Page 100 - MEGIN Book Of Abstracts - 2023
P. 100

Predicting the progression of mild cognitive       Paris, F-75013, France; Institut du Cerveau et de la Moelle
            impairment using machine learning: A systematic,   épinière, ICM, Paris, F-75013, France; Inserm, U 1127, Paris,
            quantitative and critical review (2021)            F-75013, France; CNRS, UMR 7225, Paris, F-75013, France;
                                                               Sorbonne Université, Paris, F-75013, France
                          Ansart, Manon; Epelbaum, Stéphane; Bassignana,
            Giulia; Bône, Alexandre; Bottani, Simona; Cattai, Tiziana;   ABSTRACT We performed a systematic review of stud-
            Couronné, Raphaël; Faouzi, Johann; Koval, Igor; Louis,   ies focusing on the automatic prediction of the pro-
            Maxime; Thibeau-Sutre, Elina; Wen, Junhao; Wild,   gression of mild cognitive impairment to Alzheimer's
            Adam; Burgos, Ninon; Dormont, Didier; Colliot, Olivier;   disease (AD) dementia, and a quantitative analysis of
            Durrleman, Stanley                                 the methodological choices impacting performance.
                                                               This review included 172 articles, from which 234
            Institut du Cerveau et de la Moelle épinière, ICM, Paris,   experiments were extracted. For each of them, we
            F-75013, France; Inserm, U 1127, Paris, F-75013, France; CNRS,   reported the used data set, the feature types, the
            UMR 7225, Paris, F-75013, France; Sorbonne Université, Paris,   algorithm type, performance and potential method-
            F-75013, France; Inria, Aramis project-team, Paris, F-75013,   ological issues. The impact of these characteristics on
            France. Electronic address: [email protected]; Institut du   the performance was evaluated using a multivariate
            Cerveau et de la Moelle épinière, ICM, Paris, F-75013, France;   mixed effect linear regressions. We found that us-
            Inserm, U 1127, Paris, F-75013, France; CNRS, UMR 7225, Paris,   ing cognitive, fluorodeoxyglucose-positron emission
            F-75013, France; Sorbonne Université, Paris, F-75013, France;   tomography or potentially electroencephalography
            Inria, Aramis project-team, Paris, F-75013, France; Institute of   and magnetoencephalography variables significantly
            Memory and Alzheimer's Disease (IM2A), Centre of excellence   improved predictive performance compared to not
            of neurodegenerative disease (CoEN), National Reference   including them, whereas including other modalities,
            Center for Rare or Early Dementias, Department of Neurology,   in particular T1 magnetic resonance imaging, did not
            Pitié-Salpêtrière Hospital, AP-HP, Boulevard de l'hôpital, Paris,   show a significant effect. The good performance of
            F-75013, France; Institut du Cerveau et de la Moelle épinière,   cognitive assessments questions the wide use of imag-
            ICM, Paris, F-75013, France; Inserm, U 1127, Paris, F-75013,   ing for predicting the progression to AD and advocates
            France; CNRS, UMR 7225, Paris, F-75013, France; Sorbonne   for exploring further fine domain-specific cognitive
            Université, Paris, F-75013, France; Inria, Aramis project-team,   assessments. We also identified several methodological
            Paris, F-75013, France; Dept. of Information Engineering,   issues, including the absence of a test set, or its use for
            Electronics and Telecommunication, Sapienza University   feature selection or parameter tuning in nearly a fourth
            of Rome, Italy; Institut du Cerveau et de la Moelle épinière,   of the papers. Other issues, found in 15% of the studies,
            ICM, Paris, F-75013, France; Inserm, U 1127, Paris, F-75013,   cast doubts on the relevance of the method to clinical
            France; CNRS, UMR 7225, Paris, F-75013, France; Sorbonne   practice. We also highlight that short-term predictions
            Université, Paris, F-75013, France; Inria, Aramis project-team,   are likely not to be better than predicting that sub-
            Paris, F-75013, France; AP-HP, Pitié-Salpêtrière hospital,   jects stay stable over time. These issues highlight the
            Department of Neuroradiology, Paris, France; Institut du   importance of adhering to good practices for the use of
            Cerveau et de la Moelle épinière, ICM, Paris, F-75013, France;   machine learning as a decision support system for the
            Inserm, U 1127, Paris, F-75013, France; CNRS, UMR 7225, Paris,   clinical practice.
            F-75013, France; Sorbonne Université, Paris, F-75013, France;
            Inria, Aramis project-team, Paris, F-75013, France; Institute of   Keywords: Alzheimer’s disease, Automatic prediction,
            Memory and Alzheimer's Disease (IM2A), Centre of excellence   Cognition, Mild cognitive impairment, Progression, Quan-
            of neurodegenerative disease (CoEN), National Reference   titative review
            Center for Rare or Early Dementias, Department of Neurology,
            Pitié-Salpêtrière Hospital, AP-HP, Boulevard de l'hôpital, Paris,   Medical image analysis (2021), Vol. 67 (33091740) (20
            F-75013, France; AP-HP, Pitié-Salpêtrière hospital, Department   citations)
            of Neuroradiology, Paris, France; Inria, Aramis project-team,







             ontents         Index                       79
               C
   95   96   97   98   99   100   101   102   103   104   105