染色體的自動辨識

Abstract

此篇論文的目的在於針對染色體影像研發一有效且可靠的分析系統，自動化得到該染色體影像的組型，以利於醫療人員的判讀與診斷工作。一個人類細胞內共有22對體染色體及1對性染色體，隨著組別遞增，各組染色體的長度相對地有變短的趨勢。染色體辨識的另一個重要特徵為染色體黑白紋理分佈，此紋理分佈同樣也是依其組別而有所不同。以這些特徵為基礎，染色體自動化辨識的研究主要分成以下幾個步驟。在中軸擷取階段，我們以方向性切片方式，先求得切片中點，再由這些中點推算染色體的中軸，而染色體紋理的灰階值曲線圖，即為沿著中軸斷面，以特定統計方式求得各中軸點的灰階值所組成。得到一個細胞內所有染色體的灰階值曲線圖後，經過標準化的動作，便可計算兩兩染色體之間的關連性，最後由關連性的數值來判斷染色體的配對以及辨識結果。實驗結果顯示我們所提出的染色體影像自動化辨識流程具有一定的正確性和效率。

In this paper we present our approach for automatic recognition of human chromosomes. The method we used integrates several novel algorithms to automatically obtain the karyotype of chromosomes of metaphase cell spreads. The characteristics that according to the classes chromosomes are gradually shortened by lengths and each of the 24 classes possesses unique banding patterns can be used to perform chromosome recognition. In the proposed approach, we first obtain the longitudinal axes from chromosome images, and then extract features from band-pattern profiles along the obtained longitudinal axes. After normalization, the features are correlated to obtain the matching scores for every two aligned profiles. Experiments show that the proposed algorithm is efficient and very robust.