基於影像處理MRI子宮肌瘤影像自動檢測和分割以進行HIFU刀手術規劃;Automatic detection and segmentation of MRI uterine fibroid images based on image processing for HIFU surgery planning

NCUIR > College of Electrical Engineering & Computer Science > Executive Master of Communication Engineering > Electronic Thesis & Dissertation > Item 987654321/92908

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/92908

Title:	基於影像處理MRI子宮肌瘤影像自動檢測和分割以進行HIFU刀手術規劃;Automatic detection and segmentation of MRI uterine fibroid images based on image processing for HIFU surgery planning
Authors:	曾冠慈;Tseng, Kuan-Tzu
Contributors:	通訊工程學系在職專班
Keywords:	影像處理;子宮肌瘤;MRI;HIFU;image processing;uterine fibroids;MRI;HIFU
Date:	2023-08-15
Issue Date:	2024-09-19 16:27:40 (UTC+8)
Publisher:	國立中央大學
Abstract:	由MRI影像準確定位出子宮肌瘤對於高強度聚焦超聲 (HIFU) 治療至關重要，但由於個體間子宮肌瘤形狀及大小有巨大的差異、相鄰器官跟組織的對比度較低以及子宮肌瘤的數目不詳，因此要準確提取出子宮肌瘤位置及大小並不容易，目前主要還是以醫師花時間依靠經驗判斷並使用軟體手工標註出子宮肌瘤的部分。為了解決這個問題，與長庚醫院林吉晉醫師團隊合作，提供去識別化之腹腔MRI影像(IRB案號：202201307B0C502)，提出了一套影像處理方法。首先使用4種不同的影像處理組合方式增加MRI影像的器官組織對比度，獲得4種不同差異化的影像，使用這些影像分別進行K均值分類，再依子宮肌瘤影像特徵由這些分類組合尋找包含子宮肌瘤分類以及非子宮肌瘤的其他組織器官分類，最後經過平滑處理、去除噪點、分離及判斷子宮肌瘤等步驟獲取最終結果。此方法長庚醫院林吉晉醫師與林育駿放射師團隊傳統手工標註子宮肌瘤進行比較，且使用30個案例進行驗證。 ;Accurate localization of uterine fibroids from MRI images is crucial for High-Intensity Focused Ultrasound (HIFU) treatment. However, due to significant variations in the shape and size of uterine fibroids among individuals, low contrast with adjacent organs and tissues, and the unknown number of fibroids, it is challenging to precisely extract the position and size of uterine fibroids. Currently, the main approach relies on physicians spending time and using their experience to manually label the uterine fibroid regions in software. To address this issue, we collaborated with Dr. Lin Gi-Gin′s team from Chang Gung Hospital and provided de-identified abdominal MRI images (IRB number: 202201307B0C502) to propose an image processing method. Firstly, four different combinations of image processing techniques were used to enhance the organ tissue contrast in the MRI images, resulting in four differentiated images. These images were then subjected to K-means clustering individually. Based on the characteristics of uterine fibroid images, we identified the clustering combination that includes the classification of uterine fibroids and other organ tissues as non-uterine fibroids. Finally, a series of post-processing steps including smoothing, noise removal, separation, and fibroid determination were applied to obtain the ultimate results. This method was compared to the traditional manual labeling of uterine fibroids by Dr. Lin Gi-Gin and the radiology team, and it was validated using 30 cases.
Appears in Collections:	[Executive Master of Communication Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	35	View/Open

社群 sharing

Loading...