Software for Automatic Detection and Monitoring of Fluorescent Lesions in Mice S. Foucher, M. Lalonde, L. Gagnon Computer Research Institute of Montreal, Montreal (QC), Canada {sfoucher, mlalonde, lgagnon}@crim.ca A.-M. Steff World Anti-Doping Agency, Montreal (QC), Canada ann-muriel.steff@wada-ama.org Abstract Detection and monitoring of fluorescent lesions in mice is usually performed manually with the help of an image manipulation commercial software. The task is often daunting due to the huge amount of images and the degradation of the fluorescent signal over time. We developed a software aiming at automatically detecting lesions based on a color analysis and segmentation of the images. The software offers tools to assess the fluorescence decay rate over time using robust regression techniques. 1. Introduction The main goal of the proposed software is to automatically detect and estimate surfaces of fluorescent endometrial fragments in mice based on a color analysis and segmentation of images. Detection is particularly difficult for a human operator at the end of the fluorescence lifetime when the signal to noise ratio is weak and the lesion surfaces small. A second purpose was to offer tools to assess the fluorescence decay rate over time using robust regression techniques. In nature, green fluorescent protein (GFP) is produced by Aequorea victoria, the Pacific Northwest jellyfish. The protein has become of great interest to cell and molecular biologists because it can reveal gene expression in living cells. This is done by linking the gene for GFP to the gene whose expression you are interested in. When that gene is turned on in a cell, not only is its protein synthesized, but GFP is synthesized as well. Illuminating the cells with near-ultraviolet light causes the cell to fluoresce a bright green. In this way, the experimenter can see when and where the gene is expressed in the living organism. In the case of experimental endometriosis, human endometrial tissue is genetically modified to express GFP and transplanted into immunodeficient mice. This model represents a valuable preclinical tool for testing the efficiency of new drugs targeting endometriosis [1-2]. The development of endometriotic lesions can be monitored through direct visualization of fluorescent tissue in the living animal. Several pictures are taken with the help of a digital camera over a period of 20 days. The developed software aims at identify in each image four types of objects (tissue, fluorescent endometrial fragments,