Stem Cell Classification

Abstract: Machine learning and neural networks haverecently become hot topics in many research areas. They havealready proved to be useful in the fields of medicine andbiotechnology. In these areas, they can be used to facilitatecomplicated and time consuming analysis processes. Animportant application is image recognition of cells, tumours etc.,which also is the focus of this paper.Our project was to construct both Fully Connected NeuralNetworks and Convolutional Neural Networks with the ability torecognize pictures of muscular stem cells (MuSCs). We wanted toinvestigate if the intensity values in each pixel of the images weresufficient to use as indata for classification.By optimizing the structure of our networks, we obtained goodresults. Using only the pixel values as input, the pictures werecorrectly classified with up to 95.1% accuracy. If the image sizewas added to the indata, the accuracy was as best 97.9 %.The conclusion was that it is sensible and practical to use pixelintensity values as indata to classification programs. Importantrelationships exist and by adding some other easily accessiblecharacteristics, the success rate can be compared to a human’sability to classify these cells.