Predict Performance Metrics Research Articles

Limits of a single surrogate model development methodology to represent housing stocks

Surrogate models for predicting housing stock thermal comfort, and operational cost, energy or emissions have increasingly gained popularity due to the potential to identify solutions to accelerate and optimise construction. In addition to the potential opportunity to progress towards climate goals and increase resource efficiencies by identifying areas of interest, they are much less time consuming than brute-force simulations of millions of scenarios. Although they are popular due to their promise, analysis of scalability limits for surrogate model development methodologies have been left out. Without this analysis, assurance that these may be used over differing housing stock scales cannot be provided. This work uses black-box model interpretation as well as common predictive performance metrics to assess the scalability of the development methodology presented.The novelty of this work stems from its comparison of the performance of a development methodology through training a learning algorithm with a progressively varied dataset to illustrate its scalability. This validates the use of a single surrogate model to represent numerous bottom-up archetypes used to represent a housing stock.

AI-Powered Post-Discharge Monitoring to Prevent Patients Readmissions and Reduce Workforce Burden

Abstract Background Patient readmission poses a significant burden on healthcare systems, straining resources, negatively impacting patient treatment, and it is often a Public Health problem. Traditional post-discharge follow-up methods, such as phone calls, often lack efficiency and personalization. Artificial intelligence (e.g., Machine Learning (ML)) offers a transformative approach that can potentially improve post-discharge care. Methods This scoping review aimed to map existing research on AI-powered solutions for post-discharge monitoring, focusing on their application in risk stratification and their potential for reducing hospital readmissions within a public health framework. Following PRISMA-ScR guidelines, a search was conducted across four databases using keywords related to AI, post-discharge, and risk stratification. Results Studies published between 2018 and January 2024 were included if they described the development and application of an ML model in a post-discharge setting for risk stratification. Sixteen studies met the inclusion criteria. Tree-based algorithms, particularly XGBoost, were the predominant ML approach, with promising performance metrics for readmission risk prediction. Conclusions However, limitations were identified, including restricted generalizability due to data source limitations and a lack of real-time implementation in many studies. This review highlights the potential of AI in post-discharge monitoring, particularly the use of ML for risk stratification, as a promising tool for public health contact tracing. While challenges remain regarding generalizability and real-world implementation, the findings suggest AI holds immense promise for improving post-discharge care, potentially reducing readmissions, and optimizing resource allocation. Future research with broader datasets and real-time applications can further solidify the role of AI in revolutionizing post-discharge care within public health systems. Key messages • This study shows the potential of AI and risk stratification in post-discharge monitoring, improving public health outcomes by reducing hospital readmissions and optimizing the healthcare workforce. • This review highlights the value of AI in public health, particularly for improving post-discharge care through effective prediction of high-risk readmissions.

PRO-SMOTEBoost: An adaptive SMOTEBoost probabilistic algorithm for rebalancing and improving imbalanced data classification

In the field of data mining and machine learning, dealing with imbalanced datasets is one of the most complex problems. The class imbalance issue significantly affects the classification of minority classes when using common classification algorithms. These algorithms often prioritize improving the performance of the majority class at the expense of the minority class, leading to misclassifying negative instances as positive ones. To address this problem, the Synthetic Minority Over-sampling Technique (SMOTE) has gained popularity to rebalance imbalanced data for classification. However, in this paper, we propose two algorithms to enhance the performance of imbalanced classification further. The first algorithm is PRO-SMOTE, an improvement over SMOTE. PRO-SMOTE relies on conditional probabilities to effectively rebalance imbalanced classes and improve the predictive performance metrics satisfactorily and reliably. By considering conditional probabilities, PRO-SMOTE can reduce the majority classes and optimally increase the minority class. Second, the PRO-SMOTEBoost algorithm, in turn, is based on the PRO-SMOTE to overcome classification anomalies and problems encountered by machine learning algorithms during classification, especially the weak ones. PRO-SMOTEBoost aims to maximize predictive precision to the greatest extent possible by combining the strengths of PRO-SMOTE with boosting techniques. Evaluating these algorithms using traditional machine learning algorithms such as Random Forests, C4.5, Naive Bayes, and Support Vector Machines has demonstrated excellent classification results. The performance metrics, encompassing F1-score, G-means, Precision, Accuracy, Recall, AUC-ROC, and Precision-Recall-curves, achieved by the proposed algorithm demonstrate a range that extends from over 90% to a flawless score of 100%. Compared to using these traditional algorithms individually, the utilization of PRO-SMOTEBoost has shown a significant improvement of 10% to 40% in performance metrics. Overall, the proposed algorithms, PRO-SMOTE and PRO-SMOTEBoost, offer effective solutions to address the challenges posed by imbalanced datasets. They provide improved predictive metrics and demonstrate their superiority when compared to traditional even modern machine learning algorithms.

493 Predicting pellet quality using multiple linear regression with Principal Component Analysis (PCA)

Abstract Pellet quality is a crucial key performance indicator (KPI) for commercial feed manufacturing, which influences both the efficiency of the feed mill and downstream performance of animals fed these diets. However, due to the complexity of feed manufacturing and the large number of factors involved in the manufacturing process, controlling pellet quality is an ongoing challenge for the feed industry. Previous studies have mainly explored the impact of a few factors on pellet quality under experimental settings, and empirical equations have been seldomly developed to reflect the relationship between the factors and pellet quality under the commercial feed mill settings. This study aimed to establish a relationship between pellet quality and factors collected under the settings of a commercial feed mill. The data were collected from Trouw Nutrition Canada’s feed mill located in St. Marys, Ontario (Plant 2), between December 15, 2021, and December 6, 2022. During this period, 2,691 observations were collected, with each observation representing an individual batch of pelleted feed. A total of 75 factors were recorded, including 4 factors associated with the general information of each batch, 10 manufacturing parameters, 41 feed ingredients, 8 factors regarding the nutrient composition of each diet, and 12 environmental factors. Pellet Durability Index (PDI), which was the response variable, was determined for each batch using the Holmen method. The data were randomly split into an 80% subset for training and a 20% subset for testing. The training subset was used to construct the model via a 5-fold cross-validation, while the testing subset was withheld as an independent dataset to evaluate the generalization performance of the model. The response variable (PDI) was transformed (tPDI) using the Box-Cox method to meet a normal distribution assumption. To avoid multicollinearity, Principal Component Analysis (PCA) was used to reduce the dimensionality of the numeric factors before building the multiple linear regression model. The model prediction performance was evaluated on both the training subset (using 5-fold cross-validation) and the testing subset, and the prediction performance metrics were consistent between the two subsets (Mean Absolute Error = 1.94 ± 0.102 vs. 2.02; Root Mean Square Prediction Error = 2.47 ± 0.111 vs. 2.58; Mean Square Prediction Error = 6.12 ± 0.538 vs. 6.68; Concordance Correlation Coefficient = 0.538 ± 0.0231 vs. 0.490; Pearson Correlation Coefficient = 0.606 ± 0.0247 vs. 0.553, respectively). Most feed ingredients and nutrient compositions showed either positive or negative loadings on Component 1 (17.87% of total variance), and outdoor/indoor environmental factors were positively loaded on Component 2 (14.21% of total variance). The model developed in this study could help commercial feed mills better understand how various factors impact pellet quality and optimize the manufacturing processes of pelleted feeds.

Ensemble of naive Bayes, decision tree, and random forest to predict air quality

<p>Air quality prediction is an important research issue because air quality can affect many areas of life. This study aims to predict air quality using the ensemble method and compare the results with the prediction results using a single method. The proposed ensemble method is built from three singlesupervised methods: naïve Bayes, decision trees, and random forests. The results show that the ensemble method performs better than the single methods. The ensemble method achieves the highest performance with scores of 99.89% accuracy, 79.6% precision, 79.81% recall, and 79.7% F1-score. The performance comparison between single and ensemble models is expected to provide information on the percentage increase in predictive model performance metrics from the single to ensemble methods.</p>

(Invited) Atomic-Scale Modeling of Charge-Transfer Kinetics at LixCoO2 Cathode-Electrolyte Interfaces

In the search for Li-ion batteries with improved performance, interface engineering is a comparably underexplored area with significant opportunities for innovation. Ion transport across interfaces is, for instance, a significant bottleneck for fast charging/discharging. At electrode interfaces, it is often essential to capture the coupled motion of ions and electrons in order to provide accurate computational input for interfacial design. In this work we employ the recently proposed theory of coupled ion-electron transfer kinetics[1,2] in conjugation with constrained and constant-potential density functional theory to examine Li+ transfer across battery electrolyte-cathode interfaces at the atomic scale. Here, we focus on the common Li x CoO2 intercalation-type cathode material and liquid-state organic electrolytes. We establish trends in transport kinetics for different states-of-charge (SOC, i.e., varied degree of lithiation, x), types of electrolytes, and counterions that closely reproduces experimentally observed trends. From our modeling, we can identify a few key descriptors to predict performance metrics in Li-ion transport of a given electrolyte-cathode combination. These include the Li-ion free energy of adsorption, the free energies of reorganization of the cathode material and electrolyte upon the addition/removal of electrons, as well as the electronic coupling between electron acceptor and donor states. As these quantities can be relatively straightforwardly estimated for any given electrode-electrolyte pair, we envisage broad applications of our modeling approach in the interfacial design for batteries and related fields.[1] Fraggedakis et al., Electrochimica Acta, 367, 137432 (2021)[2] Bazant, Faraday Discussion, 246, 60 (2023)

Kinetics prediction of normal knee and undergone total knee arthroplasty during squatting based on extreme gradient boosting

Following an arthroplasty, biomechanical evaluation of the knee joint is crucial because it plays a significant role in patient rehabilitation and physician follow-up. The sixty-two subjects, consisting of 31 with normal knees and 31 who had undergone total knee arthroplasty, evaluated the kinetics of the knee joints during squatting using inverse dynamics analysis. Artificial intelligence based on the XGBoost model was also used to predict kinetics data during group squatting. The result revealed that the maximum force, moment, and quadriceps force in the normal knee were higher than in the TKA group during squatting. The maximum resultant joint force magnitude in the normal and TKA groups revealed an average of 4.35 ± 1.00 and 3.08 ± 1.20 times body weight, respectively. Kinetics of the knee joint in a normal knee significantly differed (p < 0.05) from undergoing TKA, especially in the AP-direction force. Additionally, the resultant moment that included the quadriceps muscle force in the normal group was higher than in the TKA groups. The extreme gradient boosting algorithm (XGBoost) model has prediction performance metrics, which are considered to have high correlation performance in the case of kinetics.

Evaluating electrical power yield of photovoltaic solar cells with k-Nearest neighbors: A machine learning statistical analysis approach

The increasing demand for sustainable and renewable energy solutions reflects the critical importance of advancing photovoltaic (PV) technology and its operational efficiency. In response, this study introduces a novel application of the k-Nearest Neighbor (k-NN) algorithm to assess the reliability and applicability of solar panel simulation data which aimed to classify current states of partial shading, open, and short circuit conditions, alongside regression-based analysis for predicting specific operating parameters. In this research, a published dataset that involved various PV module configurations under different environmental conditions was tested and evaluated. The k-NN technique was applied to both classify the operational status and predict performance metrics of the modules. The diagnosis model demonstrated an accuracy of 99.2 % and an F1 score of 99.2 %, indicating a high degree of reliability in identifying the operational states of PV modules. Concurrently, the regression model exhibited a Root Mean Square Error (RMSE) of 0.036 and an R2 value of unity that showcased its effectiveness in predicting the operational parameters based on the simulation data. The concluded results are further enriching the effectiveness of simulation-based data generation to be endorsed and implemented before jumping into real experimental applications, in addition to highlighting the potential applicability of k-NN and machine learning for PV cells productivity statistical analysis.

Detecting defects that reduce breakdown voltage using machine learning and optical profilometry

Semiconductor wafer manufacturing relies on the precise control of various performance metrics to ensure the quality and reliability of integrated circuits. In particular, GaN has properties that are advantageous for high voltage and high frequency power devices; however, defects in the substrate growth and manufacturing are preventing vertical devices from performing optimally. This paper explores the application of machine learning techniques utilizing data obtained from optical profilometry as input variables to predict the probability of a wafer meeting performance metrics, specifically the breakdown voltage (Vbk). By incorporating machine learning techniques, it is possible to reliably predict performance metrics that cause devices to fail at low voltage. For diodes that fail at a higher (but still below theoretical) breakdown voltage, alternative inspection methods or a combination of several experimental techniques may be necessary.

Predicting mortality and recurrence in colorectal cancer: Comparative assessment of predictive models

IntroductionColorectal cancer (CRC), also known as colorectal cancer, is a significant disease marked by high fatality rates, ranking as the third leading cause of global mortality. The main objective of this study was to assess the accuracy of predictive models in predicting both mortality events and the probability of disease recurrence. MethodA retrospective analysis was conducted on a cohort of 284 individuals diagnosed with colorectal cancer between 2001 and 2017. Demographic and clinical data, including gender, disease stage, age at diagnosis, recurrence status, and treatment details, were meticulously recorded. We rigorously evaluated various predictive models, including Decision Trees, Random Forests, Random Survival Forests (RSF), Gradient Boosting, mboost, Deep Learning Neural Network (DLNN), and Cox regression. Performance metrics, such as sensitivity, positive predictive value (PPV), specificity, area under the receiver operating characteristic curve (ROC area), and overall accuracy, were calculated for each model to predict mortality and disease recurrence. The analysis was performed using R version 4.1.3 software and the Python programming language. ResultsFor mortality prediction, the mboost model demonstrated the highest sensitivity at 96.9% (95% CI: 0.83–0.99) and an ROC area of 0.88. It also exhibited high specificity at 80% (95% CI: 0.59–0.93), a positive predictive value of 86.1% (95% CI: 0.70–0.95), and an overall accuracy of 89% (95% CI: 0.78–0.96). Random Forests showed perfect sensitivity of 100% (95% CI: 0.85–1) but had low specificity at 0% (95% CI: 0–0.52) and poor overall accuracy (50%). On the other hand, DLNN had the lowest performance metrics for mortality prediction, with a sensitivity of 24% (95% CI: 0.222–0.268), specificity of 75% (95% CI: 0.73–0.77), and a lower positive predictive value of 42% (95% CI: 0.38–0.45). The Gradient Boosting model showed the best performance in predicting recurrence, achieving perfect sensitivity of 100% (95% CI: 0.87–1) and high specificity at 92.9% (95% CI: 0.76–0.99). It also had a high positive predictive value of 93.3% (95% CI: 0.77–0.99). Gradient Boosting, with an ROC area of 96.4%, and mboost, with an ROC area of 75%, demonstrated remarkable performance. DLNN had the lowest performance metrics for recurrence prediction, with sensitivity at 1.75% (95% CI: 0.01–0.02), specificity at 98% (95% CI: 0.97–0.98), and a lower positive predictive value at 52.6% (95% CI: 0.39–0.65). ConclusionIn summary, the mboost model demonstrated outstanding performance in predicting mortality, achieving exceptional results across various evaluation metrics. Random Forests exhibited perfect sensitivity but showed poor specificity and overall accuracy. The DLNN model displayed the lowest performance metrics for mortality prediction. In terms of recurrence prediction, the Gradient Boosting model outperformed other models with perfect sensitivity, high specificity, and positive predictive value. The DLNN model had the lowest performance metrics for recurrence prediction. Overall, the results emphasize the effectiveness of the mboost and Gradient Boosting models in predicting mortality and recurrence in colorectal cancer patients.

Integration of risk factor polygenic risk score with disease polygenic risk score for disease prediction

Polygenic risk score (PRS) is useful for capturing an individual’s genetic susceptibility. However, previous studies have not fully exploited the potential of the risk factor PRS (RFPRS) for disease prediction. We explored the potential of integrating disease-related RFPRSs with disease PRS to enhance disease prediction performance. We constructed 112 RFPRSs and analyzed the association of RFPRSs with diseases to identify disease-related RFPRSs in 700 diseases, using the UK Biobank dataset. We uncovered 6157 statistically significant associations between 247 diseases and 109 RFPRSs. We estimated the disease PRSs of 70 diseases that exhibited statistically significant heritability, to generate RFDiseasemetaPRS—a combined PRS integrating RFPRSs and disease PRS—and compare the prediction performance metrics between RFDiseasemetaPRS and disease PRS. RFDiseasemetaPRS showed better performance for Nagelkerke’s pseudo-R2, odds ratio (OR) per 1 SD, net reclassification improvement (NRI) values and difference of R2 considered by variance of R2 in 31 out of 70 diseases. Additionally, we assessed risk classification between two models by examining OR between the top 10% and remaining 90% individuals for the 31 diseases; RFDiseasemetaPRS exhibited better R2, NRI and OR than disease PRS. These findings highlight the importance of utilizing RFDiseasemetaPRS, which can provide personalized healthcare and tailored prevention strategies.

Expert-augmented machine learning to accelerate the discovery of copolymers for anion exchange membrane

We constructed an automated machine learning pipeline to screen the broad chemical space of candidate copolymers for anion exchange membranes (AEMs). The pipeline utilizes a genetic algorithm integrated twelve machine learning algorithms, to screen copolymers that are potentially capable to improve AEM performance through the evolution of polymer hydrophobic and hydrophilic backbones and cation groups. An exhaustive data mining accumulated 749 experimentally reported AEMs, then the pipeline automatically generated over 172 million hypothetical copolymers, and 2519 potential candidates were screened out according to their predicted performance metrics including the OH− conductivity, the conductivity – dimensional stability trade-off coefficient, and the LUMO of cations in Pareto frontier set. Expert experience was adopted in the construction of robust regression and classification predictive models, and the importance of physical, chemical and topological features were illustrated from Shapley additive explanations (SHAP) analysis. We have released a standalone software that integrates these predictive models and the screened copolymer candidates with predicted AEM performance metrics at https://github.com/polySML/polySML-AEM. This study can facilitate the development of advanced AEMs through rational design of copolymers.

Modeling influence of weather variables on energy consumption in an agricultural research institute in Ibadan, Nigeria

&lt;abstract&gt; &lt;p&gt;Climate change is having a significant impact on weather variables like temperature, humidity, precipitation, solar radiation, daylight duration, wind speed, etc. These weather variables are key indicators that affect electricity demand and consumption. Hence, understanding the significance of weather elements on energy needs and consumption is important to be able to adapt, strategize, and predict the effect of the changing climate on the required energy of an organization. This study aims to investigate the relationship between changing weather elements and electricity consumption, employing Multivariate Linear Regression (MLR), Support Vector Regressions (SVR), and Artificial Neural Network (ANN) models to predict the effect of weather changes on energy consumption. The following approaches were engaged for this study: Creating a catalog of weather elements and parameters of energy need or its consumption; analyzing and correlating electrical power consumption to weather factors; and developing prediction models—MLR, SVR, and ANN to predict the significance of the change in the variables of weather on the electrical energy consumption. Among the weather variables considered, temperature emerged as the most influential factor affecting electricity consumption, displaying the highest correlation. The monthly total pattern for electricity use for the case study area followed a similar pattern as the mean apparent temperature. Of the three models (MLR, SVR, and ANN) developed in this study, the ANN model yielded the best predictive performance, with Mean Square Error (MSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE) of 2.733%, 1.292%, and 4.66%, respectively. Notably, the ANN model outperformed the other models (MLR and SVR) by more than 20% across the predictive performance metrics employed.&lt;/p&gt; &lt;/abstract&gt;

Humidification potential optimization of various membranes for proton exchange membrane fuel cell: Experiments and deep learning assisted metaheuristics

This study addresses the durability and performance challenges in Proton Exchange Membrane Fuel Cell (PEMFC) technology, primarily focusing on optimal humidity management. Acknowledging that, beyond technical aspects, the broader commercialization of PEMFCs is critically influenced by factors such as the cost and availability of hydrogen, this research aims to provide a comprehensive solution to enhance PEMFC applicability. Utilizing Nafion (NR-212), reverse osmosis (RO), and pervaporation (PV) membranes, the study optimizes five key performance metrics: pressure drop (∆P), dew point approach temperature (DPAT), water recovery ratio (WRR), Water Flux (J), and coefficient of performance (COP). These optimizations are conducted considering variables like temperature, humidity, flowrate, and membrane material. A deep neural network (DNN) model, incorporating Bayesian surrogacy with Gaussian process, gradient boost regression trees, and random forest, is developed using experimental data. With a correlation coefficient of 0.986, the model demonstrates high accuracy in predicting performance metrics, subsequently guiding genetic algorithms for effective PEMFC humidity control. The results show significant improvements in all metrics, with optimal values achieved for NR-212, RO, and PV membranes. This study thus presents a novel, practical deep learning approach, considering both technological advancements and external economic factors, for enhancing PEMFC operations.

Reliable AI models can reveal key processes of heat recovery steam generator operation in air pollutant emission

Artificial intelligence (AI) algorithms have been applied to air pollution prediction modeling, but assessment of their trustworthiness is lacking. This study aims to examine whether AI-based models can identify key operational processes of a Heat Recovery Steam Generator (HRSG) facility and predict the emission of target air pollutants, including nitrogen oxides (NOx), sulfur oxides (SOx), and total suspended particles (TSP). First, key operational variables are selected based on the feature importance of the Random Forest models (that is, data-driven input selection). Secondly, Bidirectional Long Short-Term Memory-based AutoEncoder (BiLSTM-AE) and Random Forest (RF) models with the data-driven input selection are trained and evaluated using multiple predictive performance metrics. Then, the BiLSTM-AE and RF models are trained and evaluated using the operational variables selected by experts. Results demonstrate that the data-driven and expertise-based methods select five common operational variables that are key factors to predict variations of target output variables. The results from the multi-metric evaluations show a skillful prediction of BiLSTM-AE and RF models for NOx and TSP, but not SOx. For TSP prediction, RF is more sensitive to the input selection method than BiLSTM-AE. This study underscores the dependency of the prediction skill of AI algorithm-based models on the target air pollutant and input selection method, and suggests the trustworthiness of selected key operational variables by RF's feature importance.

A study of machine learning to predict NRDS severity based on lung ultrasound score and clinical indicators.

To develop predictive models for neonatal respiratory distress syndrome (NRDS) using machine learning algorithms to improve the accuracy of severity predictions. This double-blind cohort study included 230 neonates admitted to the neonatal intensive care unit (NICU) of Yantaishan Hospital between December 2020 and June 2023. Of these, 119 neonates were diagnosed with NRDS and placed in the NRDS group, while 111 neonates with other conditions formed the non-NRDS (N-NRDS) group. All neonates underwent lung ultrasound and various clinical assessments, with data collected on the oxygenation index (OI), sequential organ failure assessment (SOFA), respiratory index (RI), and lung ultrasound score (LUS). An independent sample test was used to compare the groups' LUS, OI, RI, SOFA scores, and clinical data. Use Least Absolute Shrinkage and Selection Operator (LASSO) regression to identify predictor variables, and construct a model for predicting NRDS severity using logistic regression (LR), random forest (RF), artificial neural network (NN), and support vector machine (SVM) algorithms. The importance of predictive variables and performance metrics was evaluated for each model. The NRDS group showed significantly higher LUS, SOFA, and RI scores and lower OI values than the N-NRDS group (p < 0.01). LUS, SOFA, and RI scores were significantly higher in the severe NRDS group compared to the mild and moderate groups, while OI was markedly lower (p < 0.01). LUS, OI, RI, and SOFA scores were the most impactful variables for the predictive efficacy of the models. The RF model performed best of the four models, with an AUC of 0.894, accuracy of 0.808, and sensitivity of 0.706. In contrast, the LR, NN, and SVM models have lower AUC values than the RF model with 0.841, 0.828, and 0.726, respectively. Four predictive models based on machine learning can accurately assess the severity of NRDS. Among them, the RF model exhibits the best predictive performance, offering more effective support for the treatment and care of neonates.

Computer vision model for the detection of canine pododermatitis and neoplasia of the paw.

Artificial intelligence (AI) has been used successfully in human dermatology. AI utilises convolutional neural networks (CNN) to accomplish tasks such as image classification, object detection and segmentation, facilitating early diagnosis. Computer vision (CV), a field of AI, has shown great results in detecting signs of human skin diseases. Canine paw skin diseases are a common problem in general veterinary practice, and computer vision tools could facilitate the detection and monitoring of disease processes. Currently, no such tool is available in veterinary dermatology. Digital images of paws from healthy dogs and paws with pododermatitis or neoplasia were used. We tested the novel object detection model Pawgnosis, a Tiny YOLOv4 image analysis model deployed on a microcomputer with a camera for the rapid detection of canine pododermatitis and neoplasia. The prediction performance metrics used to evaluate the models included mean average precision (mAP), precision, recall, average precision (AP) for accuracy and frames per second (FPS) for speed. A large dataset labelled by a single individual (Dataset A) used to train a Tiny YOLOv4 model provided the best results with a mean mAP of 0.95, precision of 0.86, recall of 0.93 and 20 FPS. This novel object detection model has the potential for application in the field of veterinary dermatology.

Development and validation of age-specific risk prediction models for primary ovarian insufficiency in long-term survivors of childhood cancer: a report from the Childhood Cancer Survivor Study and St Jude Lifetime Cohort

Female survivors of childhood cancer are at risk for primary ovarian insufficiency (POI), defined as the cessation of gonadal function before the age of 40 years. We aimed to develop and validate models to predict age-specific POI risk among long-term survivors of childhood cancer. To develop models to predict age-specific POI risk for the ages of 21-40 years, we used data from the Childhood Cancer Survivor Study (CCSS). Female survivors aged 18 years or older at their latest follow-up, with self-reported menstrual history information and free of subsequent malignant neoplasms within 5 years of diagnosis, were included. We evaluated models that used algorithms based on statistical or machine learning to consider all predictors, including cancer treatments. Cross-validated prediction performance metrics (eg, area under the receiver operating characteristic curve [AUROC]) were compared to select the best-performing models. For external validation of the models, we used data from 5-year survivors in the St Jude Lifetime Cohort (SJLIFE) with ovarian status clinically ascertained using hormone measurements (menopause defined by follicle stimulating hormone >30 mIU/mL and oestradiol <17 pg/mL) and medical chart or questionnaire review. We also evaluated an SJLIFE-based polygenic risk score for POI among 1985 CCSS survivors with genotype data available. 7891 female CCSS survivors (922 with POI) were included in the development of the POI risk prediction model, and 1349 female SJLIFE survivors (101 with POI) were included in the validation study. Median follow-up from cancer diagnosis was 23·7 years (IQR 18·3-30·0) in CCSS and 15·1 years (10·4-22·9) in SJLIFE. Between the ages of 21 and 40 years, POI prevalence increased from 7·9% (95% CI 7·3-8·5) to 18·6% (17·3-20·0) in CCSS and 7·3% (5·8-8·9) to 14·9% (11·6-19·1) in SJLIFE. Age-specific logistic regression models considering ovarian radiation dosimetry or prescribed pelvic and abdominal radiation dose, along with individual chemotherapy predictors, performed well in CCSS. In the SJLIFE validation, the prescribed radiation dose model performed well (AUROC 0·88-0·95), as did a simpler model that considered any exposures to pelvic or abdominal radiotherapy or alkylators (0·82-0·90). Addition of the polygenic risk predictor significantly improved the average positive predictive value (from 0·76 [95% CI 0·63-0·89] to 0·87 [0·80-0·94]; p=0·029) among CCSS survivors treated with ovarian radiation and chemotherapy. POI risk prediction models using treatment information showed robust prediction performance in adult survivors of childhood cancer. Canadian Institutes of Health Research, US National Cancer Institute.

Development and Validation of a Model to Quantify Injury Severity in Real Time

Quantifying injury severity is integral to trauma care benchmarking, decision-making, and research, yet the most prevalent metric to quantify injury severity-Injury Severity Score (ISS)- is impractical to use in real time. To develop and validate a practical model that uses a limited number of injury patterns to quantify injury severity in real time through 3 intuitive outcomes. In this cohort study for prediction model development and validation, training, development, and internal validation cohorts comprised 223 545, 74 514, and 74 514 admission encounters, respectively, of adults (age ≥18 years) with a primary diagnosis of traumatic injury hospitalized more than 2 days (2017-2018 National Inpatient Sample). The external validation cohort comprised 3855 adults admitted to a level I trauma center who met criteria for the 2 highest of the institution's 3 trauma activation levels. Three outcomes were hospital length of stay, probability of discharge disposition to a facility, and probability of inpatient mortality. The prediction performance metric for length of stay was mean absolute error. Prediction performance metrics for discharge disposition and inpatient mortality were average precision, precision, recall, specificity, F1 score, and area under the receiver operating characteristic curve (AUROC). Calibration was evaluated using calibration plots. Shapley addictive explanations analysis and bee swarm plots facilitated model explainability analysis. The Length of Stay, Disposition, Mortality (LDM) Injury Index (the model) comprised a multitask deep learning model trained, developed, and internally validated on a data set of 372 573 traumatic injury encounters (mean [SD] age = 68.7 [19.3] years, 56.6% female). The model used 176 potential injuries to output 3 interpretable outcomes: the predicted hospital length of stay, probability of discharge to a facility, and probability of inpatient mortality. For the external validation set, the ISS predicted length of stay with mean absolute error was 4.16 (95% CI, 4.13-4.20) days. Compared with the ISS, the model had comparable external validation set discrimination performance (facility discharge AUROC: 0.67 [95% CI, 0.67-0.68] vs 0.65 [95% CI, 0.65-0.66]; recall: 0.59 [95% CI, 0.58-0.61] vs 0.59 [95% CI, 0.58-0.60]; specificity: 0.66 [95% CI, 0.66-0.66] vs 0.62 [95%CI, 0.60-0.63]; mortality AUROC: 0.83 [95% CI, 0.81-0.84] vs 0.82 [95% CI, 0.82-0.82]; recall: 0.74 [95% CI, 0.72-0.77] vs 0.75 [95% CI, 0.75-0.76]; specificity: 0.81 [95% CI, 0.81-0.81] vs 0.76 [95% CI, 0.75-0.77]). The model had excellent calibration for predicting facility discharge disposition, but overestimated inpatient mortality. Explainability analysis found the inputs influencing model predictions matched intuition. In this cohort study using a limited number of injury patterns, the model quantified injury severity using 3 intuitive outcomes. Further study is required to evaluate the model at scale.

Comparison of known spawner abundance from fence counts to visual counts for simplified spawner estimation methods

Many salmon species are monitored by visual counts of spawners in streams; however, there are few data sets where abundance is known and compared to estimates derived from visual counts. We used spawner fences to obtain known kokanee ( Oncorhynchus nerka) spawner abundance (14 stream-years) on streams that are monitored with annual visual surveys (7 to 9 counts per year) and incorporated similar published data from pink salmon ( Oncorhynchus gorbuscha) (11 stream-years). We investigated the performance of several simplified expansion factor estimation methods with survey life and observer efficiency as unknown nuisance parameters. All visual indices of kokanee and pink salmon spawners from live ground counts were highly correlated to abundance from fence counts ( r2 ≥ 0.96 and 0.89, respectively). Application of cross-validation on out-of-sample data for both species showed that mean% error could range from 13% to 53% on a previously unsampled stream depending on the species, counting method, and visual index used. Predictive performance metrics were less sensitive to counting frequency than observer efficiency and associated variability, which was influenced by the counting method (aerial versus ground surveys).