Thesis title: Tecniche di Machine Learning per l’identificazione di cellule di epatocarcinoma da immagini in immunofluorescenza
The method used to realize the automatic detection algorithm of hepatocellular carcinoma involves the segmentation of images of liver tissue obtained with a confocal microscope. Every single cell present in the analyzed tissue is segmented and inserted into a new image with a black background; the image obtained is processed by means of a wavelet transform, thus obtaining four different images of the same cell, each of which highlights different morphological characteristics of the same. From each of the four images the co-occurrence matrix is extracted to characterize its texture and on each co-occurrence matrix statistical functions that characterize the morphology as a whole are applied ; at the same time as the wavelet transform, the images are filtered with 4 Line detection filters, thus obtaining 4 images of the same cell with different details highlighted. From the filtered images Run-length matrices will be extracted and from them the statistical functions (features). At the end of this process each cell, present in the image of the initial report, will be represented by an array of numbers, which will be input to a neural network for automatic detection.