使用縮時顯微鏡影像預測誘導性多能幹細胞形成過程之時間遞歸神經網路/METHOD AND APPARATUS FOR PREDICTING CELL REPROGRAMMING

專利類型

發明

專利國別 (專利申請國家)

美國

專利申請案號

16/719,905

專利證號

US 11,138,737 B2

專利獲證名稱

使用縮時顯微鏡影像預測誘導性多能幹細胞形成過程之時間遞歸神經網路/METHOD AND APPARATUS FOR PREDICTING CELL REPROGRAMMING

專利所屬機關 (申請機關)

中原大學

獲證日期

2021/10/05

技術說明

近來，已經開發出諸如卷積神經網絡（CNN）之類的加工學習方法來檢測顯微圖像中的iPS細胞集落。 然而，這些方法僅識別已經被重編程為iPS細胞的細胞，或者正在經歷重編程過程的細胞。 在早期選擇和識別具有重編程潛力的細胞仍然是一個挑戰。 鑑於前述內容，在現有技術中需要一種能夠預測細胞的重編程狀態，即，預測細胞從分化狀態重編程為分化程度較低的狀態（例如，被重編程為iPS的概率）的方法），以提高選擇過程的速度和成本。 該方法包括以下步驟以建立訓練好的LSTM模型： （1）選擇LSTM訓練影像的區域作為LSTM模板影像； （2）將訓練後的CNN模型應用於LSTM模板影像，以計算LSTM模板圖像的每個像素的多個類別的概率； （3）通過重複步驟（1）和（2），從多個LSTM訓練影像中產生包括多個LSTM模板影像的LSTM訓練集； 和 （4）使用在LSTM訓練集的LSTM模板圖像的每個像素處的多個類別的概率作為輸入來訓練LSTM體系結構以產生訓練後的LSTM網絡。 Recently, machining learning methods, such as convolution neural network (CNN), have been developed to detect colonies of iPS cells in microscopy images. Nonetheless, these methods merely identify the cells that have already reprogrammed into iPS cells, or the cells undergoing the reprogramming process. There is still a challenge to select and identify cells with reprogramming potential at an early stage. In view of the foregoing, there exists in the related art a need for a method capable of predicting the reprogramming status of cells, i.e., predicting the probability of cells reprogrammed from a differentiated state to a less differentiated state (e.g., being reprogrammed into iPS cells), so as to improve the speed and cost of the selection process. This method further comprises the following steps to establish the trained LSTM model: (1) selecting a region of a LSTM training image as a LSTM template image; (2) applying the trained CNN model to the LSTM template image to calculate the respective probabilities of the plurality of classes for every pixel of the LSTM template image; (3) producing a LSTM training set comprising a plurality of LSTM template images from a plurality of LSTM training images by repeating steps (1) and (2); and (4) using the probabilities of the plurality of classes at each pixel of LSTM template images of the LSTM training set as inputs to train a LSTM architecture to produce the trained LSTM network.

備註

連絡單位 (專責單位/部門名稱)

產學合作暨專利技轉中心

連絡電話

(03)2651830

網址