請用此 Handle URI 來引用此文件：
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/79487完整後設資料紀錄
| DC 欄位 | 值 | 語言 |
|---|---|---|
| dc.contributor.advisor | 洪士灝(Shih-Hao Hung) | |
| dc.contributor.author | Jia-Yan Lin | en |
| dc.contributor.author | 林佳彥 | zh_TW |
| dc.date.accessioned | 2022-11-23T09:01:42Z | - |
| dc.date.available | 2026-10-06 | |
| dc.date.available | 2022-11-23T09:01:42Z | - |
| dc.date.copyright | 2021-11-08 | |
| dc.date.issued | 2021 | |
| dc.date.submitted | 2021-10-08 | |
| dc.identifier.citation | Hans Pinckaers, Bram van Ginneken, and Geert Litjens. Streaming convolutional neural networks for endtoend learning with multimegapixel images. IEEE Transactions on Pattern Analysis and Machine Intelligence, page 1–1, 2021. Adam. Goode, Benjamin. Gilbert, Jan. Harkes, Drazen. Jukic, and Mahadev. Satyanarayanan. OpenSlide: A vendorneutral software foundation for digital pathology, 2013. Camelyon17 challenge. https://camelyon17.grand-challenge.org/Data/, 2017. pyspy. https://github.com/benfred/py-spy. cairo. https://www.cairographics.org/. Pantanowitz L. Farahani N, Parwani A. Whole slide imaging in pathology: advantages, limitations, and emerging perspectives, 2015. Inc Aperio Technologies. Digital slides and thirdparty data interchange. https://web.archive.org/web/20120420105738/http://www.aperio.com/documents/api/Aperio_Digital_Slides_and_Third-party_data_interchange.pdf. Libtiff. https://libtiff.gitlab.io/libtiff/. Openjpeg. http://www.openjpeg.org/. Alex Clark. Pillow (pil fork) documentation, 2015. Numpy. https://numpy.org/. Opencv. https://opencv.org/. Vlad Krasnov. On the dangers of intel’s frequency scaling. 2017. wikichip.org. Intel xeon gold 6148. https://en.wikichip.org/wiki/intel/xeon_gold/6148. pybind11. https://github.com/pybind/pybind11. tensorflow. https://github.com/tensorflow/tensorflow. Cancer Genome Atlas Research Network, J N Weinstein, E A Collisson, G B Mills,K R Shaw, B A Ozenberger, K Ellrott, I Shmulevich, C Sander, and J M Stuart. The cancer genome atlas pancancer analysis project. Nat Genet, 45(10):1113–1120, October 2013. memory_profiler. https://github.com/pythonprofilers/memory_profiler. Structural similarity index —skimage v0.19.0.dev0 docs. https://scikit-image.org/docs/dev/auto_examples/transform/plot_ssim.html. nvjpeg2000. https://docs.nvidia.com/cuda/nvjpeg2000/userguide.html. Highthroughput jpeg 2000. https://jpeg.org/jpeg2000/htj2k.html. | |
| dc.identifier.uri | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/79487 | - |
| dc.description.abstract | "病理切片是訂定癌症治療方針的重要依據。要檢查數位病理切片是否有腫瘤，對於醫生來說是一件費時費力的事情。檢查一張影像可能需要好幾分鐘的時間，而且要判斷得準確仰賴醫生多年的臨床經驗。為了解決此問題，我們使用卷積神經網路(Convolutional Neural Network, CNN) 來辨識病理切片，並且使用OpenSlide來讀取醫療影像。然而，由於病理切片影像的解析度非常高，進行CNN判讀之前的讀取與處理影像的過程佔據了大部分的時間(61.7%)，成為效能瓶頸。 在此論文中，我們提出了優化讀取醫療影像的方法。首先，我們使用py-spy來分析OpenSlide的效能，並且發現主要的效能瓶頸在於將塊狀圖片做渲染並組合起來成完整影像的運算所使用的cairo程式庫，因此我們重新實作其中組合塊狀圖片的功能，在效能上獲得顯著的改善。其次，我們發現OpenSlide的工作流程中有多餘的色彩轉換，而且在後續的處理中需要額外的資料結構轉換及資料複製，於是重新實作了一個讀圖程式，並使用x86處理機的AVX2向量指令集加速色彩空間轉換。最後，我們將程式以多執行緒平行化，達到可擴展的加速。綜合以上的效能優化，在開啟32個執行緒之下，與原版OpenSlide相比可以達到62.9倍的加速，讀取一張影像所需的時間由80.1秒降為1.27秒，因此完整的病理切片影像判讀的過程獲得2.55倍的加速。" | zh_TW |
| dc.description.provenance | Made available in DSpace on 2022-11-23T09:01:42Z (GMT). No. of bitstreams: 1 U0001-0610202113163100.pdf: 2869958 bytes, checksum: 2aee65f15ced28ebcfbd23d73c9165e5 (MD5) Previous issue date: 2021 | en |
| dc.description.tableofcontents | Acknowledgements i 摘要 ii Abstract iii Contents v List of Figures vii List of Tables viii Chapter 1 Introduction 1 Chapter 2 Background 4 2.1 WholeSlide Images 4 2.2 The SVS Image Format 4 2.3 OpenSlide 5 Chapter 3 Methodology 8 3.1 Identifying the Performance Bottleneck 8 3.1.1 Analyzing OpenSlide with pyspy 9 3.1.2 Modifying OpenSlide By Remove Rendering 10 3.2 Improving Workflow 10 3.2.1 Improving Color Space Conversion 11 3.2.2 Reducing Data Copies 12 3.3 Parallelization 13 Chapter 4 Evaluation 15 4.1 Experimental Setup 15 4.2 Effects Of Optimization 16 4.3 Image Similarity 18 4.4 Parallel Speedup 18 4.5 The Effects on CNN inferences 21 4.5.1 Image Preprocessing 22 4.5.2 EndtoEnd CNN Inferences 22 Chapter 5 Conclusion and Future Work 24 References 26 | |
| dc.language.iso | en | |
| dc.title | 超解析度醫療影像之卷積神經網路前處理流程的效能優化 | zh_TW |
| dc.title | Performance Optimization of Convolutional Neural Network Preprocessing Workflow for Super-Resolution Medical Images | en |
| dc.date.schoolyear | 109-2 | |
| dc.description.degree | 碩士 | |
| dc.contributor.oralexamcommittee | 逄愛君(Hsin-Tsai Liu),梁文耀(Chih-Yang Tseng),葉肇元,郭大維 | |
| dc.subject.keyword | 醫療影像前處理,平行運算,OpenSlide,卷積神經網路,醫療影像辨識, | zh_TW |
| dc.subject.keyword | medical image preprocessing,parallel computing,OpenSlide,convolutional neural network,Medical image classification, | en |
| dc.relation.page | 27 | |
| dc.identifier.doi | 10.6342/NTU202103580 | |
| dc.rights.note | 同意授權(全球公開) | |
| dc.date.accepted | 2021-10-12 | |
| dc.contributor.author-college | 電機資訊學院 | zh_TW |
| dc.contributor.author-dept | 資訊工程學研究所 | zh_TW |
| dc.date.embargo-lift | 2026-10-06 | - |
| 顯示於系所單位： | 資訊工程學系 | |
文件中的檔案：
| 檔案 | 大小 | 格式 | |
|---|---|---|---|
| U0001-0610202113163100.pdf 此日期後於網路公開 2026-10-06 | 2.8 MB | Adobe PDF |
系統中的文件，除了特別指名其著作權條款之外，均受到著作權保護，並且保留所有的權利。