Application of data mining technology to diagnostic imaging support of dementia using FDG-PET.

数据挖掘技术在 FDG-PET 痴呆症诊断成像支持中的应用。

• 批准号: 23591818
• 负责人: ISHIZU Koichi
• 金额: $ 3.33万
• 依托单位: Kyoto University
• 依托单位国家: 日本
• 项目类别: Grant-in-Aid for Scientific Research (C)
• 财政年份: 2011
• 资助国家: 日本
• 起止时间: 2011 至 2013
• 项目状态: 已结题
• 来源: https://kaken.nii.ac.jp/grant/KAKENHI-PROJECT-23591818/
• 关键词: FDG-PET; 早期認知症; 多変量解析; データマイニング; 診断支援; 予測分析; SVM; k-index法; 認知症; 画像診断支援; クラス判別; alzheimer; MCI; multivariate analysis; classifier; decision making; k-index method

The purpose of the present work was preliminary construction of diagnosis support system of dementia using FDG-PET image. FDG-PET images of Japanese 23 healthy volunteers and 58 MCI patients were used. 116 ROIs were automatically placed on each FDG image and region to whole brain uptake ratio was used for 2-class clasification. Support Vector Machine and k-index method, the reserch leader original, were used for the classification and area under the curve of ROC analysis was 71.1% and 74.3% respectively. It was seemed to be good results with pictures only consider the low accuracy of clinical diagnosis between MCI and normal. The inverse correlation was observed in FDG uptake in the cerebral cortex and the patient's body weight in the brain FDG-PET images of cancer a number of patients.

本工作的目的是利用FDG-PET图像初步构建痴呆症诊断支持系统。使用日本 23 名健康志愿者和 58 名 MCI 患者的 FDG-PET 图像。 116 个 ROI 自动放置在每个 FDG 图像上，并且区域与全脑摄取比率用于 2 类分类。采用研究领军原创的支持向量机和k指数方法进行分类，ROC分析曲线下面积分别为71.1%和74.3%。仅考虑MCI与正常人临床诊断准确率较低的情况，用图片看来似乎是不错的结果。在许多癌症患者的大脑 FDG-PET 图像中观察到大脑皮层中 FDG 的摄取量与患者体重呈负相关。