NEURAL NET-BASED IDENTIFICATION OF CELLS EXPRESSING THE P300 TUMOR-RELATED ANTIGEN USING FLUORESCENCE IMAGE-ANALYSIS

We report on preliminary investigations of the use of an image analysi s system to perform preliminary algorithmic classification of images o f fluorochrome-labeled cells followed by capture of gray-level images of potentially abnormal cells for analysis by a neural network. Cells were labeled with an antibody against a bladder cancer tumor-associate d antigen, and the neural net was used to distinguish true-positive ce lls from negative cells, false-positive cells (autofluorescent or nons pecific labeling), and cell-sized artifacts. Gray-level cell images we re digitized and processed for analysis by a feed-forward neural netwo rk using back-propagation. The network was trained and tested with two independent image sets, Various network configurations and activation functions were investigated, including a sinusoidal activation functi on. At high power, the network agreed completely with the human observ er's classification. At low power, a strong clustering of cells classi fied by the network with expert classification was seen, while the neu ral network showed roughly 75% concordance with the human observer, In addition, a set of four features extracted from raw cell. images were investigated The features were: shape factor, texture, area, and aver age pixel intensity, A network trained with these features performed b etter than one operating with gray-level Images, We conclude that usin g neural networks to recognize and classify images captured by an imag e analysis microscope is feasible. (C) 1997 Wiley-Liss, Inc.