提升乳癌篩檢效率之批次排程框架;A Batch Scheduling Framework for Improved Diagnostic Efficiency in Breast Cancer Screening

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/95726

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/95726

Title:	提升乳癌篩檢效率之批次排程框架;A Batch Scheduling Framework for Improved Diagnostic Efficiency in Breast Cancer Screening
Authors:	本尼;Akumalla, Brahma reddy
Contributors:	資訊工程學系
Keywords:	ConvNext;變壓器;卷積層;注意力;乳房腫塊;ConvNext;transformer;convolution layers;attention;Breast mass
Date:	2024-08-01
Issue Date:	2024-10-09 17:12:28 (UTC+8)
Publisher:	國立中央大學
Abstract:	深度學習在醫學診斷中的應用，特別是對覆雜的腫塊型乳腺癌，取得了顯著進展。一種采用批量調度器進行動態批量大小調整的新型訓練策略顯示出4\%的性能提升。通過結合注意力機制和雙優化器策略，進一步提高了F1分數和性能的強健性，與傳統方法相比表現更優。此外，一種結合卷積層和變壓器模型的新型特征選擇機制在實驗中表現優於現有的注意力機制。這一創新，加上集成了預訓練權重的“ConvNext Tiny”模型，大大增強了乳腺癌檢測系統的強健性，突顯了這些方法在改進醫學診斷程序中的潛力。;The application of deep learning in medical diagnostics, particularly for complex mass-type breast cancers, has seen significant advancements. A novel training strategy that employs a batch scheduler for dynamic batch size adjustment has demonstrated a performance improvement of 4\%. Further enhancements were achieved by incorporating attention mechanisms and dual optimizer strategies, resulting in superior F1 scores and robust performance compared to traditional methods.Additionally, a novel feature selection mechanism combining convolution layers and a transformer model outperformed established attention mechanisms in experimental trials. This innovation, along with the integration of the ′ConvNext Tiny′ model with pre-trained weights, substantially enhanced the robustness of the breast cancer detection system, underscoring the potential of these methodologies for improving medical diagnostic procedures.
Appears in Collections:	[Graduate Institute of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	65	View/Open

社群 sharing

Loading...