研究生: |
林柏元 Bo-Yuan Lin |
---|---|
論文名稱: |
弱監督學習結合隨機注意力、修改後全卷積網路和變量多實例學習方法在婦科癌症中的應用 Weakly Supervised Learning Incorporating Shuffle Attention, Modified-FCN, and VarMIL Approaches in application to Gynecological Cancer |
指導教授: |
王靖維
Ching-Wei Wang |
口試委員: |
王靖維
Ching-Wei Wang 許昕 Hsin Hsiu 許維君 Wei-Chun Hsu 趙載光 Tai-Kuang Chao |
學位類別: |
碩士 Master |
系所名稱: |
應用科技學院 - 醫學工程研究所 Graduate Institute of Biomedical Engineering |
論文出版年: | 2023 |
畢業學年度: | 111 |
語文別: | 中文 |
論文頁數: | 97 |
中文關鍵詞: | 子宮內膜癌 、微衛星不穩定性高 、全玻片影像 、Modified CLAM 、深度學習 |
外文關鍵詞: | endometrial cancer, MSI-H, WSIs, Modified CLAM, Deep learning |
相關次數: | 點閱:182 下載:0 |
分享至: |
查詢本校圖書館目錄 查詢臺灣博碩士論文知識加值系統 勘誤回報 |
儘管基於形態特徵的子分型對於子宮內膜癌(endometrial cancer, EC)的風險
預測和治療計劃是一種有用的工具,但這種方法往往高度主觀,難以驗證,且在
預測治療反應方面具有有限的效用。自從分子分類被應用於 EC 後,對於不同的
EC亞型,特別是微衛星不穩定性高(microsatellite instability MSI-high, MSI-H)的
免疫治療逐漸受到關注。MSI 是由於 DNA 不配對修復機制缺陷而導致某些腫瘤
易感突變的狀態。確定 EC 的 MSI 狀態很重要,因為它是一個重要的治療預測指
標。美國國家網絡結直腸癌指南(NCCN guidelines)推薦使用 pembrolizumab 和
nivolumab 治療晚期或復發的 MSI-H/mismatch repair deficient(dMMR)EC 患者。
儘管評估 MSI 對於免疫治療很重要,但由於額外的成本和時間,無法對所有癌症
進行 MSI 檢測。最近,基於深度學習(Deep learning, DL)的形態學分析從全玻
片影像(whole slide images, WSIs)中得到了發展。在這裡,我們旨在評估 DL 在
基於H&E染色的 WSIs 中預測 MSI 狀態的應用,使用了來自癌症基因組圖譜計畫
(The Cancer Genome Atlas)和三軍總醫院數據集中的與 MSI 相關的 EC 樣本。DL
的方法有潛力從 EC 的 H&E 染色的 WSIs 中直接預測 MSI 狀態,這有助於針對性
的免疫治療評估。DL方法在病理學中的應用不僅有助於提高診斷準確性,還可以
應用於個體化治療決策的精準醫學。
本研究引入了 Shuffle Attention-ResNet 作為第一個方法,改進了 create patches
的方法,採用了 modify-FCN 為第二個方法,將 MIL 網路進行了改進,引入了
varMIL 方法作為第三個方法,並將其與 MIL RNN [1], TOAD [2],CLAM [3] 做比
較,其分別測試在四個不同的測試集,TMA core 測試集、TCGA NGS WSIs 測試
集、TCGA PCR WSIs 測試集以及NDMC的 WSIs 測試集。
Although subtype classification based on morphological features is a useful tool for risk prediction and treatment planning in endometrial cancer (EC), it is often highly subjective, difficult to validate, and has limited utility in predicting treatment response. Since molecular classification has been applied to EC, the use of immune therapy, particularly for microsatellite instability-high (MSI-H) EC, has gained attention. MSI is a state of certain tumors prone to mutations due to defects in DNA mismatch repair mechanisms. Determining the MSI status in EC is crucial as it serves as an important predictive indicator for treatment. The National Comprehensive Cancer Network (NCCN) guidelines recommend the use of pembrolizumab and nivolumab for the treatment of advanced or recurrent MSI-H/mismatch repair-deficient (dMMR) EC patients. However, MSI testing cannot be performed on all cancers due to additional costs and time constraints.
Recently, morphological analysis based on deep learning (DL) has emerged for the analysis of whole-slide images (WSIs). In this study, our aim was to evaluate the application of DL in predicting MSI status from H\&E-stained WSIs in EC, using samples from The Cancer Genome Atlas and Tri-Service General Hospital dataset that are associated with MSI. DL methods have the potential to directly predict the MSI status from H\&E-stained WSIs of EC, which can aid in targeted immune therapy assessment. The application of DL methods in pathology not only improves diagnostic accuracy but also enables precision medicine in individualized treatment decision-making.
In conclusion, our study focuses on the application of DL in predicting MSI status from H\&E-stained WSIs of EC. This research demonstrates the potential of DL methods to provide valuable insights for personalized treatment decisions, particularly in the assessment of immune therapy. The integration of DL into pathology holds promise for enhancing diagnostic accuracy and advancing the field of precision medicine.
This study introduces three novel methods to improve the computational pathology on whole-slide images. The first method involves the incorporation of Shuffle Attention-ResNet, which enhances the network architecture to focus on crucial regions within the whole-slide images. The second method improves the creation of patches by employing modified-FCN, a contour-drawing technique for cells, instead of using the conventional OpenCV approach.
Furthermore, the third method addresses the MIL network by implementing the varMIL method from the DeepSMILE research, specifically for the prediction of MSI status in colorectal and breast cancer. These three methods were compared against existing approaches, namely MIL RNN\cite{campanella2019clinical}, TOAD\cite{lu2021ai},and CLAM\cite{lu2021data}, to evaluate their effectiveness.
To assess the performance of the proposed methods, four distinct test datasets were utilized, namely the TMA core test set, TCGA NGS WSIs test set, TCGA PCR WSIs test set, and NDMC's WSIs test set.
By conducting comprehensive evaluations on these datasets, the study aimed to demonstrate the potential advantages and contributions of the introduced methods in terms of data efficiency, weakly supervised learning, and accuracy for computational pathology on whole-slide images. The comparison against existing techniques further provides insights into the relative strengths and limitations of each approach in different testing scenarios.
Overall, the research contributes to the field of computational pathology by presenting novel methods that potentially enhance the accuracy and efficiency of whole-slide image analysis. The evaluation on diverse test datasets offers valuable insights into the generalization and applicability of these methods in real-world scenarios, paving the way for future advancements in computational pathology research and clinical applications.
參 考 文 獻
[1] G. Campanella, M. G. Hanna, L. Geneslaw, A. Miraflor, V. Werneck Krauss Silva,
K. J. Busam, E. Brogi, V. E. Reuter, D. S. Klimstra, and T. J. Fuchs, “Clinical-grade
computational pathology using weakly supervised deep learning on whole slide im-
ages,” Nature medicine, vol. 25, no. 8, pp. 1301–1309, 2019.
[2] M. Y. Lu, T. Y. Chen, D. F. Williamson, M. Zhao, M. Shady, J. Lipkova, and
F. Mahmood, “Ai-based pathology predicts origins for cancers of unknown pri-
mary,” Nature, vol. 594, no. 7861, pp. 106–110, 2021.
[3] M. Y. Lu, D. F. Williamson, T. Y. Chen, R. J. Chen, M. Barbieri, and F. Mahmood,
“Data-efficient and weakly supervised computational pathology on whole-slide im-
ages,” Nature biomedical engineering, vol. 5, no. 6, pp. 555–570, 2021.
[4] O. Raglan, I. Kalliala, G. Markozannes, S. Cividini, M. J. Gunter, J. Nautiyal,
H. Gabra, E. Paraskevaidis, P. Martin-Hirsch, K. K. Tsilidis, et al., “Risk factors
for endometrial cancer: An umbrella review of the literature,” International journal
of cancer, vol. 145, no. 7, pp. 1719–1730, 2019.
[5] S. F. Lax, E. S. Pizer, B. M. Ronnett, and R. J. Kurman, “Comparison of estrogen
and progesterone receptor, ki-67, and p53 immunoreactivity in uterine endometrioid
carcinoma and endometrioid carcinoma with squamous, mucinous, secretory, and
ciliated cell differentiation,” Human pathology, vol. 29, no. 9, pp. 924–931, 1998.
[6] J. V. Bokhman, “Two pathogenetic types of endometrial carcinoma,” Gynecologic
oncology, vol. 15, no. 1, pp. 10–17, 1983.
[7] M. A. Voss, R. Ganesan, L. Ludeman, K. McCarthy, R. Gornall, G. Schaller, W. Wei,
and S. Sundar, “Should grade 3 endometrioid endometrial carcinoma be considered
a type 2 cancer—a clinical and pathological evaluation,” Gynecologic oncology,
vol. 124, no. 1, pp. 15–20, 2012.
[8] R. Hong, W. Liu, D. DeLair, N. Razavian, and D. Fenyö, “Predicting endometrial
cancer subtypes and molecular features from histopathology images using multi-
resolution deep learning models,” Cell Reports Medicine, vol. 2, no. 9, p. 100400,
2021.
[9] D. A. Levine, C. G. A. R. N. G. sequencing centres: Broad Institute Getz Gad 1
Gabriel Stacey B. 1 Cibulskis Kristian 1 Lander Eric 1 Sivachenko Andrey 1 Sougnez
Carrie 1 Lawrence Mike 1, W. U. in St Louis Kandoth Cyriac 2 Dooling David 2
Fulton Robert 2 Fulton Lucinda 2 Kalicki-Veizer Joelle 2 McLellan Michael D. 2 O'
Laughlin Michelle 2 Schmidt Heather 2 Wilson Richard K. 2 Ye Kai 2 Ding Li 2
Mardis Elaine R. 2, U. of Southern California & Johns Hopkins Baylin Stephen B.
21 Bootwalla Moiz S. 22 Lai Phillip H. 22 Triche Jr Timothy J. 22 Van Den Berg
David J. 22 Weisenberger Daniel J. 22 Laird Peter W. 22 Shen Hui 22, I. for Sys-
tems Biology Reynolds Sheila M. 23 Shmulevich Ilya 23, et al., “Integrated genomic
characterization of endometrial carcinoma,” Nature, vol. 497, no. 7447, pp. 67–73,
2013.
[10] A. K. Green, J. Feinberg, and V. Makker, “A review ofimmune checkpoint blockade
therapy in endometrial cancer,” American Society ofClinical Oncology Educational
Book, vol. 40, pp. 238–244, 2020.
[11] G. F. Fleming, “Second-line therapy for endometrial cancer: the need for better op-
tions,” Obstetrical & Gynecological Survey, vol. 71, no. 7, pp. 406–408, 2016.
[12] M. Vicky, H. MacKay, I. Ray-Coquard, D. A. Levine, S. N. Westin, A. Daisuke, and
O. Ana, “Endometrial cancer (primer),” Nature Reviews: Disease Primers, vol. 7,
no. 1, 2021.
[13] S. Richman, “Deficient mismatch repair: read all about it,” International journal of oncology, vol. 47, no. 4, pp. 1189–1202, 2015.
[14] A. S. Bruegl, K. L. Ring, M. Daniels, B. M. Fellman, D. L. Urbauer, and R. R. Broad-
dus, “Clinical challenges associated with universal screening for lynch syndrome–
associated endometrial cancerchallenges with universal screening for lynch syn-
drome,” Cancer Prevention Research, vol. 10, no. 2, pp. 108–115, 2017.
[15] J. C. Watkins, E. J. Yang, M. G. Muto, C. M. Feltmate, R. S. Berkowitz, N. S.
Horowitz, S. Syngal, M. B. Yurgelun, A. Chittenden, J. L. Hornick, et al., “Uni-
versal screening for mismatch-repair deficiency in endometrial cancers to identify
patients with lynch syndrome and lynch-like syndrome,” International Journal of
Gynecological Pathology, vol. 36, no. 2, pp. 115–127, 2017.
[16] K. Kemp, J. Griffiths, S. Campbell, and K. Lovell, “An exploration of the follow-
up up needs of patients with inflammatory bowel disease,” Journal of Crohn’s and
Colitis, vol. 7, no. 9, pp. e386–e395, 2013.
[17] N. Stjepanovic, L. Moreira, F. Carneiro, F. Balaguer, A. Cervantes, J. Balmaña,
and E. Martinelli, “Hereditary gastrointestinal cancers: Esmo clinical practice guide-
lines for diagnosis, treatment and follow-up,” Annals of Oncology, vol. 30, no. 10,
pp. 1558–1571, 2019.
[18] M. A. Rodriguez-Bigas, C. R. Boland, S. R. Hamilton, D. E. Henson, S. Srivastava,
J. R. Jass, P. M. Khan, H. Lynch, T. Smyrk, M. Perucho, et al., “A national cancer
institute workshop on hereditary nonpolyposis colorectal cancer syndrome: meeting
highlights and bethesda guidelines,” Journal ofthe National Cancer Institute, vol. 89,
no. 23, pp. 1758–1762, 1997.
[19] Y. Song, Y. Gu, X. Hu, M. Wang, Q. He, and Y. Li, “Endometrial tumors with msi-
h and dmmr share a similar tumor immune microenvironment,” OncoTargets and
therapy, vol. 14, p. 4485, 2021.
[20] S. Zhao, M. Choi, J. D. Overton, S. Bellone, D. M. Roque, E. Cocco, F. Guzzo,
D. P. English, J. Varughese, S. Gasparrini, et al., “Landscape of somatic single-
nucleotide and copy-number mutations in uterine serous carcinoma,” Proceedings
of the National Academy of Sciences, vol. 110, no. 8, pp. 2916–2921, 2013.
[21] T. E. Soumerai, M. T. Donoghue, C. Bandlamudi, P. Srinivasan, M. T. Chang,
D. Zamarin, K. A. Cadoo, R. N. Grisham, R. E. O’Cearbhaill, W. P. Tew,
et al.,
“Clinical utility of prospective molecular characterization in advanced endometrial
cancermolecular characterization in advanced endometrial cancer,” Clinical Cancer
Research, vol. 24, no. 23, pp. 5939–5947, 2018.
[22] J. Liñares-Blanco, A. Pazos, and C. Fernandez-Lozano, “Machine learning analysis
of tcga cancer data,” PeerJ Computer Science, vol. 7, p. e584, 2021.
[23] N. Coudray, P. S. Ocampo, T. Sakellaropoulos, N. Narula, M. Snuderl, D. Fenyö,
A. L. Moreira, N. Razavian, and A. Tsirigos, “Classification and mutation prediction
from non–small cell lung cancer histopathology images using deep learning,” Nature
medicine, vol. 24, no. 10, pp. 1559–1567, 2018.
[24] C.-L. Chen, C.-C. Chen, W.-H. Yu, S.-H. Chen, Y.-C. Chang, T.-I. Hsu, M. Hsiao,
C.-Y. Yeh, and C.-Y. Chen, “An annotation-free whole-slide training approach
to pathological classification of lung cancer types using deep learning,” Nature
communications, vol. 12, no. 1, p. 1193, 2021.
[25] C.Wang, Y. Kim, H. Shin, and S. D. Min, “Preliminary clinical application oftextile
insole sensor for hemiparetic gait pattern analysis,” Sensors, vol. 19, no. 18, p. 3950,
2019.
[26] P. Khosravi, E. Kazemi, M. Imielinski, O. Elemento, and I. Hajirasouliha, “Deep
convolutional neural networks enable discrimination ofheterogeneous digital pathol-
ogy images,” EBioMedicine, vol. 27, pp. 317–328, 2018.
[27] J. Noorbakhsh, S. Farahmand, A. Foroughi Pour, S. Namburi, D. Caruana, D. Rimm,
M. Soltanieh-Ha, K. Zarringhalam, and J. H. Chuang, “Deep learning-based cross-
classifications reveal conserved spatial behaviors within tumor histological images,”
Nature communications, vol. 11, no. 1, p. 6367, 2020.
[28] T. Durand, N. Thome, and M. Cord, “Weldon: Weakly supervised learning of deep
convolutional neural networks,” in Proceedings ofthe IEEE conference on computer
vision and pattern recognition, pp. 4743–4752, 2016.
[29] X. Wang, Y. Yan, P. Tang, X. Bai, and W. Liu, “Revisiting multiple instance neural
networks,” Pattern Recognition, vol. 74, pp. 15–24, 2018.
[30] M. Ilse, J. Tomczak, and M.Welling, “Attention-based deep multiple instance learn-
ing,” in International conference on machine learning, pp. 2127–2136, PMLR, 2018.
[31] Z. Wang, J. Poon, S. Sun, and S. Poon, “Attention-based multi-instance neural
network for medical diagnosis from incomplete and low quality data,” in 2019
International joint conference on neural networks (IJCNN), pp. 1–8, IEEE, 2019.
[32] M. Zaheer, S. Kottur, S. Ravanbakhsh, B. Poczos, R. R. Salakhutdinov, and A. J.
Smola, “Deep sets,” Advances in neural information processing systems, vol. 30,
2017.
[33] C. R. Qi, H. Su, K. Mo, and L. J. Guibas, “Pointnet: Deep learning on point sets
for 3d classification and segmentation,” in Proceedings of the IEEE conference on
computer vision and pattern recognition, pp. 652–660, 2017.
[34] G. Liu, J. Wu, and Z.-H. Zhou, “Key instance detection in multi-instance learning,”
in Asian Conference on Machine Learning, pp. 253–268, PMLR, 2012.
[35] H. D. Couture, J. S. Marron, C. M. Perou, M. A. Troester, and M. Niethammer, “Mul-
tiple instance learning for heterogeneous images: Training a cnn for histopathol-
ogy,” in Medical Image Computing and Computer Assisted Intervention–MICCAI
2018: 21st International Conference, Granada, Spain, September 16-20, 2018,
Proceedings, Part II 11, pp. 254–262, Springer, 2018.
[36] O. Z. Kraus, J. L. Ba, and B. J. Frey, “Classifying and segmenting microscopy images
with deep multiple instance learning,” Bioinformatics, vol. 32, no. 12, pp. i52–i59,
2016.
[37] C. Zhang, J. Platt, and P. Viola, “Multiple instance boosting for object detection,”
Advances in neural information processing systems, vol. 18, 2005.
[38] L. Berrada, A. Zisserman, and M. P. Kumar, “Smooth loss functions for deep top-k
classification,” arXiv preprint arXiv:1802.07595, 2018.
[39] K. Crammer and Y. Singer, “On the algorithmic implementation ofmulticlass kernel-
based vector machines,” Journal of machine learning research, vol. 2, no. Dec,
pp. 265–292, 2001.
[40] O. Russakovsky, J. Deng, H. Su, J. Krause, S. Satheesh, S. Ma, Z. Huang, A. Karpa-
thy, A. Khosla, M. Bernstein, et al., “Imagenet large scale visual recognition chal-
lenge,” International journal of computer vision, vol. 115, pp. 211–252, 2015.
[41] Y. Schirris, E. Gavves, I. Nederlof, H. M. Horlings, and J. Teuwen, “Deepsmile:
Contrastive self-supervised pre-training benefits msi and hrd classification directly
from h&e whole-slide images in colorectal and breast cancer,” Medical Image
Analysis, vol. 79, p. 102464, 2022.
[42] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,”
in Proceedings of the IEEE conference on computer vision and pattern recognition,
pp. 770–778, 2016.
[43] Q.-L. Zhang and Y.-B. Yang, “Sa-net: Shuffle attention for deep convolutional neu-
ral networks,” in ICASSP 2021-2021 IEEE International Conference on Acoustics,
Speech and Signal Processing (ICASSP), pp. 2235–2239, IEEE, 2021.
[44] R. R. Selvaraju, M. Cogswell, A. Das, R. Vedantam, D. Parikh, and D. Batra, “Grad-
cam: Visual explanations from deep networks via gradient-based localization,” in
Proceedings of the IEEE international conference on computer vision, pp. 618–626,
2017.
[45] C.-W. Wang, Y.-C. Lee, C.-C. Chang, Y.-J. Lin, Y.-A. Liou, P.-C. Hsu, C.-C. Chang,
A.-K.-O. Sai, C.-H. Wang, and T.-K. Chao, “A weakly supervised deep learning
method for guiding ovarian cancer treatment and identifying an effective biomarker,”
Cancers, vol. 14, no. 7, p. 1651, 2022.
[46] Y.-J. Lin, T.-K. Chao, M.-A. Khalil, Y.-C. Lee, D.-Z. Hong, J.-J. Wu, and C.-W.
Wang, “Deep learning fast screening approach on cytological whole slides for thyroid
cancer diagnosis,” Cancers, vol. 13, no. 15, p. 3891, 2021.
[47] C.-W. Wang, Y.-A. Liou, Y.-J. Lin, C.-C. Chang, P.-H. Chu, Y.-C. Lee, C.-H. Wang,
and T.-K. Chao, “Artificial intelligence-assisted fast screening cervical high grade
squamous intraepithelial lesion and squamous cell carcinoma diagnosis and treat-
ment planning,” Scientific Reports, vol. 11, no. 1, p. 16244, 2021.
[48] M.-A. Khalil, Y.-C. Lee, H.-C. Lien, Y.-M. Jeng, and C.-W. Wang, “Fast segmen-
tation of metastatic foci in h&e whole-slide images for breast cancer diagnosis,”
Diagnostics, vol. 12, no. 4, p. 990, 2022.
[49] P. Jia, X. Yang, L. Guo, B. Liu, J. Lin, H. Liang, J. Sun, C. Zhang, and K. Ye,
“Msisensor-pro: fast, accurate, and matched-normal-sample-free detection of mi-
crosatellite instability,” Genomics, Proteomics & Bioinformatics, vol. 18, no. 1,
pp. 65–71, 2020.
[50] Y. Tolkach, T. Dohmgörgen, M. Toma, and G. Kristiansen, “High-accuracy prostate
cancer pathology using deep learning,” Nature Machine Intelligence, vol. 2, no. 7,
pp. 411–418, 2020.
[51] C.-W. Wang, Y.-C. Lee, Y.-J. Lin, C.-C. Chang, C.-H. Wang, T.-K. Chao, et al.,
“Interpretable attention-based deep learning ensemble for personalized ovarian can-
cer treatment without manual annotations,” Computerized Medical Imaging and
Graphics, vol. 107, p. 102233, 2023.