- 學位論文
類神經網路應用於觸診乳房硬塊之初步研究
Preliminary Study of Using Neural Network on Breast Mass Palpation
摘要
觸診利用在乳癌防治上是即方便又不花費時間的方法，可是台灣女性對於乳房的自我檢查卻沒有很重視。本文針對乳癌觸診探測法進行研究，以側向探測搭配正向探測，並對正向力和側向力的數據進行curve fitting。將curve fitting所求得之特徵參數作為類神經網路的輸入向量進行腫瘤大小、深度的推算。本文使用61個樣本當作類神經網路的訓練，12個樣本當作驗證。初步實驗結果從類神經網路訓練的樣本分別觀察腫瘤大小和腫瘤深度的準確度有72%。而從驗證的樣本觀察其腫瘤大小、深度也有50%以上的準確度。顯示利用類神經網路探測乳房腫瘤大小、深度有它的潛力存在，若能改善類神經網路訓練樣本時誤差快速收斂，則訓練和驗證樣本的準確度必定能提升。
並列摘要
Palpation diagnosis is a convenient and time-saving way among several diagnosis methods for breast cancer detection. However, self breast examination does not raise much significance for women in Taiwan. In this study, we calculated the detected forces as feature parameters in the lateral in associated with normal directions in the process of longitudinal test and normal test of tumors by curve fitting. And the derived parameters are thus fed to the neural network as input vectors so as to computerize the size and depth of tumors. In this study, we applied 61 samples during training process and 12 samples during the testing process in our neural network. The experimental results depict that the both of the accuracy is 72% in tumor size and depth during training process. And we could observed that the accuracy is more than 50% from the tumor size and depth during testing process. These results show that utilizing the neural network for recognizing size and depth of tumors is potentially detectable. And the accuracy of identifying various training and testing samples could be increased if the time needed for convergence during the training process could be decreased.