Predicting Alzheimer's Disease Progression Siv Multi-modal Deep Learning Approach

Garam Lee^1,2, Kwansik Nho^3,4ua al

Alzheimer tus kab mob(AD) yog ib qho kev mob neurodegenerative uas tshwm sim los ntawm kev poob qis hauv kev paub txog kev ua haujlwm uas tsis muaj kev kho mob-hloov kho. Nws yog ib qho tseem ceeb rau kev kho raws sij hawm kom paub meejADnyob rau hauv nws cov theem ua ntej ua ntej kev kho mob tshwm sim. Kev paub tsis meej me ntsis (MCI) yog ib theem nruab nrab ntawm kev paub txog cov neeg laus thiab cov lausAD. Txhawm rau kwv yees hloov dua siab tshiab los ntawm MCI mus rau qhov yuav tshwm sim AD, peb siv txoj hauv kev kawm tob, ib qho kev sib txuas ntawm cov neural multimodal. Peb tau tsim ib qho kev sib koom ua ke uas sib koom ua ke tsis yog tsuas yog hla ntu ntu neuroimaging biomarkers ntawm lub hauv paus tab sis kuj tseem muaj cov kua dej cerebrospinal ntev (CSF) thiab kev paub txog kev ua tau zoo biomarkers tau los ntawmAlzheimer's DiseaseNeuroimaging Initiative cohort (ADNI). Lub moj khaum tau muab tso ua ke longitudinal multi-domain cov ntaub ntawv. Peb cov txiaj ntsig tau pom tias 1) peb tus qauv twv ua ntej rau MCI hloov dua siab tshiab rauADyielded mus txog 75 feem pua qhov tseeb (chaw nyob rau hauv qhov nkhaus (AUC)=0.83) thaum siv cov ntaub ntawv ib qho kev sib cais; thiab 2) peb cov qauv kev twv ua ntej tau ua tiav qhov kev ua tau zoo tshaj plaws nrog 81 feem pua qhov raug (AUC=0.86) thaum koom nrog cov ntaub ntawv longitudinal multi-domain. Ib txoj hauv kev sib sib zog nqus ntau qhov kev kawm muaj peev xwm txheeb xyuas cov neeg muaj kev pheej hmoo ntawm kev txhim kho AD uas yuav tau txais txiaj ntsig zoo tshaj plaws los ntawm kev sim tshuaj lossis raws li txoj hauv kev stratification hauv kev sim tshuaj.

Hu rau:joanna.jia@wecistanche.com/ WhatsApp: 008618081934791

Protect brain function and prevent Alzheimer's disease

cistanche extract hmoov pdfrauAD

Alzheimer tus kab mob(AD) yog ib qho irreversible, kev loj hlob neurodegenerative teeb meem tshwm sim los ntawm txawv txav ntawm amyloid plaques thiab neurofibrillary tangles nyob rau hauv lub hlwb, ua rau muaj teeb meem nrog nco, xav, thiab coj cwj pwm.ADyog hom kev dementia ntau tshaj plaws uas tsis muaj kev kho mob-hloov kho. Kwv yees li 5.7 lab tus neeg Asmeskas nyob nrog AD hauv 2018. Los ntawm 2050, tus lej no tau kwv yees nce mus txog ze li ntawm 14 lab 1 lab. Tam sim no muaj kev kho mob qeeb tsuas yog kev loj hlob ntawm AD thiab tsis muaj kev kho mob uas tsim los txog tam sim no tuaj yeem kho tus neeg mob uas twb nyob hauv AD. Yog li, nws yog ib qho tseem ceeb rau kev kho mob raws sij hawm thiab kev ncua sij hawm los tsim cov tswv yim rau kev kuaj pom AD thaum ntxov ua ntej kev kho mob tshwm sim. Yog li ntawd, lub tswv yim ntawm kev paub tsis meej me ntsis (MCI) tau qhia. MCI, ib daim ntawv prodromal ntawmAD, tau txhais los piav txog cov neeg uas muaj cov tsos mob me me ntawm lub hlwb tsis ua haujlwm tab sis tseem tuaj yeem ua haujlwm txhua hnub. Cov neeg mob nyob rau theem ntawm MCI muaj kev pheej hmoo ntau ntxiv mus rau dementia1-4. Qee tus neeg mob hauv lawv cov theem MCI tau hloov mus rau AD nyob rau hauv ib qho kev txwv ntawm lub sij hawm qhov rais tom qab lub hauv paus, thaum qee qhov tsis yog. Nws tau tshaj tawm tias cov neeg mob MCI tau nce mus rau AD ntawm tus nqi ntawm 10 feem pua mus rau 15 feem pua ntawm ib xyoos thiab 80 feem pua ntawm cov neeg mob MCI yuav tau hloov mus rau AD tom qab kwv yees li rau xyoo tom qab 5,6. Nws yog ib lub ntsiab lus tsis tu ncua ntawm cov kws tshawb fawb txog AD txhawm rau txheeb xyuas biomarkers uas faib cov neeg mob nrog MCI uas tom qab tau nce mus rau AD (MCI converter) los ntawm cov neeg uas muaj MCI uas tsis tau nce mus rau AD (MCI tsis hloov pauv).

Ntau yam kev kawm tshuab tau siv los txheeb xyuas biomarkers rau MCI hloov dua siab tshiab twv thiab txhim kho lawv cov kev ua tau zoo. Kev them nyiaj yug vector tshuab (SVM) yog ib txoj hauv kev uas nquag siv los daws cov teeb meem kev faib tawm. Ntau cov kev tshawb fawb tau siv SVM rau MCI kev hloov pauv kev kwv yees 7-12. Kev kawm ntau yam nrog rau SVM tau siv los txheeb xyuasAD-Cov yam ntxwv tseem ceeb, qhia txog qhov raug 73.9 feem pua, 68.6 feem pua rhiab, thiab 73.6 feem pua qhov tseeb7. Rau kev siv cov kev kawm ntxiv, ib qho kev hloov pauv kev kawm los siv cov kev pab cuam xws liADthiab kev paub txog cov neeg laus laus (CN) cov kev kawm, nrog rau cov kev kawm MCI, tau pom 79.4 feem pua qhov raug, 84.5 feem pua rhiab, thiab 72.7 feem pua ntawm qhov tseeb 8. Ib qho kev txheeb cais tawm (LDA) tau siv raws li cov ntaub ntawv cortex thickness uas qhia txog 63 feem pua rhiab thiab 76 feem pua qhov tshwj xeeb13. Tsis tas li ntawd, kev sib koom ua ke ntawm cov ntaub ntawv ntau hom kev txhim kho kev ua tau zoo rau MCI kev hloov pauv kev kwv yees los ntawm kev rho tawm cov ntsiab lus AD ntsig txog biomarkers los ntawm txhua qhov qauv. Cerebrospinal kua (CSF), MRI, thiab kev paub txog kev ua tau zoo biomarkers tau ua ke, ua rau 68.5 feem pua qhov raug 53.4 feem pua rhiab, thiab 77 feem pua ntawm qhov tshwj xeeb 14,15. Nrog rau MRI thiab CSF biomarkers, APOE ε4 xwm txheej tau sib xyaw ua ke16.

Table 1. Subject demographics at baseline visit.

Table 2. Tree experimental schemes depending on training dataset composition.

Nyob rau hauv txoj kev tshawb no, txhawm rau kwv yees MCI rau AD hloov dua siab tshiab, peb tau npaj ib txoj kev sib txuas ntawm cov neural network multimodal, kev kawm tob, raws li kev sib koom ua ke ntawm cov ntaub ntawv pej xeem, longitudinal CSF biomarkers, longitudinal cognitive performance, thiab cross-sectional neuroimaging biomarkers ntawm baseline tau los ntawm lubAlzheimer's DiseaseNeuroimaging Initiative cohort (ADNI). Peb txoj kev kawm tob tob tuaj yeem suav nrog cov ntaub ntawv ntev ntev ntau lub npe thiab siv cov ntaub ntawv sib txawv-ntev ntev los ntes cov yam ntxwv ntawm lub cev ntawm ntau lub sijhawm. Tshwj xeeb, cov qauv tsis sib tshooj, nrog rau cov qauv sib tshooj ntawm txhua cov ntaub ntawv, tuaj yeem siv los tsim cov qauv kev twv ua ntej.

Cov txiaj ntsig

Cov neeg koom nrog kawm.

Txhua tus neeg siv hauv kev tshuaj xyuas yog cov neeg koom nrogAlzheimer's DiseaseNeuroimaging Initiative (ADNI) 17,18. Lub hom phiaj tag nrho ntawm ADNI yog txhawm rau kuaj seb puas muaj kev sib nqus magnetic resonance imaging (MRI), txoj hauj lwm emission tomography (PET), lwm cov cim kev lom neeg, thiab kev soj ntsuam thiab kev ntsuam xyuas neuropsychological tuaj yeem ua ke los ntsuas qhov kev loj hlob ntawm MCI thiab thaum ntxov AD. Cov ntaub ntawv pej xeem, cov ntaub ntawv nyoos neuroimaging scan, APOE genotype, CSF ntsuas, cov qhab nia xeem neuropsychological, thiab cov ntaub ntawv kuaj mob tau tshaj tawm los ntawm ADNI cov ntaub ntawv khaws cia (http://adni.loni.usc.edu). Cov ntaub ntawv pom zoo tau txais rau txhua yam kev kawm, thiab txoj kev tshawb fawb tau pom zoo los ntawm lub koom haum saib xyuas kev cuam tshuam ntawm txhua qhov chaw tau txais cov ntaub ntawv (rau cov ntaub ntawv tshiab, saib http://adni.loni.usc.edu/wp-content/ themes/freshnews-dev-v2/documents/policy/ADNI_Kev lees paub_Sau feem pua 205-29-18.pdf). Txhua txoj hauv kev tau ua raws li cov lus qhia thiab cov cai. Hauv txoj kev tshawb no, tag nrho ntawm 1,618 ADNI cov neeg koom nrog hnub nyoog 55 txog 91 xyoo tau siv, uas suav nrog 415 kev paub txog cov neeg laus laus tswj hwm (CN), 865 MCI (307 MCI converter, thiab 558 MCI tsis hloov pauv), thiab 338 AD cov neeg mob (Table 1).

Peb siv plaub hom sib txawv, lossis cov qauv ntawm cov ntaub ntawv: cov ntaub ntawv pej xeem, cov tshuaj neuroimaging phenotypes ntsuas los ntawm MRI, kev txawj ntse, thiab kev ntsuas CSF. Cov ntaub ntawv pej xeem suav nrog hnub nyoog, poj niam txiv neej, xyoo kawm ntawv, thiab APOE ε4 xwm txheej. Kev paub txog kev ua tau zoo suav nrog cov qhab nia sib xyaw rau kev ua haujlwm ua haujlwm (ADNI-EF) thiab kev nco (ADNI-MEM) tau los ntawm ADNI neuropsychological roj teeb siv cov lus teb txoj kev xav raws li tau piav qhia hauv lwm qhov19. CSF biomarkers rau AD suav nrog amyloid- 1–42 peptide (A 1–42), tag nrho tau (t-tau), thiab tau phosphorylated ntawm threonine 181 (p-tau). AD-related neuroimaging biomarkers ntsuas los ntawm MRI muaj xws li hippocampal ntim thiab entrorhinal cortical thickness.

To prevent Alzheimer's disease

cistanche GINSENG tubulosa

Kev sim teeb tsa.

Txhawm rau ntsuas qhov kev ua tau zoo thiab kev ua tau zoo ntawm peb cov txheej txheem kev kawm longitudinal multi-modal sib sib zog nqus, peb siv peb lub tswv yim thiab muab piv rau lawv cov kev ua yeeb yam (Table 2). Hauv kev sim hu ua "baseline", 4 cov ntaub ntawv hloov pauv ntawm qhov kev mus ntsib hauv paus (kev paub txog kev ua tau zoo, CSF, cov ntaub ntawv pej xeem, thiab MRI) tau muab tso ua ke. Hauv "ib qho modal", tsuas yog cov ntaub ntawv kev paub txog kev paub ntev ntev tau siv rau tus neeg twv ua ntej (peb sim tag nrho lwm cov kev hloov pauv, thiab kev ua tau zoo nrog cov qhab nia kev txawj ntse yog qhov zoo tshaj plaws). Thaum kawg, plaub qhov kev hloov pauv ntawm cov ntaub ntawv ntev tau muab tso ua ke thiab siv rau kev cob qhia cov chav kawm hauv qhov kev sim ua cim ua "tso tseg". Table 3 qhia cov ntsiab lus txheeb cais ntawm txhua qhov kev hloov pauv ntawm cov ntaub ntawv thiab hyperparameters siv rau kev cob qhia GRUs.

Rau kev cob qhia peb cov qauv, cov ncauj lus hauv CN thiab AD pawg tau siv nrog rau MCI-C thiab MCI-NC. Txoj kev no yog txhawb los ntawm 8,10,11,20–22. Tey siv CN thiab AD cov kev kawm rau kev cob qhia cov chav kawm xws li SVM23 lossis hauv zos linear embedding (LLE) 24, thiab tom qab ntawd tus classifier yog siv rau kev faib MCI-C thiab MCI-NC. Hauv peb qhov kev sim, CN thiab AD tau siv los ua cov ntaub ntawv pabcuam rau kev cob qhia ua ntej, thiab tom qab ntawd MCI-C thiab MCI-NC kuj tau siv rau kev cob qhia.

Table 3. Summary statistics for data and hyperparameters.

Figure 1. An example using longitudinal data for MCI conversion prediction.

Peb tau soj ntsuam cov classifier ntawm MCI cov neeg mob los kwv yees qhov kev hloov dua siab tshiab tom qab Δt los ntawm lub hauv paus (6, 12, 18, thiab 24 lub hlis) raws li qhia hauv daim duab 1. Vim qhov xwm ntawm peb cov ntaub ntawv tus qauv loj muaj rau kev cob qhia txawv dua Δt ( Daim duab 2). Piv txwv li, yog tias AD tshwm sim ntxov los ntawm kev mus ntsib hauv paus, ces peb muaj cov qauv kev cob qhia tsawg dua vim tias peb muaj cov ntaub ntawv me me los kwv yees. Ntawm txhua lub sijhawm kwv yees (Δt), peb tau khiav 5-fold cross-validation 10 zaug uas txhua qhov quav muaj qhov sib piv ntawm MCI-C thiab MCI-NC cov ntsiab lus. MCI cov qauv tau muab faib ua 5 pawg, thiab ib pawg tau raug xaiv rau kev sim, thaum cov khoom seem ntxiv tau siv rau kev cob qhia.

Kev sib piv ntawm kev kwv yees ntawm MCI rau AD hloov dua siab tshiab siv cov ntaub ntawv hla ntu ntawm cov ntaub ntawv hauv qab thiab cov ntaub ntawv ntev.Txhawm rau ntsuas qhov zoo ntawm kev siv cov ntaub ntawv ntev, peb thawj zaug piv cov kev ua yeeb yam ntawm ob lub tswv yim: "pib" thiab "thov" (Figs 3 thiab 4). Intuitively, cov ntaub ntawv los ntawm ntau lub sijhawm cov ntsiab lus muaj cov ntaub ntawv ntau dua li cov ntaub ntawv ntawm ib lub sijhawm. Yog li, GRU tshuaj xyuas qhov kev hloov pauv ntawm lub sijhawm hauv kev paub txog kev ua tau zoo thiab CSF kom rho tawm cov yam ntxwv (uas tsis muaj nyob hauv cov ntaub ntawv mus ntsib hauv paus) rau qhov tseeb MCI kev twv twv txiaj. Raws li pom nyob rau hauv Table 4(a,b), tus qauv twv ua ntej raws li cov ntaub ntawv ntev qhia tau hais tias kev ua tau zoo dua li cov qauv siv cov ntaub ntawv hla ntu ntawm lub hauv paus. Hauv particular, rhiab heev yog ib qho kev ntsuas tseem ceeb rau txoj haujlwm twv ua ntej uas txheeb xyuas qhov tseeb qhov zoo yog qhov tseem ceeb25. Hauv kev twv ua ntej ntawm MCI hloov dua siab tshiab, ib qho kev faib tawm uas muaj qhov muaj txiaj ntsig zoo dua yog siv tau rau kev kho raws sijhawm.

Figure 3. Predictive performances with

Figure 4. ROC curves from the

Kev sib piv ntawm kev kwv yees ntawm MCI rau AD hloov dua siab tshiab siv ib qho qauv thiab cov ntaub ntawv multimodal.Rau kev ntsuas qhov ua tau zoo ntawm kev sib koom ua ke ntawm cov ntaub ntawv multimodal, peb piv cov kev ua yeeb yam ntawm "them" thiab "ib qho modal" thwmsim. Daim duab 3 qhia qhov tseeb ntawm "thov" thiab cov qauv nrog ib qho qauv ntawm cov ntaub ntawv. Peb tshem tawm qhov tseeb ntawm tus qauv nrog cov ntaub ntawv pej xeem vim tias qhov kev twv ua haujlwm tau qis dhau. Cov qauv siv kev txawj ntse tau pom tias yog qhov tseeb tshaj plaws ntawm cov qauv uas siv txhua qhov kev hloov pauv ntawm cov ntaub ntawv. Txawm hais tias tus qauv loj rau cov ntaub ntawv neuroimaging loj dua li cov kev paub txog kev paub thiab CSF biomarkers (Fig. 3), tus qauv nrog cov ntaub ntawv neuroimaging pom qhov tseeb tsawg dua. Qhov no yog vim kev paub txog kev ua tau zoo yog cov ntaub ntawv ntev uas siv qhov zoo ntawm kev muab cov ntaub ntawv ze dua rau MCI cov ntsiab lus hloov dua siab tshiab. Txawm li cas los xij, tus qauv nrog kev paub txog kev ua tau zoo qhia tau hais tias muaj qhov sib txawv ntawm qhov kev nkag siab zoo rau kev kwv yees 18 thiab 24 lub hlis. Nws tau pom tias tus qauv tsuas yog nrog kev paub txog kev ua tau zoo tsis yog qhov ntsuas ruaj khov rau lub sijhawm ntev ntawm kev kwv yees thaum sib koom ua ke nrog lwm cov biomarkers tuaj yeem txo qhov kev sib txawv siab.

Table 4. Prediction performance based on Δt over diferent schemes.

Kev sib tham

Peb tau npaj ib qho kev sib koom ua ke rau kev kwv yees ntawm MCI rau AD hloov dua siab tshiab siv txoj kev kawm sib sib zog nqus, tshwj xeeb tshaj yog, ntau qhov kev rov ua dua neural network. Peb txoj kev siv sijhawm kom zoo dua ntawm qhov ntev thiab ntau qhov xwm txheej ntawm cov ntaub ntawv muaj los tshawb pom cov qauv uas tsis yog kab ntsig cuam tshuam nrog MCI kev nce qib. Txhawm rau ntsuas qhov zoo ntawm peb txoj kev npaj, peb piv cov kev ua tau zoo los ntawm peb lub tswv yim: "baseline", "single modal", thiab "proposed". Raws li tau pom nyob rau hauv daim duab 4, "baseline" thiab "single modal" nrog kev paub txog kev kuaj biomarkers qhia cov kev ua yeeb yam zoo sib xws hauv lub sijhawm kwv yees. Siv cov ntaub ntawv longitudinal los yog sib txuas cov ntaub ntawv multimodal yog txoj hauv kev zoo rau kev nce kev kwv yees lub zog yog li, nws zoo li ntuj rau kev sib txuas cov ntaub ntawv longitudinal multimodal ("thov") los qhia qhov ua tau zoo tshaj plaws. Hauv Table 5, raws li tau kwv yees lub sijhawm ntxiv, kev ntseeg tau ntawm kev txhim kho kev ua haujlwm tau qis dua vim tsis muaj cov qauv zoo. Txawm li cas los xij, qhov tshwj xeeb ntawm cov qauv npaj tau pom tias muaj kev ua tau zoo dua li cov txheej txheem sib tw tsis tu ncua. Tsis tas li ntawd, qhov kev kwv yees tau tshwm sim ntawm peb tus qauv tau muab piv rau cov kev tshawb fawb yav dhau los nrog kev kawm tshuab (Table 6). Peb txoj kev qhia tau tias muaj peev xwm kwv yees piv tau txawm tias peb muaj qhov tsis sib npaug ntawm cov qauv zoo thiab tsis zoo. Tshwj xeeb, qhov rhiab heev ntawm peb cov qauv qhia kev ua tau zoo dua thaum qhov tshwj xeeb qis dua. Ntxiv mus, Te balanced accuracy26, uas yog ib qho kev ntsuas ntawm qhov raug txiav txim siab qhov rhiab heev thiab qhov tshwj xeeb qhia tau hais tias 0.82 rau peb cov qauv thiab 0.81 for27.

Qhov zoo tshaj plaws ntawm peb txoj hauv kev yog tias cov ntaub ntawv tsis tu ncua ntev tuaj yeem siv tau. Ib qho ntawm cov teeb meem loj thaum cuam tshuam nrog cov ntaub ntawv ntev yog tias yuav tsum tau ua cov kauj ruam ua ntej rau kev tuav qhov sib txawv-ntev ntawm cov ntaub ntawv sib txuas thiab tsis muaj qhov tseem ceeb. Hauv cov kev tshawb fawb yav dhau los, qhov ntev ntawm lub sij hawm cov ntsiab lus tau sau los ntawm kev noj cov ntaub ntawv uas poob rau hauv ib lub sijhawm. Tsis tas li ntawd, ib qho ntxiv feature extraction theem yuav tsum tau tsim ib tug tas-loj feature sawv cev. Hauv thawj kauj ruam kev cob qhia, cais GRU Cheebtsam ua cov txheej txheem encoding, qhov ntev cov ntaub ntawv hloov mus rau hauv vector uas muaj AD-sensitive nta. Tank rau cov qauv ntawm GRU, peb txoj hauv kev muaj peev xwm lees paub qhov ntev ntawm cov ntaub ntawv raws li kev nkag mus yam tsis tau ua ua ntej.

Tsis tas li ntawd, peb txoj kev tuaj yeem siv tag nrho cov kev kawm uas muaj los ntawm txhua tus qauv rau kev cob qhia peb cov chav kawm. Qhov no yog qhov zoo tshaj plaws nyob rau hauv lub ntsej muag ntawm cov ntaub ntawv scarcity. Raws li pom hauv daim duab 2, tus naj npawb ntawm cov ncauj lus nrog CSF cov ntaub ntawv yog qhov tsawg tshaj plaws hauv cov qauv sib tshooj. Cov txheej txheem ib txwm siv tau tsuas yog cov qauv sib tshooj thaum cov qauv tsis sib tshooj raug tso tseg. Hauv peb qhov xwm txheej, cov qauv tsis sib tshooj ua rau muaj kev cob qhia tus kheej GRU feem nws yog rau kev kawm sawv cev zoo dua. Tsis tas li ntawd, cov ntaub ntawv modality ntxiv tau yooj yim muab tso rau hauv cov qauv. Contrary to the kernel-based integration, the concatenation-based integration method can incorporate other domains of data such as multi-modal neuroimaging and genomic data yam tsis muaj kev paub ua ntej. Yog li, tom ntej no, peb yuav sib sau ua ke ntau cov qauv neuroimaging thiab genomics cov ntaub ntawv rau kev kawm cov yam ntxwv uas yuav muaj txiaj ntsig zoo hauv kev twv ua ntej MCI rau AD hloov dua siab tshiab.

Txawm hais tias muaj qee qhov kev ua tau zoo raws li tau piav qhia saum toj no, peb txoj hauv kev muaj qee qhov kev txwv. Hauv thawj kauj ruam kev cob qhia, cov tswv yim ntawm txhua qhov kev hloov pauv tau hloov mus rau hauv ib qho kev qhia vector uas zoo rau MCI hloov dua siab tshiab los ntawm ib qho kev hloov pauv nkaus xwb. Yog lawm, cov yam ntxwv uas tsis cuam tshuam rau AD kev nce qib nrog kev hwm rau tib lub qauv yuav raug lim tawm. Txawm li cas los xij, yog tias muaj cov yam ntxwv uas tsis tuaj yeem muab rho tawm los ntawm ib qho qauv tab sis tsuas yog tuaj yeem piav qhia los ntawm kev sib xyaw ntawm ntau hom ntawm cov ntaub ntawv ces cov no kuj yuav raug lim tawm. Qhov no yog vim qhov tsis muaj nyob hauv GRUs tsis tau hloov kho tawm tsam qhov kev kwv yees zaum kawg. Nyob rau hauv lwm yam lus, parameter optimization rau lub thib ob kev cob qhia kauj ruam tsis cuam tshuam rau cov tsis nyob rau hauv txhua GRU rau feature extraction, yog li txhua GRU tsis tuaj yeem kawm los ntawm qhov kev kwv yees zaum kawg raws li qhov sib xyaw ua ke. Yuav kom daws tau qhov teeb meem no, peb yuav txuas GRUs mus rau logistic regressions ntawm kauj ruam thib ob kom GRU kawm feature sawv cev los ntawm ntau hom thiab ib qho qauv. Tsis tas li ntawd, peb npaj yuav hloov kho cov qauv ntawm peb cov qauv ua rau nws ua tau rau tus kheej GRU Cheebtsam kom rho tawm cov yam ntxwv sib xyaw. Tam sim no peb tab tom tshawb nrhiav qhov muaj peev xwm ua qhov txuas ntxiv rau txoj haujlwm no.

to relieve AD

Cov txheej txheem

Rov ua dua Neural Network.

Recurrent Neural Network (RNN) yog ib chav kawm sib sib zog nqus kev kawm architecture siv thaum cov ntaub ntawv sib txuas tuaj yeem txiav txim siab. Hauv kev ua cov lus ntuj (NLP), kev paub txog kev hais lus, thiab kev kuaj pom tsis zoo nyob rau hauv lub sijhawm series, RNN tau nrov siv rau kev tshuaj xyuas cov kab lus ntawm cov lus thiab lub sijhawm series data28. Qhov zoo ntawm kev thov RNN yog qhov sib txawv-ntev ntu tuaj yeem ua tiav los siv cov qauv hauv lub cev zais

nyob rau hauv qhov muab ua ntu zus. Hauv kev tsom xam kev xav, piv txwv li, lub hom phiaj yog cais cov kev xav (zoo lossis phem) ntawm kab lus. Tus classifier yuav tsum coj ib kab lus (ib ntu ntawm cov lus) ua ib qho kev tawm tswv yim, nkag siab txog cov ntsiab lus hauv nws, thiab xa rov qab qhov kev xav kom raug raws li qhov tso tawm29. Rau txoj haujlwm twv ua ntej uas kuaj pom thawj qhov kev kuaj mob plawv tsis ua haujlwm hauv 30, RNN siv sijhawm ntawm cov ntaub ntawv kho mob hauv hluav taws xob (EHRs) siv 12 mus rau 18- lub hlis soj ntsuam lub qhov rais. Nyob rau hauv cov rooj plaub no qhov twg qhov sib txawv-ntev cov tswv yim yuav tsum tau hais nrog RNN yog tus neeg sib tw tsim nyog siv.

Table 5. Performance comparison between diferent models. P-values are calculated using a paired t-test between the proposed and each competing method.

Table 6. A list of previous models that train a classifer mainly using MCI samples.

Ib qho RNN ua tiav ib lub ntsiab lus ntawm cov lus qhia ib ntus ntawm ib lub sijhawm thiab hloov kho nws lub cim xeeb uas muaj cov ntaub ntawv hais txog keeb kwm ntawm tag nrho cov ntsiab lus yav dhau los ntawm ntu 31. Lub xeev zais yog sawv cev raws li Euclidean vector (piv txwv li, ib ntu ntawm cov lej tiag) thiab tau hloov kho recursively los ntawm cov tswv yim ntawm cov kauj ruam muab thiab tus nqi dhau los ntawm lub xeev zais (Fig. 5). Piv txwv tias peb muaj N tus naj npawb ntawm cov kev kawm, txhua qhov uas muaj ib ntus ... ... x xxx {,,,,, } nn hauv Tn 1 2 qhov twg xin yog cov ntaub ntawv sau tseg ntawm n-th qauv thiab t-cov khoom. nyob rau hauv ib theem zuj zus thiab T yog qhov ntev ntawm qhov sib lawv liag. Qhov sib thooj ntawm cov zis yog recursively xam raws li:

where ui is the i-th element of the vector u. In equation (2), we abuse notation to express elementwise application of the above expression.

Hauv peb cov qauv, qhov kawg tso zis ib ntus muab los ntawm RNN raug kho raws li qhov tshwm sim vector rau kev faib tawm, thiab kev ua haujlwm hla-enttropy poob (sib npaug (3)) yog siv los ntsuas seb "nyob deb" peb qhov kev kwv yees n-th yog los ntawm n-th hauv av qhov tseeb daim ntawv lo nyob rau hauv. Qhov ntawd yog, peb xaiv qhov zoo tshaj plaws tsis muaj ⁎ ⁎ ⁎ W, W, W hxy uas txo qhov cross-entropy poob ntawm cov ntaub ntawv muab (sib npaug (4)). Lub algorithm peb siv los txhim kho qhov tsis zoo yog Backpropagation Through Time (BPTT)32, uas hloov kho qhov hnyav hauv RNN kom txo tau qhov poob haujlwm. Txawm li cas los xij, thaum txoj haujlwm yuav tsum tau ua raws cov lus qhia ntev los ua tiav, kev cob qhia RNN yog qhov nyuaj33. Qhov no yog hu ua qhov teeb meem nyob ntev. Cov kev hloov pauv ntawm RNN xws li Long Short-Term Memory (LSTM) thiab Gated Recurrent Unit (GRU) tau tsim thiab siv los daws qhov teeb meem no34,35. Hauv cov qauv npaj, peb siv GRU rau txhua qhov qauv ntawm cov ntaub ntawv los ua ntau lub sijhawm cov ntsiab lus ntawm cov tswv yim. Cov qauv ncauj lus kom ntxaws ntawm GRU tau piav qhia hauv qhov ntxiv.

Multi-modal GRU rau MCI hloov dua siab tshiab twv ua ntej.

Peb qhov teeb meem tuaj yeem suav hais tias yog cov ntaub ntawv sib txuas ua ke. Lub hom phiaj ntawm kev faib tawm yog kwv yees seb tus neeg twg nrog MCI ntawm lub hauv paus tau hloov mus rau AD lossis tsis siv cov ntaub ntawv sib lawv liag, uas muaj plaub yam xws li kev txawj ntse, CSF, thiab MRI biomarkers nrog rau cov ntaub ntawv pej xeem. Txawm hais tias cov ntaub ntawv pej xeem thiab MRI biomarkers tsis yog cov ntaub ntawv ntev peb yuav suav tias yog cov ntaub ntawv ntev-ib ntu.

Txhawm rau siv GRU-raws li kev faib tawm algorithm rau peb qhov teeb meem, peb yuav tsum tsim tus qauv uas tuaj yeem suav nrog plaub tus qauv ntawm cov ntaub ntawv. Lub tswv yim tseem ceeb ntawm peb cov qauv yog cais tsim GRU feature extractors rau txhua tus qauv thiab sib xyaw ua ke ntawm plaub qhov tshwj xeeb vectors thaum kawg. Peb tus qauv yog suav nrog ob theem kev cob qhia: (1) kawm ib zaug GRU rau txhua qhov kev hloov pauv ntawm cov ntaub ntawv, thiab (2) kawm txog kev sib koom ua ke sawv cev los ua qhov kev kwv yees zaum kawg. Ntawm thawj kauj ruam kev cob qhia, ib qho GRU raug cob qhia cais rau txhua qhov kev hloov pauv uas lub hom phiaj kev faib tawm yog kwv yees hloov pauv mus rau AD los ntawm MCI. Siv GRUs yog qhov tseem ceeb los coj cov ntaub ntawv ntev thiab hloov lawv mus rau hauv vector loj. Qhov no zoo ib yam li txoj hauv kev uas tau hais tseg hauv 36 uas qhia cov tswv yim sib lawv liag rau hauv qhov ntev-ntev sawv cev. Hauv kauj ruam thib ob, MCI hloov dua siab tshiab yog kwv yees raws li plaub lub vectors tsim los ntawm txhua qhov GRU. Rau kev sib koom ua ke plaub-vectors, peb xaiv cov ntaub ntawv sib koom ua ke-raws li kev sib koom ua ke, uas yog lub tswv yim yooj yim tshaj plaws los muab ntau qhov chaw ntawm cov ntaub ntawv rau hauv ib qho vector37. Rau qhov kev twv ua ntej zaum kawg, l1-grularized logistic regression38 yog siv rau kev faib tawm ntawm MCI-C thiab MCI-NC. Lub ntsiab lus ntawm peb txoj kev npaj tau piav qhia hauv daim duab 6.

Xaus

Ntawm no, peb tau npaj ntau txoj kev kawm sib sib zog nqus los kawm txog kev kwv yees ntawm MCI rau AD hloov dua siab tshiab siv qhov kev paub ntev ntev thiab CSF biomarkers nrog rau cov kab ke hla ntu neuroimaging thiab cov ntaub ntawv pej xeem ntawm lub hauv paus. Peb tau siv ntau GRUs los siv cov ntaub ntawv longitudinal multi-domain thiab txhua yam kev kawm nrog txhua cov ntaub ntawv hloov pauv. Peb cov txiaj ntsig tau pom tias peb tau ua tiav qhov kev kwv yees qhov tseeb ntawm MCI rau AD hloov dua siab tshiab los ntawm kev sib txuas cov ntaub ntawv longitudinal multi-domain. Ib txoj hauv kev sib sib zog nqus ntau qhov kev kawm muaj peev xwm txheeb xyuas cov neeg muaj kev pheej hmoo ntawm kev txhim kho AD uas yuav tau txais txiaj ntsig zoo tshaj plaws los ntawm kev sim tshuaj lossis raws li txoj hauv kev stratification hauv kev sim tshuaj.

Cistanche can prevent and treat Alzheimer's disease, click here for sample

Kuv tuaj yeem yuav cistanche ntxiv qhov twgtiv thaiv thiab khoAlzheimer tus kab mob, nyem qhov no rau ib qho qauv

Cov ntaub ntawv

1. Alzheimer's, A. 2015 Alzheimer's kab mob tseeb thiab cov duab. Alzheimer's Dement 11, 332–384 (2015).

2. Albert, MS et al. Te diagnitive impairment vim Alzheimer's disease: cov lus pom zoo los ntawm National Institute on Aging-Alzheimer's Association workgroups on diagnostic guide for Alzheimer's disease. Alzheimer's Dement 7, 270–279

3. Jack, CR Jr. et al. Qhia txog cov lus pom zoo los ntawm National Institute on Aging-Alzheimer's Association workgroups on diagnostic guide for Alzheimer's disease. Alzheimer's Dement 7, 257–262,

4. Sperling, RA et al. Mus rau kev txhais cov theem preclinical ntawm Alzheimer's tus kab mob: cov lus pom zoo los ntawm National Institute on Aging-Alzheimer's Association workgroups txog cov txheej txheem kuaj mob rau Alzheimer's disease. Alzheimer's Dement 7, 280–292.

5. Petersen, RC et al. Kev puas hlwb me me: kev kho tus cwj pwm thiab qhov tshwm sim. Archives ntawm neurology 56, 303–308 (1999).

6. Tábuas-Pereira, M. et al. Prognosis of Early-Onset vs. Late-Onset Mild Cognitive Impairment: Sib piv ntawm Tus Nqi Hloov Pauv thiab Nws cov kwv yees. Geriatrics 1, 11 (2016).

7. Zhang, D., Shen, D. & Alzheimer's Disease Neuroimaging, I. Multi-modal multi-task kev kawm rau kev sib koom ua ke ntawm ntau yam regression thiab kev faib tawm txawv txav hauv Alzheimer's disease. Neuroimage 59, 895–907, ib.

8. Cheng, B., Liu, M., Zhang, D., Munsell, BC & Shen, D. Domain Transfer Learning for MCI Conversion Prediction. IEEE Trans Biomed Eng 62, 1805–1817.

9. Zhang, D., Shen, D. & Initiative, ASDN Predicting yav tom ntej soj ntsuam kev hloov ntawm MCI cov neeg mob siv longitudinal thiab multimodal biomarkers. PloS one 7, e33182 (2012).

10. Nho, K. et al. Tsis Siv Neeg Prediction of Conversion los ntawm Me Me Kev Paub Txog Kev Tsis Txaus Siab rau Kev Mob Alzheimer's Disease siv cov qauv sib nqus resonance imaging. AMIA Annu Symp Proc 2010, 542–546 (2010).

11. Wee, CY, Yap, PT, Shen, D. & Alzheimer's Disease Neuroimaging, I. Kev kwv yees ntawm Alzheimer's kab mob thiab kev txawj ntse me me uas siv cov qauv cortical morphological. Hum Brain Mapp 34, 3411–3425,

12. Wolz, R. et al. Multi-txoj kev tsom xam ntawm MRI cov duab nyob rau hauv thaum ntxov diagnostics ntawm Alzheimer tus kab mob. PLoS Ib 6, e25446

13. Cho, Y., Seong, JK, Jeong, Y., Shin, SY & Alzheimer's Disease Neuroimaging, I. Kev faib tawm ntawm tus kheej rau Alzheimer's tus kab mob raws li kev kawm nce ntxiv uas siv qhov ntau zaus sawv cev ntawm cov ntaub ntawv cortical thickness. Neuroimage 59, 2217–2230, https://doi.org/10.1016/j.neuroimage.2011.09.085 (2012).

14. Kim, D. et al. Ib daim duab-raws li kev sib koom ua ke ntawm Multimodal Brain Imaging Cov Ntaub Ntawv rau Kev Tshawb Fawb Txog Kev Tsis Txaus Siab Thaum Ntxov Me Me (E-MCI). Multimodal Brain Image Anal (2013) 8159, 159–169,

15. Ewers, M. et al. Kev kwv yees ntawm kev hloov dua siab tshiab los ntawm kev puas siab puas ntsws me me mus rau Alzheimer's disease dementia raws li biomarkers thiab kev kuaj mob neuropsychological. Neurobiol Hnub nyoog 33, 1203–1214,

(2012).

16. Heister, D. et al. Kev kwv yees MCI cov txiaj ntsig nrog kev kho mob MRI thiab CSF biomarkers. Neurology 77, 1619–1628

17. Saykin, AJ et al. Kev tshawb nrhiav caj ces ntawm ntau MCI thiab AD phenotypes hauv ADNI: Kev nce qib, cov hauv kev, thiab cov phiaj xwm. Alzheimer's & dementia: phau ntawv journal ntawm Alzheimer's Association 11, 792-814 (2015).

18. Saykin, AJ et al. Alzheimer's Disease Neuroimaging Initiative biomarkers raws li phenotypes ntau: Genetics core aims, kev nce qib, thiab cov phiaj xwm. Alzheimer's & dementia: Phau ntawv Journal ntawm Alzheimer's Association 6, 265–273 (2010).

19. Nho, K. et al. Voxel thiab nto-raws li topography ntawm lub cim xeeb thiab cov thawj coj tsis txaus nyob rau hauv me me kev txawj ntse impairment thiab Alzheimer's kab mob. Brain imaging thiab cwj pwm 6, 551–567 (2012).

20. Falahati, F., Westman, E. & Simmons, A. Multivariate data analysis and machine learning in Alzheimer's disease with a focus on structural magnetic resonance imaging. J Alzheimers Dis 41, 685–708.

Yuav Ua Li Cas Saib Xyuas Cov Neeg Laus Nrog Alzheimer's Disease?

Yuav ua li cas kom paub tseeb tias koj muaj Alzheimer's Disease?