Cross Validation Research Articles

Leveraging Deep Learning and Generative AI for Predicting Rheological Properties and Material Compositions of 3D Printed Polyacrylamide Hydrogels.

Artificial intelligence (AI) has the ability to predict rheological properties and constituent composition of 3D-printed materials with appropriately trained models. However, these models are not currently available for use. In this work, we trained deep learning (DL) models to (1) predict the rheological properties, such as the storage (G') and loss (G") moduli, of 3D-printed polyacrylamide (PAA) substrates, and (2) predict the composition of materials and associated 3D printing parameters for a desired pair of G' and G". We employed a multilayer perceptron (MLP) and successfully predicted G' and G" from seven gel constituent parameters in a multivariate regression process. We used a grid-search algorithm along with 10-fold cross validation to tune the hyperparameters of the MLP, and found the R2 value to be 0.89. Next, we adopted two generative DL models named variational autoencoder (VAE) and conditional variational autoencoder (CVAE) to learn data patterns and generate constituent compositions. With these generative models, we produced synthetic data with the same statistical distribution as the real data of actual hydrogel fabrication, which was then validated using Student's t-test and an autoencoder (AE) anomaly detector. We found that none of the seven generated gel constituents were significantly different from the real data. Our trained DL models were successful in mapping the input-output relationship for the 3D-printed hydrogel substrates, which can predict multiple variables from a handful of input variables and vice versa.

Concrete carbonation depth prediction model based on a gradient-boosting decision tree and different metaheuristic algorithms

Carbonation is a significant factor contributing to the instability of concrete structures. This study proposes two prediction models based on the gradient boosting decision tree (GBDT) to address the challenge of predicting concrete carbonation depth. This study integrates GBDT with two metaheuristic algorithms, particle swarm optimization algorithm (PSO) and sparrow search optimization algorithm (SSA), forming two hybrid models, PSO-GBDT and SSA-GBDT, aimed at enhancing prediction accuracy. Six influencing parameters (FA, t, w/b, B, RH, and CO2) were selected as input features to train and evaluate the hybrid models, with the concrete carbonation depth was used as the output. A database containing 883 groups of cases was established. Finally, three classic models, GBDT, ANN and SVR, were compared against the two hybrid models. To enhance model generalization, all models were subjected to five-fold cross validation. Four evaluation indicators (RMSE, R2, MAE, and VAF) and Taylor diagrams were employed to comprehensively assess these models. The results indicated that the two hybrid models exhibited superior prediction performance in the dataset (training set: R2, VAF>0.98, testing set: R2, VAF>0.96). The SSA-GBDT model achieved the highest prediction performance (RMSE= 2.7008, R2= 0.9639, MAE= 1.7691, VAF= 0.9627). Additionally, the SSA-GBDT model identified exposure time (t) as the feature with the greatest impact. This study demonstrates the applicability of the SSA-GBDT models in predicting concrete carbonation depth, offering a novel approach for improving accuracy in such predictions.

Comprehensive Approach to TB Diagnosis Powered by Machine Learning in Endemic Countries

Abstract Introduction/Objective Tuberculosis (TB), a leading cause of mortality worldwide, is primarily diagnosed through methods that face numerous challenges, including lengthy result times and diagnostic inaccuracy, particularly in the context of drug-resistant strains and latent infections. This study proposes a novel approach that integrates multiple diagnostic modalities with machine learning to enhance TB diagnosis accuracy. Methods/Case Report The Machine learning model was tailored for the classification of radiological images, specifically CT scans, using original DICOM images to preserve valuable voxel information. The study involved a curated dataset of CT scans from Pakistan, TB-positive (smear+/Culture+ n=72; smear+/culture- n= 3; smear-/Culture+ n=16; smear-/Culture- n=14; clinical and radiological diagnosis only n= 12) and healthy (n=103) scans from COVID-CTSet were included in the training to assess the specificity of the model. Data preprocessing included noise reduction, histogram matching, and data augmentation techniques to ensure a robust training dataset. The methodology also incorporated using k-fold cross validation and a custom learning rate scheduling strategy to optimize the model’s training process. To further test its robustness, the model was evaluated by swapping random patches and analyzing 60 mm patches from the entire CT scan for local feature categorization. Results (if a Case Study enter NA) The model has demonstrated outstanding performance, with sensitivity and specificity exceeding 95% on our internal dataset (n=117) and aligning well with clinical, smear, culture, and anti-TB antibody findings. It also showed promising results on the external NIH TB Portal’s dataset (n= 1,460). The model is robust to variations in CT scanner image quality and noise profiles, maintaining strong generalization across datasets. Even when analyzing random 60mm patches, effectively classifying all local features Conclusion Our study enhances tuberculosis diagnostic methods in endemic regions by blending machine learning with conventional diagnostic techniques. This approach improves TB detection, particularly the early and accurate identification in areas impacted by drug-resistant strains and asymptomatic latent infections. By integrating radiological and clinico-pathological data, we aim to develop a comprehensive diagnostic score. Moving forward, we plan to further integrate this method into clinical decision support systems and adapt the approach to other infectious diseases.

Cubic spline estimation for non parametric uncertain differential equation

. In the history of researching to estimate the unknown parameters that were in the uncertain differential equation (UDE), the problem of parameter estimation with known functional forms is often studied. However, in practical situations, its functional form is often unknown. In order to deal with this problem, this article proposes the cubic spline method to approximate the autonomous UDE, and perform non parametric estimation on it. The cross validation is introduced to determine the number of term (J), which is in the approximate cubic spline. In addition, the uncertain hypothesis testing is given to verify the rationality of this method. Finally, some numerical examples are given. Then this method is applied to a case study of the Beijing Air Quality Index, and a comparative analysis is given to verify the practicability and superiority of the cubic spline method.

Depth Video-Based Secondary Action Recognition in Vehicles via Convolutional Neural Network and Bidirectional Long Short-Term Memory with Spatial Enhanced Attention Mechanism

Secondary actions in vehicles are activities that drivers engage in while driving that are not directly related to the primary task of operating the vehicle. Secondary Action Recognition (SAR) in drivers is vital for enhancing road safety and minimizing accidents related to distracted driving. It also plays an important part in modern car driving systems such as Advanced Driving Assistance Systems (ADASs), as it helps identify distractions and predict the driver’s intent. Traditional methods of action recognition in vehicles mostly rely on RGB videos, which can be significantly impacted by external conditions such as low light levels. In this research, we introduce a novel method for SAR. Our approach utilizes depth-video data obtained from a depth sensor located in a vehicle. Our methodology leverages the Convolutional Neural Network (CNN), which is enhanced by the Spatial Enhanced Attention Mechanism (SEAM) and combined with Bidirectional Long Short-Term Memory (Bi-LSTM) networks. This method significantly enhances action recognition ability in depth videos by improving both the spatial and temporal aspects. We conduct experiments using K-fold cross validation, and the experimental results show that on the public benchmark dataset Drive&Act, our proposed method shows significant improvement in SAR compared to the state-of-the-art methods, reaching an accuracy of about 84% in SAR in depth videos.

Estimation of the Unemployment Rate in Indonesia Using Nonparametric Spline Regression

The COVID-19 pandemic impact is not only on health but on all aspects of life, one of the socio-economic aspects is unemployment. The linear regression be used as a method in modeling the unemployment rate frequently. Whereas linear regression is inappropriate for data that have trends or irregular relationship patterns such as unemployment rate data. Therefore, this research used of the nonparametric spline regression method. It can produce more flexible curves according to the flatline of data based on optimum knots at the optimum value of the cross validation. There are several factors suggested that affect the Unemployment Rate (TPT ), including the number of positive COVID cases, labor force participation rate, economic growth and human development index. The purpose of this study is to find out if there is a difference between the open unemployment rate (TPT) in Indonesia before and during the pandemic. The Data used are secondary data obtained from Badan Pusat Statistika (BPS). The calculations results show that there are differences between the models of the unemployment rate before and during the pandemic. Generally there are three significant variables in the before the pandemic model: variables of economic growth (PE) and the human development index (IPM), while during the pandemic the significant variable are the labor force participation rate (TPAK) and economic growth (PE).

Comparative Analysis of Cross-Validation Techniques: LOOCV, K-folds Cross-Validation, and Repeated K-folds Cross-Validation in Machine Learning Models

Effective model evaluation is crucial for robust machine learning, and cross-validation techniques play a significant role. This study compares Repeated k-folds Cross Validation, k-folds Cross Validation, and Leave-One-Out Cross Validation (LOOCV) on imbalanced and balanced datasets across four models: Support Vector Machine (SVM), K-Nearest Neighbors (K-NN), Random Forest (RF), and Bagging, both with and without parameter tuning. On imbalanced data without parameter tuning, Repeated k-folds cross-validation demonstrated strong performance for SVM with a sensitivity of 0.541 and balanced accuracy of 0.764. K-folds Cross Validation showed a higher sensitivity of 0.784 for RF and a balanced accuracy of 0.884. In contrast, LOOCV achieved notable sensitivity for RF and Bagging at 0.787 and 0.784, respectively, but at the cost of lower precision and higher variance, as detailed in Table 1. When parameter tuning was applied to balanced data, the performance metrics improved. Sensitivity for SVM reached 0.893 with LOOCV and balanced accuracy for Bagging increased to 0.895. Stratified k-folds provided enhanced precision and F1-Score for SVM and RF. Notably, processing times varied significantly, with k-folds being the most efficient with SVM taking 21.480 seconds and Repeated k-folds showing higher computational demands where RF took approximately 1986.570 seconds in model processing, as shown in Table 4. This analysis underscores that while k-folds and repeated k-folds are generally efficient, LOOCV and balanced approaches offer enhanced accuracy for specific models but require greater computational resources. The choice of cross-validation technique should thus be tailored to the dataset characteristics and computational constraints to ensure optimal model evaluation.

Integrating proximal geophysical sensing and machine learning for digital soil mapping: Spatial prediction and model evaluation using a small dataset

Geophysical methods support soil security by providing non-invasive tools to assess soil properties, monitor degradation, and guide sustainable management strategies. However, studies focusing the spatial prediction of geophysical data remain limited. In this research, we aimed to model and predict the spatial distribution of soil geophysical properties using parent material and terrain attributes with machine learning algorithms. In addition, we tested the nested leave-one-out cross validation (nested-LOOCV) method to deal with datasets with limited size. We performed a geophysical survey using three types of sensors (radiometric, magnetic and electric methods). The random forest (RF) and support vector machine (SVM) algorithms presented the best results, with RF showing higher performance for K40 and magnetic susceptibility, and SVM had higher performance for eU, eTh and apparent electrical conductivity. Parent materials and digital elevation model were the most significant variables for the modelling. The nested-LOOCV method proved to be adequate for small soil dataset. Machine learning techniques are potential tools for modelling soil geophysical variables. The combination with computational techniques shows the great relevance of geophysical measurements for the estimation of soil properties related to fertility and soil genesis.

Radiomics-based machine learning role in differential diagnosis between small renal oncocytoma and clear cells carcinoma on contrast-enhanced CT: A pilot study

PurposeTo investigate the potential role of radiomics-based machine learning in differentiating small renal oncocytoma (RO) from clear cells carcinoma (ccRCC) on contrast-enhanced CT (CECT). Material and methodsFifty-two patients with small renal masses who underwent CECT before surgery between January 2016 and December 2020 were retrospectively included in the study. At pathology examination 39 ccRCC and 13 RO were identified. All lesions were manually delineated unenhanced (B), arterial (A) and venous (V) phases. Radiomics features were extracted using three different fixed bin widths (bw) of 25 HU, 10 HU, and 5 HU from each phase (B, A, V), and with different combinations (B+A, B+V, B+A+V, A+V), leading to 21 different datasets. Montecarlo Cross Validation technique was used to quantify the estimator performance. The final model built using the hyperparameter selected with Optuna was trained again on the training set and the final performance evaluation was made on the test set. ResultsThe A+V bw 10 achieved the greater median (IQR) balanced accuracy considering all the models of 0.70 (0.64–0.75), while A bw 10 considering only the monophasic ones. The A bw 10 model achieved a median (IQR) sensitivity of 0.60 (0.40–0.60), specificity of 0.80 (0.73–0.87), AUC-ROC of 0.77 (0.66–0.84), accuracy of 0.75 (0.70–0.80), and a Phi Coefficient of 0.38 (0.20–0.47). None of the nine models with the lowest mean balanced accuracy values implemented features from A. ConclusionThe A bw 10 model was identified as the most efficient mono-phasic model in differentiating small RO from ccRCC.

Motor Insurance Claim Frequency Prediction Using XGBoost

Insurance claim frequency modelling is an important task for non-life insurers, this together with other variables forms an important part of product pricing and risk management. The traditional frequency models such as Poisson, Negative Binomial and Zero Inflated models have several weakness such as scalibility issues, overdispersion and independence assumptions for large datasets and therefore not ideal to use when dealing with complex and unstructured data. Extreme gradient Boosting Algorithm (XGBoost) is an ensemble learning which has the capacity to effectively handle big complex and unstructured insurance data. XGBoost creates tree-based models by iteratively fitting decision trees to the residuals of the previous predictions, effectively reducing the error in each iteration. This research utilized and explored the XGBoost algorithm to process motor insurance claims large dataset in-order to predict the frequencies of insurance claims, that is 0,1,2,and 3. Using this algorithm we aim to enhance the accuracy of predictions that will yield better estimates for improved risk assessment and pricing of insurance products. Cross validation was performed to assess the true performance of our model. Cross validation results showed that XGBoost models for the claim frequency had a RMSE estimate of 0.949, MAE of 0.7741 and RSQ 0.781. This demonstrated a strong predictive performance, with an RMSE of 0.949 and an MAE of 0.7741, indicating a low average error in the predictions.The RSQ of 0.781 suggests further that the model explained a significant proportion of variability of the insurance data. Our model was evaluated with a confusion matrix. The results of the confusion matrix showed that for the frequency 0 99.59% of cases were correctly predicted, frequency 1 94.01% were correctly predicted and frequency 2 84.80% and finally frequency 3 only 40.96% of the observations were correctly predicted.These results highlights the potential of XGBoost as a robust modeling technique for handling big data and accurately predicting insurance claim frequency. The results corroborate with other studies that XGBoost is an invaluable tool for insurance companies.

Radial polarisation patterns identify macular damage: a machine learning approach

ABSTRACT Clinical relevance Identifying polarisation-modulated patterns may be an effective method for both detecting and monitoring macular damage. Background The aim of this work is to determine the effectiveness of polarisation-modulated patterns in identifying macular damage and foveolar involvement using a methodology that involved feature selection, Naïve Bayes supervised machine learning, cross validation, and use of an interpretable nomogram. Methods A cross-sectional study involving 520 eyes was undertaken, encompassing both normal and abnormal cases, including those with age-related macular disease, diabetic retinopathy or epiretinal membrane. Macular damage and foveolar integrity were assessed using optical coherence tomography. Various polarisation-modulated geometrical and optotype patterns were employed, along with traditional methods for visual function measurement, to complete perceptual detection and identification measures. Other variables assessed included age, sex, eye (right, left) and ocular media (normal, pseudophakic, cataract). Redundant variables were removed using a Fast Correlation-Based Filter. The area under the receiver operating characteristic curve and Matthews correlation coefficient were calculated, following 5-fold stratified cross validation, for Naïve Bayes models describing the relationship between the selected predictors of macular damage and foveolar involvement. Results Only radially structured polarisation-modulated patterns and age emerged as predictors of macular damage and foveolar involvement. All other variables, including traditional logMAR measures of visual acuity, were identified as redundant. Naïve Bayes, utilising the Fast Correlation-Based Filter selected features, provided a good prediction for macular damage and foveolar involvement, with an area under the receiver operating curve exceeding 0.7. Additionally, Matthews correlation coefficient showed a medium size effect for both conditions. Conclusions Radially structured polarisation-modulated geometric patterns outperform polarisation-modulated optotypes and standard logMAR acuity measures in predicting macular damage, regardless of foveolar involvement.

Statistical modelling of a tractor tractive performance during ploughing operation on a tropical Alfisol

Abstract Tractor is the most prominent off-road agricultural machinery that is significant to the global food security. The tractive modelling of tyre–soil interaction and agricultural implement dynamics is a complex phenomenon that require holistic approach. Terramechanics techniques such as empirical, semi-empirical, analytical, and numerical methods such as finite element models and discrete element models have gained traction in tractive performance studies. Some of these approaches are premised on large arrays of variables for modelling tractive performance based on the soil–tyre and tools interactions. In this study, soft computing in R software domain was used to model the tractor tractive performance during ploughing operations on a tropical Alfisol. The research farm at the National Centre for Agricultural Mechanization was used for the field experiment. The experimental design was a nested-factorial under a Randomized Complete Block Design having three replications. The input factors were tractor power size, T, (60, 65, and 70 hp); tyre inflation pressure, P, (83, 124, and 165 kPa); implement configuration, I, (2 and 3 bottoms disc plough); and operational speed, S, (6.31, 7.90, 9.47, 11.05, and 12.63 km/h). Standard procedures were followed to obtain the measured parameters in the field, which were statistically analysed. Correlation analysis and analysis of variance of the measured parameters at 5% significance level were established. Multiple linear regression was used to develop the model, validated using the 10-fold cross-validation method. The results revealed that the evaluated variables have a range of 1.56–7.79 kN, 5.15–27.20%, 9.10–32.00 cm, 4.50–13.94%, 1.31–1.67 g/cm3, 95.89–207.78 kPa, and 98.67–295.56 for draught, wheel slip, depth of cut, moisture content, bulk density, cone index (CI), and shear stress, respectively. A positive correlation exists between the towing force (TF) and the measured variables except for the shear stress and CI. The final developed model has seven variables for predicting TF with a 6.5% error and an average of 0.4735 cross validation root mean square error. The model quality of fit achieved an R Adj 2 = 0.8754 {R}_{\text{Adj}}^{2}=0.8754 which satisfactorily described the response variable. The study provides insights into tractive dynamic systems modelling of machine, tractive medium (soil), and agricultural tools anchored on soft computing approach. Its adoption will assist in quality ploughing operation integrating the variables established in the model.

Detecting Patient Position Using Bed-Reaction Forces for Pressure Injury Prevention and Management

A key best practice to prevent and treat pressure injuries (PIs) is to ensure at-risk individuals are repositioned regularly. Our team designed a non-contact position detection system that predicts an individual’s position in bed using data from load cells under the bed legs. The system was originally designed to predict the individual’s position as left-side lying, right-side lying, or supine. Our previous work suggested that a higher precision for detecting position (classifying more than three positions) may be needed to determine whether key bony prominences on the pelvis at high risk of PIs have been off-loaded. The objective of this study was to determine the impact of categorizing participant position with higher precision using the system prediction F1 score. Data from 18 participants was collected from four load cells placed under the bed legs and a pelvis-mounted inertial measurement unit while the participants assumed 21 positions. The data was used to train classifiers to predict the participants’ transverse pelvic angle using three different position bin sizes (45°, ~30°, and 15°). A leave-one-participant-out cross validation approach was used to evaluate classifier performance for each bin size. Results indicated that our prediction F1 score dropped as the position category precision was increased.

Development of a Machine‐Learning Model for Diagnosis of Pancreatic Cancer from Serum Samples Analyzed by Thermal Liquid Biopsy

Pancreatic ductal adenocarcinoma (PDAC) poses a considerable diagnostic and therapeutic challenge due to the lack of specific biomarkers and late diagnosis. Early detection is crucial for improving prognosis, but current techniques are insufficient. An innovative approach based on differential scanning calorimetry (DSC) of blood serum samples, thermal liquid biopsy (TLB), combined with machine‐learning (ML) analysis, may offer a more efficient method for diagnosing PDAC. Serum samples from a cohort of 212 PDAC patients and 184 healthy controls are studied. DSC thermograms are analyzed using ML models. The generated models are built applying algorithms based on penalized regression, resampling, categorization, cross validation, and variable selection. The ML‐based model demonstrates outstanding ability to discriminate between PDAC patients and control subjects, with a sensitivity of 90% and an area under the ROC receiver operating characteristic curve of 0.83 in the training and test groups. Application of the model to an independent validation cohort of 113 PDAC patients confirms its robustness and utility as a diagnosis tool. The application of ML to serum TLB data emerges as a promising methodology for early diagnosis, representing a significant advance for detecting and managing PDAC, envisaging a minimally invasive and more efficient methodology for identifying biomarkers.

Fusion of spectral and topographic features for land use mapping using a machine learning framework for a regional scale application.

This study investigated the dynamics of land use and land cover (LULC) modelling, mapping, and assessment in the Kegalle District ofSri Lanka, where policy decision-making is crucial in agricultural development where LULC temporal datasets are not readily available.Employing remotely sensed datasets and machine learning algorithms, the work presented here aims to compare the accuracy of three classification approaches in mapping LULC categories across the time in the study area primarily using the Google Earth Engine (GEE). Three classifiers namely random forest (RF), support vector machines (SVM), and classification and regression trees (CART) were used in LULC modelling, mapping, and change analysis. Different combinations of input features were investigated to improve classification performance. Developed models were optimised using the grid search cross-validation (CV) hyperparameter optimisation approach. It was revealed that the RF classifier constantly outstrips SVM and CART in terms of accuracy measures, highlighting its reliability in classifying the LULC. Land cover changes were examined for two periods, from 2001 to 2013 and 2013 to 2022, implying major alterations such as the conversion of rubber and coconut areas to built-up areas and barren lands. For suitable classification with higher accuracy, the study suggests utilising high spatial resolution satellite data, advanced feature selection approaches, and a combination of several spatial and spatial-temporal data sources. The study demonstrated practical applications of derived temporal LULC datasets for land management practices in agricultural development activities in developing nations.

Estimating and Predicting the Rate of Kidney Function Decline over 10 years in the General Population.

We aimed to estimate the rate of kidney function decline over 10 years in the general population and develop a machine learning model to predict it. We used the JMDC database from 2012 to 2021, which includes company employees and their family members in Japan, where annual health checks are mandated for people aged 40-74 years. We estimated the slope (average change) of estimated glomerular filtration rate (eGFR) over a period of 10 years. Then, using the annual health-check results and prescription claims for the first five years from 2012 to 2016 as predictor variables, we developed an XGBoost model, evaluated its prediction performance with the root mean squared error (RMSE), R2, and area under the receiver operating characteristic curve (AUROC) for rapid decliners (defined as the slope <-3 ml/min/1.73 m2/year) using 5-fold cross validation, and compared these indicators with those of (i) the simple application of the eGFR slope from 2012 to 2016 and (ii) the adjusted linear regression model. We included 126,424 individuals (mean age, 45.2 years; male, 82.4%; mean eGFR, 79.0 ml/min/1.73 m2 in 2016). The mean slope was -0.89 (standard deviation, 0.96) ml/min/1.73 m2/year. The predictive performance of the XGBoost model (RMSE, 0.78; R2, 0.35; and AUROC, 0.89) was better than that of either the simple application of the eGFR slope from 2012 to 2016 (RMSE, 1.94; R2, -3.03; and AUROC, 0.79) or the adjusted linear regression model (RMSE, 0.81; R2, 0.30; and AUROC, 0.87). We estimated the rate of kidney function decline over 10 years in the general population as well as demonstrated that application of machine learning to annual health-check and claims data provides better predictive performance compared to traditional methods.

Noninvasive assessment of single kidney glomerular filtration rate using multiple diffusion weighted imaging models.

This study aimed to assess single kidney glomerular filtration rate (GFR) using various diffusion weighted imaging (DWI) models. We reviewed adult patients with kidney diseases who underwent magnetic resonance imaging (MRI) examination from February 2021 to December 2023. DWI with 13 b-values was performed using 3.0-T scanners. Diffusion parameters were calculated with multi-slice ROIs positioned in renal parenchyma using four DWI models, including monoexponential model (MEM), diffusion kurtosis imaging (DKI), stretched exponential model (SEM), and intravoxel incoherent motion (IVIM). The split GFRs were measured by 99mTc-DTPA scintigraphy using Gates' method. Four different regression algorithms including the linear regression, regression tree, Gaussian regression and support vector machine (SVM) regression were employed to predict the GFR value based on different diffusion parameters. The leave-one-out cross validation was used to evaluate prediction ability of different models, and the performance of each model was quantified using the root mean square error (RMSE) and correlation coefficient. Fifteen (male/female, 10/5; age, 41.60±10.83 years) patients were included in this study. Among the four DWI models, the IVIM parameters with SVM regression model achieved the best performance with 0.184 RMSE and 0.789 correlation coefficient ( ). The parameters combining the four DWI models with SVM regression algorithm achieved the best performance in this study, with 0.171 RMSE and 0.815 correlation coefficient ( ). The DWI characteristics are able to serve as imaging biomarkers for assessing the function of single kidney. The integration of DWI into clinical practice could contribute to the advancement of non-invasive diagnostic methodologies.

Predicting response to immunotherapy in oral squamous cell carcinoma via a CT-based radiomics model

BackgroundTo investigate whether radiomics models derived from pretreatment CT could help to predict response to immunotherapy in oral squamous cell carcinoma (OSCC).MethodsRetrospectively, a total of 40 patients with measurable OSCC were included. The patients were divided into responder group and non-responder group according to the comparison of pre-treatment and post-treatment CT findings. Radiomics features were extracted from pre-treatment CT images, and optimal features were selected by univariate analysis and the least absolute shrinkage and selection operator (LASSO) regression analysis. Neural network, support vector machine, random forest and logistic regression models were used to predict response to immunotherapy in OSCC, and leave-one-out cross validation was employed to assess the performance of the classifiers. The area under the curve (AUC), accuracy, sensitivity and specificity were calculated to quantify the predictive efficacy.ResultsA total of 7 features were selected to build models upon machine learning methods. By comparing different machine learning based models, the neural network model achieved the best predictive ability, with an AUC of 0.864, an accuracy of 82.5%, a sensitivity of 82.5%, and a specificity of 82.5%.ConclusionsThe pretreatment CT-based radiomics model showed good performance in predicting response to immunotherapy in OSCC. Pretreatment CT-based radiomics model might provide an alternative approach for the selection of patients who benefit from immunotherapy.

Generalized Additive Models

Generalized additive models are generalized linear models in which the linear predictor includes a sum of smooth functions of covariates, where the shape of the functions is to be estimated. They have also been generalized beyond the original generalized linear model setting to distributions outside the exponential family and to situations in which multiple parameters of the response distribution may depend on sums of smooth functions of covariates. The widely used computational and inferential framework in which the smooth terms are represented as latent Gaussian processes, splines, or Gaussian random effects is reviewed, paying particular attention to the case in which computational and theoretical tractability is obtained by prior rank reduction of the model terms. An empirical Bayes approach is taken, and its relatively good frequentist performance discussed, along with some more overtly frequentist approaches to model selection. Estimation of the degree of smoothness of component functions via cross validation or marginal likelihood is covered, alongside the computational strategies required in practice, including when data and models are reasonably large. It is briefly shown how the framework extends easily to location-scale modeling, and, with more effort, to techniques such as quantile regression. Also covered are the main classes of smooths of multiple covariates that may be included in models: isotropic splines and tensor product smooth interaction terms.

Prediction of pathological complete response to chemotherapy for breast cancer using deep neural network with uncertainty quantification.

The I-SPY 2 trial is a national-wide, multi-institutional clinical trial designed to evaluate multiple new therapeutic drugs for high-risk breast cancer. Previous studies suggest that pathological complete response (pCR) is a viable indicator of long-term outcomes of neoadjuvant chemotherapy for high-risk breast cancer. While pCR can be assessed during surgery after the chemotherapy, early prediction of pCR before the completion of the chemotherapy may facilitate personalized treatment management to achieve an improved outcome. Notably, the acquisition of dynamic contrast-enhanced magnetic resonance (DCEMR) images at multiple time points during the I-SPY 2 trial opens up the possibility of achieving early pCR prediction. In this study, we investigated the feasibility of the early prediction of pCR to neoadjuvant chemotherapy using multi-time point DCEMR images and clinical data acquired in the I-SPY2 trial. The prediction uncertainty was also quantified to allow physicians to make patient-specific decisions on treatment plans based on the level of associated uncertainty. The dataset used in our study included 624 patients with DCEMR images acquired at 3 time points before the completion of the chemotherapy: pretreatment (T0), after 3 cycles of treatment (T1), and after 12 cycles of treatment (T2). A convolutional long short-term memory (LSTM) network-based deep learning model, which integrated multi-time point deep image representations with clinical data, including tumor subtypes, was developed to predict pCR. The performance of the model was evaluated via the method of nested 5-fold cross validation. Moreover, we also quantified prediction uncertainty for each patient through test-time augmentation. To investigate the relationship between predictive performance and uncertainty, the area under the receiver operating characteristic curve (AUROC) was assessed on subgroups of patients stratified by the uncertainty score. By integrating clinical data and DCEMR images obtained at three-time points before treatment completion, the AUROC reached 0.833 with a sensitivity of 0.723 and specificity of 0.800. This performance was significantly superior (p<0.01) to models using only images (AUROC=0.706) or only clinical data (AUROC=0.746). After stratifying the patients into eight subgroups based on the uncertainty score, we found that group #1, with the lowest uncertainty, had a superior AUROC of 0.873. The AUROC decreased to 0.637 for group #8, which had the highest uncertainty. The results indicate that our convolutional LSTM network-based deep learning model can be used to predict pCR earlier before the completion of chemotherapy. By combining clinical data and multi-time point deep image representations, our model outperforms models built solely on clinical or image data. Estimating prediction uncertainty may enable physicians to prioritize or disregard predictions based on their associated uncertainties. This approach could potentially enhance the personalization of breast cancer therapy.