The Dental Health Unit, School of Dentistry, The University of Manchester, M15 6SE, UK.
The Dental Health Unit, School of Dentistry, The University of Manchester, M15 6SE, UK.
J Dent. 2018 Jul;74 Suppl 1:S34-S41. doi: 10.1016/j.jdent.2018.04.021.
To assess a novel method of automatic fluorosis detection and classification from white light and fluorescent images.
Dental images from 1,729 children living in two fluoridated and two non-fluoridated UK cities were utilised. A novel detection and classification algorithm was applied to each image and TF scores were obtained using thresholding criteria. These were compared to clinical reference standard images. Comparisons between reference and automated assessments were undertaken to record correct and incorrect classifications and the ability of the system to separate the fluoridated and non-fluoridated populations.
The automated system performed well and was able to differentiate the two populations (P < 0.0001) to the same degree as the reference standard. When using the highest score from the clinical assessment the agreement between automated and clinical assessments was 0.56 (Kappa SE = 0.0160, p < 0.0001).
Assessment of dental fluorosis is typically undertaken by clinical examiners in epidemiological studies. The training and calibration of such examiners is complex and time consuming and the assessments are subject to bias - frequently because of the examiner's awareness of the water fluoridation status of subjects. The use of remote scoring using photographs has been advocated but still requires trained examiners. This study has shown that image-processing methodologies applied to white light and fluorescent images could automatically score fluorosis and statistically separate fluoridated and non-fluoridated areas. The system requires further refinement to manage confounding factors such as the presence of non-fluoride opacities and tooth stain.
评估一种从白光和荧光图像自动检测和分类氟斑牙的新方法。
利用来自英国两个氟化和两个非氟化城市的 1729 名儿童的牙齿图像。将一种新的检测和分类算法应用于每个图像，并使用阈值标准获得 TF 分数。将这些与临床参考标准图像进行比较。对参考和自动评估进行比较，以记录正确和错误的分类以及系统区分氟化和非氟化人群的能力。
自动系统性能良好，能够区分这两个人群（P<0.0001），与参考标准的程度相同。当使用临床评估中的最高分数时，自动评估和临床评估之间的一致性为 0.56（Kappa SE=0.0160，p<0.0001）。
氟斑牙的评估通常由流行病学研究中的临床检查者进行。这种检查者的培训和校准很复杂且耗时，评估容易受到偏见的影响-通常是因为检查者对研究对象的水氟化物状态有了认识。已经提倡使用照片进行远程评分，但仍需要经过培训的检查者。本研究表明，应用于白光和荧光图像的图像处理方法可以自动评分氟斑牙，并从统计学上区分氟化区和非氟化区。该系统需要进一步改进，以管理混杂因素，如非氟化物不透明和牙齿染色的存在。
