计算机应用研究
計算機應用研究
계산궤응용연구
APPLICATION RESEARCH OF COMPUTERS
2015年
5期
1554-1558,1575
,共6页
磁共振成像%脑肿瘤%模板匹配%自适应%检测
磁共振成像%腦腫瘤%模闆匹配%自適應%檢測
자공진성상%뇌종류%모판필배%자괄응%검측
magnetic resonance imaging%brain tumor%template marching%adaptive%detection
为快速有效地检测脑肿瘤，提出一种基于3D 自适应模板匹配算法的脑肿瘤快速检测方法。采用改进的 BET（brain extraction tool）算法从磁共振颅脑图像中提取出脑实质；再从脑实质中提取出包含所有肿瘤结构的3D 感兴趣区域，并采用圆形度等特征对这些3D 感兴趣区域进行筛选，筛选后的3D 感兴趣区域可能是脑肿瘤。以每个3D 感兴趣区域的中间层为基本层建立3D 模板，将建立的3D 模板与原图像中相应位置的3D 感兴趣区域进行匹配，根据匹配特征确定相应的阈值，将高于阈值的3D 感兴趣区域标记为肿瘤区域。为评价算法的性能，采用包含124个肿瘤（3～15 mm）的23个临床病例对该方法进行测试，利用 ROC（receiver operating character-istic）曲线对测试结果进行分析，结果显示，该方法的敏感性率为88．7097％，假阳性为16．03％。与近年来报道的模板匹配方法相比，检测性能有明显的提高。
為快速有效地檢測腦腫瘤，提齣一種基于3D 自適應模闆匹配算法的腦腫瘤快速檢測方法。採用改進的 BET（brain extraction tool）算法從磁共振顱腦圖像中提取齣腦實質；再從腦實質中提取齣包含所有腫瘤結構的3D 感興趣區域，併採用圓形度等特徵對這些3D 感興趣區域進行篩選，篩選後的3D 感興趣區域可能是腦腫瘤。以每箇3D 感興趣區域的中間層為基本層建立3D 模闆，將建立的3D 模闆與原圖像中相應位置的3D 感興趣區域進行匹配，根據匹配特徵確定相應的閾值，將高于閾值的3D 感興趣區域標記為腫瘤區域。為評價算法的性能，採用包含124箇腫瘤（3～15 mm）的23箇臨床病例對該方法進行測試，利用 ROC（receiver operating character-istic）麯線對測試結果進行分析，結果顯示，該方法的敏感性率為88．7097％，假暘性為16．03％。與近年來報道的模闆匹配方法相比，檢測性能有明顯的提高。
위쾌속유효지검측뇌종류，제출일충기우3D 자괄응모판필배산법적뇌종류쾌속검측방법。채용개진적 BET（brain extraction tool）산법종자공진로뇌도상중제취출뇌실질；재종뇌실질중제취출포함소유종류결구적3D 감흥취구역，병채용원형도등특정대저사3D 감흥취구역진행사선，사선후적3D 감흥취구역가능시뇌종류。이매개3D 감흥취구역적중간층위기본층건립3D 모판，장건립적3D 모판여원도상중상응위치적3D 감흥취구역진행필배，근거필배특정학정상응적역치，장고우역치적3D 감흥취구역표기위종류구역。위평개산법적성능，채용포함124개종류（3～15 mm）적23개림상병례대해방법진행측시，이용 ROC（receiver operating character-istic）곡선대측시결과진행분석，결과현시，해방법적민감성솔위88．7097％，가양성위16．03％。여근년래보도적모판필배방법상비，검측성능유명현적제고。
This paper presented a three-dimensional adaptive template matching algorithm to detect brain tumors from magne-tic resonance images quickly.First,it removed skull and other non-brain tissues by the improved BET algorithm.Then it extrac-ted the structures that contained all small tumors as ROIs(region of interest).After that,it screened all the ROIs by the circu-lar degree and other features.Then it created a three-dimensional template conformed to tumor characteristics for each ROI. Finally,it marched the three-dimensional templates with the original images to calculate the similarity coefficient.Then it deter-mined the threshold according to the matching characteristic.After that,it marked the three-dimensional ROI with the similarity coefficient which was higher than the threshold value as the tumor region.To evaluate the performance of the algorithm,this paper used 23 clinical cases which contained 1 24 tumors(3 ~1 5 mm)in different size to test the system,and used ROC curve to analysis the test results.According to the ROC curve,the sensibility reaches 88.7097% and the false position is 1 6.03%. Compared to other template matching methods,the algorithm has been significantly improved.
