Back-propagation Artificial Neural Network Model Research Articles

Comparative Study of Back-Propagation Artificial Neural Network Models for Predicting Salinity Parameters Based on Spectroscopy Under Different Surface Conditions of Soda Saline–Alkali Soils

Comparative Study of Back-Propagation Artificial Neural Network Models for Predicting Salinity Parameters Based on Spectroscopy Under Different Surface Conditions of Soda Saline–Alkali Soils

A Novel Neural Network Prediction Model for Mineral Content Based on the Data of Elemental Logging and Well Logging

The mineral content of shale plays an important role in the study of reservoir characteristics. Obtaining high-precision mineral content data from a simple and straightforward method is helpful for the study of reservoir properties. Due to the complex and varied elemental combination of minerals and the lithological diversity in different formations, there are limitations in the traditional statistical methods for predicting mineral content. In this study, a multi-source fusion dataset is constructed first using a detailed workflow to process multi-source data from elemental logging and conventional well logging from the Funing shale formations in China. A hybrid neural network model consisting of a back-propagation artificial neural network (BP-ANN) and gated recurrent unit (GRU) structure, with a custom constraint loss function, is developed to predict the mineral contents. The results show that the predictive performance of this hybrid multi-source model performs statistically significantly better than that of the BP-ANN and GRU models using single source data. In addition, the accuracy of the hybrid model is further improved by using the custom constraint loss function. This method provides high-resolution distributions of mineral content compared to the X-ray powder diffraction testing method. It will help enhance the understanding of the formation.

Prediction of 3D printability and rheological properties of different pineapple gel formulations based on LF-NMR

In this study, a pineapple-starch-xanthan gum system was prepared using fresh pineapple juice, maize starch, and xanthan gum (XG). The feasibility of using low-field nuclear magnetic resonance (LF-NMR) to predict pineapple gels' rheological properties and printability was evaluated. Results indicated that as maize starch and XG increased, the gel transformed from unable to support printed models to a stable shape, eventually becoming too viscous for printing. Principal component analysis and Fisher discriminant analysis classified the gels into four categories based on their rheological properties, aligning with the actual printing results. Pearson correlation analysis showed a strong correlation between the LF-NMR parameters and the rheological properties of gels. The partial least squares (PLS) and back-propagation artificial neural network (BP-ANN) models constructed using the LF-NMR parameters can effectively predict the rheological properties of pineapple gels. Therefore, LF-NMR is a valuable, non-destructive method for quickly assessing pineapple gels' 3D printing suitability.

Prediction of soil heavy metal content around mine tailings using multiple methods combined with transformed hyperspectral reflectance data

The extensive accumulation of tailings can potentially cause heavy metal contamination in the surrounding farmland soil. Accurately predicting the spatial distribution of heavy metals in farmland soil is crucial for assessing the potential environmental hazards of tailings.This study focuses on the spatial distribution and the quantitative prediction of heavy metals (chromium (Cr), vanadium (V), and copper (Cu)) in soils surrounding mine tailings using advanced spectral data analysis and multiple prediction models. The original hyperspectral reflectance data were processed using first-order differential (FD), second-order differential (SD), reciprocal logarithmic (LR), and continuum removal (CR) transformations to highlight the positions of characteristic bands. Multiple linear regression (MLR), stepwise linear regression (SLR), partial least squares regression (PLSR), random forest (RF), and back propagation artificial neural network (BP-ANN) models were used to establish inversion models for Cr, V, and Cu based on bands with high correlation coefficients. The performance of the inversion models was evaluated using the coefficient of determination (R2), root mean square error (RMSE), mean absolute error (MAE), and residual predictive deviation (RPD). The results indicate that the raw hyperspectral data from the measured soil exhibit a weak response to heavy metal content in the study area. However, applying FD, SD, and CR transformations significantly enhances the sensitivity of soil spectral data to heavy metal concentrations, facilitating subsequent modeling. Among these, the SD transformation is particularly beneficial for modeling the Cr and Cu elements in the soil. For the V element, the FD transformation yields data that are more suitable for modeling. Regarding the inversion models based on the measured spectral data, the BP-ANN model exhibited the best predictive performance. Specifically, when combined with SD spectral data, the BP-ANN achieved the highest predictive accuracy for Cu content (R² = 0.85, RPD = 2.12). The RF model demonstrated the next best performance, with its optimal inversion model also utilizing SD spectral data for predicting Cu content (R² = 0.76, RPD = 1.90). On the other hand, the MLR model exhibited the poorest performance and is unsuitable for predicting heavy metal content in the region using the measured spectral data. This study highlights the potential of spectral data in environmental monitoring and provides a technical reference for the inversion assessment and regulation of heavy metals in farmlands surrounding tailing sites.

Indirect prediction of the 3D printability of polysaccharide gels using multiple machine learning (ML) models

In this paper, the capabilities of NIR spectroscopy and LF-NMR data were compared for rapidly predicting the rheological properties of polysaccharide gels and assessing their printability. Seven machine learning (ML) models were established for rheological property prediction based on partial least squares regression (PLSR), support vector regression (SVR), back propagation artificial neural network (BPANN), one-dimensional convolutional neural network (1D CNN), recurrent neural network (RNN), long short-term memory (LSTM), and Transformer. The results showed that among the seven models, the SVR, BPANN, and 1D CNN models based on NIR spectroscopy effectively predicted the rheological parameters of polysaccharide gels, with the highest R2 in the prediction set reaching 0.9796 and the highest RPD reaching 7.0708. For most polysaccharide gels, using the LF-NMR relaxation time distribution curves provided better predictions of rheological properties than using transverse relaxation time and peak area. Among the seven models, the PLSR, SVR, 1D CNN, and Transformer models effectively predicted the rheological characteristics based on LF-NMR parameters, with the highest R2 in the prediction set reaching 0.9869 and the highest RPD reaching 8.7220. This study successfully established a prediction system for the rheological behaviors and 3D printing performance of polysaccharide gels using NIR spectroscopy and LF-NMR data combined with ML methods, achieving an intelligent assessment of the 3D printing behavior of polysaccharide gels.

Efficient production of pullulan by Aureobasidium pullulans using a multi-objective optimization strategy with orthogonal experimental design coupling artificial neural network and genetic algorithm

Efficient pullulan production has long been a central research focus. This study used maltodextrin as the carbon source for pullulan production by Aureobasidium pullulans fermentation. A hybrid optimization approach, integrating orthogonal experimental design (OED), backpropagation artificial neural network (BP-ANN), and elite strategy non-dominated sequential genetic algorithm-II (NSGA-II), was developed. Range analysis based on OED revealed that MgSO4·7H2O significantly affects production but less impacts molecular weight, while pH notably influences molecular weight with a lesser effect on production, underscoring the need for multi-objective optimization. The BP-ANN model showed strong predictive capabilities, with goodness-of-fit values of 0.984 and 0.980 for production and molecular weight, respectively. Using this model as the fitness function for the optimization algorithm enhanced efficiency. Taking cost factors into account, the BP-ANN-NSGA-II algorithm identified the optimal fermentation medium conditions, resulting in a 6.89 % increase in production, a 368.97 % increase in molecular weight, and a 42.49 % reduction in cost. The maximum comprehensive optimization efficiency is 63.73 %, and the multi-objective optimization is better than the single objective optimization. This method significantly guides the improvement of pullulan fermentation optimization efficiency.

Identification of key risk factors for venous thromboembolism in urological inpatients based on the Caprini scale and interpretable machine learning methods

PurposeTo identify the key risk factors for venous thromboembolism (VTE) in urological inpatients based on the Caprini scale using an interpretable machine learning method.MethodsVTE risk data of urological inpatients were obtained based on the Caprini scale in the case hospital. Based on the data, the Boruta method was used to further select the key variables from the 37 variables in the Caprini scale. Furthermore, decision rules corresponding to each risk level were generated using the rough set (RS) method. Finally, random forest (RF), support vector machine (SVM), and backpropagation artificial neural network (BPANN) were used to verify the data accuracy and were compared with the RS method.ResultsFollowing the screening, the key risk factors for VTE in urology were “(C1) Age,” “(C2) Minor Surgery planned,” “(C3) Obesity (BMI > 25),” “(C8) Varicose veins,” “(C9) Sepsis (< 1 month),” (C10) “Serious lung disease incl. pneumonia (< 1month) ” (C11) COPD,” “(C16) Other risk,” “(C18) Major surgery (> 45 min),” “(C19) Laparoscopic surgery (> 45 min),” “(C20) Patient confined to bed (> 72 h),” “(C18) Malignancy (present or previous),” “(C23) Central venous access,” “(C31) History of DVT/PE,” “(C32) Other congenital or acquired thrombophilia,” and “(C34) Stroke (< 1 month.” According to the decision rules of different risk levels obtained using the RS method, “(C1) Age,” “(C18) Major surgery (> 45 minutes),” and “(C21) Malignancy (present or previous)” were the main factors influencing mid- and high-risk levels, and some suggestions on VTE prevention were indicated based on these three factors. The average accuracies of the RS, RF, SVM, and BPANN models were 79.5%, 87.9%, 92.6%, and 97.2%, respectively. In addition, BPANN had the highest accuracy, recall, F1-score, and precision.ConclusionsThe RS model achieved poorer accuracy than the other three common machine learning models. However, the RS model provides strong interpretability and allows for the identification of high-risk factors and decision rules influencing high-risk assessments of VTE in urology. This transparency is very important for clinicians in the risk assessment process.

Artificial Neural Network Model for Predicting Mechanical Strengths of Economical Ultra-High-Performance Concrete Containing Coarse Aggregates: Development and Parametric Analysis.

Ultra-high-performance concrete with coarse aggregates (UHPC-CA) has the advantages of high strength, strong shrinkage resistance and a lower production cost, presenting a broad application prospect in civil engineering construction. In view of the difficulty in establishing a mathematical model to accurately predict the mechanical properties of UHPC-CA, the back-propagation artificial neural network (BP-ANN) method is used to fully consider the various influential factors of the compressive strength (CS) and flexural strength (FS) of UHPC-CA in this paper. By taking the content of cement (C), silica fume (SF), slag, fly ash (FA), coarse aggregate (CA), steel fiber, the water-binder ratio (w/b), the sand rate (SR), the cement type (CT), and the curing method (CM) as input variables, and the CS and FS of UHPC-CA as output objectives, the BP-ANN model with three layers has been well-trained, validated and tested with 220 experimental data in the studies published in the literature. Four evaluating indicators including the determination coefficient (R2), the root mean square error (RMSE), the mean absolute percentage error (MAPE), and the integral absolute error (IAE) were used to evaluate the prediction accuracy of the BP-ANN model. A parametric study for the various influential factors on the CS and FS of UHPC-CA was conducted using the BP-ANN model and the corresponding influential mechanisms were analyzed. Finally, the inclusion levels for the CA, steel fiber, and the dimensionless parameters of the W/B and sand rate were recommended to obtain the optimal strength of UHPC-CA.

Identification of Euphorbiae pekinensis Radix and its counterfeit and adulterated products based on DNA barcode, UPLC-Q-TOF-MS, UPLC fingerprint, and chemometrics.

Euphorbiae pekinensis Radix (EPR) is a traditional Chinese herb commonly used to treat edema, pleural effusion, and ascites. However, counterfeit and adulterated products often appear in the market because of the homonym phenomenon, similar appearance, and artificial forgery of Chinese herbs. This study comprehensively evaluated the quality of EPR using multiple methods. The DNA barcode technique was used to identify EPR, while the UPLC-Q-TOF-MS technique was utilized to analyze the chemical composition of EPR. A total of 15 tannin and phenolic acid components were identified. Furthermore, UPLC fingerprints of EPR and its common counterfeit products were established, and unsupervised and supervised pattern recognition models were developed using these fingerprints. The backpropagation artificial neural network and counter-propagation artificial neural network models accurately identified counterfeit and adulterated products, with a counterfeit ratio of more than 25%. Finally, the contents of the chemical markers 3,3'-di-O-methyl ellagic acid-4'-O-β-D-glucopyranoside, ellagic acid, 3,3'-di-O-methyl ellagic acid-4'-O-β-d-xylopyranoside, and 3,3'-di-O-methyl ellagic acid were determined to range from 0.05% to 0.11%, 1.95% to 8.52%, 0.27% to 0.86%, and 0.10% to 0.42%, respectively. This proposed strategy offers a general procedure for identifying Chinese herbs and distinguishing between counterfeit and adulterated products.

A multiobjective optimization framework based on FEA, ANN, and NSGA-II to optimize the process parameters of tube-to-tubesheet joint

This study presents a multiobjective optimization framework that integrates Artificial Neural Network (ANN) and Non-dominated Sorting Genetic Algorithm-II (NSGA-II) for the optimization of rolling process parameters of tube-to-tubesheet joint (TTT-joint). During the rolling process, both beneficial contact pressure and detrimental tensile residual stress are generated within the joint. The primary objective of this framework is to minimize the tensile residual stress while maximizing the contact pressure in the TTT-joint. To achieve this, a backpropagation ANN model is trained to rapidly estimate the residual stress and contact pressure for various sets of rolling process parameters. For training purposes, a series of nonlinear elastoplastic finite element (FE) simulations are performed to generate the input database for the neural network. A detailed parametric study is performed based on the axisymmetric FE model of the TTT-joint. The trained neural network is then incorporated into the NSGA-II optimization algorithm to find the fitness function and optimized process parameters. The contact pressure and residual stress predicted by the proposed ANN-NSGA-II framework are validated by finite element analysis (FEA) using the optimized parameters. The present analysis established that the proposed methodology can be applied in practical engineering problems to obtain the process parameters that yield the maximum contact pressure and minimum tensile residual stress in the TTT-joint.

Design and optimization of handheld alloy analysis instrument based on microjoule high pulse repetition frequency LIBS.

We briefly describe the design of a handheld metal detection instrument based on microjoule high repetition frequency laser-induced breakdown spectroscopy. The instrument uses a Raspberry Pi as the control core and a laser with a frequency of 10kHz and a single pulse energy of 100 µJ as the excitation source. In addition, a mini-putter is built into the instrument to move the laser, allowing the ablation of the sample surface line area without external auxiliary equipment. The excitation-generated plasma radiation is collected by a simple optical path and transmitted directly to the spectrometer. We also constructed and trained a Backpropagation Artificial Neural Network (BP-ANN) model based on 12 different grades of alloys and transplanted the feedback process of the BP-ANN to the Raspberry Pi, which realized the rapid classification of the 12 alloys with >95% classification accuracy on the handheld instrument.

Artificial intelligence-based precise prediction of anthropometric data for female garment pattern-making

Anthropometric data form the cornerstone of garment pattern-making. This article introduces an artificial intelligence-driven approach, employing a back-propagation artificial neural network (BP-ANN), to predict the anthropometric data essential for crafting patterns for women’s upper tops. The model adeptly processes minimal critical data from women’s upper bodies, yielding projected dimensions that are arduous to manually measure yet crucial for tailoring body-fitting tops. Utilising a three-dimensional body scanner for accurate anthropometric data collection from 196 women in Sichuan Province, China, our study compares the BP-ANN model with a Linear Regression (LR) model. Results demonstrate superior predictive accuracy for BP-ANN. Notably, the BP-ANN model excels in efficiency and accuracy, particularly in challenging anthropometric parameters. The findings underscore the transformative potential of AI-based models in optimizing garment production processes, offering a precise alternative to traditional methods. This research contributes valuable insights for the integration of AI technology in advancing pattern-making practices.

Enhancing riverine load prediction of anthropogenic pollutants: Harnessing the potential of feed-forward backpropagation (FFBP) artificial neural network (ANN) models

Assessing riverine pollutant loads is a more realistic method for analysing point and non-point anthropogenic pollution sources throughout a watershed. This study compares numerous mathematical modelling strategies for estimating riverine loads based on the chosen water quality parameters: Biochemical Oxygen Demand (BOD), Chemical Oxygen Demand (COD), Suspended Solids (SS), and Ammoniacal Nitrogen (NH3–N). A riverine load model was developed by employing various input variables including river flow and pollutant concentration values collected at several monitoring sites. Among the mathematical modelling methods employed are artificial neural networks with feed-forward backpropagation algorithms and radial basis functions. The classical multiple linear regression (MLR) statistical model was used for the comparison. Four widely used statistical performance assessment metrics were adopted to evaluate the performance of the various developed models: the root mean square error (RMSE), mean absolute error (MAE), mean relative error (MRE), and coefficient of determination (R2). The considerable number of errors (with RMSE, MAE, and MRE) discovered in estimating riverine loads using the multiple linear regression (MLR) statistical model can be attributed to the nonlinear relationship between the independent variables (Q and Cx) and dependent variables (W). The feed-forward neural network model with a backpropagation algorithm and Bayesian regularisation training algorithm outperformed the radial basis neural network. This finding implies that, in addition to suspended sediment loads, riverine loads may be predicted using an artificial neural network using pollutant concentration (Cx) and river discharge (Q) as input variables. Other geographical and temporal fluctuation characteristics that may impact river water quality, on the other hand, may be incorporated as input variables to enhance riverine load prediction. Finally, riverine load analyses were successfully conducted to reduce the riverine load.

Dynamics optimization of small branch pipes in nuclear power plants based on machine learning algorithms

Vibration fatigue failure of small branch pipes poses a great threat to the safe operation of nuclear power plants. However, the transient and wide-band vibration problems are not adequately considered in ASME and RCC-M code, resulting in repeated fatigue failures. In addition, the current research mainly focuses on vibration test methods and fatigue analysis methods, neglecting the study of pipeline vibration characteristics. Therefore, innovative approaches were essential for effectively managing complex dynamic loads. In this study, an innovative approach combining backpropagation artificial neural networks (BP-ANN) and non-dominated sorting Genetic Algorithm II (NSGA-II) was proposed to optimize the vibration of these pipes. The goal was mitigating vibration-induced failures by enhancing operational stability. The methodology progressed through several key stages. Firstly, BP-ANN was utilized for regression analysis, correlating pipe characteristics to vibration effects. Through regression analysis, the complex interrelationships governing the pipes' dynamic behavior was revealed. Subsequently, based on the regression model, NSGA-II was used to derive an optimal combination of design parameters to minimize the vibration response. The proposed technique was validated on an L-shaped cantilevered pipe via finite element simulations and physical experiments. The analysis case shows that the BP-ANN model demonstrated excellent accuracy in predicting vibration responses. Meanwhile, NSGA-II successfully revealed the trade-offs between conflicting objectives, generating a Pareto-optimal set balancing stability under different excitation directions. This study highlights the potential of machine learning methods for dynamic optimization of small branch pipes in nuclear power plants, and verifies the accuracy and effectiveness of the method through experiments. The research results provide a new idea for small branch pipe design and vibration control of nuclear power plant, which contributes to enhancing the safety and reliability of small branch pipes.

Prediction and visualization of moisture content in Tencha drying processes by computer vision and deep learning.

It is important to monitor and control the moisture content throughout the Tencha drying processing procedure so that its quality is ensured. Workers often rely on their senses to perceive the moisture content, leading to relative subjectivity and low reproducibility. Traditional drying methods, which are used for measuring moisture content, are destructive to samples. This research was conducted using computer vision combined with deep learning to detect moisture content during the Tencha drying process. Different color space components of Tencha drying sample images were first extracted by computer vision. The color components were preprocessed using MinMax and Z score. Subsequently, one-dimensional convolutional neural networks (1D-CNN), partial least squares, and backpropagation artificial neural networks models were built and compared. The 1D-CNN model and Z scorepreprocessing achieved superior predictive accuracy, with correlation coefficient of prediction (Rp) = 0.9548 for moisture content. The migration of moisture content during the Tencha drying process was eventually visualized by mapping its spatial and temporal distributions. The results indicated that computer vision combined with 1D-CNN was feasible for moisture prediction during the Tencha drying process. This study provides technical support for the industrial and intelligent production of Tencha. © 2024 Society of Chemical Industry.

Classification of Banana Types Based on The Geometrical Attributes using Artificial Neural Network Method

Banana (Musa paradisiaca) is one of the important horticultural commodities. This study aims to measure the physical and geometrical parameters of three different bananas (Muli, Ambon, and Kepok) and to develop prediction equations using an Artificial Neural Network (ANN) model. In this study the backpropagation ANN model with supervised learning method was used. The ANN model had one output node, two hidden layers, and network architecture of 8 inputs, namely fruit weight and volume, projected area and roundness of the fruit, cross section, peel color, and geometric mean fruit cross section diameter. The data for building the model and testing the model were respectively 70% and 30% of the 150 data number in total. The results showed that the best ANN model structure for estimating Muli, Ambon and Kepok bananas was purelin-logsig-logsig with an RMSE value of 0.0077 and an R2 of 0.9999. This shows that the ANN model is highly robust to predict the banana types. Using the built model, the accuracy of the prediction results is 100%. Keywords: Artificial Neural Network, Banana fruits, Geometry attribute.

Application of Northern Goshawk Back-Propagation Artificial Neural Network in the Prediction of Monohydroxycarbazepine Concentration in Patients with Epilepsy.

A northern goshawk back-propagation artificial neural network (NGO-BPANN) model was established to predict monohydroxycarbazepine (MHD) concentration in patients with epilepsy. The data were collected from 108 Han Chinese patients with epilepsy on oxcarbazepine monotherapy. The results of 14 genotype variates were selected as the input layer in the first BPANN model, and the variables that had a more significant impact on the plasma concentration of MHD were retained. With demographic characteristics and clinical laboratory test results, the genotypes of SCN1A rs2298771 and SCN2A rs17183814 were used to construct the BPANN model. The BPANN model was comprehensively validated and used to predict the MHD plasma concentration of five patients with epilepsy in our hospital. The model demonstrated favorable fitness metrics, including a mean squared error of 0.00662, a gradient magnitude of 0.00753, an absence of validation tests amounting to zero, and a correlation coefficient of 0.980. Sex, BMI, and the genotype SCN1A rs2298771 were ranked highest by the absolute mean impact value (MIV), which is primarily associated with the concentration of MHD. The test group exhibited a range of - 20.84% to 31.03% bias between the predicted and measured values, with a correlation coefficient of 0.941 between the two. With BPANN, the MHD nadir concentration could be predicted precisely. The NGO-BPANN model exhibits exceptional predictive capability and can be a practical instrument for forecasting MHD concentration in patients with epilepsy. www.chiCTR-OOC-17012141 .

Formation and prediction of heterocyclic amines and N-nitrosamines in smoked sausages using back propagation artificial neural network.

Heterocyclic amines (HAs) and N-nitrosamines (NAs) are formed easily during the thermal processing of food, and epidemiological studies have demonstrated that consuming HAs and NAs increases the risk of cancer. However, there are few studies on the application of back propagation artificial neural network (BP-ANN) models to simultaneously predict the content of HAs and NAs in sausages. This study aimed to investigate the effects of cooking time and temperature, smoking time and temperature, and fat-to-lean ratio on the formation of HAs and NAs in smoked sausages, and to predict their total content based on the BP-ANN model. With an increase in processing time, processing temperature and fat ratio, the content of HAs and NAs in smoked sausages increased significantly, while the content of HA precursors and nitrite residues decreased significantly. The optimal network topology of the BP-ANN model was 5-11-2, the correlation coefficient values for training, validation, testing and all datasets were 0.99228, 0.99785, 0.99520 and 0.99369, respectively, and the mean squared error value of the best validation performance was 0.11326. The bias factor and the accuracy factor were within acceptable limits, and the predicted values approximated the true values, indicating that the model has good predictive performance. The contents of HAs and NAs in smoked sausages were significantly influenced by the cooking conditions, smoking conditions and fat ratio. The BP-ANN model has high application value in predicting the contents of HAs and NAs in sausages, which provides a theoretical basis for the suppression of carcinogen formation. © 2024 Society of Chemical Industry.

Predictive Quality Analytics of Surface Roughness in Turning Operation Using Polynomial and Artificial Neural Network Models

The variability of the material properties of steel from different suppliers causes problems in achieving the required surface quality after turning. Therefore, the manufacturer needs to estimate the resulting quality before starting production, especially if it is an expensive, small-batch production from stainless steel. Predictive models will make it possible to estimate the surface roughness from the mechanical properties of steel and thus support decision making about supplier selection or acceptance of a material supply. This research presents a step-by-step decision-making procedure, which enables the trained staff to make quick decisions based on commonly available information in the Mill Test Certificate (MTC). A new multivariate second-order polynomial model and feedforward backpropagation artificial neural network (ANN) models have been developed using input variables from the MTC: Tensile Strength, Yield Strength, Elongation, and Hardness. Models were used to enhance the methodological robustness in formulating the decision if the predicted surface roughness is outside the required range, even before accepting the delivery. Both models can accurately predict surface roughness, while the ANN model is more accurate than the polynomial model; however, the predictive model is sensitive to the accuracy of the input data, and the model’s prediction is valid only under precisely defined conditions.

Assessing the impact of occlusal plane rotation on facial aesthetics in orthodontic treatment: a machine learning approach

BackgroundAdequate occlusal plane (OP) rotation through orthodontic therapy enables satisfying profile improvements for patients who are disturbed by their maxillomandibular imbalance but reluctant to surgery. The study aims to quantify profile improvements that OP rotation could produce in orthodontic treatment and whether the efficacy differs among skeletal types via machine learning.Materials and methodsCephalometric radiographs of 903 patients were marked and analyzed by trained orthodontists with assistance of Uceph, a commercial software which use artificial intelligence to perform the cephalometrics analysis. Back-propagation artificial neural network (BP-ANN) models were then trained based on collected samples to fit the relationship among maxillomandibular structural indicators, SN-OP and P-A Face Height ratio (FHR), Facial Angle (FA). After corroborating the precision and reliability of the models by T-test and Bland-Altman analysis, simulation strategy and matrix computation were combined to predict the consequent changes of FHR, FA to OP rotation. Linear regression and statistical approaches were then applied for coefficient calculation and differences comparison.ResultsThe regression scores calculating the similarity between predicted and true values reached 0.916 and 0.908 in FHR, FA models respectively, and almost all pairs were in 95% CI of Bland-Altman analysis, confirming the effectiveness of our models. Matrix simulation was used to ascertain the efficacy of OP control in aesthetic improvements. Intriguingly, though FHR change rate appeared to be constant across groups, in FA models, hypodivergent group displayed more sensitive changes to SN-OP than normodivergent, hypodivergent group, and Class III group significantly showed larger changes than Class I and II.ConclusionsRotation of OP could yield differently to facial aesthetic improvements as more efficient in hypodivergent groups vertically and Class III groups sagittally.