Improving the accuracy of multiple sclerosis classification via robust feature selection based on quantitative tractography
Maria Petracca1,2, Alberto Azzari3, Antonella Mensi3, Nicole Graziano2, Alessandro Daducci3, Manuele Bicego3, Matilde Inglese2,4,5,6, and Simona Schiavi3,4
1Department of Human Neurosciences, Sapienza University of Rome, Rome, Italy, 2Department of Neurology, Icahn School of Medicine at Mount Sinai, New York, NY, United States, 3Department of Computer Science, University of Verona, Verona, Italy, 4Department of Neuroscience, Rehabilitation, Ophthalmology, Genetics, Maternal and Child Health (DINOGMI), University of Genoa, Genoa, Italy, 5Department of Radiology, Icahn School of Medicine at Mount Sinai, New York, NY, United States, 6IRCCS Ospedale Policlinico San Martino, Genoa, Italy
Improving image-based classification accuracy in multiple sclerosis while characterizing biological relevant features remains an open challenge. To this aim we applied a robust feature selection (FS) procedure based on a leave-one-out cross-validation scheme on structural connectivity features derived from connectomes computed with convex optimization modelling for microstructure informed tractography. We computed classification accuracy for different classifiers (NN, KNN, SVM-LIN, SVM-RBF, RF) before and after the application of the FS procedure. The highest overall accuracy (91%) was obtained using the FS procedure on the whole connectome. The biological meaningfulness of the selected features is supported by their correlations with clinical scores.
This abstract and the presentation materials are available to members only;
a login is required.