臺灣博碩士論文加值系統

科技日新月異，各式各樣的電子產品充斥在生活中，改善了生活的方便性，但也造成了許多文明病的猖獗。許多統計資料顯示過度使用電子產品會導致頸椎病患急遽的增長，加上病患醫療資訊來源繁多，數量龐大，包括了文字、影像等，使得醫療院所必需改善其診療速度。為了幫助病患能得到更好、更快、更精確的診斷結果和治療，醫療院所需要效率更高、準確性更高的輔助工具來判斷病人所罹患的病症並提升醫療服務。這樣的輔助工具能夠讓醫師有更多充裕時間去了解病人的病情狀況，不僅病人受惠，也能提升醫療水準和品質。
頸椎磁振造影能提供高解析之醫學影像，以利於醫師判讀頸椎骨解剖位置之異常並能診斷出病人頸椎損傷的病況程度。因臨床醫學影像數量繁多，醫師需花大量時間進行影像判讀，且可能會發現有人為判讀差異的狀況產生。為了提升醫師判讀影像之效率和降低人為差異，本研究旨在建立一套輔助系統來協助醫師於頸椎磁振造影之判讀。醫師以頸椎骨嚴重程度、彎曲度以及整齊度，這三項重要指標來診斷頸椎病況程度。因此，本研究透過蒐集大量病人的頸椎骨磁振造影影像，並以頸椎骨解剖位置為測量基準，測量方式包括，方式(A)：彎曲距離（curvature distance）、方式(B)：軸椎終板前後徑寬度（anteroposterior diameter of superior endplate of T1 vertebra）、方式(C)：各節椎體之高度（vertebral height）、方式(D)：Powers ratio （顱底-後弓距離／枕後點-前弓距離）、方式(E)：脊髓前後徑距離（APDcanal－anteroposterior diameter of canal）、方式(F)：脊椎前後徑距離（APDcord－anteroposterior diameter of cervical cord）、方式(G)：脊髓前後徑距離／脊椎前後徑距離（CCR－cord canal ratio）、方式(H)：椎間盤高度（DH－disk height）。並將此量測資料利用基因規劃法（Genetic Programming, GP）與臨床醫師判讀頸椎之三個重要指標數值進行預測模式之訓練與測試。進一步的，再將本研究所提出之基因規劃預測模型與隨機森林演算法（Random forest）進行實驗比較與驗證。
由實驗結果而知，基因規劃法預測頸椎嚴重程度的正確率為71%，權重調整後正確率為91%，而平均十次後出現次數為前三名的預測參數有第四椎體之前面高度、第五椎體之後面高度、第七椎體之中間高度、第二椎體之脊髓前後徑距離、第四椎體之脊髓前後徑距離、第五椎體到第六椎體之椎間盤高度，而隨機森林演算法正確率為69%，權重調整後正確率為91%，所預測參數為Powers ratio、第三椎體之脊髓前後徑距離、第四椎體之脊髓／脊椎管比例、第六椎體之脊髓／脊椎管比例、第三椎體到第四椎體之椎間盤高度；基因規劃法預測頸椎彎曲度的正確率為66%，權重調整後正確率為90%，而平均十次後出現次數為前三名的預測參數有第七椎體之前面高度、第七椎體之後面高度、第三椎體之脊椎管前後徑距離、第二椎體之脊髓前後徑距離、第五椎體之脊髓／脊椎管比例，而隨機森林演算法正確率為69%，權重調整後正確率為89%，所預測參數為第六節椎體之彎曲距離；最後基因規劃法預測頸椎整齊度的正確率為70%，權重調整後為91%，而平均十次後出現次數為前三名的預測參數有椎體上終板前後徑長度、第三椎體之前面高度、第五椎體之後面高度、第二椎體之脊椎管前後徑距離、第三椎體之脊椎管前後徑距離、第四椎體之脊椎管前後徑距離，而隨機森林演算法正確率為69%，權重調整後正確率為92%，所預測參數為第五椎體之彎曲距離、第三椎體之前面高度、第二椎體之脊髓／脊椎管比例，由上述兩種方法得知，使用基因規劃法正確率有比隨機森林演算法正確率高，而判斷的規則裡也可發現兩者有預測到同樣的參數，而權重調正後正確率都有大幅提升，可知預測出的結果還是有可信度，但分析還需更精確正確率才能夠更高，找出更能預測頸椎病況的準則，使預測項目能夠提供臨床醫師對於病人頸椎骨病況之初步了解，而後更有效率且準確地進行診斷。

Rapid changes in technology, a wide range of electronic products flooding in life, improved the lives of convenience, but also caused many diseases of civilization rampant. Many statistics show that excessive use can cause cervical spondylosis electronic products grew sharply, with many patients medical information sources, a huge number, including text, images, etc., so that the necessary medical facilities to improve their treatment speed. To help patients get better, faster, more accurate diagnosis and treatment, medical clinics need higher efficiency, higher accuracy aids to determine the patient suffering from the disease and improve health services. Such aid can allow physicians to have more ample time to understand the disease status of the patient, not only benefit the patient, but also to enhance the health and quality standards.
Cervical MRI provides high resolution medical imaging, the physician in order to facilitate the interpretation of the anatomical location of abnormal cervical vertebra and cervical spine injury patients can diagnose the extent of the condition. Due to the vast amount of clinical and medical imaging, physicians need to spend a lot of time for image interpretation and may find someone to produce differences in interpretation of the situation. In order to enhance efficiency and reduce imaging physician interpretation artificial differences, this study aims to establish a support system to assist physicians in the interpretation of cervical MRI. Physicians with the severity of cervical vertebrae, cervical curvature and whether in a straight line, these three important indicator of the extent to diagnose cervical vertebrae condition. Therefore, this study collected through a large number of patients of cervical vertebra MRI images, and cervical vertebra anatomical position measurement reference, including the measurement, mode (A): bending distance (curvature distance), mode (B): Final vertebral axis front and rear plates path width (anteroposterior diameter of superior endplate of T1 vertebra), mode (C): the sections of the vertebral body height (vertebral height), mode (D): Powers ratio (base of the skull - the posterior arch distance / posterior point - anterior arch distance), mode (E): before and after spinal cord diameter distance (APDcanal-anteroposterior diameter of canal), mode (F): before and after spinal diameter distance (APDcord-anteroposterior diameter of cervical cord), mode (G): spinal anteroposterior diameter from front to back / spine diameter distance (CCR-cord canal ratio), mode (H): disc height (DH-disk height). And this measurement data using genetic programming (Genetic Programming, GP) interpretation and clinicians three important indicators of the value of the cervical vertebrae to train and test the prediction mode. Further, in this study the gene and then put forward the planning and forecasting model Random Forest algorithm (Random forest) were compared with experimental verification.
From the experimental results with knowledge, genetic programming method to predict the severity of the cervical accuracy was 71%, after the correct weight adjustment was 91%, while the average after ten times the number of occurrences for the prediction parameters are the top three in front of the fourth vertebrae height, height at the back of the fifth vertebra, middle height seventh vertebrae, the spinal cord before and after the second vertebra Trail from around the spinal vertebrae Trail from the fourth, fifth vertebra to the vertebral disc height sixth and the random forest algorithm accuracy was 69%, after the correct weight adjustment was 91%, the forecast parameters for the Powers ratio, before and after spinal vertebrae Trail third from fourth vertebrae of the spinal cord / spinal canal ratio, the first six vertebrae of the spinal cord / spinal canal ratio, the third vertebral disc height to the fourth vertebrae; gene programming cervical curvature prediction accuracy rate of 66%, after the correct weight adjustment was 90%, while the average ten times after the top three predictive parameters for the number of occurrences are seventh in front of the vertebral body height, back height seventh vertebrae, the spinal canal around a third of the diameter from the vertebrae, the spinal cord before and after the second vertebra Trail distance, fifth the spinal vertebrae / spinal canal ratio, and random forest algorithm accuracy was 69%, after the weight adjustment right 89% of the forecast parameter is bent away from the vertebrae VI; and finally gene planning Predicting cervical uniformity the accuracy rate of 70%, after the weight was adjusted to 91%, while the average after ten times the number of occurrences of the top three predictive parameters before and after the endplate path length, the height of the front third of the vertebral body, the fifth vertebrae the height at the back, around the spinal canal anteroposterior diameter before and after the second vertebra from the third vertebrae of the spinal canal diameter from the fourth vertebra of the spinal canal diameter distance, and random forest algorithm accuracy was 69%, weight adjustment after the correct rate of 92%, the forecast parameters for the bending distance of the fifth vertebra, the height of the front third vertebrae, the spinal cord of a second vertebral / spinal canal ratio, known from the above two methods, the use of genetic programming the correct rate higher than random forest algorithms correct, but judge rules can also be found in both forecast to have the same parameters, weighting alignment correct rate has increased dramatically, we can see the predicted results or credible degrees, but a more precise analysis needs to be able to correct the rate higher, more predictable guidelines identify cervical condition, so that the project can provide clinicians forecast for a preliminary understanding of the condition of patients with cervical vertebra, then more efficient and accurate diagnosis .

目錄

摘要 ii
ABSTRACT iv
誌 謝 vii
目錄 viii
表目錄 x
圖目錄 xi
第一章 緒論 1
1.1研究背景與動機 1
1.2研究目的 2
1.3研究架構 4
1.4研究範圍與限制 6
第二章 文獻探討 7
2.1頸椎骨 7
2.2磁振造影 9
2.3基因規劃法 10
2.4隨機森林 14
第三章 研究架構 19
3.1研究架構 19
3.2資料收集與處理 20
3.3設定初始條件 26
3.4資料匯入 27
3.5基因規劃法模型建立與分析 27
3.5.1產生初代族群 28
3.5.2適應函數 30
3.5.3母代挑選機制 31
3.5.4演化運算操作 31
3.5.5終止條件 34
3.5.6正確率權重調整 34
3.6隨機森林分群法 34
3.6.1建立二分式決策樹 34
3.6.2起始分隔尋找 35
3.6.3決策樹建立 35
3.6.4節點錯誤率計算 35
3.6.5決策樹錯誤率計算 36
3.6.6分支決策樹確認 36
3.6.7分支樹評估 36
3.6.8最佳分支樹評估 36
3.6.9投票表決 36
第四章 實驗設計與分析結果 37
4.1實驗環境 37
4.2病人頸椎骨嚴重程度判斷 38
4.2.1適應函數 39
4.2.2母代挑選機制 40
4.2.3演化運算 40
4.2.4終止條件 43
4.2.5正確率權重調整 43
4.3病人頸椎彎曲度判斷 44
4.4病人頸椎整齊度判斷 44
4.5實驗資料 44
4.6實驗結果 45
4.6.1病人頸椎嚴重程度預測 45
4.6.2病人頸椎彎曲度預測 46
4.6.3病人頸椎整齊度預測 47
第五章 結論與未來研究方向 49
5.1結論 49
5.2未來研究方向 51
參考文獻 52

參考文獻

[1]陳祐德，智慧型手機等科技產品對身心之影響，臺北市立成功高級中學，2012年。
[2]台南醫院，肩頸動動好姿態健康低頭真輕快，衛生福利部，2014年。
[3]黃彥凱，使用頸椎參數化有限元素模型評估頸椎融合術對椎節力學之影響 國立臺灣大學，碩士論文，2013年。
[4]蘇皓琮，退化性頸椎脊髓病變患者接受手術及術後運動訓練的姿勢控制能力變化 長庚大學，碩士論文，2014年。
[5]楊樹榮、張家瑞，以基因規劃法建構路基土壤在反覆載重下之塑性模式，明新科技大學土木工程系，碩士論文，2009年。
[6]林勁伍，利用核醣核酸的二級結構及基因規劃法找尋其共同結構元，國立交通大 學，資訊科學系，碩士論文，2005年。
[7]陳昶憲、鍾侑達、方唯鈞，遺傳規劃在河川演算之應用，臺灣水利第53卷第4期
2005年。
[8]趙李英記，隨機森林運用於白血病基因分類，文化大學國企所，The 9th International Conference on Knowledge Community KC，2013年。
[9]陳樹衡、池秉聰、陳俊元，經濟學中的創新行為建模: 遺傳規劃的觀點 國立政治大學，2005年。
[10]葉名山、李旻錡、劉欣憲，以基因規劃法建構交通事故責任判別之決策支援系統
學術論文研討會，2009年。
[11]蕭宜昌，結合決策樹與基因規劃法於資料分類之研究，屏東科技大學，碩士論文 2007年。
[12]霍學軍，早期股骨頭缺血性壞死的綜合影像診斷，山東德州市中醫院 ，中國傷殘醫學，12期 (2013 / 12 / 24) ，P59 - 59,60，2013年。
[13]黃錦源，電腦斷層(CT)於急診之利用與相關因素之探討，中國醫藥大學，碩士論文2013年。

[14]洪愷均，建立在 MRI 影像之基礎的脊椎自動切割方法，中興大學，碩士論文
2014年。
[15]許時榮，結合動態X光與立體像素骨模型量測活體三維頸椎骨運動，國立臺灣大學碩士論文，2007年。
[16]吳錫昆，攝影動作分析系統對頸椎局部角度測量之效度及頸椎局部動作之探討
國立成功大學，碩士論文，2007年。
[17]張哲肇，模擬僵直性脊椎炎之頸椎有限元素分析，國立成功大學，碩士論文
2015年。
[18]張家維，以生理訊號及量表探討頸椎牽引之角度偏移對療效的影響，中原大學
碩士論文，2015年。
[19]簡嘉志，利用隨機森林演算法做車牌辨識，暨南大學，碩士論文，2013年。
[20]蘇韻璇，利用單形體體積生長演算法進行腦部核磁共振影像分類，中興大學
碩士論文，2015年。
[21]黃永隆，骨肉瘤核磁共振影像細胞壞死率分析之研究，元智大學，2010年。
[22]謝忠和，以機器學習技術改造傳統臨床決策輔具工具—應用隨機森林、支援向量機器與類神經網路於急性闌尾炎診斷與乳癌風險評估，國立陽明大學，博士論文
2012年。
[23]徐啟庭，以霍夫隨機森林建構人臉表情辨識技術，國立清華大學，碩士論文
2012年。
[24]P. T. Saunders ,” Alan Turing and biology”, IEEE Computer Society,2002.
[25]John R. Koza. “Genetic Programming: On the Programming of Computers by Means
of Natural Selection.” MIT Press, 1992.
[26]Addisu Mesfin, R., Wajeeh Bakhsh, Tapanut Chuntarapas and K. Daniel Riew ,”Cervical Scoliosis: Clinical and Radiographic Outcomes”, Global Spine J. 2016
[27]C., Moonen, P., Bandettini, “Functional MRI”. Springer, Berlin,1999.
[28]Yu, X., Wadghiri, Y.Z., Sanes, D.H., Turnbull,” In vivo auditory brain
mapping in mice with Mn-enhanced MRI”,2005.
[29]P.Jezzard, P.M., Matthews, S.M, Smith, “Functional MRI A Introduction to
Methods.”, Oxford, UK,2001.
[30]Maureen Donohue, “Cervical MRI Scan, Maureen Donohue Medically Reviewed”,2015.
[31]JL., Antevil,Sise, MJ., Sack, DI., “Spiral computed tomography for the initialevaluation of spine trauma：new standard of caIe.”,2006.
[32]L., Breiman, A. Cutler, “Random Forests,Machine Learning”, Vol. 45, October, 2001.
[33]Q., C., Meng , Eng., Ocean, T. J. Feng ; Z. Chen ; C. J. Zhou, ”Genetic algorithms encoding study and a sufficient convergence condition of Gas”, IEEE,1999.
[34]T. K. Ho,”The random subspace method for constructing decision forests”, IEEE Trans. on Pattern Analysis and Machine Intelligence, Vol. 20, No. 8 ,pp. 832-844., 1998
[35]D.E., Goldberg, “Genetic algorithms in search optimization and machine learning. Addison-Wesley, Reading”, MA, USA, 1989.
[36]B.J. Park, H.R.,Choi, H.S., Kim, “A hybrid genetic algorithm for the job shop scheduling problems”, Computers & Industrial Engineering, Vol.45, 597-613, 2003.
[37]RJ., Gardocki, “Other disorders of the spine”, In: ST., Canale, JH.,Beaty,” Campbells Operative Orthopaedics.”, 12th ed. Philadelphia, PA: Elsevier Modby, 2012.
[38]R., Chou, A., Qaseem, DK., Owens, Shekelle P., “ Diagnostic Imaging for Low Back Pain: Advice for High-Value Health Care From the American College of Physicians.”, Ann Intern Med. 2011
[39]JL., Even, MS., Eskander, WF.,Donaldson, “Cervical spine injuries. In: Miller MD, Thompson SR”, eds.DeLee and DrezsOrthopaedic Sports Medicine.4th ed. Philadelphia, PA: Elsevier Saunders, 2015.
[40]T., Baccetti, L., Franchi, J.A., McNamara Jr, “The Cervical Vertebral Maturation (CVM) Method for the Assessment of Optimal Treatment Timing in Dentofacial Orthopedics.” Seminars in Orthodontics, 2005.
[41]T,, Tin KamHoA, T., Bell Laboratories, “Random Decision Forests”,1995
[42] Maureen Donohue, “MD, Cervical MRI Scan, Maureen DonohueMedically Reviewed “,2015.
[43]XP., Zhong, YX., Chen, Li., ZY1, ZW., Shen, KM., Kong, RH., Wu , “Cervical spinal functional magnetic resonance imaging of the spinal cord injured patient during electrical stimulation.”,2016.
[44]N., Siauve, GE., Chalouhi, B., Deloison, M., Alison, O., Clement, Y., Ville,” Salomon LJ,Functional imaging of the human placenta with magnetic resonance,Am J Obstet Gynecol.” 2015.
[45]FW., Guermazi, Roemer, H., Alizai, CS., Winalski, G., Welsch, M., Brittberg, S., Trattnig” MR Imaging after Knee Cartilage Repair Surgery, Radiology,” 2015.