LDGAN: Longitudinal-Diagnostic Generative Adversarial Network for Disease Progression Prediction with Missing Structural MRI

Zhenyuan Ning, Yu Zhang, Yongsheng Pan, Tao Zhong, Mingxia Liu, Dinggang Shen

    Research output: Chapter in Book/Report/Conference proceedingConference contribution

    6 Citations (Scopus)

    Abstract

    Predicting future progression of brain disorders is fundamental for effective intervention of pathological cognitive decline. Structural MRI provides a non-invasive solution to examine brain pathology and has been widely used for longitudinal analysis of brain disorders. Previous studies typically use only complete baseline MRI scans to predict future disease status due to the lack of MRI data at one or more future time points. Since temporal changes of each brain MRI are ignored, these methods would result in sub-optimal performance. To this end, we propose a longitudinal-diagnostic generative adversarial network (LDGAN) to predict multiple clinical scores at future time points using incomplete longitudinal MRI data. Specifically, LDGAN imputes MR images by learning a bi-directional mapping between MRIs of two adjacent time points and performing clinical score prediction jointly, thereby explicitly encouraging task-oriented image synthesis. The proposed LDGAN is further armed with a temporal constraint and an output constraint to model the temporal regularity of MRIs at adjacent time points and encourage the diagnostic consistency, respectively. We also design a weighted loss function to make use of those subjects without ground-truth scores at certain time points. The major advantage of the proposed LDGAN is that it can impute those missing scans in a task-oriented manner and can explicitly capture the temporal characteristics of brain changes for accurate prediction. Experimental results on both ADNI-1 and ADNI-2 datasets demonstrate that, compared with the state-of-the-art methods, LDGAN can generate more reasonable MRI scans and efficiently predict longitudinal clinical measures.

    Original languageEnglish
    Title of host publicationMachine Learning in Medical Imaging - 11th International Workshop, MLMI 2020, Held in Conjunction with MICCAI 2020, Proceedings
    EditorsMingxia Liu, Chunfeng Lian, Pingkun Yan, Xiaohuan Cao
    PublisherSpringer Science and Business Media Deutschland GmbH
    Pages170-179
    Number of pages10
    ISBN (Print)9783030598600
    DOIs
    Publication statusPublished - 2020
    Event11th International Workshop on Machine Learning in Medical Imaging, MLMI 2020, held in conjunction with the 23rd International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2020 - Lima, Peru
    Duration: 2020 Oct 42020 Oct 4

    Publication series

    NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
    Volume12436 LNCS
    ISSN (Print)0302-9743
    ISSN (Electronic)1611-3349

    Conference

    Conference11th International Workshop on Machine Learning in Medical Imaging, MLMI 2020, held in conjunction with the 23rd International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2020
    Country/TerritoryPeru
    CityLima
    Period20/10/420/10/4

    Bibliographical note

    Publisher Copyright:
    © 2020, Springer Nature Switzerland AG.

    ASJC Scopus subject areas

    • Theoretical Computer Science
    • General Computer Science

    Fingerprint

    Dive into the research topics of 'LDGAN: Longitudinal-Diagnostic Generative Adversarial Network for Disease Progression Prediction with Missing Structural MRI'. Together they form a unique fingerprint.

    Cite this