Ossowska Agata, Kusiak Aida, Świetlik Dariusz
Department of Periodontology and Oral Mucosa Diseases, Medical University of Gdansk, Orzeszkowej 18 St., 80-208 Gdansk, Poland.
Division of Biostatistics and Neural Networks, Medical University of Gdansk, Debinki 1 St., 80-211 Gdansk, Poland.
J Clin Med. 2022 Aug 10;11(16):4667. doi: 10.3390/jcm11164667.
Periodontitis is an inflammatory disease of the tissues surrounding the tooth that results in loss of periodontal attachment detected as clinical attachment loss (CAL). The mildest form of periodontal disease is gingivitis, which is a necessary condition for periodontitis development. We can distinguish also some modifying factors which have an influence on the rate of development of periodontitis from which the most important are smoking and poorly controlled diabetes. According to the new classification from 2017, we can identify four stages of periodontitis and three grades of periodontitis. Grades tell us about the periodontitis progression risk and may be helpful in treatment planning and motivating the patients. Artificial neural networks (ANN) are widely used in medicine and in dentistry as an additional tool to support clinicians in their work. In this paper, ANN was used to assess grades of periodontitis in the group of patients. Gender, age, nicotinism approximal plaque index (API), bleeding on probing (BoP), clinical attachment loss (CAL), and pocket depth (PD) were taken into consideration. There were no statistically significant differences in the clinical periodontal assessment in relation to the neural network assessment. Based on the definition of the sensitivity and specificity in medicine we obtained 85.7% and 80.0% as a correctly diagnosed and excluded disease, respectively. The quality of the neural network, defined as the percentage of correctly classified patients according to the grade of periodontitis was 84.2% for the training set. The percentage of incorrectly classified patients according to the grade of periodontitis was 15.8%. Artificial neural networks may be useful tool in everyday dental practice to assess the risk of periodontitis development however more studies are needed.
牙周炎是牙齿周围组织的一种炎症性疾病，会导致牙周附着丧失，表现为临床附着丧失（CAL）。牙周疾病最轻微的形式是牙龈炎，它是牙周炎发展的必要条件。我们还可以区分一些影响牙周炎发展速度的调节因素，其中最重要的是吸烟和控制不佳的糖尿病。根据2017年的新分类，我们可以识别牙周炎的四个阶段和三个等级。等级能告诉我们牙周炎的进展风险，可能有助于治疗计划制定和激励患者。人工神经网络（ANN）在医学和牙科中被广泛用作辅助临床医生工作的工具。在本文中，人工神经网络被用于评估一组患者的牙周炎等级。考虑了性别、年龄、烟瘾、邻面菌斑指数（API）、探诊出血（BoP）、临床附着丧失（CAL）和牙周袋深度（PD）。临床牙周评估与神经网络评估之间没有统计学上的显著差异。根据医学中敏感性和特异性的定义，我们分别得到85.7%和80.0%的正确诊断疾病和排除疾病的比例。神经网络质量，即根据牙周炎等级正确分类患者的百分比，训练集为84.2%；根据牙周炎等级错误分类患者的百分比为15.8%。人工神经网络可能是日常牙科实践中评估牙周炎发展风险的有用工具，但还需要更多研究。
