Thank you for visiting Nature.com. The version of browser you are using has limited CSS support. For best results, we recommend using a newer version of your browser (or turning off compatibility mode in Internet Explorer). In the meantime, to ensure ongoing support, we are showing the site without styling or JavaScript.
Teeth are considered the most accurate indicator of the age of the human body and are often used in forensic age assessment. We aimed to validate data mining-based dental age estimates by comparing the estimation accuracy and classification performance of the 18-year threshold with traditional methods and data mining-based age estimates. A total of 2657 panoramic radiographs were collected from Korean and Japanese citizens aged 15 to 23 years. They were divided into a training set, each containing 900 Korean radiographs, and an internal test set containing 857 Japanese radiographs. We compared the classification accuracy and efficiency of traditional methods with test sets of data mining models. The accuracy of the traditional method on the internal test set is slightly higher than that of the data mining model, and the difference is small (mean absolute error <0.21 years, root mean square error <0.24 years). The classification performance for the 18-year cutoff is also similar between traditional methods and data mining models. Thus, traditional methods can be replaced by data mining models when performing forensic age assessment using the maturity of second and third molars in Korean adolescents and young adults.
Dental age estimation is widely used in forensic medicine and pediatric dentistry. In particular, because of the high correlation between chronological age and dental development, age assessment by dental developmental stages is an important criterion for assessing the age of children and adolescents1,2,3. However, for young people, estimating dental age based on dental maturity has its limitations because dental growth is almost complete, with the exception of the third molars. The legal purpose of determining the age of young people and adolescents is to provide accurate estimates and scientific evidence of whether they have reached the age of majority. In the medico-legal practice of adolescents and young adults in Korea, age was estimated using Lee’s method, and a legal threshold of 18 years was predicted based on the data reported by Oh et al 5 .
Machine learning is a type of artificial intelligence (AI) that repeatedly learns and classifies large amounts of data, solves problems on its own, and drives data programming. Machine learning can discover useful hidden patterns in large volumes of data6. In contrast, classical methods, which are labor-intensive and time-consuming, may have limitations when dealing with large volumes of complex data that are difficult to process manually7. Therefore, many studies have been conducted recently using the latest computer technologies to minimize human errors and efficiently process multidimensional data8,9,10,11,12. In particular, deep learning has been widely used in medical image analysis, and various methods for age estimation by automatically analyzing radiographs have been reported to improve the accuracy and efficiency of age estimation13,14,15,16,17,18,19,20. For example, Halabi et al 13 developed a machine learning algorithm based on convolutional neural networks (CNN) to estimate skeletal age using radiographs of children’s hands. This study proposes a model that applies machine learning to medical images and shows that these methods can improve diagnostic accuracy. Li et al14 estimated age from pelvic X-ray images using a deep learning CNN and compared them with regression results using ossification stage estimation. They found that the deep learning CNN model showed the same age estimation performance as the traditional regression model. Guo et al.’s study [15] evaluated the age tolerance classification performance of CNN technology based on dental orthophotos, and the results of the CNN model proved that humans outperformed its age classification performance.
Most studies on age estimation using machine learning use deep learning methods13,14,15,16,17,18,19,20. Age estimation based on deep learning is reported to be more accurate than traditional methods. However, this approach provides little opportunity to present the scientific basis for age estimates, such as the age indicators used in the estimates. There is also a legal dispute over who conducts the inspections. Therefore, age estimation based on deep learning is difficult to accept by administrative and judicial authorities. Data mining (DM) is a technique that can discover not only expected but also unexpected information as a method for discovering useful correlations between large amounts of data6,21,22. Machine learning is often used in data mining, and both data mining and machine learning use the same key algorithms to discover patterns in data. Age estimation using dental development is based on the examiner’s assessment of the maturity of the target teeth, and this assessment is expressed as a stage for each target tooth. DM can be used to analyze the correlation between dental assessment stage and actual age and has the potential to replace traditional statistical analysis. Therefore, if we apply DM techniques to age estimation, we can implement machine learning in forensic age estimation without worrying about legal liability. Several comparative studies have been published on possible alternatives to traditional manual methods used in forensic practice and EBM-based methods for determining dental age. Shen et al23 showed that the DM model is more accurate than the traditional Camerer formula. Galibourg et al24 applied different DM methods to predict age according to the Demirdjian criterion25 and the results showed that the DM method outperformed the Demirdjian and Willems methods in estimating the age of the French population.
To estimate the dental age of Korean adolescents and young adults, Lee’s method 4 is widely used in Korean forensic practice. This method uses traditional statistical analysis (such as multiple regression) to examine the relationship between Korean subjects and chronological age. In this study, age estimation methods obtained using traditional statistical methods are defined as “traditional methods.” Lee’s method is a traditional method, and its accuracy has been confirmed by Oh et al. 5; however, the applicability of age estimation based on the DM model in Korean forensic practice is still questionable. Our goal was to scientifically validate the potential usefulness of age estimation based on the DM model. The purpose of this study was (1) to compare the accuracy of two DM models in estimating dental age and (2) to compare the classification performance of 7 DM models at the age of 18 years with those obtained using traditional statistical methods Maturity of second and third molars in both jaws.
Means and standard deviations of chronological age by stage and tooth type are shown online in Supplementary Table S1 (training set), Supplementary Table S2 (internal test set), and Supplementary Table S3 (external test set). The kappa values for intra- and interobserver reliability obtained from the training set were 0.951 and 0.947, respectively. P values and 95% confidence intervals for kappa values are shown in online supplementary table S4. The kappa value was interpreted as “almost perfect”, consistent with the criteria of Landis and Koch26.
When comparing mean absolute error (MAE), the traditional method slightly outperforms the DM model for all genders and in the external male test set, with the exception of multilayer perceptron (MLP). The difference between the traditional model and the DM model on the internal MAE test set was 0.12–0.19 years for men and 0.17–0.21 years for women. For the external test battery, the differences are smaller (0.001–0.05 years for men and 0.05–0.09 years for women). Additionally, the root mean square error (RMSE) is slightly lower than the traditional method, with smaller differences (0.17–0.24, 0.2–0.24 for the male internal test set, and 0.03–0.07, 0.04–0.08 for external test set). ). MLP shows slightly better performance than Single Layer Perceptron (SLP), except in the case of the female external test set. For MAE and RMSE, the external test set scores higher than the internal test set for all genders and models. All MAE and RMSE are shown in Table 1 and Figure 1.
MAE and RMSE of traditional and data mining regression models. Mean absolute error MAE, root mean square error RMSE, single layer perceptron SLP, multilayer perceptron MLP, traditional CM method.
Classification performance (with a cutoff of 18 years) of the traditional and DM models was demonstrated in terms of sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and area under the receiver operating characteristic curve (AUROC) 27 (Table 2, Figure 2 and Supplementary Figure 1 online). In terms of the sensitivity of the internal test battery, traditional methods performed best among men and worse among women. However, the difference in classification performance between traditional methods and SD is 9.7% for men (MLP) and only 2.4% for women (XGBoost). Among DM models, logistic regression (LR) showed better sensitivity in both sexes. Regarding the specificity of the internal test set, it was observed that the four SD models performed well in males, while the traditional model performed better in females. The differences in classification performance for males and females are 13.3% (MLP) and 13.1% (MLP), respectively, indicating that the difference in classification performance between models exceeds sensitivity. Among the DM models, the support vector machine (SVM), decision tree (DT), and random forest (RF) models performed best among males, while the LR model performed best among females. The AUROC of the traditional model and all SD models was greater than 0.925 (k-nearest neighbor (KNN) in men), demonstrating excellent classification performance in discriminating 18-year-old samples28. For the external test set, there was a decrease in classification performance in terms of sensitivity, specificity and AUROC compared to the internal test set. Moreover, the difference in sensitivity and specificity between the classification performance of the best and worst models ranged from 10% to 25% and was larger than the difference in the internal test set.
Sensitivity and specificity of data mining classification models compared to traditional methods with a cutoff of 18 years. KNN k nearest neighbor, SVM support vector machine, LR logistic regression, DT decision tree, RF random forest, XGB XGBoost, MLP multilayer perceptron, traditional CM method.
The first step in this study was to compare the accuracy of dental age estimates obtained from seven DM models with those obtained using traditional regression. MAE and RMSE were evaluated in internal test sets for both sexes, and the difference between the traditional method and the DM model ranged from 44 to 77 days for MAE and from 62 to 88 days for RMSE. Although the traditional method was slightly more accurate in this study, it is difficult to conclude whether such a small difference has clinical or practical significance. These results indicate that the accuracy of dental age estimation using the DM model is almost the same as that of the traditional method. Direct comparison with results from previous studies is difficult because no study has compared the accuracy of DM models with traditional statistical methods using the same technique of recording teeth in the same age range as in this study. Galibourg et al24 compared MAE and RMSE between two traditional methods (Demirjian method25 and Willems method29) and 10 DM models in a French population aged 2 to 24 years. They reported that all DM models were more accurate than traditional methods, with differences of 0.20 and 0.38 years in MAE and 0.25 and 0.47 years in RMSE compared to the Willems and Demirdjian methods, respectively. The discrepancy between the SD model and traditional methods shown in the Halibourg study takes into account numerous reports30,31,32,33 that the Demirdjian method does not accurately estimate dental age in populations other than the French Canadians on which the study was based. in this study. Tai et al 34 used the MLP algorithm to predict tooth age from 1636 Chinese orthodontic photographs and compared its accuracy with the results of the Demirjian and Willems method. They reported that MLP has higher accuracy than traditional methods. The difference between the Demirdjian method and the traditional method is <0.32 years, and the Willems method is 0.28 years, which is similar to the results of the present study. The results of these previous studies24,34 are also consistent with the results of the present study, and the age estimation accuracy of the DM model and the traditional method are similar. However, based on the presented results, we can only cautiously conclude that the use of DM models to estimate age may replace existing methods due to the lack of comparative and reference previous studies. Follow-up studies using larger samples are needed to confirm the results obtained in this study.
Among the studies testing the accuracy of SD in estimating dental age, some showed higher accuracy than our study. Stepanovsky et al 35 applied 22 SD models to panoramic radiographs of 976 Czech residents aged 2.7 to 20.5 years and tested the accuracy of each model. They assessed the development of a total of 16 upper and lower left permanent teeth using the classification criteria proposed by Moorrees et al 36 . The MAE ranges from 0.64 to 0.94 years and the RMSE ranges from 0.85 to 1.27 years, which are more accurate than the two DM models used in this study. Shen et al23 used the Cameriere method to estimate the dental age of seven permanent teeth in the left mandible in eastern Chinese residents aged 5 to 13 years and compared it with ages estimated using linear regression, SVM and RF. They showed that all three DM models have higher accuracy compared to the traditional Cameriere formula. The MAE and RMSE in Shen’s study were lower than those in the DM model in this study. The increased precision of the studies by Stepanovsky et al. 35 and Shen et al. 23 may be due to the inclusion of younger subjects in their study samples. Because age estimates for participants with developing teeth become more accurate as the number of teeth increases during dental development, the accuracy of the resulting age estimation method may be compromised when study participants are younger. Additionally, MLP’s error in age estimation is slightly smaller than SLP’s, meaning that MLP is more accurate than SLP. MLP is considered slightly better for age estimation, possibly due to the hidden layers in MLP38. However, there is an exception for the outer sample of women (SLP 1.45, MLP 1.49). The finding that the MLP is more accurate than the SLP in assessing age requires additional retrospective studies.
The classification performance of the DM model and the traditional method at an 18-year threshold was also compared. All tested SD models and traditional methods on the internal test set showed practically acceptable levels of discrimination for the 18-year-old sample. Sensitivity for men and women was greater than 87.7% and 94.9%, respectively, and specificity was greater than 89.3% and 84.7%. The AUROC of all tested models also exceeds 0.925. To the best of our knowledge, no study has tested the performance of the DM model for 18-year classification based on dental maturity. We can compare the results of this study with the classification performance of deep learning models on panoramic radiographs. Guo et al.15 calculated the classification performance of a CNN-based deep learning model and a manual method based on Demirjian’s method for a certain age threshold. The sensitivity and specificity of the manual method were 87.7% and 95.5%, respectively, and the sensitivity and specificity of the CNN model exceeded 89.2% and 86.6%, respectively. They concluded that deep learning models can replace or outperform manual assessment in classifying age thresholds. The results of this study showed similar classification performance; It is believed that classification using DM models can replace traditional statistical methods for age estimation. Among the models, DM LR was the best model in terms of sensitivity for the male sample and sensitivity and specificity for the female sample. LR ranks second in specificity for men. Moreover, LR is considered to be one of the more user-friendly DM35 models and is less complex and difficult to process. Based on these results, LR was considered the best cutoff classification model for 18-year-olds in the Korean population.
Overall, the accuracy of age estimation or classification performance on the external test set was poor or lower compared to the results on the internal test set. Some reports indicate that classification accuracy or efficiency decreases when age estimates based on the Korean population are applied to the Japanese population5,39, and a similar pattern was found in the present study. This deterioration trend was also observed in the DM model. Therefore, to accurately estimate age, even when using DM in the analysis process, methods derived from native population data, such as traditional methods, should be preferred5,39,40,41,42. Since it is unclear whether deep learning models can show similar trends, studies comparing classification accuracy and efficiency using traditional methods, DM models, and deep learning models on the same samples are needed to confirm whether artificial intelligence can overcome these racial disparities in limited age. assessments.
We demonstrate that traditional methods can be replaced by age estimation based on the DM model in forensic age estimation practice in Korea. We also discovered the possibility of implementing machine learning for forensic age assessment. However, there are clear limitations, such as the insufficient number of participants in this study to definitively determine the results, and the lack of previous studies to compare and confirm the results of this study. In the future, DM studies should be conducted with larger numbers of samples and more diverse populations to improve its practical applicability compared with traditional methods. To validate the feasibility of using artificial intelligence to estimate age in multiple populations, future studies are needed to compare the classification accuracy and efficiency of DM and deep learning models with traditional methods in the same samples.
The study used 2,657 orthographic photographs collected from Korean and Japanese adults aged 15 to 23 years. The Korean radiographs were divided into 900 training sets (19.42 ± 2.65 years) and 900 internal test sets (19.52 ± 2.59 years). The training set was collected at one institution (Seoul St. Mary’s Hospital), and the own test set was collected at two institutions (Seoul National University Dental Hospital and Yonsei University Dental Hospital). We also collected 857 radiographs from another population-based data (Iwate Medical University, Japan) for external testing. Radiographs of Japanese subjects (19.31 ± 2.60 years) were selected as the external test set. Data were collected retrospectively to analyze the stages of dental development on panoramic radiographs taken during dental treatment. All data collected were anonymous except for gender, date of birth and date of radiograph. Inclusion and exclusion criteria were the same as previously published studies 4 , 5 . The actual age of the sample was calculated by subtracting the date of birth from the date the radiograph was taken. The sample group was divided into nine age groups. The age and sex distributions are shown in Table 3 This study was conducted in accordance with the Declaration of Helsinki and approved by the Institutional Review Board (IRB) of Seoul St. Mary’s Hospital of the Catholic University of Korea (KC22WISI0328). Due to the retrospective design of this study, informed consent could not be obtained from all patients undergoing radiographic examination for therapeutic purposes. Seoul Korea University St. Mary’s Hospital (IRB) waived the requirement for informed consent.
Developmental stages of bimaxillary second and third molars were assessed according to Demircan criteria25. Only one tooth was selected if the same type of tooth was found on the left and right sides of each jaw. If homologous teeth on both sides were at different developmental stages, the tooth with the lower developmental stage was selected to account for uncertainty in the estimated age. One hundred randomly selected radiographs from the training set were scored by two experienced observers to test interobserver reliability after precalibration to determine dental maturity stage. Intraobserver reliability was assessed twice at three-month intervals by the primary observer.
The sex and developmental stage of the second and third molars of each jaw in the training set were estimated by a primary observer trained with different DM models, and the actual age was set as the target value. SLP and MLP models, which are widely used in machine learning, were tested against regression algorithms. The DM model combines linear functions using the developmental stages of the four teeth and combines these data to estimate age. SLP is the simplest neural network and does not contain hidden layers. SLP works based on threshold transmission between nodes. The SLP model in regression is mathematically similar to multiple linear regression. Unlike the SLP model, the MLP model has multiple hidden layers with nonlinear activation functions. Our experiments used a hidden layer with only 20 hidden nodes with nonlinear activation functions. Use gradient descent as the optimization method and MAE and RMSE as the loss function to train our machine learning model. The best obtained regression model was applied to the internal and external test sets and the age of the teeth was estimated.
A classification algorithm was developed that uses the maturity of four teeth on the training set to predict whether a sample is 18 years old or not. To build the model, we derived seven representation machine learning algorithms6,43: (1) LR, (2) KNN, (3) SVM, (4) DT, (5) RF, (6) XGBoost, and (7) MLP. LR is one of the most widely used classification algorithms44. It is a supervised learning algorithm that uses regression to predict the probability of data belonging to a certain category from 0 to 1 and classifies the data as belonging to a more likely category based on this probability; mainly used for binary classification. KNN is one of the simplest machine learning algorithms45. When given new input data, it finds k data close to the existing set and then classifies them into the class with the highest frequency. We set 3 for the number of neighbors considered (k). SVM is an algorithm that maximizes the distance between two classes by using a kernel function to expand the linear space into a non-linear space called fields46. For this model, we use bias = 1, power = 1, and gamma = 1 as hyperparameters for the polynomial kernel. DT has been applied in various fields as an algorithm for dividing an entire data set into several subgroups by representing decision rules in a tree structure47. The model is configured with a minimum number of records per node of 2 and uses the Gini index as a measure of quality. RF is an ensemble method that combines multiple DTs to improve performance using a bootstrap aggregation method that generates a weak classifier for each sample by randomly drawing samples of the same size multiple times from the original dataset48. We used 100 trees, 10 tree depths, 1 minimum node size, and Gini admixture index as node separation criteria. The classification of new data is determined by a majority vote. XGBoost is an algorithm that combines boosting techniques using a method that takes as training data the error between the actual and predicted values of the previous model and augments the error using gradients49. It is a widely used algorithm due to its good performance and resource efficiency, as well as high reliability as an overfitting correction function. The model is equipped with 400 support wheels. MLP is a neural network in which one or more perceptrons form multiple layers with one or more hidden layers between the input and output layers38. Using this, you can perform non-linear classification where when you add an input layer and get a result value, the predicted result value is compared to the actual result value and the error is propagated back. We created a hidden layer with 20 hidden neurons in each layer. Each model we developed was applied to internal and external sets to test classification performance by calculating sensitivity, specificity, PPV, NPV, and AUROC. Sensitivity is defined as the ratio of a sample estimated to be 18 years of age or older to a sample estimated to be 18 years of age or older. Specificity is the proportion of samples under 18 years of age and those estimated to be under 18 years of age.
The dental stages assessed in the training set were converted into numerical stages for statistical analysis. Multivariate linear and logistic regression were performed to develop predictive models for each sex and derive regression formulas that can be used to estimate age. We used these formulas to estimate tooth age for both internal and external test sets. Table 4 shows the regression and classification models used in this study.
Intra- and interobserver reliability was calculated using Cohen’s kappa statistic. To test the accuracy of DM and traditional regression models, we calculated MAE and RMSE using the estimated and actual ages of the internal and external test sets. These errors are commonly used to evaluate the accuracy of model predictions. The smaller the error, the higher the accuracy of the forecast24. Compare the MAE and RMSE of internal and external test sets calculated using DM and traditional regression. Classification performance of the 18-year cutoff in traditional statistics was assessed using a 2 × 2 contingency table. The calculated sensitivity, specificity, PPV, NPV, and AUROC of the test set were compared with the measured values of the DM classification model. Data are expressed as mean ± standard deviation or number (%) depending on data characteristics. Two-sided P values <0.05 were considered statistically significant. All routine statistical analyzes were performed using SAS version 9.4 (SAS Institute, Cary, NC). The DM regression model was implemented in Python using Keras50 2.2.4 backend and Tensorflow51 1.8.0 specifically for mathematical operations. The DM classification model was implemented in the Waikato Knowledge Analysis Environment and the Konstanz Information Miner (KNIME) 4.6.152 analysis platform.
The authors acknowledge that data supporting the study’s conclusions can be found in the article and supplementary materials. The datasets generated and/or analyzed during the study are available from the corresponding author on reasonable request.
Ritz-Timme, S. et al. Age assessment: state of the art to meet the specific requirements of forensic practice. internationality. J. Legal medicine. 113, 129–136 (2000).
Schmeling, A., Reisinger, W., Geserik, G., and Olze, A. Current status of forensic age assessment of living subjects for criminal prosecution purposes. Forensics. medicine. Pathology. 1, 239–246 (2005).
Pan, J. et al. A modified method for assessing the dental age of children aged 5 to 16 years in eastern China. clinical. Oral survey. 25, 3463–3474 (2021).
Lee, S.S. etc. Chronology of the development of second and third molars in Koreans and its application for forensic age assessment. internationality. J. Legal medicine. 124, 659–665 (2010).
Oh, S., Kumagai, A., Kim, S. Y. and Lee, S. S. Accuracy of age estimation and estimation of the 18-year threshold based on the maturity of second and third molars in Koreans and Japanese. PLoS ONE 17, e0271247 (2022).
Kim, JY, et al. Preoperative machine learning-based data analysis can predict sleep surgery treatment outcome in patients with OSA. the science. Report 11, 14911 (2021).
Han, M. et al. Accurate age estimation from machine learning with or without human intervention? internationality. J. Legal medicine. 136, 821–831 (2022).
Khan, S. and Shaheen, M. From Data Mining to Data Mining. J.Information. the science. https://doi.org/10.1177/01655515211030872 (2021).
Khan, S. and Shaheen, M. WisRule: The First Cognitive Algorithm for Association Rule Mining. J.Information. the science. https://doi.org/10.1177/01655515221108695 (2022).
Shaheen M. and Abdullah U. Karm: Traditional data mining based on context-based association rules. calculate. Matt. continue. 68, 3305–3322 (2021).
Muhammad M., Rehman Z., Shaheen M., Khan M. and Habib M. Deep learning based semantic similarity detection using text data. inform. technologies. control. https://doi.org/10.5755/j01.itc.49.4.27118 (2020).
Tabish, M., Tanoli, Z., and Shahin, M. A system for recognizing activity in sports videos. multimedia. Tools Applications https://doi.org/10.1007/s11042-021-10519-6 (2021).
Halabi, S.S. et al. RSNA Machine Learning Challenge in Pediatric Bone Age. Radiology 290, 498–503 (2019).
Li, Y. et al. Forensic age estimation from pelvic X-rays using deep learning. EURO. radiation. 29, 2322–2329 (2019).
Guo, Y. C., et al. Accurate age classification using manual methods and deep convolutional neural networks from orthographic projection images. internationality. J. Legal medicine. 135, 1589–1597 (2021).
Alabama Dalora et al. Bone age estimation using different machine learning methods: a systematic literature review and meta-analysis. PLoS ONE 14, e0220242 (2019).
Du, H., Li, G., Cheng, K., and Yang, J. Population-specific age estimation of African Americans and Chinese based on pulp chamber volumes of first molars using cone-beam computed tomography. internationality. J. Legal medicine. 136, 811–819 (2022).
Kim S., Lee Y.H., Noh Y.K., Park F.K. and Oh K.S. Determining age groups of living people using artificial intelligence-based images of first molars. the science. Report 11, 1073 (2021).
Stern, D., Payer, C., Giuliani, N., and Urschler, M. Automatic age estimation and majority age classification from multivariate MRI data. IEEE J. Biomed. Health Alerts. 23, 1392–1403 (2019).
Cheng, Q., Ge, Z., Du, H. and Li, G. Age estimation based on 3D pulp chamber segmentation of first molars from cone beam computed tomography by integrating deep learning and level sets. internationality. J. Legal medicine. 135, 365–373 (2021).
Wu, W.T., et al. Data mining in clinical big data: common databases, steps, and methods models. World. medicine. resource. 8, 44 (2021).
Yang, J. et al. Introduction to Medical Databases and Data Mining Technologies in the Big Data Era. J. Avid. Basic medicine. 13, 57–69 (2020).
Shen, S. et al. Camerer’s method for estimating tooth age using machine learning. BMC Oral Health 21, 641 (2021).
Galliburg A. et al. Comparison of different machine learning methods for predicting dental age using the Demirdjian staging method. internationality. J. Legal medicine. 135, 665–675 (2021).
Demirdjian, A., Goldstein, H. and Tanner, J. M. A new system for assessing dental age. snort. biology. 45, 211–227 (1973).
Landis, J. R., and Koch, G. G. Measures of observer agreement on categorical data. Biometrics 33, 159–174 (1977).
Bhattacharjee S, Prakash D, Kim C, Kim HK and Choi HK. Textural, morphological and statistical analysis of two-dimensional magnetic resonance imaging using artificial intelligence techniques for differentiation of primary brain tumors. Health information. resource. https://doi.org/10.4258/hir.2022.28.1.46 (2022).
Post time: Jan-04-2024