裘馨氏肌肉失養症(Duchenne Muscular Dystrophy, DMD)為我國罕見肌肉神經疾病之一，隨著疾病惡化患者會逐漸失去自主行動能力且連帶著許多併發症，患者平均壽命約莫為26歲上下。而超音波成像具備非侵入性與實時顯示之特性，適合醫事人員於長期追蹤DMD患者病況發展，以便即早治療延長患者壽命。本研究基於聚類演算法對DMD之腓腸肌肌群超音波影像進行特徵重建且根據DMD患者之行走動態功能與病況嚴重程度等項目進行分類，透過量化分析經聚類重建之腓腸肌肌群超音波影像於不同分類分佈之差異，來觀察患者病況於不同階段之惡化程度且結合機器、深度學習實現自動輔助判讀。採用之原始數據集由DMD患者經腓腸肌肌群超音波成像檢查所拍攝共85筆資料所構成。
DMD數據集分別以「針對腓腸肌區域(Gastrocnemius only, GO)」、「涵蓋腓腸肌及周圍肌群(Gastrocnemius and peripheral muscles, GP)」、「涵蓋腓腸肌及周圍肌群重建結果提取腓腸肌區域(Gastrocnemius of GP reconstructed, GPR)」等三個項目進行分析研究。採用K-means、Fuzzy C-means(FCM)聚類演算法對DMD數據集進行紋理重建，將每一影像重建為具七類類別紋理特徵影像，並採用其中六類類別作為主要分析項目，經分析後實驗結果於” GO”、“GPR”組別呈現之顯著差異趨勢且皆達到p <0.05。透過建立機器學習模型，執行DMD病症之行走功能、嚴重程度自動判讀任務，高斯貝葉斯方法(Gaussian Naive Bayes, GNB)、K-近鄰演算法(K-Nearest Neighbor Classification, KNN)模型於行走功能分類任務中達到86.78%之辨識準確率。決策樹(Decision Tree, DT)模型於嚴重程度分類中達到83.80% 的識別準確率。此外，另以建立深度捲積神經網路模型為深度學習模型之主要架構，同樣執行行走功能、嚴重程度自動輔助判讀任務，並使用資料擴增以提升訓練模型辨識效能。VGG16、19模型於行走功能分類項目中皆達到98.53%之準確率。VGG19模型於嚴重程度分類項目中達到92.64%之準確率。從實驗結果來看，本研究所採用K-means、FCM聚類演算法所重建DMD之腓腸肌肌群超音波成像特徵紋理，確實有助於量化分析DMD患者之腓腸肌肌群超音波成像於不同階段之惡化程度，並且結合機器、深度學習技術實現可自動且準確輔助辨別DMD病症，以協助醫事人員使用超音波影像長期追蹤DMD病患病況惡化分析追蹤之目標。

Duchenne Muscular Dystrophy (DMD) is a rare neuromuscular disease. As the disease develops, patients gradually lose their ability to move independently, and with many complications, the average lifespan of patients is about 26 years old. Ultrasound imaging has the characteristics of non-invasive and real-time display, which is suitable for medical personnel to track the development of DMD patients, and to prolong the long-term life of patients with early treatment. In this research, a clustering algorithm was used to reconstruct the characteristics of the gastrocnemius muscle group in ultrasound images, and the patients were classified according to the ambulatory function and the severity of the disease in DMD. By analyzing the different classification distributions of the ultrasonic gastrocnemius and peripheral muscles reconstructed images using clustering, the deterioration degree of the patient's condition at various stages is observed, and the automatic auxiliary identifies realized by machine and deep learning.
The original data set was composed of 85 records of data taken by DMD patients through ultrasound imaging of the gastrocnemius muscle group. The data set was mainly analyzed and studied with three projects, namely "Gastrocnemius only (GO)," "Gastrocnemius and peripheral muscles (GP)," and "Gastrocnemius of GP reconstructed (GPR)." K-means and Fuzzy C-means (FCM) clustering algorithms were used to reconstruct the texture of the DMD dataset. Each image was reconstructed with seven categories of texture features, and six categories were used as the primary analysis items. After the analysis, the experimental results in the "GO" and "GPR" groups show a significant trend of differences, and both reach p < 0.05. By establishing a machine learning model, the task of automatically identifying the ambulatory function and severity of DMD disease is performed. According to the experimental results, in the classification of ambulatory function, the Gaussian Naive Bayes (GNB) and K-Nearest Neighbor Classification (KNN) models achieved an identification accuracy of 86.78%. The Decision Tree (DT) model achieves an identification accuracy of 83.80% in the severity classification. A deep convolutional neural network model is established as the main structure of the deep learning model while performing automatic auxiliary interpretation tasks of ambulatory function and severity and through data augmentation to improve the recognition performance of the trained model. Both VGG16 and 19 models achieved 98.53% accuracy in the classification of ambulatory function. The VGG19 model achieved 92.64% accuracy in severity classification.
In terms of overall results, the K-means and FCM clustering algorithms were used in this study to reconstruct the characteristic texture of the gastrocnemius muscle group in DMD, which is indeed helpful in quantitatively analyzing the deterioration of the gastrocnemius muscle group in DMD patients at different stages. Combined with the machine and deep learning technology, it can automatically and accurately assist in identifying DMD diseases to help medical personnel use ultrasonic images to track the long-term deterioration of DMD diseases.

[1] EMERY, Alan EH; MUNTONI, Francesco; QUINLIVAN, Rosaline. Duchenne muscular dystrophy. Oxford Monographs on Medical G, 2015.
[2] BUSHBY, Katharine, et al. Diagnosis and management of Duchenne muscular dystrophy, part 1: diagnosis, and pharmacological and psychosocial management. The Lancet Neurology, 2010, 9.1: 77-93.
[3] BIGGAR, W. Douglas. Duchenne muscular dystrophy. Pediatrics in Review, 2006,27.3: 83-88.
[4] GOWERS, William Richard. Pseudo-hypertrophic muscular paralysis. J. & A. Churchill, 1879.
[5] DUCHENNE, Guillaume-Benjamin. De la paralysie musculaire pseudo-hypertro
phique ou paralysie myo-sclérosique. P. Asselin, 1868.
[6] SUSSMAN, Michael. Duchenne muscular dystrophy. JAAOS-Journal of the American Academy of Orthopaedic Surgeons, 2002, 10.2: 138-151.
[7] X-linked Recessive inheritance From http://ghr.nlm.nih.gov/ Re-creation of entire image in svg format by drsrisenthil.
[8] BUSHBY, Katharine, et al. Diagnosis and management of Duchenne muscular dystrophy, part 1: diagnosis, and pharmacological and psychosocial management. The Lancet Neurology, 2010, 9.1: 77-93.
[9] SUH, Mi Ri, et al. Multiplex ligation-dependent probe amplification in X-linked recessive muscular dystrophy in Korean subjects. Yonsei medical journal, 2017, 58.3: 613-618.
[10] SAPOZHNIKOV, Oleg A., et al. Acoustic holography as a metrological tool for characterizing medical ultrasound sources and fields. The Journal of the Acoustical Society of America, 2015, 138.3: 1515-1532.
[11] MASON, Timothy J. Therapeutic ultrasound an overview. Ultrasonics sonochemistry, 2011, 18.4: 847-852.
[12] DIETRICH, Christoph F., et al. Medical student ultrasound education: a WFUMB position paper, part I. Ultrasound in medicine & biology, 2019, 45.2: 271-281.
[13] PILLEN, Sigrid; VAN ALFEN, Nens. Skeletal muscle ultrasound. Neurological research, 2011, 33.10: 1016-1024.
[14] HECKMATT, JohnZ; DUBOWITZ, Victor; LEEMAN, Sidney. Detection of pathological change in dystrophic muscle with B-scan ultrasound imaging. The Lancet, 1980, 315.8183: 1389-1390.
[15] HECKMATT, J. Z.; PIER, N.; DUBOWITZ, V. Measurement of quadriceps muscle
thickness and subcutaneous tissue thickness in normal children by real‐time ultrasound imaging. Journal of clinical ultrasound, 1988, 16.3: 171-176.
[16] REEVES, Neil D.; MAGANARIS, Constantinos N.; NARICI, Marco V. Ultrasonographic assessment of human skeletal muscle size. European journal of applied physiology, 2004, 91.1: 116-118.
[17] PILLEN, Sigrid, et al. Skeletal muscle ultrasound: correlation between fibrous tissue and echo intensity. Ultrasound in medicine & biology, 2009, 35.3: 443-446.
[18] REIMERS, Carl D., et al. Muscular ultrasound in idiopathic inflammatory myopathies of adults. Journal of the neurological sciences, 1993, 116.1: 82-92.
[19] HECKMATT, J. Z.; PIER, N.; DUBOWITZ, V. Real‐time ultrasound imaging of muscles. Muscle & Nerve: Official Journal of the American Association of Electrodiagnostic Medicine, 1988, 11.1: 56-65.
[20] ZUBERI, S. M., et al. Muscle ultrasound in the assessment of suspected neuromuscular disease in childhood. Neuromuscular Disorders, 1999, 9.4: 203-207.
[21] OJALA, Timo; PIETIKAINEN, Matti; HARWOOD, David. Performance evaluation of texture measures with classification based on Kullback discrimination of distributions. In: Proceedings of 12th international conference on pattern recognition. IEEE, 1994. p. 582-585.
[22] TAN, Chuanqi, et al. A survey on deep transfer learning. In: International conference on artificial neural networks. Springer, Cham, 2018. p. 270-279.
[23] WENG, Wen-Chin, et al. Evaluation of muscular changes by ultrasound Nakagami imaging in Duchenne muscular dystrophy. Scientific Reports, 2017, 7.1: 1-11.
[24] EJAZ, Khurram, et al. Hybrid segmentation method with confidence region detection for tumor identification. IEEE Access, 2020, 9: 35256-35278.
[25] NITHYA, A., et al. Kidney disease detection and segmentation using artificial neural network and multi-kernel k-means clustering for ultrasound images. Measurement, 2020, 149: 106952.
[26] LEE, Sun Joo, et al. Efficient Fuzzy Image Stretching for Automatic Ganglion Cyst Extraction Using Fuzzy C-Means Quantization. Applied Sciences, 2021, 11.24: 12094.
[27] BUSHBY, Katharine, et al. Diagnosis and management of Duchenne muscular dystrophy, part 2: implementation of multidisciplinary care. The Lancet Neurology, 2010, 9.2: 177-189.
[28] OJALA, Timo; PIETIKÄINEN, Matti; HARWOOD, David. A comparative study of texture measures with classification based on featured distributions. Pattern recognition, 1996, 29.1: 51-59.
[29] PILLEN, Sigrid; ARTS, Ilse MP; ZWARTS, Machiel J. Muscle ultrasound in neuromuscular disorders. Muscle & Nerve: Official Journal of the American Association of Electrodiagnostic Medicine, 2008, 37.6: 679-693.
[30] HECKMATT, J. Z.; LEEMAN, S.; HEALEY, A. J. Characterisation of muscle disease in children. In: 1994 Proceedings of IEEE Ultrasonics Symposium. IEEE, 1994. p. 1471-1474.
[31] KLINGLER, Werner, et al. The role of fibrosis in Duchenne muscular dystrophy.
Acta Myologica, 2012, 31.3: 184.
[32] LIU, Zichao, et al. Multiple-surface-approximation-based FCM with interval memberships for bias correction and segmentation of brain MRI. IEEE Transactions on Fuzzy Systems, 2019, 28.9: 2093-2106.
[33] HARTIGAN, John A.; WONG, Manchek A. Algorithm AS 136: A k-means clustering algorithm. Journal of the royal statistical society. series c (applied statistics), 1979, 28.1: 100-108.
[34] BEZDEK, James C.; EHRLICH, Robert; FULL, William. FCM: The fuzzy c-means clustering algorithm. Computers & geosciences, 1984, 10.2-3: 191-203.
[35] GRIMM, Laurence G.; YARNOLD, Paul R. Reading and understanding multivariate statistics. American Psychological Association, 1995.
[36] Luis, V. [Algorithms] - Logistic Regression . Retrieved September 29, 2021, from http://luisevalencia.com/algorithms-logistic-regression/
[37] JOYCE, James. Bayes’ theorem. 2003.
[38] RUSSELL, Stuart; NORVIG, Peter. Artificial intelligence: a modern approach.2002.
[39] Prateek, M. OpenGenus IQ: Computing Expertise & Legacy — Gaussian Naive Bayes . from https://iq.opengenus.org/gaussian-naive-bayes/
[40] SLOVIC, Paul; FISCHHOFF, Baruch; LICHTENSTEIN, Sarah. Behavioral decision theory. Annual review of psychology, 1977, 28.1: 1-39.
[41] MAIMON, Oded Z.; ROKACH, Lior. Data mining with decision trees: theory and applications. World scientific, 2014.
[42] Chung-yi. Introduction to ML (16) Decision Tree. Retrieved September 27, 2019,
from https://medium.com/chung-yi/ml%E5%85%A5%E9%96%80%E5%8D%81%E5%85%AD-%E6%B1%BA%E7%AD%96%E6%A8%B9-decision-tree-59e5fb6a0f56
[43] HO, Tin Kam. Random decision forests. In: Proceedings of 3rd international conference on document analysis and recognition. IEEE, 1995. p. 278-282.
[44] PAL, Mahesh. Random forest classifier for remote sensing classification. International journal of remote sensing, 2005, 26.1: 217-222..
[45] A Chung-yi. Introduction to ML (16) Random Forest . Retrieved September 28, 2019, from https://medium.com/chung-yi/ml%E5%85%A5%E9%96%80-%E5%8D%81
%E4%B8%83-%E9%9A%A8%E6%A9%9F%E6%A3%AE%E6%9E%97-random-forest-6afc24871857
[46] FIX, Evelyn; HODGES, Joseph Lawson. Discriminatory analysis. Nonparametric discrimination: Consistency properties. International Statistical Review/Revue Internationale de Statistique, 1989, 57.3: 238-247.
[47] A Navlani, A. KNN Classification Tutorial using Scikit-learn. Retrieved August 3, 2018, from https://www.datacamp.com/community/tutorials/machine-learning-python
[48] CORTES, Corinna; VAPNIK, Vladimir. Support-vector networks. Machine learning, 1995, 20.3: 273-297.
[49] AURIA, Laura; MORO, Rouslan A. Support vector machines (SVM) as a technique for solvency analysis. 2008.
[50] Wikipedia. Support vector machine. Retrieved March 3, 2022, from https://zh.wikipedia.org/wiki
[51] LECUN, Yann, et al. Gradient-based learning applied to document recognition. Proceedings of the IEEE, 1998, 86.11: 2278-2324.
[52] YAN, Le Cun; YOSHUA, B.; GEOFFREY, H. Deep learning. nature, 2015, 521.7553: 436-444.
[53] SIMONYAN, Karen; ZISSERMAN, Andrew. Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556, 2014.
[54] KRIZHEVSKY, Alex; SUTSKEVER, Ilya; HINTON, Geoffrey E. Imagenet classification with deep convolutional neural networks. Advances in neural information processing systems, 2012, 25.
[55] TAN, Chuanqi, et al. A survey on deep transfer learning. In: International conference on artificial neural networks. Springer, Cham, 2018. p. 270-279.
[56] REFAEILZADEH, Payam; TANG, Lei; LIU, Huan. Cross-validation. Encyclopedia of database systems, 2009, 5: 532-538.
[57] Dev, SZ. K-Fold Cross Validation. Retrieved February 11, 2018, from http://www.szdev.com/blog/ AI/model-selection-k-fold-cross-validation/
[58] BAKKER, Jan PJ, et al. Predictive factors of cessation of ambulation in patients with Duchenne muscular dystrophy. American journal of physical medicine & rehabilitation, 2002, 81.12: 906-912.
[59] YAN, Dong, et al. Clinical evaluation of duchenne muscular dystrophy severity using ultrasound small-window entropy imaging. Entropy, 2020, 22.7: 715.
[60] LIAO, Ai-Ho, et al. Deep Learning of Ultrasound Imaging for Evaluating Ambulatory Function of Individuals with Duchenne Muscular Dystrophy. Diagnostics, 2021, 11.6: 963.
[61] 劉士鋐。「基於深度卷積神經網路於裘馨氏肌肉萎縮症之超音波影像自動檢測」。碩士論文，國立臺灣科技大學醫學工程研究所，2020。https://hdl.handle.net/11296/9u85s4。