應用模糊整合類神經網路於疾病診斷之研究—以肝病為例

Abstract

本研究主要是利用Krishnapuram與Lee所提之模糊整合類神經網路模式，使用模糊集合理論之整合運算子（加權綜合平均模式與混合模式）作為網路節點之運算函數，建構一疾病診斷之評估模式。為了有效率地設定網路學習參數值，本研究應用田口方法，求得合適之參數值。本研究使用80筆肝病檢驗資料為訓練樣本，另60筆檢驗資料為測試樣本，進行模式之訓練與測試。結果顯示，混合模式之表現較佳，判斷受測者是否有肝病之正確率高達96.67％。本研究之結果可輔助醫師在疾病的診斷上給予協助，藉以減少醫師主觀的判斷所產生的偏差，進而提昇醫療品質。最後，本研究之結果將與倒傳遞類神經網路模式進行診斷成效之優劣比較。

This study applies Fuzzy Aggregation Network (FAN) to diagnosis of liver diseases. FAN uses fuzzy aggregation operators, such as weighted generalized means and hybrid model, in fuzzy set methods as the activation functions of the nodes. In order to effectively determine the parameters of the network learning, Taguchi method was used to obtain better parameter settings. The model trained with the inspection data of liver diseases from 80 patients, of which 41 are diseased liver patients and 39 are healthy liver patients. Other 60 patients are further applied as testing examples, of which 25 are diseased liver patients and 35 are healthy liver patients. Experimental result shows that the better diagnostic performance is made using the hybrid model, in comparison to the weighted generalized means and backpropagation neural network. The hybrid model is able to correctly differentiate 96.67% of testing examples. Furthermore, the developed model can be of great assistance to physicians with the diagnosis of liver diseases, to reducing diagnostic errors and improving the medical treatment quality.