Department of Dentistry, State University of Paraiba, Rua Baraúnas, 351, Bairro Universitário, Campina Grande, Paraíba, Brasil.
Rua Aprígio Veloso, Federal University of Campina Grande, RuAprígio Veloso, 882, Bairro Universitário, Campina Grande, Paraíba, Brazil.
Dentomaxillofac Radiol. 2023 Apr;52(4):20220363. doi: 10.1259/dmfr.20220363. Epub 2023 Mar 29.
This study aimed to assess and compare age estimation on panoramic radiography using the Kvaal method and machine learning (ML).
554 panoramic radiographs were selected from a Brazilian practice. To estimate age using the Kvaal method, the following measurements were performed on the upper left central incisors and canines: tooth, pulp and root length; root and pulp width at three different levels: at the enamel-cementum junction (ECJ); midpoint between the enamel-cementum junction and; at the mid root level. For ML age estimation, radiomic, semantic and the radiomic-semantic attribute extractions were assessed. Nineteen semantic and 14 radiomic attributes and a single set of 33 semantic-radiomic attributes were extracted. Logistic Regression, Linear Regression, KNN, SVR, Decision Tree Reg, Random Forest Reg, Gradient Boost Reg e XG Boosting Reg were used for ML classification. For the Kvaal method, Mann-Whitney test, Spearman correlation coefficient, Student's -test and linear regression with its respective coefficient of determination were used to estimate age and to assess data variability.
Mean absolute error (MAE) and standard error estimate (SEE) were assessed. For the Kvaal method, upper incisors presented higher precision than canines (R²: 0.335, SSE: 7.108). Males presented better MAE and SEE values (5.29,6.96) than females (5.69,7.37). The radiomic-semantic attributes presented superior precision (MAE: 4.77) than the radiomic and semantic (MAE: 5.23) attributes. The XG Boosting Reg classifier performed better than the other six assessed classifiers (MAE: 4.65). ML (MAE: 4.77 presented higher age estimation precision than the Kvaal method (MAE: 5.68).
The use of ML on panoramic radiographs can improve age estimation.
本研究旨在评估和比较使用 Kvaal 法和机器学习（ML）对全景片进行年龄估计的方法。
从巴西的一家诊所中选择了 554 张全景片。为了使用 Kvaal 法估计年龄，在上颌左中切牙和侧切牙上进行了以下测量：牙齿、牙髓和根长；在釉牙骨质界（ECJ）、釉牙骨质界和根尖中点之间的三个不同水平处的牙髓和根宽；在釉牙骨质界和根尖中点之间的三个不同水平处的牙髓和根宽。对于 ML 年龄估计，评估了放射组学、语义和放射组学语义属性提取。提取了 19 个语义属性和 14 个放射组学属性以及一组 33 个语义放射组学属性。逻辑回归、线性回归、KNN、SVR、决策树回归、随机森林回归、梯度提升回归和 XGBoosting 回归用于 ML 分类。对于 Kvaal 法，使用曼-惠特尼检验、Spearman 相关系数、学生 t 检验和线性回归及其相应的决定系数来估计年龄并评估数据变异性。
评估了平均绝对误差（MAE）和标准误差估计（SEE）。对于 Kvaal 法，上颌切牙比侧切牙具有更高的精度（R²：0.335，SSE：7.108）。男性的 MAE 和 SEE 值（5.29，6.96）优于女性（5.69，7.37）。放射组学语义属性的精度优于放射组学和语义属性（MAE：5.23）（MAE：4.77）。XG Boosting 回归分类器的性能优于其他六种评估的分类器（MAE：4.65）。ML（MAE：4.77）比 Kvaal 法（MAE：5.68）的年龄估计精度更高。
在全景片中使用 ML 可以提高年龄估计的精度。
