1
|
van Veen R, Tamboli NRB, Lövdal S, Meles SK, Renken RJ, de Vries GJ, Arnaldi D, Morbelli S, Clavero P, Obeso JA, Oroz MCR, Leenders KL, Villmann T, Biehl M. Subspace corrected relevance learning with application in neuroimaging. Artif Intell Med 2024; 149:102786. [PMID: 38462286 DOI: 10.1016/j.artmed.2024.102786] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [MESH Headings] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 07/10/2023] [Revised: 01/12/2024] [Accepted: 01/21/2024] [Indexed: 03/12/2024]
Abstract
In machine learning, data often comes from different sources, but combining them can introduce extraneous variation that affects both generalization and interpretability. For example, we investigate the classification of neurodegenerative diseases using FDG-PET data collected from multiple neuroimaging centers. However, data collected at different centers introduces unwanted variation due to differences in scanners, scanning protocols, and processing methods. To address this issue, we propose a two-step approach to limit the influence of center-dependent variation on the classification of healthy controls and early vs. late-stage Parkinson's disease patients. First, we train a Generalized Matrix Learning Vector Quantization (GMLVQ) model on healthy control data to identify a "relevance space" that distinguishes between centers. Second, we use this space to construct a correction matrix that restricts a second GMLVQ system's training on the diagnostic problem. We evaluate the effectiveness of this approach on the real-world multi-center datasets and simulated artificial dataset. Our results demonstrate that the approach produces machine learning systems with reduced bias - being more specific due to eliminating information related to center differences during the training process - and more informative relevance profiles that can be interpreted by medical experts. This method can be adapted to similar problems outside the neuroimaging domain, as long as an appropriate "relevance space" can be identified to construct the correction matrix.
Collapse
Affiliation(s)
- Rick van Veen
- Bernoulli Institute for Mathematics, Computer Science and Artificial Intelligence, University of Groningen, The Netherlands.
| | - Neha Rajendra Bari Tamboli
- Bernoulli Institute for Mathematics, Computer Science and Artificial Intelligence, University of Groningen, The Netherlands.
| | - Sofie Lövdal
- Bernoulli Institute for Mathematics, Computer Science and Artificial Intelligence, University of Groningen, The Netherlands; Department of Nuclear Medicine and Molecular Imaging, University Medical Center Groningen, The Netherlands.
| | - Sanne K Meles
- Department of Neurology, University Medical Center Groningen, The Netherlands.
| | - Remco J Renken
- Department of Biomedical Sciences of Cells & Systems, Cognitive Neuroscience Center, University Medical Center Groningen, The Netherlands.
| | | | - Dario Arnaldi
- Department of Neuroscience, University of Genoa, Italy; IRCCS Ospedale Policlinico San Martino, Genoa, Italy.
| | - Silvia Morbelli
- IRCCS Ospedale Policlinico San Martino, Genoa, Italy; Department of Health Sciences, University of Genoa, Italy.
| | - Pedro Clavero
- Servicio de Neurología, Complejo Hospitalario de Navarra, Pamplona, Spain.
| | - José A Obeso
- Académico de Número Real Academia Nacional de Medicina de España, Spain.
| | - Maria C Rodriguez Oroz
- Neurology Department, Clínica Universidad de Navarra, Spain; Neuroscience Program, Center for Applied Medical Research, Universidad de Navarra, Pamplona, Spain; Navarra Institute for Health Research, Pamplona, Spain
| | - Klaus L Leenders
- Department of Nuclear Medicine and Molecular Imaging, University Medical Center Groningen, The Netherlands.
| | - Thomas Villmann
- Saxon Institute for Computational Intelligence and Machine Learning, University of Applied Sciences Mittweida, Germany.
| | - Michael Biehl
- Bernoulli Institute for Mathematics, Computer Science and Artificial Intelligence, University of Groningen, The Netherlands; SMQB, Inst. of Metabolism and Systems Research, College of Medical and Dental Sciences, Birmingham, United Kingdom.
| |
Collapse
|
2
|
Wang J, Xue L, Jiang J, Liu F, Wu P, Lu J, Zhang H, Bao W, Xu Q, Ju Z, Chen L, Jiao F, Lin H, Ge J, Zuo C, Tian M. Diagnostic performance of artificial intelligence-assisted PET imaging for Parkinson's disease: a systematic review and meta-analysis. NPJ Digit Med 2024; 7:17. [PMID: 38253738 PMCID: PMC10803804 DOI: 10.1038/s41746-024-01012-z] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [Grants] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 09/14/2023] [Accepted: 01/10/2024] [Indexed: 01/24/2024] Open
Abstract
Artificial intelligence (AI)-assisted PET imaging is emerging as a promising tool for the diagnosis of Parkinson's disease (PD). We aim to systematically review the diagnostic accuracy of AI-assisted PET in detecting PD. The Ovid MEDLINE, Ovid Embase, Web of Science, and IEEE Xplore databases were systematically searched for related studies that developed an AI algorithm in PET imaging for diagnostic performance from PD and were published by August 17, 2023. Binary diagnostic accuracy data were extracted for meta-analysis to derive outcomes of interest: area under the curve (AUC). 23 eligible studies provided sufficient data to construct contingency tables that allowed the calculation of diagnostic accuracy. Specifically, 11 studies were identified that distinguished PD from normal control, with a pooled AUC of 0.96 (95% CI: 0.94-0.97) for presynaptic dopamine (DA) and 0.90 (95% CI: 0.87-0.93) for glucose metabolism (18F-FDG). 13 studies were identified that distinguished PD from the atypical parkinsonism (AP), with a pooled AUC of 0.93 (95% CI: 0.91 - 0.95) for presynaptic DA, 0.79 (95% CI: 0.75-0.82) for postsynaptic DA, and 0.97 (95% CI: 0.96-0.99) for 18F-FDG. Acceptable diagnostic performance of PD with AI algorithms-assisted PET imaging was highlighted across the subgroups. More rigorous reporting standards that take into account the unique challenges of AI research could improve future studies.
Collapse
Affiliation(s)
- Jing Wang
- Huashan Hospital & Human Phenome Institute, Fudan University, Shanghai, China
- Department of Nuclear Medicine/PET Center, Huashan Hospital, Fudan University, Shanghai, China
| | - Le Xue
- Department of Nuclear Medicine, the Second Hospital of Zhejiang University School of Medicine, Hangzhou, Zhejiang, China
| | - Jiehui Jiang
- Institute of Biomedical Engineering, School of Life Science, Shanghai University, Shanghai, China
| | - Fengtao Liu
- Department of Neurology, Huashan Hospital, Fudan University, Shanghai, China
- National Clinical Research Center for Aging and Medicine, & National Center for Neurological Disorders, Huashan Hospital, Fudan University, Shanghai, China
| | - Ping Wu
- Department of Nuclear Medicine/PET Center, Huashan Hospital, Fudan University, Shanghai, China
| | - Jiaying Lu
- Department of Nuclear Medicine/PET Center, Huashan Hospital, Fudan University, Shanghai, China
| | - Huiwei Zhang
- Department of Nuclear Medicine/PET Center, Huashan Hospital, Fudan University, Shanghai, China
| | - Weiqi Bao
- Department of Nuclear Medicine/PET Center, Huashan Hospital, Fudan University, Shanghai, China
| | - Qian Xu
- Department of Nuclear Medicine/PET Center, Huashan Hospital, Fudan University, Shanghai, China
| | - Zizhao Ju
- Department of Nuclear Medicine/PET Center, Huashan Hospital, Fudan University, Shanghai, China
| | - Li Chen
- Department of Ultrasound Medicine, Huashan Hospital, Fudan University, Shanghai, China
| | - Fangyang Jiao
- Department of Nuclear Medicine/PET Center, Huashan Hospital, Fudan University, Shanghai, China
| | - Huamei Lin
- Department of Nuclear Medicine/PET Center, Huashan Hospital, Fudan University, Shanghai, China
| | - Jingjie Ge
- Department of Nuclear Medicine/PET Center, Huashan Hospital, Fudan University, Shanghai, China.
| | - Chuantao Zuo
- Huashan Hospital & Human Phenome Institute, Fudan University, Shanghai, China.
- Department of Nuclear Medicine/PET Center, Huashan Hospital, Fudan University, Shanghai, China.
- National Clinical Research Center for Aging and Medicine, & National Center for Neurological Disorders, Huashan Hospital, Fudan University, Shanghai, China.
| | - Mei Tian
- Huashan Hospital & Human Phenome Institute, Fudan University, Shanghai, China.
- Department of Nuclear Medicine/PET Center, Huashan Hospital, Fudan University, Shanghai, China.
| |
Collapse
|
3
|
van Veen R, Meles SK, Renken RJ, Reesink FE, Oertel WH, Janzen A, de Vries GJ, Leenders KL, Biehl M. FDG-PET combined with learning vector quantization allows classification of neurodegenerative diseases and reveals the trajectory of idiopathic REM sleep behavior disorder. COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE 2022; 225:107042. [PMID: 35970056 DOI: 10.1016/j.cmpb.2022.107042] [Citation(s) in RCA: 6] [Impact Index Per Article: 3.0] [Reference Citation Analysis] [Abstract] [Key Words] [MESH Headings] [Track Full Text] [Subscribe] [Scholar Register] [Received: 04/01/2022] [Revised: 07/11/2022] [Accepted: 07/27/2022] [Indexed: 06/15/2023]
Abstract
BACKGROUND AND OBJECTIVES 18F-fluorodeoxyglucose (FDG) positron emission tomography (PET) combined with principal component analysis (PCA) has been applied to identify disease-related brain patterns in neurodegenerative disorders such as Parkinson's disease (PD), Dementia with Lewy Bodies (DLB) and Alzheimer's disease (AD). These patterns are used to quantify functional brain changes at the single subject level. This is especially relevant in determining disease progression in idiopathic REM sleep behavior disorder (iRBD), a prodromal stage of PD and DLB. However, the PCA method is limited in discriminating between neurodegenerative conditions. More advanced machine learning algorithms may provide a solution. In this study, we apply Generalized Matrix Learning Vector Quantization (GMLVQ) to FDG-PET scans of healthy controls, and patients with AD, PD and DLB. Scans of iRBD patients, scanned twice with an approximate 4 year interval, were projected into GMLVQ space to visualize their trajectory. METHODS We applied a combination of SSM/PCA and GMLVQ as a classifier on FDG-PET data of healthy controls, AD, DLB, and PD patients. We determined the diagnostic performance by performing a ten times repeated ten fold cross validation. We analyzed the validity of the classification system by inspecting the GMLVQ space. First by the projection of the patients into this space. Second by representing the axis, that span this decision space, into a voxel map. Furthermore, we projected a cohort of RBD patients, whom have been scanned twice (approximately 4 years apart), into the same decision space and visualized their trajectories. RESULTS The GMLVQ prototypes, relevance diagonal, and decision space voxel maps showed metabolic patterns that agree with previously identified disease-related brain patterns. The GMLVQ decision space showed a plausible quantification of FDG-PET data. Distance traveled by iRBD subjects through GMLVQ space per year (i.e. velocity) was correlated with the change in motor symptoms per year (Spearman's rho =0.62, P=0.004). CONCLUSION In this proof-of-concept study, we show that GMLVQ provides a classification of patients with neurodegenerative disorders, and may be useful in future studies investigating speed of progression in prodromal disease stages.
Collapse
Affiliation(s)
- Rick van Veen
- Bernoulli Institute for Mathematics, Computer Science and Artificial Intelligence, University of Groningen, Groningen, the Netherlands; Data Science Department, Software Competence Center Hagenberg, Hagenberg, Austria.
| | - Sanne K Meles
- Department of Neurology, University Medical Center Groningen, University of Groningen, Groningen, the Netherlands
| | - Remco J Renken
- Department of Biomedical Sciences of Cells & Systems, University of Groningen, University Medical Center Groningen, Cognitive Neuroscience Center, Groningen, the Netherlands
| | - Fransje E Reesink
- Department of Neurology, University Medical Center Groningen, University of Groningen, Groningen, the Netherlands
| | - Wolfgang H Oertel
- Department of Neurology, Philipps-Universität Marburg, Marburg, Germany; Institute for Neurogenomics, Helmholtz Center for Health and Environment, Munich, Germany
| | - Annette Janzen
- Department of Neurology, Philipps-Universität Marburg, Marburg, Germany
| | | | - Klaus L Leenders
- Department of Nuclear Medicine and Molecular Imaging, University Medical Center Groningen, University of Groningen, Groningen, the Netherlands
| | - Michael Biehl
- Bernoulli Institute for Mathematics, Computer Science and Artificial Intelligence, University of Groningen, Groningen, the Netherlands; SMQB, Institute of Metabolism and Systems Research, College of Medical and Dental Sciences, Birmingham, United Kingdom
| |
Collapse
|
4
|
Galgani A, Vergallo A, Campese N, Lombardo F, Pavese N, Petrozzi L, LoGerfo A, Franzini M, Cecchetti D, Puglisi-Allegra S, Busceti CL, Siciliano G, Tognoni G, Baldacci F, Lista S, Hampel H, Fornai F, Giorgi FS. Biological determinants of blood-based cytokines in the Alzheimer's Disease clinical continuum. J Neurochem 2022; 163:40-52. [PMID: 35950445 DOI: 10.1111/jnc.15686] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 01/11/2022] [Revised: 05/31/2022] [Accepted: 08/06/2022] [Indexed: 11/29/2022]
Abstract
Converging translational and clinical research strongly indicates that altered immune and inflammatory homeostasis (neuroinflammation) plays a critical pathophysiological role in Alzheimer's disease (AD), across the clinical continuum. A dualistic role of neuroinflammation may account for a complex biological phenomenon, representing a potential pharmacological target. Emerging blood-based pathophysiological biomarkers, such as cytokines (Cyt) and interleukins (ILs) have been studied as indicators of neuroinflammation in AD. However, inconsistent results have been reported, probably due to lack of standardization of assays with methodological and analytical differences. We used machine-learning and a cross-validation-based statical workflow to explore and analyze the potential impact of key biological factors, such as age, sex, apolipoproteinE (APOE) genotype (the major genetic risk factor for late-onset AD) on Cyt. A set of Cyt was selected based on previous literature, and we investigated any potential association in a pooled cohort of cognitively healthy, mild cognitive impairment (MCI), and AD-like dementia patients. We also performed explorative analyses to extrapolate preliminary clinical insights. We found a robust sex effect on IL12 and an APOE-related difference in IL10, with the latter being also related to the presence of advanced cognitive decline. IL1β was the variable most significantly associated with MCI-to-dementia conversion over a 2.5 year-clinical follow-up. Albeit preliminary, our data support further clinical research to understand whether plasma Cyt may represent reliable and non-invasive tools serving the investigation of neuroimmune and inflammatory dynamics in AD and to foster biomarker-guided pathway-based therapeutic approaches, within the precision medicine development framework.
Collapse
Affiliation(s)
- A Galgani
- Neurology Unit, Pisa University Hospital, Pisa, Italy
| | - A Vergallo
- Sorbonne University, Alzheimer Precision Medicine (APM), AP-HP, Pitié-Salpêtrière Hospital, Paris, France
| | - N Campese
- Neurology Unit, Pisa University Hospital, Pisa, Italy
| | - F Lombardo
- U.O.C. "Risonanza Magnetica Specialistica e Neuroradiologia", Fondazione "G. Monasterio"- National Research Council/Tuscany Region, Pisa, Italy
| | - N Pavese
- Clinical Ageing Research Unit, Newcastle University, Newcastle upon Tyne, UK.,Institute of clinical Medicine, PET Centre, Aarhus University
| | - L Petrozzi
- Neurology Unit, Pisa University Hospital, Pisa, Italy
| | - A LoGerfo
- Neurology Unit, Pisa University Hospital, Pisa, Italy
| | - M Franzini
- Department of Translational Research and of New Surgical and Medical Technologies, University of Pisa, Pisa, Italy
| | - D Cecchetti
- Department of Translational Research and of New Surgical and Medical Technologies, University of Pisa, Pisa, Italy
| | | | | | - G Siciliano
- Neurology Unit, Pisa University Hospital, Pisa, Italy
| | - G Tognoni
- Neurology Unit, Pisa University Hospital, Pisa, Italy
| | - F Baldacci
- Neurology Unit, Pisa University Hospital, Pisa, Italy
| | - S Lista
- Sorbonne University, Alzheimer Precision Medicine (APM), AP-HP, Pitié-Salpêtrière Hospital, Paris, France.,Memory Resources and Research Center (CMRR), Neurology Department, Gui de Chauliac University Hospital, Montpellier, France
| | - H Hampel
- Sorbonne University, Alzheimer Precision Medicine (APM), AP-HP, Pitié-Salpêtrière Hospital, Paris, France
| | - F Fornai
- Department of Translational Research and of New Surgical and Medical Technologies, University of Pisa, Pisa, Italy.,IRCCS Neuromed, Pozzilli, Italy
| | - F S Giorgi
- Neurology Unit, Pisa University Hospital, Pisa, Italy.,Department of Translational Research and of New Surgical and Medical Technologies, University of Pisa, Pisa, Italy
| |
Collapse
|
5
|
Enhancing the Speed of the Learning Vector Quantization (LVQ) Algorithm by Adding Partial Distance Computation. CYBERNETICS AND INFORMATION TECHNOLOGIES 2022. [DOI: 10.2478/cait-2022-0015] [Citation(s) in RCA: 1] [Impact Index Per Article: 0.5] [Reference Citation Analysis] [Abstract] [Track Full Text] [Subscribe] [Scholar Register] [Indexed: 11/20/2022]
Abstract
Abstract
Learning Vector Quantization (LVQ) is one of the most widely used classification approaches. LVQ faces a problem as when the size of data grows large it becomes slower. In this paper, a modified version of LVQ, which is called PDLVQ is proposed to accelerate the traditional version. The proposed scheme aims to avoid unnecessary computations by applying an efficient Partial Distance (PD) computation strategy. Three different benchmark datasets are used in the experiments. The comparisons have been done between LVQ and PDLVQ in terms of runtime and in result, it turns out that PDLVQ shows better efficiency than LVQ. PDLVQ has achieved up to 37% efficiency in runtime compared to LVQ when the dimensions have increased. Also, the enhanced algorithm (PDLVQ) shows clear enhancement to decrease runtime when the size of dimensions, the number of clusters, or the size of data becomes increased compared with the traditional one which is LVQ.
Collapse
|
6
|
Changes in Temporal and Spatial Patterns of Intrinsic Brain Activity and Functional Connectivity in Upper-Limb Amputees: An fMRI Study. Neural Plast 2021; 2021:8831379. [PMID: 33981337 PMCID: PMC8088358 DOI: 10.1155/2021/8831379] [Citation(s) in RCA: 6] [Impact Index Per Article: 2.0] [Reference Citation Analysis] [Abstract] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 08/30/2020] [Revised: 01/04/2021] [Accepted: 04/05/2021] [Indexed: 11/20/2022] Open
Abstract
Background Amputation in adults is a serious procedure or traumatic outcome, one that leads to a possible “remapping” of limb representations (somatotopy) in the motor and sensory cortex. The temporal and spatial extent underlying reorganization of somatotopy is unclear. The aim of this study was to better understand how local and global structural plasticity in sensory-motor cortical networks changes temporally and spatially after upper-limb amputation. Methods We studied 8 healthy nonamputee control subjects and 16 complete upper-limb amputees. Resting-state MRI (rs-fMRI) was used to measure local and large-scale relative differences (compared to controls) in both the amplitude of low-frequency fluctuations (ALFF) and degree of centrality (DC) at 2 months, 6 months, and 12 months after traumatic amputation. Results In amputees, rs-fMRI scans revealed differences in spatial patterns of ALFF and DC among brain regions over time. Significant relative increases in ALFF and DC were detected not only in the sensory and motor cortex but also in related cortical regions believed to be involved in cognition and motor planning. We observed changes in the magnitude of ALFFs in the pre- and postcentral gyrus and primary sensory cortex, as well as in the anterior cingulate, parahippocampal gyrus, and hippocampus, 2 months after the amputation. The regional distribution of increases/decreases in ALFFs and DC documented at 2-month postamputation was very different from those at 6 and 12-month postamputation. Conclusion Local and wide-spread changes in ALFFs in the sensorimotor cortex and cognitive-related brain regions after upper-limb amputation may imply dysfunction not only in sensory and motor function but also in areas responsible for sensorimotor integration and motor planning. These results suggest that cortical reorganization after upper extremity deafferentation is temporally and spatially more complicated than previously appreciated, affecting DC in widespread regions.
Collapse
|