具有自動特徵擷取功能的生物影像辨識與分析系統

Abstract

本計晝擬設計並實做一套具有自動特徵擷取功能的生物影像識別與分析系統，旨 在提供研究生物醫學的學者，可以在不具影像處理與識別領域專業知識的情況下，利用 這套生物影像系統的協助，藉由高通量與自動分析技術，提升分子細胞影像研究的效 能。本系統將包含5個子系統：（1)可接受各種生物影像整批輸入且含影像前處理功能的 輸入系統，（2)能由影像產生眾多候選特徵的程式庫，（3)具有自動特徵擷取功能的核心 模組，從候選特徵(包含使用者自行提供之特徵)經由最佳化分類精確度選取一組最佳特 徵，（4)使用所選取特徵且可進行獨力測試的分類器，（5)含特徵重要貢獻排名與解釋的 分析報告。各子系統已有雛型，經由5套影像分析問題(線蟲和老鼠影像老化研究、肺 癌細胞及神經細胞影像的識別與分析)測試良好。本計晝擬改良並整合5個子系統，並 利用來研究預測癌細胞轉移能力。

Epithelial-mesenchymal transition (EMT)是一種藉由完全分化之上皮細胞轉變成間 質細胞的過程增加轉移及侵犯能力，而此能力的增加是主要扮演癌細胞轉移之重要角 色。我們將藉由EMT活化建立創新型癌細胞轉移影像評估預測系統。其主要是利用 TGF-p能於各種生物系統疾病生理中，藉由活化主要訊號傳遞及調控因子連接龐大調控 網路，開始啟動及維持EMT。我們將先利用先前之基因晶片研究中挑選出重要EMT markers，利用TGF-p的誘發，藉由EMT促使癌細胞轉移及侵犯，並利用免疫螢光染色 及共軛膠顯微鏡擷取相關影像，再經由本生物影像系統來識別、分析與預測，並將利用 其他癌細胞株影像驗證本系統的預測效能。

本計晝將是結合生物實驗與生物資訊系統的跨領域研究，藉由與共同主持人的專長合 作，目的是建立一套方便實用的生物影像分析系統。

This project aims to design and implement a bioimage recognition and analysis system with automatic feature selection. The biomedical researchers without expertise of image processing and recognition can advance their research of molecular and cellular images by utilizing this bioimage system with abilities of high-throughput and automatic analysis. The bioimage system mainly consists of five subsystems: 1) an input system with preprocessing functions which can accept various types of bioimages using a batch job, 2) a program library to generate numerous candidate features from input images, 3) the core subroutine for automatic feature selection from the generated problem-independent features and user-provided problem-dependent features (if any) by optimizing classification accuracy, 4) the obtained classifier with the identified optimal feature set for independent tests, and 5) analysis report including the feature ranking and feature explanation. We have all prototypes of the five subsystems which perform well on the evaluation using five bioimage analysis problems (C. elegans muscle, Terminal bulb and mouse liver aging, lung cancer image, and neuron image). This project will improve and integrate these five prototypes, and further investigate the cancer metastasis by utilizing this system.

Epithelial-mesenchymal transition (EMT), a process whereby fully differentiated epithelial cells undergo transition to a mesenchymal phenotype giving rise to invasion and migration, is increasingly recognized as playing an important role in cancer metastasis. In this project, we will develop a novel cancer during invasion and metastasis image analysis and prediction system following the EMT activated. First, we will select the EMT markers according to previous studies from microarray analysis. Secondly, we will use TGF-P to treat cancer cells undergo EMT during invasion and migration. The images will be produced by immunoflorescence stain and picked up by laser screening confocal microscope. Following, we will analyze these images by using the developed system for recognition, analysis and prediction. Additionally, several independent cell line images will be used to validate the prediction performance of this bioimage system.

This project is a multidiscipline research combining biological experiments and bioinformatics system. Collaborating with the co-PIs having complementary expertise, this project aims to establish a useful and easily-used bioimage analysis system.