WEKO3
アイテム
腹部CT像上の複数オブジェクトのセグメンテーションのための統計的手法に関する研究
https://doi.org/10.18997/00008906
https://doi.org/10.18997/00008906e77ad3c8-a67e-4450-a569-0a15b20b3549
| 名前 / ファイル | ライセンス | アクション |
|---|---|---|
kou_k_546.pdf (4.5 MB)
|
|
| アイテムタイプ | 学位論文 = Thesis or Dissertation(1) | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| 公開日 | 2022-06-10 | |||||||||
| 資源タイプ | ||||||||||
| 資源タイプ識別子 | http://purl.org/coar/resource_type/c_db06 | |||||||||
| 資源タイプ | doctoral thesis | |||||||||
| タイトル | ||||||||||
| タイトル | Study on Statistical Method for Segmentation of Multi-object on Abdominal CT Images | |||||||||
| 言語 | en | |||||||||
| タイトル | ||||||||||
| タイトル | 腹部CT像上の複数オブジェクトのセグメンテーションのための統計的手法に関する研究 | |||||||||
| 言語 | ja | |||||||||
| 言語 | ||||||||||
| 言語 | eng | |||||||||
| 著者 |
呉, 佳奇
× 呉, 佳奇
|
|||||||||
| 抄録 | ||||||||||
| 内容記述タイプ | Abstract | |||||||||
| 内容記述 | Computer aided diagnosis （CAD） is the use of a computer-generated output as an auxiliary tool for the assistance of efficient interpretation and accurate diagnosis. Medical image segmentation has an essential role in CAD in clinical applications. Generally, the task of medical image segmentation involves multiple objects, such as organs or diffused tumor regions. Moreover, it is very unfavorable to segment these regions from abdominal Computed Tomography （CT） images because of the overlap in intensity and variability in position and shape of soft tissues. In this thesis, a progressive segmentation framework is proposed to extract liver and tumor regions from CT images more efficiently, which includes the steps of multiple organs coarse segmentation, fine segmentation, and liver tumors segmentation. Benefit from the previous knowledge of the shape and its deformation, the Statistical shape model （SSM） method is firstly utilized to segment multiple organs regions robustly. In the process of building an SSM, the correspondence of landmarks is crucial to the quality of the model. To generate a more representative prototype of organ surface, a k-mean clustering method is proposed. The quality of the SSMs, which is measured by generalization ability, specificity, and compactness, was improved. We furtherly extend the shapes correspondence to multiple objects. A non-rigid iterative closest point surface registration process is proposed to seek more properly corresponded landmarks across the multi-organ surfaces. The accuracy of surface registration was improved as well as the model quality. Moreover, to localize the abdominal organs simultaneously, we proposed a random forest regressor cooperating intensity features to predict the position of multiple organs in the CT image. The regions of the organs are substantially restrained using the trained shape models. The accuracy of coarse segmentation using SSMs was increased by the initial information of organ positions. Consequently, a pixel-wise segmentation using the classification of supervoxels is applied for the fine segmentation of multiple organs. The intensity and spatial features are extracted from each supervoxels and classified by a trained random forest. The boundary of the supervoxels is closer to the real organs than the previous coarse segmentation. Finally, we developed a hybrid framework for liver tumor segmentation in multiphase images. To deal with these issues of distinguishing and delineating tumor regions and peripheral tissues, this task is accomplished in two steps: a cascade region-based convolutional neural network （R-CNN） with a refined head is trained to locate the bounding boxes that contain tumors, and a phase-sensitive noise filtering is introduced to refine the following segmentation of tumor regions conducted by a level-set-based framework. The results of tumor detection show the adjacent tumors are successfully separated by the improved cascaded R-CNN. The accuracy of tumor segmentation is also improved by our proposed method. 26 cases of multi-phase CT images were used to validate our proposed method for the segmentation of liver tumors. The average precision and recall rates for tumor detection are 76.8% and 84.4%, respectively. The intersection over union, true positive rate, and false positive rate for tumor segmentation are 72.7%, 76.2%, and 4.75%, respectively. | |||||||||
| 言語 | en | |||||||||
| 目次 | ||||||||||
| 内容記述タイプ | TableOfContents | |||||||||
| 内容記述 | 1 Introduction||2 Literature Review||3 Statistical Shape Model Building||4 Multi-organ Segmentation||5 Liver Tumors Segmentation||6 Summary and Outlook | |||||||||
| 備考 | ||||||||||
| 内容記述タイプ | Other | |||||||||
| 内容記述 | 九州工業大学博士学位論文 学位記番号: 工博甲第546号 学位授与年月日: 令和4年3月25日 | |||||||||
| キーワード | ||||||||||
| 主題Scheme | Other | |||||||||
| 主題 | Computer tomography | |||||||||
| キーワード | ||||||||||
| 主題Scheme | Other | |||||||||
| 主題 | Computer-aided diagnosis | |||||||||
| キーワード | ||||||||||
| 主題Scheme | Other | |||||||||
| 主題 | Tumor detection | |||||||||
| キーワード | ||||||||||
| 主題Scheme | Other | |||||||||
| 主題 | Image segmentation | |||||||||
| キーワード | ||||||||||
| 主題Scheme | Other | |||||||||
| 主題 | Statistical shape model | |||||||||
| キーワード | ||||||||||
| 主題Scheme | Other | |||||||||
| 主題 | Deep learning | |||||||||
| アドバイザー | ||||||||||
| 神谷, 亨 | ||||||||||
| 学位授与番号 | ||||||||||
| 学位授与番号 | 甲第546号 | |||||||||
| 学位名 | ||||||||||
| 学位名 | 博士（工学） | |||||||||
| 学位授与年月日 | ||||||||||
| 学位授与年月日 | 2022-03-25 | |||||||||
| 学位授与機関 | ||||||||||
| 学位授与機関識別子Scheme | kakenhi | |||||||||
| 学位授与機関識別子 | 17104 | |||||||||
| 学位授与機関名 | 九州工業大学 | |||||||||
| 学位授与年度 | ||||||||||
| 内容記述タイプ | Other | |||||||||
| 内容記述 | 令和3年度 | |||||||||
| 出版タイプ | ||||||||||
| 出版タイプ | VoR | |||||||||
| 出版タイプResource | http://purl.org/coar/version/c_970fb48d4fbd8a85 | |||||||||
| アクセス権 | ||||||||||
| アクセス権 | open access | |||||||||
| アクセス権URI | http://purl.org/coar/access_right/c_abf2 | |||||||||
| ID登録 | ||||||||||
| ID登録 | 10.18997/00008906 | |||||||||
| ID登録タイプ | JaLC | |||||||||
