基於幾何深度學習之乳癌淋巴結轉移分類系統;A Breast Cancer Lymph Node Metastasis Classification System Based on Geometric Deep Learning

NCUIR > college of Health Sciences and Technology > Institute of Biomedical Engineering > Electronic Thesis & Dissertation > Item 987654321/96126

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/96126

Title:	基於幾何深度學習之乳癌淋巴結轉移分類系統;A Breast Cancer Lymph Node Metastasis Classification System Based on Geometric Deep Learning
Authors:	李昕瑜;Lee, Hsin-Yu
Contributors:	生物醫學工程研究所
Keywords:	乳癌診斷;電腦輔助診斷;深度學習;病理影像分析
Date:	2025-01-22
Issue Date:	2025-04-09 16:11:57 (UTC+8)
Publisher:	國立中央大學
Abstract:	乳癌被認為是造成女性死亡的重要原因之一，其預後與前哨淋巴結轉移狀態有高度相關，因此早期診斷可以有效提高存活率。在乳癌診斷中，通常使用組織病理學影像作為診斷的黃金標準，但乳癌組織病理學影像的分類需要由病理學家進行，這種繁瑣的檢查過程非常耗時，並且可能漏掉體形較小的轉移細胞。為了提高乳癌診斷的準確性和效率，本研究提出了一套結合幾何快速資料密度泛函轉化(g-fDDFT)與深度學習的自動分類系統，對乳癌前哨淋巴結組織影像分類。資料方面使用了CAMELYON17的病理影像開放資料庫。方法上，首先使用g-fDDFT萃取特徵，接著使用水平集方法分割出細胞核，再使用深度學習模型對轉移以及非轉移的影像進行分類。成果方面，本系統成功將處理時間從12720秒縮減至2059秒，提升效率達83.8%。在分類表現方面，F1 score達到75%，較傳統方法提升13%，但是準確度下降13%顯示出未來改進的方向。本研究證實g-fDDFT能有效處理大型病理影像，為病理診斷自動化提供新的解決方案。雖然在分類準確度方面仍有改進空間，但系統展現出的效率優勢，為未來醫學影像分析的發展開創了新的方向。;This study proposes an automated classification system combining geometric fast data density functional transformation (g-fDDFT) with deep learning for breast cancer sentinel lymph node tissue image classification. Breast cancer is considered one of the leading causes of death among women, and its prognosis is highly correlated with sentinel lymph node metastasis status, making early diagnosis crucial for improving survival rates. In breast cancer diagnosis, histopathological images are typically used as the gold standard for diagnosis. However, the classification of breast cancer histopathological images needs to be performed by pathologists, making this meticulous examination process time-consuming and potentially missing smaller metastatic cells. The CAMELYON17 pathology image open database was used for data. Methodologically, g-fDDFT was first used to extract features, followed by the level set method to segment cell nuclei, and then deep learning models were used to classify metastatic and non-metastatic images. In terms of results, the system successfully reduced processing time from 12,720 seconds to 2,059 seconds, improving efficiency by 83.8%. Regarding classification performance, F1 score reached 75%, a 15% improvement over traditional methods, though accuracy decreased by 13%, indicating areas for future improvement. This study demonstrates that g-fDDFT can effectively process large pathological images, providing a new solution for pathological diagnosis automation. While there is still room for improvement in classification accuracy, the system′s demonstrated efficiency advantages pioneer new directions for the future development of medical image analysis.
Appears in Collections:	[Institute of Biomedical Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	55	View/Open

社群 sharing

Loading...