English  |  正體中文  |  简体中文  |  全文筆數/總筆數 : 45422/58598 (78%)
造訪人次 : 2514372      線上人數 : 240
RC Version 7.0 © Powered By DSPACE, MIT. Enhanced by NTU Library IR team.
搜尋範圍 查詢小技巧:
  • 您可在西文檢索詞彙前後加上"雙引號",以獲取較精準的檢索結果
  • 若欲以作者姓名搜尋,建議至進階搜尋限定作者欄位,可獲得較完整資料
  • 進階搜尋
    請使用永久網址來引用或連結此文件: http://libir.tmu.edu.tw/handle/987654321/65005


    題名: 運用綠色深度學習技術建構非小細胞肺癌亞型病理影像低耗能分類系統
    Applying Green Deep Learning Technology to Develop a Low-Energy Consumption Classification System for Non-Small Cell Lung Cancer Subtype Pathology Images
    作者: 陳姿霓
    CHEN, TZU-NI
    貢獻者: 醫學院人工智慧醫療碩士在職專班
    林于翔
    關鍵詞: 綠色深度學習;非小細胞肺癌;分類;低耗能系統
    Green Deep Learning;Non-Small Cell Lung Cancer;Classification;Low-Energy Consumption System
    日期: 2024-07-16
    上傳時間: 2025-01-06 09:19:27 (UTC+8)
    摘要: 肺癌的發生率與死亡率逐年上升,肺癌治療指引建議依據肺癌亞型病理分類與突變基因給予個人化治療,可顯著改善肺癌病患的生存期,因此肺癌亞型病理的精準分類非常重要。近年來,深度學習輔助各種醫療影像的判讀技術蓬勃發展,並獲得了卓越的成功。然而,與其他醫學影像不同的是,一張病理影像的尺寸通常達到gigapixel等級,故需仰賴大量的人力進行標註,以及極高效能的硬體運算設備進行模型訓練。除此之外,為有效提升分類準確率,通常需要極長的模型訓練時間,此舉將會導致大量的電力能源消耗以及二氧化碳排放,將嚴重影響生態環境。有鑑於此,本研究考量生態永續與精準醫療性能之平衡,運用綠色深度學習技術,建構非小細胞肺癌亞型病理影像低耗能分類系統,並進行深度學習模型性能、能源消耗、效率比較的詳盡分析。首先,本研究提出了一種輕量化的低耗能深度學習模型GreenHisNet,在分類肺癌亞型病理影像之二分類之準確率達到了99.64%的優異表現,此外,當訓練資料量減少為30筆時,GreenHisNet之準確率與訓練時間顯著優於其他大型深度學習模型,故相當適合於低耗能的設備上,進行深度學習訓練及推理應用。本研究亦針對了5種深度學習模型,詳盡探討其不同數量的訓練資料,與對應之分類準確率關係,並透過此關係提出最佳之訓練資料數量方案,此舉可有效在維持優秀分類準確率的前提之下,大量減少訓練成本及人力標註負擔。此外,本研究還提出了一種無須標註資訊的新型分類準確率預測方法,本研究的結果顯示,在無須標註的肺癌亞型病理影像中,可以透過本研究提出之方法精準預測模型的分類準確率,透過此預測方法,未來即可根據不同醫療場景之需求,運用最適合的資料量進行模型訓練,進而達成訓練資料量優化,以及節能減碳的目標。
    The incidence and mortality rates of lung cancer are increasing annually. Treatment guidelines recommend personalized treatment based on the pathological classification of lung cancer subtypes and mutated genes, significantly improving patient survival. Recently, deep learning has been successfully applied to various medical image interpretations. However, pathological images typically reach gigapixel sizes, requiring substantial human annotation and high-performance hardware for model training. Extended training times to improve accuracy result in significant energy consumption and carbon emissions, impacting the environment.This study addresses ecological sustainability and precise medical performance using green deep learning technology to develop a low-energy classification system for non-small cell lung cancer subtype pathology images. It provides a detailed analysis of model performance, energy consumption, and efficiency. The lightweight GreenHisNet model achieves an outstanding 99.64% binary classification accuracy. With only 30 training samples, GreenHisNet outperforms other large models in accuracy and training time, making it suitable for low-energy devices.Additionally, this study examines the relationship between training data quantity and classification accuracy for five deep learning models, proposing an optimal data quantity plan to reduce costs and annotation burdens while maintaining high accuracy. A novel prediction method for classification accuracy without annotated information is also introduced. The results demonstrate that this method can accurately predict model accuracy using unannotated images. This allows for optimized data usage in different medical scenarios, achieving data quantity optimization and energy conservation goals.
    描述: 碩士
    指導教授:林于翔
    口試委員:彭徐鈞
    口試委員:蘇家玉
    口試委員:林于翔
    附註: 論文公開日期:2029-07-21
    資料類型: thesis
    顯示於類別:[人工智慧醫療碩士在職專班] 博碩士論文

    文件中的檔案:

    檔案 描述 大小格式瀏覽次數
    index.html0KbHTML0檢視/開啟


    在TMUIR中所有的資料項目都受到原著作權保護.

    TAIR相關文章

    著作權聲明 Copyright Notice
    • 本平台之數位內容為臺北醫學大學所收錄之機構典藏,包含體系內各式學術著作及學術產出。秉持開放取用的精神,提供使用者進行資料檢索、下載與取用,惟仍請適度、合理地於合法範圍內使用本平台之內容,以尊重著作權人之權益。商業上之利用,請先取得著作權人之授權。

      The digital content on this platform is part of the Taipei Medical University Institutional Repository, featuring various academic works and outputs from the institution. It offers free access to academic research and public education for non-commercial use. Please use the content appropriately and within legal boundaries to respect copyright owners' rights. For commercial use, please obtain prior authorization from the copyright owner.

    • 瀏覽或使用本平台,視同使用者已完全接受並瞭解聲明中所有規範、中華民國相關法規、一切國際網路規定及使用慣例,並不得為任何不法目的使用TMUIR。

      By utilising the platform, users are deemed to have fully accepted and understood all the regulations set out in the statement, relevant laws of the Republic of China, all international internet regulations, and usage conventions. Furthermore, users must not use TMUIR for any illegal purposes.

    • 本平台盡力防止侵害著作權人之權益。若發現本平台之數位內容有侵害著作權人權益情事者,煩請權利人通知本平台維護人員([email protected]),將立即採取移除該數位著作等補救措施。

      TMUIR is made to protect the interests of copyright owners. If you believe that any material on the website infringes copyright, please contact our staff([email protected]). We will remove the work from the repository.

    Back to Top
    DSpace Software Copyright © 2002-2004  MIT &  Hewlett-Packard  /   Enhanced by   NTU Library IR team Copyright ©   - 回饋