| Summary: | 碩士 === 國立政治大學 === 應用數學系 === 106 === Observation of cytopathic effects by virus infection is a standard method to exam the presence of viruses. Viruses can infect specific cells and cause characteristic morphological changes. When we observe cytopathic effects, we can use the unique morphology change to classify virus species. The virus identifacation can be later confirmed to immunofluorescence staining. Con- sidering the screen test is essential but labor-intensive, we use deep learning to recognize the different patterns between normal cells and virus-infected cells. We took 154 10X normal cell photographs and 532 10X influenza virus infect- ing cell photographs to train the convolutional neural network model. The model we got is able to distinguish 97.36% of training data. Then we send 400 new photographs to the model. These photographs contain both normal cell photos and virus-infected cell photos. Our model can specifically identify 99.5% of the testing data. In particular, this model differentiate positive sam- ples accurately. The accuracy of positive samples reach up to 100%. On the other hand, the accuracy of negative controls is 97.99%. Hence, we expect to use this model to reduce the timing required for this labor-intensive screening test, and identify virus more specifically.
|
|---|
