以TSK式類神經模糊系統為基礎之進化式演算法及其在醫學影像上的應用

Abstract

本論文提出一進化式演算法來訓練TSK式類神經模糊系統，並將此系統應用於醫學影像辨識上，來提高影像辨識的成功率。本進化式演算法係利用生物共生的概念，讓每個染色體或粒子相互之間產生共生關係，以改善傳統最佳化演算法的缺點，提高最佳解的搜尋與正確性。因此本論文進化式演算法會分為兩個部分來描述，其分別是：1)GIEA方法介紹與 2)MPSO演算法介紹，並且與其他方法比較，驗證本文的進化式演算法的優點。最後將進化式演算法應用在TSK類神經模糊系統來辨識三種醫學影像：1)咽喉胃酸逆流影像、2)癌症細胞影像以及3)肺結核菌影像。由最後的實驗結果，可以驗證本論文所提之進化演算法可以得到穩定的解及快速收斂，應用在三種醫學影像的辨識上可以達到90％以上的成功率。

This dissertation proposes the evolutionary algorithms to train Takagi-Sugeno-Kang type neural fuzzy system (TNFS) and applies TNFS to medical image identification to raise the accuracy rate of identification. To improve the disadvantage of the traditional evolutionary algorithms and increase the search capability, the proposed evolutionary algorithms use the concept of symbiosis of creatures to make the individual to have symbiotic relationship with each other. Therefore, the proposed evolutionary algorithms of this dissertation are divided into two parts: 1) group interaction-based evolutionary algorithm (GIEA), and 2) mutualism particle swarm optimization (MPSO). We compare with other methods to approve the advantages of the proposed evolutionary algorithms. Finally, the proposed evolutionary algorithms are applied to Takagi-Sugeno-Kang (TSK) type neural fuzzy system to identify three medical images: 1) laryngopharyngeal reflux image, 2) cancer cell colonies image, and 3) tubercle bacilli image. According to the final experimental result, it is proved that the proposed evolutionary algorithms can calculate the stable solution, and the accuracy rate of three medical images identification can achieve more than 90％.