- 學位論文
以類神經網路建立含碘對比劑不良反應風險預測評估模式
Building an Adverse Reaction Risk Assessment Prediction Model of Iodinated Contrast Media by Artificial Neural Network
摘要
在現階段放射醫學臨床檢查當中，正常人體的軟組織如血管、肌肉以及臟器因其組織密度相近，所以在經由X光照射下後的影像，並沒有辦法能夠清楚地分辨，而為了讓這些密度相似的軟組織能夠在X光下清楚的被顯影出來，就必須依靠含碘顯影劑的注射。含碘顯影劑中含有的離子因對輻射線有著不能穿透的特性，所以當含碘顯影劑經由動靜脈注射流至各組織器官、體腔或血管中時，能使人體軟組織在X光影像上產生對比的效果，讓人體的軟組織得以在X光影像中顯影，進而達到影像診斷的目的。 在含碘的離子性或非離子性顯影劑的使用上，卻不是對人體絕對安全無虞，臨床含碘顯影劑的使用上，有一定的機率會引起人體對含碘顯影劑的不良反應。約有5%的機率當人體在使用含碘顯影劑之後，會產生輕中度的顯影劑不良反應如溫熱、噁心、劇烈嘔吐、頭暈、面部或四肢水腫、皮膚起紅疹、心悸盜汗等症狀；約萬分之一的機率會出現嚴重的不良反應如休克、昏迷；而因為不良反應過於嚴重而致命的機率約為四萬分之一；在某些少數特異體質者身上，更有十萬分之一的機率會有猝死的情形發生。面對含碘顯影劑使用上的種種不良反應，目前醫學臨床上並無法在執行檢查之前事先得知。 有鑒於此，本研究希望藉由類神經網路（Artificial Neural Network）能仿效生物神經網路具有透過訓練及反覆學習的特性，來進一步的瞭解資料的特徵與型態，達到每個輸入都能對應到其所需的輸出，最後並將類神經網路實際的應用於預測含碘顯影劑引起的不良反應上，使臨床醫護人員及病患在面臨含碘顯影劑使用時，能有一輔佐參考的臨床工具，達到降低含碘顯影劑引起病患不良反應的狀況。 最後本研究中的類神經網路模型，經輸入測試組的資料後，在預測含碘顯影劑引起病患的不良反應效能方面，得到正確率98.4％、敏感度87.5％、特異度99.15％、AUC＝0.933。總結此研究中架構的含碘顯影劑引起不良反應類神經網路模型，在效能方面整體而言，不論在正確率、敏感度、特異度或AUC，都有不錯的效能，希望藉此能使類神經網路在預防含碘顯影劑引起的不良反應方面，成為一有效的臨床輔佐工具。
並列摘要
In the current clinical radiology examination the density of soft tissues are similar such as vessel, muscle or organ in the normal human body, so their x-ray images could not distinguish so clearly. In order to create a clear X-ray image in these soft tissues which have similar density, we have to depend on the injection of iodinated contrast medium. Because radiation cannot penetrating the ion which contained in iodinated contrast medium, so when the iodinated contrast medium injected into the artery or vein flowing to the soft tissues, soft tissues and vascular which in the body cavity were generated the contrast effects and developed in the X-ray image, thus achieving the purpose of diagnostic imaging. Using ionic or non-ionic iodinated contrast medium is still having some risks of damaging human body. The iodinated contrast media has the possibility of causing adverse reactions to the body in the clinical radiology examination. There is a 5% chance to occur mild to moderate adverse reactions, such as warm, nausea, vomiting, dizziness, facial or limb edema, skin rash, palpitations and sweating symptoms after using the iodinated contrast media, and about 1/10000 chance will cause severe adverse reactions such as shock, coma; more even have the 1/40000 chance the human were die for serious adverse reactions. 1 / 100000 of these people who might be a special physical patient will face the complication of sudden death. The face of all the adverse reactions about using iodinated contrast medium, the current clinical medicine could not do any checks to know in advance. This study will be using artificial neural network (ANN), which could imitate biological neural network. ANN is an adaptive system that change its structure based on external or internal information that flows through the network during the learning phase. According to this feature, the adverse reactions of the iodinated contrast media could be predicted. This information could be used in clinical purpose that could reduce the iodinated-containing contract media adverse reaction in patients during the time when the iodinated contrast medium is used. Finally, we inputted the test set data into the artificial neural network model, the result of predicting adverse reactions by iodinated contrast media of patients in this ANN model in this study is: 98.4% in accuracy, 87.5% in sensitivity, 99.15% in specificity and AUC = 0.933. In summary the model of predicting of adverse reactions of iodinated contrast media in artificial neural network has got the good result of all in accuracy, sensitivity, specificity, and AUC in this study. Consequently, to make this artificial neural network model to become an effective clinical support tool to prevent iodinated contrast media related adverse reactions.