Tumor segmentation by convolutional neural networks for PET images – Oona Rainio (Turku PET Centre)

During treatment and monitoring of cancer patients, different medical imaging methods are commonly utilized. Positron emission tomography (PET) is a nuclear medicine imaging technique based on using short-lived radioactive tracer substances to gain information about body function, metabolism, and biology. Tumor segmentation means creating a binary mask of the same size as the original image so that all the 3D pixels (called voxels) are labelled as either positive or negative based on whether they contain cancer or not. Since analyzing the 3D PET images is time-consuming work, especially if there are many patients, it is useful to find automatic machine learning solutions that help in this work and convolutional neural networks have been noted to be very efficient for this kind of image processing.