基於二維小腦模型設計心臟病分類系統

On 2D Cerebellar Model-Based Heart-Disease Classification System

李嘉陵(Jia-Ling Lee) ； 蔡坤龍(Steven Tsai)

《MC-Transaction on Biotechnology》 4卷1期 (2012/03) Pp. 30-44

https://doi.org/10.29690/MCTB.201203.0004

摘要

本論文針對葡萄牙的心臟病資料庫，將設計二維小腦模型分類系統以便將病患快速地分出無病或有病之類別。小腦模型分類系統的設計程序包含：(一)隨機選取訓練樣本及測試樣本，(二)規劃二維小腦模型分類系統，(三)訓練二維小腦模型分類系統，(四)測試二維小腦模型分類系統。本論文以區塊交集法可快速建立位址索引矩陣，並以座標計算方式可連結狀態單元及其位址索引。我們嘗試利用兩組不同的心臟病屬性分別作為小腦模型的輸入變數，並採用平均輸出誤差的學習法則來調整小腦模型的記憶細胞權重。我們將採用的二維小腦模型可簡化分類系統的設計，並快速收斂到輸出誤差小於1%以內。在測試試驗中，我們找出了兩個與心臟病相關程度較高的屬性作為輸入變數，可使得整體分類正確率達到81.5%以及有病捕捉率達到96.2%的結果，由此足以證明本分類系統僅以二維小腦模型為架構的有效性，而且，採用相關且病症傾向相同的屬性會有較佳的分類效果。

關鍵字

心臟病症 ； 小腦模型 ； 記憶細胞 ； 分類系統

並列摘要

We will propose a classification system using two dimensional Cerebellar Model (2D CM) for the heart-disease database of Porto university. The purpose of this paper is to classify some classes of those patients in the database into the correct one. The design procedure for the 2D CM classification system is as follows: (1) We select some samples whose heart disease may be absent or present from the database randomly, and compose these samples to act as a training group and an evaluated group. Each group contains both samples with or without heart disease. (2) The classification system is then schemed. (3) Based on the desired output and the learning rule, the weighting memory of CM is tuned from the training group. (4) The classification system is tested from the evaluated group finally. In this paper, we use both the blocks’ intersection method to build up address indices rapidly, and the coordinate computing method to connect states with address indices. We use two types of different attributes in the database to act as the input of the classification system, and adopt the mean output error method for the learning rule to tune the weighting memory cell of CM. The proposed framework of 2D CM classification system in this paper is simple and converges fast within 1% output error. In the evaluated trial, we find two important attributes, whose relation level with heart disease is more higher, to jointly screen the potential heart disease. The percentage of accurate classification rate can attain 81.5%, and the percentage of capture rate for those who have heart disease can attain 96.2%. It is demonstrated that the proposed classification scheme in this paper is effective for the 2D CM framework. Furthermore, we find that the better classification result can be achieved by adopting some attributes with higher relation level and with the same tendency towards heart disease.

並列關鍵字

Heart disease ； Cerebellar model ； Memory cell ； Classification system