口腔医学研究 ›› 2023, Vol. 39 ›› Issue (3): 211-216.DOI: 10.13701/j.cnki.kqyxyj.2023.03.006

• 口腔影像学研究 • 上一篇    下一篇

CBCT影像组学在颌骨成釉细胞瘤术前诊断中的应用研究

聂根定1,2, 梁燕2,3, 王立冬1,2, 马文1,2*, 黎明1,2*   

  1. 1.昆明医科大学口腔医学院/医院口腔颌面外科 云南 昆明 650106;
    2.云南省口腔医学重点实验室 云南 昆明 650106;
    3.昆明医科大学口腔医学院/医院教学管理办公室 云南 昆明 650106
  • 收稿日期:2022-09-13 出版日期:2023-03-28 发布日期:2023-03-21
  • 通讯作者: * 马文,E-mail:421991950@qq.com;黎明,E-mail:1020513890@qq.com
  • 作者简介:聂根定(1996~ ),男,云南腾冲人,硕士在读,住院医师,主要从事口腔颌面外科学研究。
  • 基金资助:
    云南省“高层次人才培养支持计划”(编号:YNWR-MY-2020-086);云南省昆医联合项目-面上项目(编号:202001AY070001-250);昆明医科大学校级教研教改项目(编号:2020-JY-Y-041、2022-JY-Y-149)

Application of CBCT Radiomics in Preoperative Diagnosis of Ameloblastoma of the Jaw

NIE Gending1,2, LIANG Yan2,3, WANG Lidong1,2, MA Wen1,2*, LI Ming1,2*   

  1. 1. Department of Oral and Maxillofacial Surgery, Kunming Medical University School and Hospital of Stomatology, Kunming 650106, China;
    2. Yunnan Key Laboratory of Stomatology, Kunming 650106, China;
    3. Teaching Management Office, Kunming Medical University School and Hospital of Stomatology, Kunming 650106, China
  • Received:2022-09-13 Online:2023-03-28 Published:2023-03-21

摘要: 目的: 探讨CBCT影像组学特征在颌骨成釉细胞瘤术前诊断中的应用价值。方法: 回顾性分析104例经病理学诊断的颌骨囊性病变的患者CBCT资料(包括成釉细胞瘤45例、牙源性颌骨囊肿59例)。勾画病变区域并提取影像组学特征,通过特征筛选建立影像组学标签,构建支持向量机、随机森林和逻辑回归分类器模型;结合常规影像学特征建立综合模型。分别用训练集、验证集数据进行训练和评价,以曲线下面积(AUC值)、准确率(ACC)评价模型的诊断性能。结果: 利用影像组学特征构建的3种模型在测试集中的准确率均为81.3%,AUC值分别为0.849(95%CI:0.707~0.991)、0.865(95%CI:0.734~0.996)和0.849(95%CI:0.703~0.995),结合常规影像学特征后的准确率分别为81.3%,81.3%和84.4%,AUC分别为0.877(95%CI:0.751~1.000)、0.873(95%CI:0.747~0.999)和0.889(95%CI:0.765~1.000)。无论在3种模型之间或影像组学模型与综合模型之间均无统计学意义。结论: 基于CBCT影像组学特征构建的预测模型在成釉细胞瘤术前诊断中具有较高的诊断性能,可用于辅助诊断成釉细胞瘤,指导治疗计划的选择。

关键词: 成釉细胞瘤, 牙源性颌骨囊肿, 影像组学

Abstract: Objective: To investigate the application value of CBCT radiomics features in preoperative diagnosis of ameloblastoma of the jaw. Methods: A retrospective analysis of the CBCT data of 104 patients pathologically diagnosed as cystic lesions of the jaw was conducted, including 45 cases of ameloblastoma and 59 cases of odontogenic cyst. The radiomics features of the lesion area were extracted by manually drawing lesion areas. Radiomics labels were established by feature screening, and support vector machine, random forest, and logistic regression classifier models were constructed. A comprehensive model was established by combining conventional radiologic characteristics. The data of the training set and validation set were used for training and evaluating, and the area under the curve (AUC) and accuracy rate (ACC) were used to evaluate the diagnostic performance of the model. Results: In the testing set, the accuracy of the three models based on radiomics features was all 81.3%, and the AUC values of the radiomics models were 0.849 (95%CI:0.707-0.991), 0.865 (95%CI:0.734-0.996), and 0.849 (95%CI:0.703-0.995), respectively. Combined with conventional radiologic features, the accuracy was 81.3%, 81.3%, and 84.4%, and the AUC values of the comprehensive models were 0.877 (95%CI:0.751-1.000), 0.873 (95%CI:0.747-0.999), and 0.889 (95%CI:0.765-1.000), respectively. However, neither between three models nor the radiomics and comprehensive models had a significant difference (P>0.05). Conclusion: The prediction model based on CBCT radiomics features has high diagnostic performance in preoperative diagnosis of ameloblastoma, which can be used to assist diagnosis of ameloblastoma and guide the selection of a treatment plan.

Key words: ameloblastoma, odontogenic cyst, radiomics