目的：應用類神經網路建立鼻咽癌病人預後預測系統，提供個人化的存活狀況預測資訊。
材料與方法：本研究篩選 1990 年至 2005 年間臺灣地區某癌症專科醫院臨床研究室資料庫的1,114 位鼻咽癌病人，選用年齡、性別、腫瘤期別、原始腫瘤侵犯程度、淋巴結擴散狀態、病理細胞切片、放射治療方式、化學治療方式、乳酸脫氫脢指數、鹼性磷酸脢指數、過往吸菸習慣、家族癌症病史等十二項變數，剔除變數資料欄位不全者 70 人以及 60 位遠端轉移者後，將整體樣本 984 人隨機挑選 75% 為訓練樣本，建立類神經網路模型預測鼻咽癌病人五年存活狀態，並對25% 測試樣本的預測準確率、靈敏度、特異度和 ROC 曲線下面積進行評估及分析。
結果：整體樣本平均年齡為 45.45 歲，五年整體存活率為 77.74%，利用 STATISTICA 軟體使用多層次類神經網路方式建立起最佳化的預測模型為 MLP 34-5-2，其訓練效能為 92.00，測試效能為87.80。與原始病人存活狀態資料進行比對分析，整體樣本中對於個別病人的存活狀態預測準確率為90.96%，靈敏度和特異度分別為 93.73% 及 81.28%，ROC 曲線下面積為 0.95，如果單以測試組樣本來看，其預測準確率為 87.80%，靈敏度和特異度分別為 92.23% 及 71.70%，ROC 曲線下面積為0.88。
結論：本研究運用類神經網路提供了一種預測個別鼻咽癌病人預後的方法，使用各項統計指標評估該類神經網路預測模型效能，顯示選用更多輸入變數資料的類神經網路，在預測個別鼻咽癌病人五年存活狀態的表現較先前研究為佳，但此方法是否適用於實際臨床病人仍有待進一步的評估與研究。

Purpose : This study wants to establish a prognostic prediction system by artificial neural network for individual patients with nasopharyngeal carcinoma.
Materials and methods : In this study, the dataset are 1,114 patients with nasopharyngeal carcinoma in one cancer center during the year from 1990 to 2005. The chosen variables include age, sex, primary tumor, regional lymph nodes, biopsy, radiotherapy, chemotherapy, lactic dehydrogenase, alkaline phosphatase, smoking, and family history. The final dataset are 984 patients excluding 70 patients with any data columns missing and 60 patients with distant metastasis. Seventy-five percent of 984 patients
are randomly selected and classified to training group. An artificial neural network is created by computer software to predict the five-year survival status of patients with nasopharyngeal carcinoma.
The performance of prediction models will be evaluated according to parameters such as accuracy, sensitivity, specificity, and the area under receiver operating characteristic curve.
Result : The average age of the patients is 45.45 years old, and the five-year overall survival rate is 77.74%. The optimized artificial neural network is MLP 34-5-2 which training performance is 92.00 and test performance is 87.80. Its accuracy is 90.96%, sensitivity is 93.73%, specificity is 81.27%, and area under the ROC curve is 0.95 for all patients. Its accuracy is 87.80%, sensitivity is 92.23%, specificity is 71.70%, and area under the ROC curve is 0.88 for test group.
Conclusion : This study shows that the prognostic prediction system established by artificial neural network has the potential to predict the five-year survival status of individual patients with nasopharyngeal carcinoma. With more input data, the performance of prediction models is better than previous researches. But this prognostic prediction system still need further study to prove it could be used for clinical patients.