149 KB19 KB2023Brecl Jakob, GregorDular, LaraMagdič, JožefRojc, BojanSavšek, LinaŠpiclin, ŽigaStr. 51-57ISSN:0353-3484COBISSID_HOST:149964803URN:URN:NBN:SI:doc-EV512K41slMedicinski razglediMedicinski razgledi (Supplement)analiza MR-slikmachine learning of prognostic modelsMR image analysisnapovedovanje poteka multiple sklerozeprognosis of multiple sclerosis disease coursestrojno učenje napovednih modelovNapovedovanje kliničnega poteka multiple skleroze iz magnetnoresonančnih slik| predhodni rezultati raziskave| preliminary study results| The prediction of multiple sclerosis disease course from magnetic resonance images|Backgrounds. Early identification of multiple sclerosis patients at risk of disease progression and/or progression to secondary progressive multiple sclerosis is an important unmet clinical need. This research aimed to train and evaluate predictive models of disability progression in patients with multiple sclerosis from MR images. Methods. The volumes of 268 brain structures and the volume and number of white matter lesions were determined from T1-enhanced and water-attenuated MR images through the automatic analysis of these images. We used these measurements to train three established classifier models, while the fourth model was based on a convolutional neural network and was trained directly on the gray values of MR images. Results. The fourth model achieved the best area under the receiver operating characteristic curve value of 74%, the accuracy of 76% and sensitivity of 51%, and the random forest classifier-based model achieved the highest sensitivity of 54%. The stated results are consistent with the results of previous research. Discussion. The ability to identify increased risk of disability progression from MR images is a promising research direction. Further improvements in predictive models would enable early intervention to prevent or delay the progression of disability and/or the transition to secondary progressive multiple sclerosisTEXTznanstveno časopisjejournalsSlovenian National E-content AggregatorNational and University Library of SloveniaDruštvo Medicinski razgledi