臺灣博碩士論文加值系統

一、 研究背景
隨著年齡層老化，慢性傷口經常需要數月到數年的時間才能癒合。亦需要醫護人員定期檢查及傷口清創，檢查傷口癒合進度並在必要時轉介傷口專家。一致和準確的傷口護理對於傷口的適當癒合至關重要，延遲就診傷口專家可能會增加下肢截肢甚至死亡的風險。然而，傷口專家的短缺，尤其是在偏鄉地區，會導致診斷晚和傷口護理不佳。此外，不必要的醫院就診增加了臨床醫生的工作量，並為患者增加了可避免的經濟負擔。患者或來訪護士可以在患者家中使用基於智能手機照片的傷口評估系統是解決這些問題有希望的解決方案。
而各式傷口圖像以另一個角度來看，就如看似相近的人臉一樣，如何從每個圖像中辨識傷口，更甚者可以自動測量傷口區域，並將數據有效輸入到電子病歷中，以加強對患者的護理。這對臨床醫師與照護的傷口護理師，甚至病患家屬都是一大福音。
二、 研究目的及方法
由於傷口常呈現出不同的深度、弧度及血色，而在取像時更常因傷口形狀及血色變異而產生各類型的取像變異，常會影響後續的分析與特徵分類效果，因此不同類別傷口定位與分類辨識以協助醫護人員進行精準醫療是有必要的。為有效進行各種傷口的潰瘍、縫合、發黑之自動判圖與病況分類，傷口精確定位是有必要的。本研究透過臨床蒐集傷口照片，再運用特殊演算法來對傷口進行定位分析。本研究創新提出自變性能量分析並搭配特徵點消除演算法，這套演算架構之目的為移除傷口區塊周遭的雜點干擾，俾有效減少後台檢測傷口時的運算負擔同時得以提升系統性能。
三、 研究發現
本研究採用高階奇異值分解(High-Order Singular Value Decomposition, HOSVD)做影像的增強，我們可以將一張圖像看成是一個由R、G、B三通道所組成的三維矩陣，針對水平、垂直等方向進行分割再進行矩陣分解，再萃取各個矩陣中的識別資訊。經過分析精準定位框選出傷口後，將傷口影像饋入深度學習進行傷口分類。透過使用不同深度學習的網路結構，對於傷口分析進行比較後發現使用RESNET50有相對最好的精準度。
四、 研究結論
透過影像增強及深度學習後，得以精準辨識傷口類別。我們再分別使用RESNET50、RESTNET101兩種網路結構去訓練，訓練後兩種模型的準確率分別為94%及86%。藉由傷口分類可使得臨床醫師及傷口護理師減少分析傷口及追蹤所耗費時間，並依據傷口連續性變化紀錄來制定個人化傷口方式。更甚者因遠距進行傷口分析，可將技術拓展至長照醫療，落實全方位的傷口護理。
五、 關鍵字
智慧醫療；慢性傷口；傷口分析；高階奇異值分解；深度學習

I. Background
As long as the growing of edging population, more and more chronic wounds are noted which may course several months or years to heal. Most of the chronic wounds need regular following up by medical staff or even transfer the patient to expertise for further debridement and management. It’s crucial to maintain the consistency and the accuracy of the wound evaluation and treatment, otherwise may increase the risk of limbs amputation rate or even mortality rate. However, due to lack of wounds expertise, especially remote area, may lead delay diagnosis and wounds treatment. Beside, some unnecessary visiting may increase the working loading of doctors and increase the economic burden of patients. Using smartphone to upload the wounds picture and artificial intelligence wounds evaluation system may be the solution of these problems.
On the other side, each of the wound picture is similar to face picture, we could consider using the technique of face recognition to identified the features of the wounds. Even more, we could measure the wounds area automatically and input the data into electronic medical record, which may reduce the labor cost of medical staff and improve the quality of wound care.
II. Purpose and Methods
Wounds images always present different depths and curves and colors, also various differences may occur while taking the images due to environmental and human factors. All these variances can influence the effeteness of further analysis and features classification. So it’s necessary to help medical staff precisely marking the wounds and classifying the wounds. In order to automated assessing the wound images, we created an unique algorithm to analysis to wound location. Also we used professors Wang’s novel illumination compensation method called adaptive high-order singular value decomposition to enhance wound images at the preprocessing step.
III. Results
After the preprocessing, total 138 wounds images were used for training and we used six different machine learning architecture for wound recognition. The results showed using RESNET-50 can reach the best accuracy rate of 94%.
IV. Conclusion
By using image enhance technique and wounds recognition algorithm, we could precisely marking the wounds area and increase the wounds features. Then we used several deep learning architecture training, and the results showed RESNET-50 and RESNET-100 have the best accuracy rate. Using the wound analysis system can reduce the clinical loading and provide individual treatment according to the continuously record. Even more, the technique can improve the telemedicine.
V. Keywords
Smart Healthcare；Chronic wound；Wound assessment；High-Order Singular Value Decomposition；Deep Learning

摘要 1
前言 4
一.傷口簡介 4
二.傷口辨識及分類 5
研究目的及方法 8
一.研究目的 8
二.傷口取像檢測問題描述 9
三.自適應傷口檢測系統 9
1.自變性能量分析演算架構 10
2.自變性能量分析預訓練流程 11
3.自變性能量分析檢測流程 12
四.傷口影像檢測系統 14
1.傷口特徵區塊影像增強 14
2.高階奇異值分解與參數分析 15
五.深度學習與傷口分類 16
研究結果 18
討論 20
一. 傷口分析 20
二. 自變性能量分析 21
三. 網路架構分析傷口圖像 21
四. 訓練結果圖像分析 23
五. 研究適用性及前景 24
六. 研究限制 25
結論 26
參考文獻 27

1.Zahia, S., Zapirain, M. B. G., Sevillano, X., González, A., Kim, P. J., & Elmaghraby, A. (2020). Pressure injury image analysis with machine learning techniques: A systematic review on previous and possible future methods. Artificial intelligence in medicine, 102, 101742.
2.Carracher, A. M., Marathe, P. H., & Close, K. L. (2018). International diabetes federation 2017.
3.Armstrong, D. G., Boulton, A. J., & Bus, S. A. (2017). Diabetic foot ulcers and their recurrence. New England Journal of Medicine, 376(24), 2367-2375.
4.Boulton, A. J., & Whitehouse, R. W. (2017). The diabetic foot.
5.Armstrong, D. G., Swerdlow, M. A., Armstrong, A. A., Conte, M. S., Padula, W. V., & Bus, S. A. (2020). Five year mortality and direct costs of care for people with diabetic foot complications are comparable to cancer. Journal of foot and ankle research, 13(1), 1-4.
6.Amin, N., & Doupis, J. (2016). Diabetic foot disease: from the evaluation of the “foot at risk” to the novel diabetic ulcer treatment modalities. World journal of diabetes, 7(7), 153.
7.Lemaster, J. W., Reiber, G. E., Smith, D. G., Heagerty, P. J., & Wallace, C. A. R. O. L. Y. N. (2003). Daily weight-bearing activity does not increase the risk of diabetic foot ulcers. Medicine and Science in Sports and Exercise, 35(7), 1093-1099.
8.Kirsner, R. S., & Vivas, A. C. (2015). Lower‐extremity ulcers: diagnosis and management. British Journal of Dermatology, 173(2), 379-390.
9.Schofield, J., Leelarathna, L., & Thabit, H. (2020). COVID-19: Impact of and on Diabetes. Diabetes Therapy, 11(7), 1429-1435.
10.Rogers, L. C., Lavery, L. A., Joseph, W. S., & Armstrong, D. G. (2020). All feet on deck—the role of podiatry during the COVID-19 pandemic: preventing hospitalizations in an overburdened healthcare system, reducing amputation and death in people with diabetes. Journal of the American Podiatric Medical Association, 0000-0000.
11.Zhao, W., Chellappa, R., Phillips, P. J., & Rosenfeld, A. (2003). Face recognition: A literature survey. ACM computing surveys (CSUR), 35(4), 399-458.
12.Ahonen, T., Hadid, A., & Pietikäinen, M. (2004, May). Face recognition with local binary patterns. In European conference on computer vision (pp. 469-481). Springer, Berlin, Heidelberg.
13.Li, S., Gong, D., & Yuan, Y. (2013). Face recognition using Weber local descriptors. Neurocomputing, 122, 272-283.
14.Kim, K. I., Jung, K., & Kim, H. J. (2002). Face recognition using kernel principal component analysis. IEEE signal processing letters, 9(2), 40-42.
15.Déniz, O., Bueno, G., Salido, J., & De la Torre, F. (2011). Face recognition using histograms of oriented gradients. Pattern recognition letters, 32(12), 1598-1603.
16.Guo, G., Li, S. Z., & Chan, K. (2000, March). Face recognition by support vector machines. In Proceedings fourth IEEE international conference on automatic face and gesture recognition (cat. no. PR00580) (pp. 196-201). IEEE.
17.Bhattacharyya, S. K., & Rahul, K. (2013). Face recognition by linear discriminant analysis. International Journal of Communication Network Security, 2(2), 31-35.
18.Masud, M., Muhammad, G., Alhumyani, H., Alshamrani, S. S., Cheikhrouhou, O., Ibrahim, S., & Hossain, M. S. (2020). Deep learning-based intelligent face recognition in IoT-cloud environment. Computer Communications, 152, 215-222.
19.Hu, G., Yang, Y., Yi, D., Kittler, J., Christmas, W., Li, S. Z., & Hospedales, T. (2015). When face recognition meets with deep learning: an evaluation of convolutional neural networks for face recognition. In Proceedings of the IEEE international conference on computer vision workshops (pp. 142-150).
20.Jordan, M. I., & Mitchell, T. M. (2015). Machine learning: Trends, perspectives, and prospects. Science, 349(6245), 255-260.
21.Jiang, M., Ma, Y., Guo, S., Jin, L., Lv, L., Han, L., & An, N. (2021). Using machine learning technologies in pressure injury management: systematic review. JMIR medical informatics, 9(3), e25704.
22.Chanussot-Deprez, C., & Contreras-Ruiz, J. (2013). Telemedicine in wound care: a review. Advances in skin & wound care, 26(2), 78-82.
23.Gardiner, S., & Hartzell, T. L. (2012). Telemedicine and plastic surgery: a review of its applications, limitations and legal pitfalls. Journal of Plastic, Reconstructive & Aesthetic Surgery, 65(3), e47-e53.
24.Le, T. N., Giap, D. B., Wang, J. W., & Wang, C. C. (2021). Tensor-Compensated Color Face Recognition. IEEE Transactions on Information Forensics and Security, 16, 3339-3354.
25.Lorente, L. S., Vega, J. M., & Velazquez, A. (2008). Generation of aerodynamics databases using high-order singular value decomposition. Journal of Aircraft, 45(5), 1779-1788.
26.Portnoy, J., Waller, M., & Elliott, T. (2020). Telemedicine in the era of COVID-19. The Journal of Allergy and Clinical Immunology: In Practice, 8(5), 1489-1491.
27.Boulton, A. J. (2021). Diabetic foot disease during the COVID-19 pandemic. Medicina, 57(2), 97.
28.DeBoard, R. H., Rondeau, D. F., Kang, C. S., Sabbaj, A., & McManus, J. G. (2007). Principles of basic wound evaluation and management in the emergency department. Emergency Medicine Clinics of North America, 25(1), 23-39.
29.Rahimi, R., Ochoa, M., Parupudi, T., Zhao, X., Yazdi, I. K., Dokmeci, M. R., ... & Ziaie, B. (2016). A low-cost flexible pH sensor array for wound assessment. Sensors and Actuators B: Chemical, 229, 609-617.
30.Grey, J. E., Enoch, S., & Harding, K. G. (2006). Wound assessment. Bmj, 332(7536), 285-288.
31.Yeong, E. K., Mann, R., Goldberg, M., Engrav, L., & Heimbach, D. (1996). Improved accuracy of burn wound assessment using laser Doppler. Journal of Trauma and Acute Care Surgery, 40(6), 956-962.
32.Romanelli, M., Dini, V., Bianchi, T., & Romanelli, P. (2007). Wound assessment by 3-dimensional laser scanning. Archives of dermatology, 143(10), 1331-1344.
33.Seo, H., Badiei Khuzani, M., Vasudevan, V., Huang, C., Ren, H., Xiao, R., ... & Xing, L. (2020). Machine learning techniques for biomedical image segmentation: an overview of technical aspects and introduction to state‐of‐art applications. Medical physics, 47(5), e148-e167.
34.Qassim, H., Verma, A., & Feinzimer, D. (2018, January). Compressed residual-VGG16 CNN model for big data places image recognition. In 2018 IEEE 8th Annual Computing and Communication Workshop and Conference (CCWC) (pp. 169-175). IEEE.
35.Xia, X., Xu, C., & Nan, B. (2017, June). Inception-v3 for flower classification. In 2017 2nd International Conference on Image, Vision and Computing (ICIVC) (pp. 783-787). IEEE.
36.Sandler, M., Howard, A., Zhu, M., Zhmoginov, A., & Chen, L. C. (2018). Mobilenetv2: Inverted residuals and linear bottlenecks. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 4510-4520).
37.Akiba, T., Suzuki, S., & Fukuda, K. (2017). Extremely large minibatch sgd: Training resnet-50 on imagenet in 15 minutes. arXiv preprint arXiv:1711.04325.
38.Baltruschat, I. M., Nickisch, H., Grass, M., Knopp, T., & Saalbach, A. (2019). Comparison of deep learning approaches for multi-label chest X-ray classification. Scientific reports, 9(1), 1-10.
39.Reddy, A. S. B., & Juliet, D. S. (2019, April). Transfer learning with ResNet-50 for malaria cell-image classification. In 2019 International Conference on Communication and Signal Processing (ICCSP) (pp. 0945-0949). IEEE.