Automated 5-year Mortality Prediction using Deep Learning and Radiomics\n Features from Chest Computed Tomography

Abstract

We propose new methods for the prediction of 5-year mortality in elderly\nindividuals using chest computed tomography (CT). The methods consist of a\nclassifier that performs this prediction using a set of features extracted from\nthe CT image and segmentation maps of multiple anatomic structures. We explore\ntwo approaches: 1) a unified framework based on deep learning, where features\nand classifier are automatically learned in a single optimisation process; and\n2) a multi-stage framework based on the design and selection/extraction of\nhand-crafted radiomics features, followed by the classifier learning process.\nExperimental results, based on a dataset of 48 annotated chest CTs, show that\nthe deep learning model produces a mean 5-year mortality prediction accuracy of\n68.5%, while radiomics produces a mean accuracy that varies between 56% to 66%\n(depending on the feature selection/extraction method and classifier). The\nsuccessful development of the proposed models has the potential to make a\nprofound impact in preventive and personalised healthcare.\n

Ähnliche Arbeiten