with an AUC of 98% (97–99%). The DT and RF models did not perform as well as the SVM model.Conclusion: Radiomics analysis using an SVM model achieved an accuracy of 93% for distinguishing OCH and OSC, which may be helpful for clinical diagnosis.