| 王之龙,高云,唐磊,孙应实,曹崑,张晓鹏.人工神经网络模型基于胃癌生物学行为的MSCT影像信息判断淋巴结转移[J].中国医学影像技术,2011,27(6):1218~1222 |
| 人工神经网络模型基于胃癌生物学行为的MSCT影像信息判断淋巴结转移 |
| Artificial neural networks model for prediction of lymph node metastases in gastric cancer basing on biological behavior information of MSCT |
| 投稿时间：2010-11-20 修订日期：2011-02-22 |
| DOI： |
| 中文关键词: 胃肿瘤 淋巴结转移 体层摄影术,X线计算机 人工神经网络 |
| 英文关键词:Stomach neoplasms Lymph node metastasis Tomography, X-ray computed Artificial neural networks |
| 基金项目:国家自然科学基金(30970825)、北京市自然科学基金(7092020)。 |
| 作者 | 单位 | E-mail | | 王之龙 | 北京大学临床肿瘤学院、北京肿瘤医院暨北京市肿瘤防治研究所医学影像科,恶性肿瘤发病机制及转化研究教育部重点实验室,北京 100142 | | | 高云 | 北京大学临床肿瘤学院、北京肿瘤医院暨北京市肿瘤防治研究所医学影像科,恶性肿瘤发病机制及转化研究教育部重点实验室,北京 100142 | | | 唐磊 | 北京大学临床肿瘤学院、北京肿瘤医院暨北京市肿瘤防治研究所医学影像科,恶性肿瘤发病机制及转化研究教育部重点实验室,北京 100142 | | | 孙应实 | 北京大学临床肿瘤学院、北京肿瘤医院暨北京市肿瘤防治研究所医学影像科,恶性肿瘤发病机制及转化研究教育部重点实验室,北京 100142 | | | 曹崑 | 北京大学临床肿瘤学院、北京肿瘤医院暨北京市肿瘤防治研究所医学影像科,恶性肿瘤发病机制及转化研究教育部重点实验室,北京 100142 | | | 张晓鹏 | 北京大学临床肿瘤学院、北京肿瘤医院暨北京市肿瘤防治研究所医学影像科,恶性肿瘤发病机制及转化研究教育部重点实验室,北京 100142 | zxp@bjcancer.org |
|
| 摘要点击次数: 2630 |
| 全文下载次数: 1096 |
| 中文摘要: |
| 目的 应用人工神经网络(ANN)分析反映胃癌生物学行为的MSCT影像学信息,建立胃癌淋巴结转移判断模型。方法 收集经手术病理证实的胃癌患者175例,患者术前均接受MSCT检查,术前未接受抗肿瘤治疗,未发现远处转移。根据手术病理是否存在淋巴结转移,分为有淋巴结转移组与无淋巴结转移组。观察测量治疗前MSCT显示的反映胃癌生物学行为的癌肿、淋巴结及临床相关指标。以统计学单因素分析筛选两组间有统计学差异的指标,进一步建立ANN和Logistic回归分析模型判断胃癌淋巴结转移。结果 175例胃癌患者中,手术病理证实共134例存在淋巴结转移,41例无淋巴结转移。单因素分析显示癌肿浆膜浸润、大体类型、最大径线、厚度、强化方式、淋巴结数目、分站、最大淋巴结短径共8项指标在有、无淋巴结转移组之间差异有统计学意义。将其作为输入指标建模,ANN模型判断淋巴结转移的总敏感度、总特异度和总准确率分别为90.30%(121/134)、82.93%(34/41)和88.57%(155/175),而Logistic回归判断淋巴结转移的总敏感度、总特异度和总准确率为85.82%(115/134)、70.73%(29/41)和82.29%(144/175)。结论 采用ANN模型,利用MSCT反映的胃癌生物学行为相关信息,可帮助术前判断患者是否存在淋巴结转移,其效能优于Logistic回归分析。 |
| 英文摘要: |
| Objective To analyze MSCT information based on the gastric cancer biological behavior using artificial neural networks (ANN) model, and to establish diagnostic model of lymph node metastases of gastric cancer. Methods Totally 175 consecutive patients with gastric cancer proved by postoperative pathology and underwent MSCT examination before surgery were included. No chemotherapy nor radiation therapy was performed before surgery. Distant metastasis was not found in the preoperative examination. All patients were divided into two groups according to the existence of lymph node metastasis on pathology. Some indicators of tumor, lymph node and clinical information which reflected the biological behavior of gastric cancer were evaluated. Artificial neural network (ANN) and logistic regression diagnostic models were established with the indicators which were significantly different between the two groups at the univariate analysis. Results There were 134 patients with lymph node metastases. Statistical differences were found in 8 indexes (serosal infiltration, tumor classification, maximum diameter, thickness, enhancement patterns, the number of lymph node, sub-station lymph nodes, maximum short diameter) between the two groups. Model was established based on the indexes. The sensitivity, specificity and accuracy of lymph node metastases with ANN was 90.30% (121/134), 82.93% (34/41) and 88.57% (155/175), respectively, whereas of logistic regression model was 85.82% (115/134), 70.73% (29/41) and 82.29% (144/175), respectively. Conclusion Based on biological behavior information of gastric cancer on MSCT, ANN model can help to diagnose lymph node metastases preoperatively. The efficacy of ANN is better than that of Logistic regression analysis. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
