Ukusetyenziswa kwe-MemTrax kunye neModeli yoMshini wokuFunda kuHlelo lweNgcaciso yeMild Cognitive Impairment

Inqaku loPhando

Ababhali: Bergeron, Michael F. | Indawo, uSara | Zhou, Xianbo | Ding, Tao | Khoshgoftaar, Taghi M. | Zhao, Feng | Du, Bo | Chen, Xinjie | Wang, Xuan | Zhong, Lianmei | Liu, Xiaolei| Ashford, J. Wesson

I-DOI: 10.3233/JAD-191340

Ijenali: Ijenali ye Izifo ze-Alzheimer's, iv. 77, hayi. I-4, iphe. 1545-1558, 2020

Abstract

imvelaphi:

Ukuxhaphaka kweziganeko kunye nokuxhaphaka kwe Isifo se-Alzheimer kunye nokukhubazeka okuncinci kwengqondo (MCI) kubangele umnxeba ophuthumayo wophando ukuze kuqinisekiswe ukuxilongwa kwangaphambili kwengqondo kunye novavanyo.

Injongo:

Eyona njongo yethu yophando yayikukujonga ukuba ngaba iimetrics zeMemTrax ezikhethiweyo kunye neendawo ezifanelekileyo kunye neempawu zeprofayili yezempilo zingasetyenziswa ngokufanelekileyo kwiimodeli eziqikelelwayo eziphuhliswe ngokufunda ngomatshini ukwahlula impilo yengqondo (eqhelekileyo ngokuchasene ne-MCI), njengoko kuya kuboniswa ngu Montreal Cognitive Assessment (MoCA).

Iindlela:

Senze uphononongo olunqamlezileyo kwi-neurology engama-259, iklinikhi yenkumbulo, kunye nezigulana zabantu abadala zangaphakathi eziqeshwe kwababini. kwizibhedlele eTshayina. Isigulana ngasinye sanikwa i-MoCA yolwimi lwesiTshayina kwaye yazilawula ngokuqhubekayo ukuqatshelwa kwe-MemTrax kwi-intanethi ye-episodic. uvavanyo lwememori kwi-intanethi kwangolo suku. Iimodeli zokwahlula kwangaphambili zakhiwe kusetyenziswa umatshini wokufunda kunye nokuqinisekiswa kwe-10-fold cross, kunye nokusebenza kwemodeli kulinganiswe ngokusebenzisa iNdawo ePhantsi kwe-Receiver Operating Characteristic Curve (AUC). Iimodeli zakhiwe kusetyenziswa iimetriki ezimbini zokusebenza kwe-MemTrax (ipesenti echanekileyo, ixesha lokuphendula), kunye neempawu ezisibhozo eziqhelekileyo kunye nembali yomntu.

iziphumo:

Xa kuthelekiswa abafundi kuzo zonke iindibaniselwano ezikhethiweyo zamanqaku e-MoCA kunye nee-thresholds, i-Naïve Bayes idla ngokuba ngoyena mfundi ugqwesileyo ngokuphumelela kuhlelo lulonke luka-0.9093. Ngaphezulu, phakathi kwabafundi abathathu abaphezulu, ukusebenza kokuhlelwa okusekelwe kwi-MemTrax kukonke bekugqwesile kusetyenziswa iimpawu ezine ezikudidi oluphezulu (0.9119) xa kuthelekiswa nokusebenzisa zonke iimpawu ezili-10 eziqhelekileyo (0.8999).

Isiphelo:

Ukusebenza kweMemTrax kunokusetyenziswa ngokufanelekileyo kwimodeli yolwahlulo lokufunda ngomatshini isicelo sokuhlola ukufumanisa ukuphazamiseka kwengqondo kwangoko.

INTSHAYELELO

Iziganeko ezivunyiweyo (nangona zingafunyaniswanga ngaphantsi) ezixhaphake ngokubanzi kunye nokuxhaphaka kunye nokunyuka okuhambelanayo kwezonyango, intlalo, kunye noluntu impilo iindleko kunye nomthwalo wesifo se-Alzheimer's (AD) kunye nokuphazamiseka kwengqondo okuncinci (MCI) kuya kuba nzima ngakumbi kubo bonke abachaphazelekayo [1, 2]. Le meko icinezelayo kunye neyokuqhatha iye yabangela ukuba kubizwe ngokungxamisekileyo ukuba uphando luqinisekise ukufumanisa kwangoko Uvavanyo lwengqondo kunye nezixhobo zovavanyo zokusetyenziswa rhoqo kwiseto yobuqu kunye nekliniki kwizigulana ezikhulileyo kwimimandla eyahlukeneyo kunye nabemi [3]. Ezi zixhobo kufuneka zibonelele ngoguqulelo olungenamthungo lweziphumo ezifundisayo kwiirekhodi zempilo zombane. Izibonelelo ziya kufezekiswa ngokwazisa izigulane kunye nokuncedisa oogqirha ekuboneni utshintsho olubalulekileyo ngaphambili kwaye ngaloo ndlela lwenza ukuba i-stratification ngokukhawuleza kwaye ifike ngexesha, ukuphunyezwa, kunye nokulandelela unyango olufanelekileyo kunye nonyango olusebenzayo kunye nokunyamekela kwabo baqala ukufumana amava. ukuhla kwengqondo [3, 4].

Isixhobo sekhompyutha seMemTrax (https://memtrax.com) luvavanyo olulula nolufutshane oluqhubekayo lokuqaphela olunokuthi luzilawule ngokwalo kwi-intanethi ukulinganisa ukusebenza kwememori ye-episodic enzima apho umsebenzisi aphendule kwimifanekiso ephindaphindiweyo kwaye kungekhona kwintetho yokuqala [5, 6]. Uphando lwakutsha nje kunye neziphumo zeziphumo ezisebenzayo ziqala ukuqhubela phambili kwaye ngokudibeneyo ukubonisa ukusebenza kwekliniki ye-MemTrax ekuqaleni kwe-AD kunye nokuhlolwa kwe-MCI [5-7]. Nangona kunjalo, uthelekiso oluthe ngqo lwenkonzo yeklinikhi kwizinto ezikhoyo impilo yokuqonda Uvavanyo kunye nemigangatho eqhelekileyo iqinisekisiwe ukwazisa imbono yobuchwephesha kunye nokuqinisekisa usetyenziso lweMemTrax ekubhaqweni kwangaphambili kunye nenkxaso yoxilongo. van der Hoek et al. [8] uthelekisa iimetrics zokusebenza ezikhethiweyo zeMemTrax (isantya sokuphendula kunye nepesenti echanekileyo) kwisimo sokuqonda njengoko kumiselwe yiMontreal. Uvavanyo lokuqonda (MoCA). Nangona kunjalo, olu phononongo lwalukhawulelwe ekunxulumaniseni ezi metrics zokusebenza kunye nophawu lwesimo sokuqonda (njengoko kumiselwe yi-MoCA) kunye nokuchaza uluhlu oluhambelanayo kunye namaxabiso anqanyuliweyo. Ngokufanelekileyo, ukwandisa kolu phando kunye nokuphucula ukusebenza kokuhlelwa kunye nokusebenza kakuhle, umbuzo wethu ophambili wophando waba:

  • Ngaba umntu okhethiweyo weMemTrax iimetriki zokusebenza kunye neendawo ezifanelekileyo kunye nempilo profile Iimpawu mazisetyenziswe ngokufanelekileyo kwimodeli eqikelelwayo ephuhliswe ngokufunda komatshini ukwahlula impilo yengqondo ngokudibeneyo (eqhelekileyo ngokuchasene ne-MCI), njengoko kuya kuboniswa ngamanqaku e-MoCA?

Okwesibini koku, besifuna ukwazi:

  • Kubandakanywa iimpawu ezifanayo, ngaba imodeli yokufunda yomatshini esekelwe kwi-MemTrax inokusetyenziswa ngokufanelekileyo kwisigulane ukuqikelela ubukhali (obuncinci ngokuchasene nobunzima) phakathi kweendidi ezikhethiweyo zokuphazamiseka kwengqondo njengoko kuya kugqitywa ngoxilongo oluzimeleyo lwekliniki?

Ukufika kunye nokusetyenziswa okusebenzayo okusebenzayo kobukrelekrele bokwenziwa kunye nokufunda koomatshini ekuhlolweni / ukubhaqwa sele kubonakalise iingenelo ezisebenzayo ezicacileyo, kunye nemodeli eqikelelwayo ekhokela oogqirha ngokufanelekileyo kuvavanyo olucelomngeni lwengqondo / yengqondo kunye nolawulo lwesigulane. Kuphononongo lwethu, sikhethe indlela efanayo kwimodeli yokuhlelwa kwe-MCI kunye nocalucalulo oluqatha lokuphazamiseka kwengqondo njengoko kungqinwa luxilongo lwezonyango olusuka kwiiseti ezintathu ezimele izigulane ezikhethiweyo zokuzithandela kunye nezigulana eziphuma kwizibhedlele ezibini zaseTshayina. Sisebenzisa imodeli eqikelelwayo yokufunda ngomatshini, sichonge abafundi abagqwesileyo kwiidathasethi/iintlanganisela zabafundi kwaye sabeka amanqaku okusikhokela ekuchazeni eyona modeli isebenzayo yeklinikhi.

Ingqikelelo yethu ibiyeyokuba imodeli esekwe kwiMemTrax eqinisekisiweyo ingasetyenziselwa ukwahlula impilo yengqondo ngokwe-dichotomously (eqhelekileyo okanye i-MCI) ngokusekwe kwikhrayitheriya yamanqaku e-MoCA, kwaye imodeli efanayo ye-MemTrax yokuxela kwangaphambili ingasetyenziswa ngempumelelo ekucaluleni ubungqongqo kwiindidi ezikhethiweyo ze ufunyaniswe ngonyango ukukhubazeka kwengqondo. Ukubonisa iziphumo ezilindelekileyo kuya kuba luncedo ekuxhaseni ukusebenza kwe-MemTrax njengesikrini sokubona kwangaphambili ukuhla kwengqondo kunye nokuhlelwa kokuphazamiseka kwengqondo. Uthelekiso oluncomekayo kushishino ekucingelwa ukuba lukumgangatho ohambelana nokulula okukhulu kunye nokukhawuleza kokusebenza kuya kuba nefuthe ekuncedeni oogqirha bamkele esi sixhobo silula, sithembekileyo, kunye nesifikelelekayo njengesikrini sokuqala ekuboneni kwangethuba (kubandakanya iprodromal) ukusilela kwengqondo. Indlela enjalo yokusebenza kunye luncedo inokukhuthaza ngakumbi ukhathalelo lwesigulane kunye nongenelelo olungcono kwangexesha kunye nolwahlulo olungcono. Ezi mbono zokucinga phambili kunye neemetrics eziphuculweyo kunye neemodeli nazo zinokuba luncedo ekunciphiseni okanye ekumiseni ukuqhubela phambili kwengqondo, kuquka i-AD kunye ne-AD-related dementias (ADRD).

IMPAHLA NENKQUBO

Inani labafundi

Phakathi kukaJanuwari 2018 kunye no-Agasti 2019, uphando olunqamlezileyo lwagqitywa kwizigulana eziqeshwe kwizibhedlele ezibini zaseChina. Ulawulo lwe-MemTrax [5] kubantu abaneminyaka eyi-21 nangaphezulu kwaye ukuqokelelwa kunye nocazululo lwezo datha zaphononongwa kwaye zamkelwa kwaye zalawulwa ngokuhambelana nemigangatho yeenqobo zokuziphatha. umntu IKomiti yoKhuseleko lweSifundo yeYunivesithi yaseStanford. I-MemTrax kunye nazo zonke ezinye iimvavanyo zolu phononongo lulonke lwenziwa ngokwe-Helsinki declaration ye-1975 kwaye ivunyiwe yiBhodi yokuHlola yeZiko leSibhedlele esiBambiseneyo sokuQala seYunivesithi yezoNyango yaseKunming, eYunnan, eChina. Umsebenzisi ngamnye unikwe i imvume yokwazi ukuba ufunde/uphonononge kwaye uvume ngokuzithandela ukuthatha inxaxheba.

Abathathi-nxaxheba baqeshwe kwi-pool yezigulane ezingaphandle kwiklinikhi ye-neurology kwisibhedlele saseYanhua (YH sub-dataset) kunye ne iklinikhi yenkumbulo kwiSibhedlele sokuQala esiManyeneyo saseKunming Medical IYunivesithi (i-XL sub-dataset) eBeijing, eChina. Abathathi-nxaxheba baphinde baqeshwe kwi-neurology (i-XL sub-dataset) kunye neyeza zangaphakathi (i-KM sub-dataset) izigulane kwiSibhedlele esiBambiseneyo sokuQala saseKunming Medical University. Iikhrayitheriya zokubandakanywa zibandakanya i-1) amadoda kunye nabasetyhini ubuncinane iminyaka eyi-21 ubudala, i-2) ukukwazi ukuthetha isiTshayina (isiMandarin), kunye ne-3) ukukwazi ukuqonda izikhokelo zomlomo kunye nezibhaliweyo. Indlela yokukhutshelwa ngaphandle ibingumbono kunye nokonakala kwemoto kuthintele abathathi-nxaxheba ukuba bagqibezele Uvavanyo lweMemTrax, kunye nokungakwazi ukuqonda imiyalelo ethile yovavanyo.

Inguqulelo yesiTshayina yeMemTrax

I-Intanethi Iqonga lovavanyo leMemTrax laguqulelwa kwisiTshayina (i-URL: https://www.memtrax.com.cn) kwaye ilungelelaniswe ngakumbi ukuba isetyenziswe nge-WeChat (i-Shenzhen Tencent Computer Systems Co. LTD., i-Shenzhen, i-Guangdong, i-China) ukuze uzilawule. Idatha igcinwe kwi-server yefu (i-Ali Cloud) ehlala e-China kwaye inikwe ilayisenisi esuka e-Alibaba (Alibaba Technology Co. Ltd., Hangzhou, Zhejiang, China) yi-SJN Biomed LTD (Kunming, Yunnan, China). Iinkcukacha ezithe ngqo kwi-MemTrax kunye neendlela zokuqinisekisa ukuvavanya ezisetyenziswe apha zichazwe ngaphambili [6]. Uvavanyo lwanikezelwa ngaphandle kwentlawulo kwizigulana.

Inkqubo yokufunda

Kwizigulane ezilaliswayo kunye nezigulana zangaphandle, iphepha lemibuzo eliqhelekileyo lokuqokelela ulwazi lwabantu kunye nolwazi lomntu olufana nobudala, isondo, iminyaka yemfundo, umsebenzi, uhlala wedwa okanye nosapho, kwaye imbali yezonyango yayilawulwa lilungu leqela lophononongo. Emva kokugqitywa koluhlu lwemibuzo, iimvavanyo ze-MoCA [12] kunye ne-MemTrax zaqhutywa (i-MoCA kuqala) ngaphandle kwemizuzu engama-20 phakathi kweemvavanyo. Ipesenti ye-MemTrax echanekileyo (MTx-% C), ixesha lokuphendula elichanekileyo (MTx-RT), kunye nomhla kunye nexesha lovavanyo zirekhodwe kwiphepha lilungu leqela lofundo-nzulu kumthathi-nxaxheba ngamnye ovavanyiweyo. Ikhweshine ezaliswe ngokupheleleyo kunye neziphumo ze-MoCA zafakwa kwi-Excel spreadsheet ngumphandi owayeqhuba iimvavanyo waza wangqinwa ngugxa wakhe phambi kokuba iifayili ze-Excel zigcinwe ukuze zihlalutywe.

Uvavanyo lweMemTrax

Uvavanyo lwe-intanethi ye-MemTrax lubandakanya imifanekiso ye-50 (i-25 eyodwa kunye ne-25 ephindaphindiweyo; iiseti ezi-5 zemifanekiso ye-5 yemifanekiso eqhelekileyo okanye izinto) eziboniswe kwi-pseudo-random order. Umthathi-nxaxheba uya (ngokwemiyalelo) ukucofa iqhosha lokuQalisa kwisikrini ukuqalisa uvavanyo kwaye aqale ukujonga uthotho lwemifanekiso kwaye kwakhona uchukumise umfanekiso okwisikrini ngokukhawuleza kangangoko xa kuvela umfanekiso ophindaphindiweyo. Umfanekiso ngamnye uvele kangangemizuzu emi-3 okanye de umfanekiso okwisikrini uchukunyiswe, nto leyo ebangele ukuba umfanekiso olandelayo uboniswe ngoko nangoko. Ukusebenzisa iwotshi yangaphakathi yesixhobo sendawo, iMTx-RT yomfanekiso ngamnye igqitywe lixesha eligqithileyo ukusuka ekubonisweni komfanekiso ukuya kuthi ga xa isikrini sichukunyiswe ngumthathi-nxaxheba ekuphenduleni ekuboniseni ukuqondwa komfanekiso njengalowo sele ubonisiwe. ngexesha lovavanyo. I-MTx-RT yarekhodwa kuwo wonke umfanekiso, kunye ne-3 epheleleyo erekhodiweyo ebonisa ukuba akukho mpendulo. I-MTx-% C ibalwe ukubonisa ipesenti yokuphinda kunye nemifanekiso yokuqala apho umsebenzisi aphendule ngokuchanekileyo (inyani yokwenene + i-negative yokwenyani eyahlulwe ngama-50). Iinkcukacha ezongezelelweyo zolawulo lwe-MemTrax kunye nokuphunyezwa, ukunciphisa idatha, idatha engavumelekanga okanye "akukho mpendulo", kunye nohlalutyo oluphambili lwedatha luchazwe kwenye indawo [6].

Uvavanyo lwe-MemTrax luchazwe ngokweenkcukacha kwaye uvavanyo lokuziqhelanisa (kunye nemifanekiso ekhethekileyo ngaphandle kwaleyo isetyenziswe kuvavanyo lokurekhoda iziphumo) yanikezelwa kubathathi-nxaxheba kwisimo esibhedlele. Abathathi-nxaxheba kwi-YH kunye ne-KM-sub-datasets bathatha uvavanyo lwe-MemTrax kwi-smartphone eyayilayishwe ngesicelo kwi-WeChat; ngelixa inani elilinganiselweyo le-XL sub-dataset yezigulane zisebenzisa i-iPad kwaye abanye basebenzisa i-smartphone. Bonke abathathi-nxaxheba bathatha uvavanyo lwe-MemTrax kunye nomphandi wophononongo ngokujonga ngokungathandabuzekiyo.

Uvavanyo lokuqonda lwaseMontreal

I-Beijing version ye-MoCA yaseTshayina (i-MoCA-BC) [13] yalawulwa kwaye yafumana amanqaku ngabaphandi abaqeqeshiweyo ngokwemiyalelo yovavanyo olusemthethweni. Ngokufanelekileyo, i-MoCA-BC ibonakaliswe njengethembekile uvavanyo lwengqondo ukuhlolwa kuwo onke amanqanaba emfundo kubantu abadala baseTshayina [14]. Uvavanyo ngalunye luthathe malunga nemizuzu eli-10 ukuya kwengama-30 ukulawulwa ngokusekelwe kubuchule bokuqonda bomthathi-nxaxheba.

Imodeli yohlelo lwe-MoCA

Bekukho izinto ezisebenzisekayo ezingama-29, kubandakanywa neeMemTrax ezimbini iimetrics zokuvavanya ukusebenza kunye neempawu ezingama-27 ezinxulumene nokubalwa kwabantu kunye nempilo ulwazi lomthathi-nxaxheba ngamnye. Isigulana ngasinye sovavanyo lwe-aggregate ye-MoCA sisetyenziswe njenge ukuhlolwa kwengqondo "ibhenchmark" yokuqeqesha iimodeli zethu zokuqikelela. Ngokufanelekileyo, ngenxa yokuba i-MoCA yayisetyenziselwa ukwenza ileyibhile yeklasi, asikwazanga ukusebenzisa amanqaku adibeneyo (okanye nawaphi na amanqaku e-moCA subset scores) njengento ezimeleyo. Senze imifuniselo yokuqala apho senza imodeli (ukuhlelwa kwempilo yengqondo echazwe yi-MoCA) i(ii)izibhedlele ezithathu zokuqala/iikliniki ezisezantsi zolwazi ngamnye kwaye emva koko zidityaniswe kusetyenziswa zonke iimpawu. Nangona kunjalo, zonke iinkcukacha ezifanayo azizange ziqokelelwe kwikliniki nganye kwezine ezimele i-sub-datasets ezintathu; ngoko, uninzi lweempawu zethu kwidathasethi edityanisiweyo (xa usebenzisa zonke iimpawu) zineziganeko eziphezulu zamaxabiso angekhoyo. Emva koko sakhe imifuziselo enedatha edityanisiweyo sisebenzisa izinto eziqhelekileyo kuphela ezikhokelele ekuphuculweni kokusebenza kokuhlelwa. Oku kusenokwenzeka ukuba kwacaciswa ngendibaniselwano yokuba nemizekelo emininzi yokusebenza kunye nokudibanisa iiseti zedatha yesigulane ezithathu kwaye akukho zimpawu zokuxhaphaka okungafanelekanga kwamaxabiso alahlekileyo (inye kuphela into kwidathasethi edibeneyo, uhlobo lomsebenzi, olunamaxabiso alahlekileyo, achaphazelayo. Izihlandlo ezithathu kuphela zezigulane), kuba ziimpawu eziqhelekileyo ezirekhodwe kuzo zontathu iziza eziqukiweyo. Okuqaphelekayo, besingenawo umqathango othile wokukhatywa kwenqaku ngalinye elingakhange liqukwe kwidatha edityanisiweyo. Nangona kunjalo, kwimodeli yethu yedatha edibeneyo edibeneyo, saqala ukusebenzisa zonke iimpawu ukusuka kwisigulana ngasinye se-sub-datasets ezintathu ezahlukeneyo. Oku kukhokelele ngokubanzi kwindlela yokusebenza ebisezantsi ngokulinganisekayo kunomfuziselo wokuqala kwiseti yedatha engaphantsi nganye. Ngaphezu koko, ngelixa ukusebenza kokuhlelwa kweemodeli ezakhiwe kusetyenziswa zonke iimpawu bezikhuthaza, kubo bonke abafundi kunye nezikim zokuhlela, ukusebenza kuphuculwe ngokuphindwe kabini kwiimodeli ezininzi xa kusetyenziswa iimpawu ezifanayo kuphela. Enyanisweni, phakathi kwezinto eziye zaphelela ekubeni ngabafundi bethu abaphezulu, zonke ngaphandle kwemodeli enye ziphuculwe ekupheliseni iimpawu ezingaqhelekanga.

I-dataset yokugqibela ehlanganisiweyo (i-YH, i-XL, kunye ne-KM edibeneyo) iquka iimeko ze-259, nganye imele umthathi-nxaxheba okhethekileyo owathatha zombini iimvavanyo ze-MemTrax kunye ne-MoCA. Kwakukho iimpawu ezizimeleyo ezi-10 ezabelwana ngazo: Iimethrikhi zokusebenza kwe-MemTrax: MTx-% C kunye ne-MTx-RT; ulwazi lwedemografi kunye nembali yonyango: iminyaka, isondo, iminyaka yemfundo, uhlobo lomsebenzi (ikhola eluhlaza okwesibhakabhaka / ikhola emhlophe), inkxaso yentlalontle (nokuba ngaba umntu ovavanyayo uhlala yedwa okanye nosapho), kwaye ewe/hayi iimpendulo zokuba ngaba umsebenzisi imbali yesifo seswekile, i-hyperlipidemia, okanye ukwenzakala kwengqondo. Iimethrikhi ezimbini ezongezelelweyo, i-aggregate score ye-MoCA kunye ne-aggregate score ye-MoCA elungelelaniswe iminyaka yemfundo [12], isetyenziswe ngokwahlukileyo ukuphuhlisa iilebhile zokuhlela ezixhomekeke, ngaloo ndlela kudala izikimu ezimbini ezihlukeneyo zemodeli eziza kusetyenziswa kwidathasethi yethu edibeneyo. Kwinguqulelo nganye (elungisiweyo kwaye engalungiswanga) yamanqaku e-MoCA, idatha yaphinda yahlulwa ngokwahlukileyo kwi-binary classification kusetyenziswa imiqobo emibini yekhrayitheriya-eyokuqala enconyiweyo [12] kunye nexabiso elinye elisetyenziswe kwaye liphakanyiswe ngabanye [8, 15]. Kwisinye isikimu sokuhlelwa kwe-threshold, isigulane sasithathwa njengengqondo yengqondo eqhelekileyo ukuba ifumene i-≥23 kuvavanyo lwe-MoCA kunye nokuba ne-MCI ukuba amanqaku angama-22 okanye aphantsi; kanti, kwifomathi yohlelo olucetyiswayo lokuqala, isigulana kwafuneka sifumane amanqaku angama-26 okanye angcono kwi-MoCA ukuze ibhalwe njengempilo yengqondo eqhelekileyo.

Idatha ehluziweyo yokwenziwa kolwahlulo lwe-MoCA

Siphinde savavanya ulwahlulo lwe-MoCA sisebenzisa iindlela ezine ezisetyenziswa ngokuqhelekileyo zokubeka amanqaku: i-Chi-Squared, iGain Ratio, iNgxowa-mali yoLwazi, kunye nokungaqiniseki kweSymmetrical. Ngombono wethutyana, sisebenzise iirenki kuyo yonke idatha edityanisiweyo sisebenzisa isikimu sethu ngasinye kwezine. Bonke amanqanaba bavumelana ngeempawu eziphezulu ezifanayo, oko kukuthi, ubudala, inani leminyaka yemfundo, kunye neemetriki zombini zeMemTrax (MTx-% C, ithetha iMTx-RT). Emva koko sakha kwakhona iimodeli sisebenzisa ubuchule bokhetho ngalunye ukuqeqesha iimodeli kwiimpawu ezine eziphezulu kuphela (bona Ukukhetha amanqaku ngezantsi).

Iziphumo ezokugqibela ezisibhozo ezahlukeneyo zenkqubo yokuhlela amanqaku e-MoCA zithiwe thaca kwiThebhile yoku-1.

Itheyibhile 1

Isishwankathelo seenguqu zesikimu somzekelo ezisetyenziselwa ukuhlelwa kwe-MoCA (Eqhelekileyo Impilo yengqondo ngokuchasene neMCI)

iModeling SchemeImpilo yeNgqondo eqhelekileyo (Udidi olubi)I-MCI (iKlasi ePositive)
Uhlengahlengiso-23 Aluhluzwanga/Aluhluzwanga101 (39.0%)158 (61.0%)
Uhlengahlengiso-26 Aluhluzwanga/Aluhluzwanga49 (18.9%)210 (81.1%)
Ingalungiswanga-23 Ayihluzwanga/Ayihluzwanga92 (35.5%)167 (64.5%)
Ingalungiswanga-26 Ayihluzwanga/Ayihluzwanga42 (16.2%)217 (83.8%)

Inani elichaphazelekayo kunye neepesenti zezigulane ezipheleleyo kwiklasi nganye ziyahlukana ngohlengahlengiso lwamanqaku emfundo (Uhlengahlengiso okanye olungalungiswanga) kunye nomda wokuhlela (i-23 okanye i-26), njengoko isetyenziswe kuzo zombini iisethi zeempawu (Azihlungwanga kwaye ziFifa).

Imodeli yovavanyo lweklinikhi esekwe kwiMemTrax

Kwiiseti zethu ezintathu zantlandlolo (YH, XL, KM), kuphela izigulane ze-XL ze-sub-dataset ziye zafunyaniswa ngokuzimeleyo ngokukhubazeka kwengqondo (okt, amanqaku abo e-MoCA awazange asetyenziswe ekusekeni ukuhlelwa kwesiqhelo ngokuchasene nokukhubazeka). Ngokukodwa, izigulana ze-XL zafunyaniswa zinazo Uvavanyo lwesifo sika-Alzheimer (AD) okanye i-vascular dementia (VaD). Ngaphakathi kwezi ndidi zoxilongo oluphambili, bekukho enye inkcazo ye-MCI. Ukuxilongwa kwe-MCI, i-dementia, i-vascular neurocognitive disorder, kunye ne-neurocognitive disorder ngenxa ye-AD yayisekelwe kwiinkqubo zokuxilonga ezicacileyo kunye nezahlukileyo ezichazwe kwi-Diagnostic and Statistical Manual of Mental Disorders: DSM-5 [16]. Ukuqwalasela olu xilongo olucokisekileyo, izikimu ezimbini zokumisela imodeli zisetyenziswe ngokwahlukileyo kwi-sub-dataset ye-XL ukwahlula inqanaba lobunzima (iqondo lokukhubazeka) kwicandelo ngalinye lokuxilongwa kweprayimari. Iinkcukacha ezisetyenzisiweyo kuzo zonke ezi zicwangciso zokuxilonga i-diagnostic (AD kunye ne-VaD) zibandakanya ulwazi lwabantu kunye nolwazi lwembali yesigulane, kunye nokusebenza kwe-MemTrax (MTx-% C, ithetha i-MTx-RT). Uxilongo ngalunye luphawulwe ngobumnene ukuba luchongiwe i-MCI; kungenjalo, kwakugqalwa njengento eqatha. Siqale saqwalasela ukubandakanya amanqaku e-MoCA kwiimodeli zokuxilongwa (ezithambileyo ngokuchasene nokuqina); kodwa sizimisele ukuba oko kuya koyisa injongo yesikimu sethu somzekelo wesibini. Apha abafundi baya kuqeqeshwa kusetyenziswa ezinye iimpawu zesigulana ezifumaneka lula kumnikezeli kunye neemetrics zokusebenza zovavanyo olulula lwe-MemTrax (endaweni ye-MoCA) ngokuchasene nereferensi "umgangatho wegolide", ukuxilongwa kwekliniki okuzimeleyo. Kwakukho iimeko ezingama-69 kwidatha yokuxilongwa kwe-AD kunye neemeko ezingama-76 ze-VaD (Itheyibhile 2). Kuzo zombini iiseti zedatha, bekukho iimpawu ezizimeleyo ezili-12. Ukongeza kwiimpawu ze-10 ezibandakanyiweyo kwi-classification yamanqaku e-MoCA, imbali yesigulane nayo iquka ulwazi kwimbali yoxinzelelo lwegazi kunye ne-stroke.

Itheyibhile 2

Isishwankathelo seenguqu zesikimu somzekelo ezisetyenziselwa ulwahlulo lobungqongqo boxilongo (Buphakathi ngokuchasene noBubi)

iModeling SchemeLuphakathi (Udidi olubi)Kakhulu (Udidi Olulungileyo)
I-MCI-AD ngokuchasene ne-AD12 (17.4%)57 (82.6%)
I-MCI-VaD ngokuchasene neVaD38 (50.0%)38 (50.0%)

Inani elichaphazelekayo kunye neepesenti zezigulane ezipheleleyo kwiklasi nganye zihlukaniswe ngoluhlu oluphambili lokuxilongwa (AD okanye i-VaD).

Statistics

Ukuthelekiswa kweempawu zabathathi-nxaxheba kunye nezinye iimpawu zamanani phakathi kwee-sub-datasets kwisicwangciso sokuhlelwa kwemodeli nganye (ukuqikelela impilo yengqondo ye-MoCA kunye nobunzima bokuxilongwa) kwenziwa ngokusebenzisa ulwimi lweprogram yePython (uguqulelo 2.7.1) [17]. Ukwahluka kwentsebenzo yemodeli ekuqaleni kunqunywe usebenzisa enye-okanye into emibini (njengoko kufanelekile) i-ANOVA kunye ne-95% yexesha lokuzithemba kunye novavanyo lwe-Tukey oluthembekileyo oluphawulekayo (HSD) ukuthelekisa iindlela zokusebenza. Olu vavanyo lweeyantlukwano phakathi kweemodeli zentsebenzo zenziwa ngokudibanisa iPython kunye ne-R (inguqulo 3.5.1) [18]. Sisebenzise le ndlela (nangona, ingaphantsi kuneyona ilungileyo) njengendlela yoncedo lwe-heuristic kule inqanaba lokuqala imodeli yokuqala yokuthelekisa ukusebenza ngokulindelekileyo kwisicelo seklinikhi esinokwenzeka. Emva koko sasebenzisa uvavanyo lwe-Bayesian olusayiniweyo lisebenzisa ukuhanjiswa kwangasemva ukufumanisa ukuba kunokwenzeka ukungafani komzekelo wokusebenza [19]. Kolu hlalutyo, sasebenzisa i-interval -0.01, 0.01, ebonisa ukuba ukuba amaqela amabini anomahluko wokusebenza ongaphantsi kwe-0.01, ayecatshangelwa ngokufanayo (ngaphakathi kwendawo yokulinganisa okusebenzayo), okanye ngenye indlela ahluke (enye ingcono kune-1.0.2). enye). Ukwenza uthelekiso lweBayesi labahlukanisi nokubala oku nokwenzeka, sisebenzise ilayibrari ye-baycomp (uguqulelo 3.6.4) kwiPython XNUMX.

Imodeli eqikelelwayo

Sakhe iimodeli eziqikelelwayo sisebenzisa iiyantlukwano ezilishumi ezipheleleyo zezicwangciso zethu zokumisela ukuqikelela (ukwahlula) isiphumo sovavanyo lwe-MoCA yesigulana ngasinye okanye ubuzaza bokuxilongwa kweklinikhi. Bonke abafundi basetyenziswa kwaye iimodeli zakhiwa kusetyenziswa iqonga lomthombo ovulekileyo lwesoftware Weka [20]. Uhlalutyo lwethu lokuqala, siqeshe i-10 esetyenziswa ngokuqhelekileyo i-algorithms yokufunda: 5-Abamelwane abasondeleyo, iinguqulelo ezimbini zeC4.5 umthi wesigqibo, i-Logistic Regression, i-Multilayer Perceptron, i-Naïve Bayes, iinguqulelo ezimbini ze-Random Forest, i-Radial Basis Function Network, kunye neVector yeNkxaso. Umatshini. Iimpawu eziphambili kunye nokuchasana kwezi ndlela zichazwe kwenye indawo [21] (jonga iSihlomelo esichaphazelekayo). Ezi zikhethwe ngenxa yokuba zibonisa iindidi ngeendidi ezahlukeneyo zabafundi nangenxa yokuba siye sabonisa impumelelo xa sizisebenzisa kuhlalutyo lwangaphambili kwidatha efanayo. Izicwangciso ze-hyper-parameter zikhethwe kuphando lwethu lwangaphambili olubonisa ukuba zomelele kwiinkcukacha ezahlukeneyo [22]. Ngokusekelwe kwiziphumo zohlalutyo lwethu lwangaphambili kusetyenziswa i-dataset efanayo edibeneyo kunye neempawu eziqhelekileyo eziye zasetyenziswa emva koko kuhlalutyo olupheleleyo, sichonge abafundi abathathu ababonelela ngokusebenza okuqinileyo kuzo zonke iindidi: Ukuguqulwa kweLogistic, i-Naïve Bayes, kunye ne-Support Vector Machine.

Ukuqinisekiswa okunqamlezileyo kunye nemodeli yemetric yokusebenza

Kuyo yonke imodeli eqikelelwayo (kubandakanywa nohlalutyo lwangaphambili), imodeli nganye yakhiwe ngokusebenzisa ukuqinisekiswa kwe-10-fold cross, kunye nokusebenza kwemodeli kulinganiswe ngokusebenzisa iNdawo ePhantsi kwe-Receiver Operating Characteristic Curve (AUC). Ukuqinisekiswa kwe-cross-validation kwaqala ngokuhlula ngokungenamkhethe nganye ye-10 yedatha yeskimu ye-modeling kwi-10 elinganayo (i-folds), isebenzisa ezisithoba kula macandelo ahlukeneyo ukuqeqesha imodeli kunye necandelo eliseleyo lokuvavanya. Le nkqubo iphindwe ngamaxesha e-10, isebenzisa icandelo elahlukileyo njengovavanyo olusetiweyo kwi-iteration nganye. Iziphumo zaye zadityaniswa ukubala iziphumo zomzekelo wokugqibela/ukusebenza. Kumfundi ngamnye/indibaniselwano yedatha, yonke le nkqubo yaphindwa izihlandlo ezili-10 ngedatha yahlulwe ngokwahlukileyo ngexesha ngalinye. Eli nyathelo lokugqibela lanciphisa i-bias, laqinisekisa ukuphindaphinda, kwaye lanceda ekumiseleni imodeli yokusebenza ngokubanzi. Lilonke (kumanqaku e-MoCA kunye nezikimu zokuhlela ubungqongqo ezidityanisiweyo), iimodeli ezingama-6,600 zakhiwa. Oku kuquka iimodeli ezingahluzwanga ezili-1,800 (iinkqubo zemodeli ezi-6 ezisetyenziswa kwidathasethi×3 yabafundi×10 ibaleka×10 imigoqo = 1,800 imifuziselo) kunye nemizekelo ehluziweyo eyi-4,800 (iinkqubo zomfuziselo ezi-4 ezisetyenziswa kwidathasethi×3 yabafundi×4 ubuchule bokukhetha uphawu×10 balekayo× Ii-folds ezili-10 = iimodeli ze-4,800).

Ukukhetha amanqaku

Kwimifuziselo ehluziweyo, ukhetho lwenqaku (usebenzisa iindlela ezine zokubeka amanqaku) lwenziwa ngaphakathi kokuqinisekiswa okunqamlezayo. Kwi-folds nganye ye-10, njenge-10% eyahlukileyo yedatha yedatha yayiyidatha yovavanyo, kuphela iimpawu ezine eziphezulu ezikhethiweyo kwidathasethi yoqeqesho ngalunye (okt, ezinye iifolda ezilithoba, okanye i-90% eseleyo yedatha epheleleyo) yasetyenziswa. ukwakha iimodeli. Asikwazanga ukuqinisekisa ukuba zeziphi iimpawu ezine ezisetyenzisiweyo kwimodeli nganye, njengoko olo lwazi alugcinwanga okanye lwenziwe lufumaneke ngaphakathi kweqonga lomzekelo esilisebenzisileyo (Weka). Nangona kunjalo, kunikwe ukulungelelaniswa ekukhethweni kwethu kokuqala kweempawu eziphezulu xa i-rankers isetyenziswe kuyo yonke idatha yedatha edibeneyo kunye nokufana okulandelayo kwintsebenzo yomzekelo, ezi mpawu ezifanayo (iminyaka yobudala, iminyaka yemfundo, i-MTx-% C, kwaye ithetha i-MTx-RT ) kusenokwenzeka ukuba zezona zixhaphakileyo eziphezulu ezine ezisetyenziswayo kunye nokhetho lwenqaku ngaphakathi kwenkqubo yokuqinisekisa.

IINKCUKACHA

Iimpawu zamanani zabathathi-nxaxheba (kubandakanywa amanqaku e-MoCA kunye neemetriki zokusebenza kwe-MemTrax) kwiiseti zedatha ezihambelanayo kwisicwangciso sohlelo ngalunye lomzekelo ukuxela kwangaphambili impilo yengqondo ebonakaliswe yi-MoCA (eqhelekileyo xa ithelekiswa ne-MCI) kunye nobunzima bokuxilongwa (ubumnene ngokuchasene nobukhulu) buboniswe kwiThebhile yesi-3.

Itheyibhile 3

Iimpawu zabathathi-nxaxheba, amanqaku e-MoCA, kunye nokusebenza kwe-MemTrax kwisicwangciso sokuhlela imodeli nganye

IQhinga loHleloubudalaEducationI-MoCA ihlengahlengisweI-MoCA ayilungiswangaMTx-% CMTx-RT
Udidi lwe-MoCA61.9 y (13.1)9.6 y (4.6)19.2 (6.5)18.4 (6.7)I-74.8% (15.0)1.4 s (0.3)
Uxilongo Ubukhali65.6 y (12.1)8.6 y (4.4)16.7 (6.2)15.8 (6.3)I-68.3% (13.8)1.5 s (0.3)

Amaxabiso abonisiwe (intsingiselo, i-SD) eyahlulwe ngobuchule bokuhlela imodeli imele idatha edityanisiweyo esetyenziselwa ukuqikelela impilo yengqondo ebonakaliswe yi-MoCA (MCI ngokuchasene nesiqhelo) kunye neseti yedatha engaphantsi kwe-XL esetyenziselwa ukuqikelela ubunzulu boxilongo (ubumnene ngokuchasene nobunzima).

Kwindibaniselwano nganye yamanqaku e-MoCA (elungisiweyo/engalungiswanga) kunye nomda (26/23), bekukho umahluko wezibalo (p = 0.000) kwi-pairwise nganye kuthelekiso (impilo yengqondo eqhelekileyo ngokuchasene ne-MCI) yobudala, imfundo, kunye nokusebenza kwe-MemTrax (MTx-% C kunye ne-MTx-RT). Isigulana ngasinye se-sub-dataset kwiklasi ye-MCI nganye kwindibaniselwano nganye yayikwi-avareji malunga ne-9 ukuya kwi-15 yeminyaka ubudala, inikwe ingxelo malunga neminyaka emihlanu embalwa yemfundo, kwaye yayinentsebenzo engaphantsi kwe-MemTrax yazo zombini iimethrikhi.

Iziphumo eziqikelelwayo zentsebenzo yeziphumo zokuhlelwa kwamanqaku e-MoCA kusetyenziswa abona bafundi bathathu baphezulu, ukuRegression kweLogistic, iNaïve Bayes, kunye noMatshini weVector yeNkxaso, zibonisiwe kwiThebhile 4. isetyenziswe kwiiseti zedatha yazo zonke izikimu zokubonisa. Kuluhlu lwedatha olungahluzwanga kunye nemodeli, ixabiso ngalinye ledatha kwiThebhile 4 libonisa imodeli yokusebenza esekelwe kwi-AUC ngokufanelekileyo intsingiselo ethathwe kwiimodeli ezili-100 (ii-10 runs×10 folds) ezakhelwe umfundi ngamnye/iskim somdibaniso, ngeyona ndlela iphakamileyo. umfundi ophumeleleyo obhalwe ngqindilili. Ngelixa umfuziselo wesethi yedatha ehluziweyo, iziphumo ezixelwe kwiTheyibhile 4 zibonisa umndilili oqhelekileyo wokuqhutywa kwemodeli ukusuka kuma-400 eemodeli zomfundi ngamnye esebenzisa indlela yokuhlela amanqaku (iindlela ezi-4 zokuhlela amanqaku×10 baleka×10 ukugotywa).

Itheyibhile 4

Intsebenzo yokuhlelwa kwamanqaku eDichotomous MoCA (AUC; 0.0–1.0) iziphumo zomfundi ngamnye kwabathathu abaqhube kakuhle kuzo zonke izikim zokumodareyitha.

Iseti yophawu olusetyenzisiweyoInqaku le-MoCACutoff ThresholdUlawulo loLungiseleloNaïve BayesInkxaso yeVector Machine
Akuhluzwanga (iimpawu ezili-10)Utshintshiwe230.88620.89130.8695
260.89710.92210.9161
Ayilungelelaniswanga230.91030.90850.8995
260.88340.91530.8994
Ihluziwe (4 iimpawu)Utshintshiwe230.89290.89540.8948
260.91880.92470.9201
Ayilungelelaniswanga230.91350.91340.9122
260.91590.92360.9177

Ukusebenzisa iiyantlukwano zeseti yenqaku, inqaku le-MoCA, kunye ne-MoCA cutoff threshold, eyona ntsebenzo iphezulu yesikim ngasinye somzekelo ibonisiwe ngesibindi (engahlukanga ngokwezibalo kunabo bonke abanye abangangeni ngesibindi kwimodeli efanelekileyo).

Ukuthelekisa abafundi kuzo zonke iindibaniselwano zeenguqulelo zamanqaku e-MoCA kunye nemigangatho (elungisiweyo/engatshintshwanga no-23/26, ngokulandelelanayo) kwidathasethi edityanisiweyo engahluzwanga (okt, kusetyenziswa iimpawu ezili-10 eziqhelekileyo), uNaïve Bayes ngokuqhelekileyo ebengoyena mfundi ugqwesileyo ngokufunda ngokubanzi. ukusebenza kokuhlelwa kwe-0.9093. Xa kuthathelwa ingqalelo abafundi abathathu abaphezulu, iimvavanyo ezihambelana neBayesian-correlated sign-rank tests zibonise ukuba kunokwenzeka (Pr) yaseNaïve Bayes eqhube kakuhle kakhulu kuLogistic Regression ibe yi-99.9%. Ngaphezu koko, phakathi kwe-Naïve Bayes kunye ne-Support Vector Machine, i-21.0% yamathuba okulingana okubonakalayo ekusebenzeni komfundi (ngoko, i-79.0% yamathuba oMshini weVector ye-Naïve Bayes ogqwesileyo kwi-Support Vector Machine), kunye ne-0.0% yokuba nokwenzeka koMshini weVector yeNkxaso oqhuba ngcono, ngokulinganayo yomeleza inzuzo yokusebenza kweNaïve Bayes. Uthelekiso olongezelelweyo lwenguqulelo yamanqaku e-MoCA kubo bonke abafundi/imidongwezi lucebise ukuba kukho i-advanteji encinane yokusebenza kusetyenziswa amanqaku e-MoCA angalungiswanga xa ethelekiswa nohlengahlengiso (0.9027 xa ithelekiswa no-0.8971, ngokulandelelanayo; Pr (ayilungiswanga > ilungisiwe) = 0.988). Ngokufanayo, uthelekiso lwe-cutoff threshold kubo bonke abafundi kunye neenguqulelo zamanqaku e-MoCA lubonise inzuzo encinci yokuhlelwa kwentsebenzo kusetyenziswa u-26 njengomgangatho wokuhlela xa uthelekiswa no-23 (0.9056 xa uthelekiswa no-0.8942, ngokulandelelanayo; Pr (26> 23) = 0.999). Okokugqibela, ukuvavanya inkqubo yokuhlelwa kweemodeli kusetyenziswa kuphela iziphumo ezihluziweyo (okt, iimpawu ezine zodidi oluphezulu kuphela), uNaïve Bayes (0.9143) ngokwamanani wayengoyena mfundi uqhube kakuhle kuzo zonke iinguqulelo/imiqobo yamanqaku e-MoCA. Nangona kunjalo, kuzo zonke iindlela zokulinganisa amanqaku zidityanisiwe, bonke abafundi abagqwesileyo baqhube ngendlela efanayo. Iimvavanyo zomgangatho osayiniweyo waseBayesi zibonise amathuba angama-100% okulingana kwepraktikhali phakathi kwesibini ngasinye sabafundi abahluziweyo. Njengedatha engahluzwanga (usebenzisa zonke iimpawu eziqhelekileyo ezili-10), kwakhona kwabakho inzuzo yokusebenza kwinguqu engalungiswanga yamanqaku e-MoCA (Pr (ayilungiswanga > ilungiswe) = 1.000), kunye nenzuzo efanayo eyahlukileyo yoluhlu lwe-26 (Pr (26> 23) = 1.000). Okuqaphelekayo, umndilili wokupasa komfundi ngamnye kwabathathu abaphezulu kuzo zonke iinguqulelo/imiqobo yamanqaku e-MoCA kusetyenziswa kuphela iimpawu ezine ezikudidi oluphezulu ugqithise kumndilili wokupasa kwakhe nawuphi na umfundi kwidatha engahluzwanga. Akumangalisi ukuba, ukusebenza kokuhlelwa kweemodeli ezihluziweyo (usebenzisa iimpawu ezine ezikumgangatho ophezulu) iyonke yayiphezulu (0.9119) kwiimodeli ezingahluzwanga (0.8999), kungakhathaliseki ukuba imodeli yendlela yokubeka ifaniswa naloo mifuziselo ngokusebenzisa zonke ezili-10 eziqhelekileyo. Iimbonakalo. Kwindlela nganye yokukhetha uphawu, bekukho i-100% enokwenzeka yenzuzo yokusebenza ngaphezulu kweemodeli ezingahluzwanga.

Ngezigulane ezithathelwa ingqalelo kwi-AD yokuxilongwa ngokuqatha, phakathi kweqela (MCI-AD ngokuchasene ne-AD) umahluko weminyaka (p = 0.004), imfundo (p = 0.028), amanqaku e-MoCA ahlengahlengisiweyo/alungiswanga (p = 0.000), kunye neMTx-% C (p = 0.008) zazibalulekile ngokwezibalo; ngelixa iMTx-RT yayingeyiyo (p = 0.097). Ngezo zigulana ziqwalaselwe ulwahlulo lobunzima bokuxilongwa kwe-VaD, phakathi kweqela (MCI-VaD ngokuchasene neVaD) umahluko kumanqaku e-MoCA ahlengahlengisiweyo / angalungiswanga (p = 0.007) kunye neMTx-% C (p = 0.026) kunye neMTx-RT (p = 0.001) zazibalulekile ngokwezibalo; kanti iminyaka (p = 0.511) kunye nemfundo (p = 0.157) kwakungekho ntlukwano phakathi kweqela.

Iziphumo zovavanyo oluqikelelweyo lweziphumo zokucalulwa kobuqatha kusetyenziswa abafundi abathathu abachongiweyo ngaphambili, uLogistic Regression, iNaïve Bayes, kunye noMatshini weVector yeNkxaso, zibonisiwe kwiThebhile 5. , abona bafundi bathathu sibachonge njengabona bancomekayo kwimodeli yethu yangaphambili baye banikezela ngeyona ndlela iqhuba kakuhle kuzo zombini ezi nkqubo zintsha zokumodareyitha. Ukuthelekisa abafundi kulo lonke udidi oluphambili loxilongo (AD kunye neVaD), akuzange kubekho mahluko wokusebenza ngokwahlukileyo phakathi kwabafundi be-MCI-VaD ngokuchasene neVaD, nangona uMshini weVector yeNkxaso ngokubanzi usebenze ngokubalaseleyo. Ngokufanayo, akukho mahluko ubalulekileyo phakathi kwabafundi be-MCI-AD ngokuchasene nokuhlelwa kwe-AD, nangona i-Naïve Bayes (NB) inenzuzo encinci yokusebenza ngaphezu kwe-Logistic Regression (LR) kunye nesininzi nje esingenakuphikiswa phezu koMshini weVector yeNkxaso, okunokwenzeka kwe-61.4% kunye ne-41.7% ngokulandelelanayo. Kuzo zombini iiseti zedatha, bekukho inzuzo yokusebenza ngokubanzi kuMatshini weVector yeNkxaso (SVM), nge Pr (SVM> LR) = 0.819 kunye Pr (SVM > NB) = 0.934. Inkqubo yethu yokuhlelwa ngokubanzi kubo bonke abafundi ekucingeni ubuzaza boxilongo kwi-sub-dataset ye-XL ibingcono kudidi lokuxilongwa kwe-VaD xa kuthelekiswa ne-AD (Pr (VAD> AD) = 0.998).

Itheyibhile 5

Intsebenzo yolwahlulo lobungqongqo lwezonyango (i-AUC; 0.0–1.0) isiphumo somfundi ngamnye kwabathathu abagqwesileyo kuzo zombini izikim zokubonisa imodeli.

iModeling SchemeUlawulo loLungiseleloNaïve BayesInkxaso yeVector Machine
I-MCI-AD ngokuchasene ne-AD0.74650.78100.7443
I-MCI-VaD ngokuchasene neVaD0.80330.80440.8338

Owona msebenzi uphezulu weskim somzekelo ngamnye uboniswe kwi ngesibindi (engahlukanga ngokwezibalo kunabanye abangangeni ngesibindi).

UKUQALA

Ukubona kwangethuba utshintsho kwimpilo yengqondo kubalulekile into eluncedo kulawulo lwempilo yomntu kunye nempilo yoluntu ngokufanayo. Ewe, ikwayinto ephambili kakhulu kwiisetingi zeklinikhi kwizigulana kwihlabathi liphela. Injongo ekwabelwana ngayo kukulumkisa izigulane, abanakekeli, kunye nababoneleli kwaye bakhawuleze unyango olufanelekileyo kunye noluneendleko eziphantsi kunye nokhathalelo lwexesha elide kwabo baqala ukufumana ukuhla kwengqondo. Ukudibanisa iiseti zedatha yezibhedlele/zeklinikhi zethu ezithathu, sichonge abafundi abathathu abakhetheke kakhulu (kunye neyona nto ibalaseleyo -Naïve Bayes) ukwakha imifuziselo eqikelelwayo kusetyenziswa. Iimethrikhi zokusebenza kwe-MemTrax ezinokuhlela ngokuthembekileyo imeko yezempilo yengqondo i-dichotomously (impilo yesiqhelo yengqondo okanye i-MCI) njengoko iya kuboniswa ngamanqaku adibeneyo e-MoCA. Okuqaphelekayo, umgangatho wokusebenza kwabo bobathathu abafundi uye waphucuka xa iimodeli zethu zisebenzise kuphela iimpawu ezine ezikudidi oluphezulu ezibandakanya ubukhulu becala ezi metrics zentsebenzo ye-MemTrax. Ngaphezu koko, siye satyhila amandla angqinisisiweyo okusetyenziswa kwabafundi abafanayo kunye neemetrics zentsebenzo ye-MemTrax kwisikimu sohlelo lwenkxaso yoxilongo ukuze kwahlulwe ubunzulu bamacandelo amabini oxilongo lwe-dementia: i-AD ne-VaD.

Uvavanyo lwenkumbulo ngundoqo ekubhaqweni kwangaphambili kwe-AD [23, 24]. Ke, lithuba lokuba iMemTrax yamkelekile, ibandakanyeke, kwaye kulula ukuyisebenzisa kwi-intanethi. uvavanyo lokuhlola inkumbulo ye-episodic kubemi ngokubanzi [6]. Ukuchaneka kokuqaphela kunye namaxesha okuphendula avela kulo msebenzi oqhubekayo wokusebenza atyhila ngokukodwa ekuchongeni ukuwohloka kwangaphambili kunye nokuguquka kunye nokusilela kwiinkqubo ze-neuroplastic ezinxulumene nokufunda, imemori kunye nokuqonda. Oko kukuthi, iimodeli ezilapha ezisekwe ubukhulu becala kwiimethrikhi zokusebenza kwe-MemTrax zinovakalelo kwaye zinokwenzeka ngokulula kwaye ngexabiso elincinci zityhila ukusilela kwebhayoloji ye-neuropathologic ngexesha lenguqu ye-asymptomatic yenqanaba ngaphambi kokulahleka okukhulu kokusebenza [25]. Ashford et al. ihlolisise iipatheni kunye nokuziphatha kokuchaneka kwememori yokuqaphela kunye nexesha lokuphendula kubasebenzisi be-intanethi abathathe inxaxheba ngokwabo ngeMemTrax [6]. Ukuhlonipha ukuba olu lwabiwo lubalulekile kwimodeli efanelekileyo kunye nokuphuhlisa izicelo zokhathalelo lwezigulane ezisebenzayo nezisebenzayo, ukucacisa ukuqatshelwa okusebenzayo kweklinikhi kunye neeprofayili zexesha lokuphendula kubalulekile ekusekeni isiseko sesiseko esibalulekileyo seklinikhi kunye nophando oluluncedo. Ixabiso elisebenzayo le-MemTrax ekuhlolweni kwe-AD kwisigaba sokuqala sokukhubazeka kwengqondo kunye nenkxaso yokuxilonga eyahlukileyo kufuneka emva koko ihlolwe ngokucokisekileyo kwimeko yekliniki apho i-comorbidities kunye nokuqonda, i-sensory, kunye ne-motor capabilities echaphazela ukusebenza kovavanyo inokuqwalaselwa. Kwaye ukwazisa imbono yobuchwephesha kunye nokukhuthaza usetyenziso lweklinikhi olusebenzayo, kunyanzelekile kuqala ukubonisa uthelekiso kuvavanyo olusekiweyo lovavanyo lwempilo yengqondo, nangona olu lwamva lunokuthi lunyanzeliswe luvavanyo olunzima, imfundo kunye nezithintelo zolwimi, kunye neempembelelo zenkcubeko [26] . Kule nkalo, uthelekiso oluncomekayo lwe-MemTrax ekusebenzeni kwezonyango kwi-MoCA edla ngokuchazwa njengomgangatho woshishino lubalulekile, ngakumbi xa kulinganiswa ubunzima obukhulu bokusebenziseka kunye nokwamkelwa kwesigulana kwe-MemTrax.

Uphononongo lwangaphambili oluthelekisa iMemTrax ne-MoCA luqaqambisa ingqiqo kunye nobungqina bokuqala obuqinisekisa uphando lwethu lomfuziselo [8]. Nangona kunjalo, olu thelekiso lwangaphambili lunxulumanise nje iimetriki ezimbini eziphambili zeMemTrax esizivavanyile kunye nemeko yokuqonda njengoko kugqitywe yi-MoCA kwaye kuchazwe amanqanaba ahlukeneyo kunye namaxabiso aphelileyo. Siye sazinzisa uvavanyo lwenkqubo yezonyango ye-MemTrax ngokuphonononga indlela esekelwe kwimodeli eqikelelweyo eya kubonelela ingqwalaselo ethe kratya yezinye iiparameters ezinokuthi zichaphazele isigulane. Ngokwahlukileyo kwabanye, asizange sifumane inzuzo kwimodeli yokusebenza kusetyenziswa ukulungiswa kwemfundo (uhlengahlengiso) kumanqaku e-MoCA okanye ekwahlukeni kwempilo yengqondo ecalula umyinge wamanqaku e-MoCA aggregate threshold ukusuka kwi-26 ecetyiswayo yokuqala ukuya kwi-23 [12, 15]. Ngapha koko, i-advanteji yokusebenza yohlelo ithandwa kusetyenziswa amanqaku e-MoCA angalungiswanga kunye nomda ophezulu.

Amanqaku aphambili ekusebenzeni kweklinikhi

Ukufunda ngoomatshini kuhlala kusetyenziswa kakhulu kwaye kusebenza kakhulu kwimodeli eqikelelweyo xa idatha ibanzi kwaye ine-multi-dimensional, oko kukuthi, xa kukho uqwalaselo oluninzi kunye noluhlu olubanzi oluhambelanayo lwexabiso eliphezulu (igalelo) iimpawu. Nangona kunjalo, ngale datha yangoku, iimodeli ezihluziweyo ezineempawu ezikhethiweyo ezine kuphela zenze ngcono kunezo zisebenzisa zonke iimpawu ezili-10 eziqhelekileyo. Oku kuphakamisa ukuba i-aggregate yedatha yesibhedlele ayizange ibe neyona nto ifanelekileyo yekliniki (ixabiso eliphezulu) iimpawu zokuhlela ngokufanelekileyo izigulane ngale ndlela. Nangona kunjalo, ugxininiso lwenqanaba logxininiso kwiimetrikhi zokusebenza zeMemTrax-MTx-% C kunye neMTx-RT-ixhasa ngamandla ukwakhiwa kweemodeli zovavanyo lwenqanaba lokuqala malunga nolu vavanyo olulula, olulula ukululawula, olunexabiso eliphantsi, kunye nokutyhila ngokufanelekileyo malunga nolu vavanyo. ukusebenza kwememori, ubuncinci ngoku njengesikrini sokuqala sohlelo lwebinary yesimo sempilo yengqondo. Ukunikezelwa koxinzelelo oluhlala lunyuka kubaboneleli kunye neenkqubo zokhathalelo lwempilo, iinkqubo zokuhlolwa kwezigulane kunye nezicelo zeklinikhi kufuneka ziphuhliswe ngokufanelekileyo ngogxininiso lokuqokelela, ukulandelela, kunye nokulinganisa ezo mpawu zesigulane kunye neemetrics zokuvavanya eziluncedo kakhulu, ezinenzuzo, kunye nezingqinisiso ezisebenzayo kuxilongo. kunye nenkxaso yolawulo lwesigulane.

Ngee-metrics ezimbini eziphambili ze-MemTrax ezingundoqo kuhlelo lwe-MCI, umfundi wethu ogqwesileyo (i-Naïve Bayes) wayenomsebenzi oqikelelweyo ophezulu kakhulu kwiimodeli ezininzi (i-AUC ngaphezu kwe-0.90) kunye ne-real-positive-positive ratio ye-fake-positive ratio esondela okanye engaphezulu kwe-4. : 1. Isicelo soguqulo lwezonyango kusetyenziswa lo mfundi siya kuthi ke ngoko sibambe (ukuhlelwa ngokuchanekileyo) uninzi lwabo banengxaki yokuqonda (cognitive deficit), ngelixa kuncitshiswa iindleko ezinxulumene nokuhlela ngempazamo umntu onempilo yesiqhelo yokuqonda njengonengxaki yokuqonda (false positive) okanye ukuphoswa kolohlelo kwabo banengxaki yokuqonda (ubuxoki obungalunganga). Nayiphi na kwezi meko zokungahlelwa kakuhle kunokubeka umthwalo ongafanelekanga ngokwasengqondweni kwintlalontle kwisigulana nakubanonopheli.

Nangona kucazululo lokuqala nolupheleleyo siye sasebenzisa bonke abafundi abalishumi kwisikim semodeli nganye, siye sajolisa iziphumo zethu kubahlalutyi abathathu ababonisa owona msebenzi unamandla ungatshintshiyo. Oku kwakhona yayikukuqaqambisa, ngokusekwe kule datha, abafundi abalindeleke ukuba baqhube ngokuxhomekeka kwinqanaba eliphezulu kwisicelo seklinikhi esisebenzayo ekumiseleni ukuhlelwa kwemeko yokuqonda. Ngaphezu koko, ngenxa yokuba olu phononongo lwalujongwe njengophando oluyintshayelelo malunga nokusebenziseka komatshini wokufunda ekuhlolweni kwengqondo kunye nale mingeni yeklinikhi ngexesha elifanelekileyo, senze isigqibo sokugcina iindlela zokufunda zilula kwaye zenziwe ngokubanzi, kunye nokulungiswa kweparameter encinci. Siyayixabisa into yokuba le ndlela inokuthi ithintele amandla okukwazi ukuqikelela ngakumbi isigulana. Ngokunjalo, ngelixa ukuqeqesha iimodeli kusetyenziswa kuphela iimpawu eziphezulu (indlela ehluziweyo) zisazisa ngakumbi malunga nezi datha (ngokukodwa kwiintsilelo kwidatha eqokelelweyo kwaye iqaqambisa ixabiso ekwandiseni ixesha elixabisekileyo lekliniki kunye nezixhobo), siyaqonda ukuba kuphambi kwexesha ukucutha. ububanzi beemodeli kwaye, ke ngoko, zonke (kunye nezinye iimpawu) kufuneka ziqwalaselwe kunye nophando lwexesha elizayo de sibe neprofayili ecacileyo yeempawu eziphambili eziza kusebenza kuluntu olubanzi. Ngaloo ndlela, sikwaqonda ngokupheleleyo ukuba idatha ebandakanyayo kunye nommeli ngokubanzi kunye nokulungiswa kwezi kunye nezinye iimodeli kuya kuba yimfuneko ngaphambi kokuba zibandakanywe kwisicelo seklinikhi esisebenzayo, ngakumbi ukulungiselela ukugula okuchaphazela ukusebenza kwengqondo okuya kufuneka kuthathelwe ingqalelo kuvavanyo olongezelelweyo lweklinikhi.

Ukusetyenziswa kwe-MemTrax kuye kwalungiswa ngakumbi yimodeli yobunzima besifo ngokusekwe kuxilongo olwahlukileyo lwezonyango. Ukusebenza kokuhlelwa okungcono xa kuqikelelwa ubunzima be-VaD (xa kuthelekiswa ne-AD) kwakungekho iyamangalisa inikwe iimpawu zeprofayili yesigulane kwiimodeli ezithile kwimpilo ye-vascular kunye nomngcipheko we-stroke, oko kukuthi, uxinzelelo lwegazi, i-hyperlipidemia, isifo seswekile, kunye (ngokuqinisekileyo) imbali ye-stroke. Nangona bekuya kunqweneleka ngakumbi kwaye kufanelekile ukuba kwenziwe uvavanyo lwezonyango olufanayo kwizigulane ezihambelanayo ezinempilo yengqondo eqhelekileyo ukuqeqesha abafundi ngale datha ibandakanya ngakumbi. Oku kuqinisekisiwe ngokukodwa, njengoko i-MemTrax ijoliswe ukuba isetyenziswe ngokuyinhloko ekubonweni kwangethuba lokusilela kwengqondo kunye nokulandelela okulandelayo kotshintsho lomntu. Kukwavakala ukuba ukuhanjiswa kwedatha okunqweneleka ngakumbi kwidathasethi ye-VaD kube negalelo ngokuyinxenye ekusebenzeni kwemodeli engcono. I-dataset ye-VaD yayilungelelaniswe kakuhle phakathi kweeklasi ezimbini, ngelixa idatha ye-AD enezigulane ezimbalwa ze-MCI zazingekho. Ngokukodwa kwiiseti zedatha ezincinci, kwanemizekelo embalwa eyongezelelweyo inokwenza umahluko onokulinganiswa. Zombini ezi mbono ziingxoxo ezinengqiqo ezisisiseko sokumahluko ekusebenzeni kobungqongqo besifo. Nangona kunjalo, ukulinganisa ukusebenza okuphuculweyo kwidathasethi yeempawu zamanani okanye iimpawu zendalo ezithe ngqo kwinkcazo yekliniki eqwalaselwayo kuphambi kwexesha. Nangona kunjalo, le noveli ibonise usetyenziso lwemodeli yolwahlulo lwe-MemTrax kwindima yenkxaso yoxilongo lweklinikhi ibonelela ngombono obalulekileyo kwaye iqinisekisa ukufuna uviwo olongezelelweyo kunye nezigulana kulo lonke ixesha lokuqhubeka kwe-MCI.

Ukuphunyezwa kunye nokusebenziseka okubonisiwe kwe-MemTrax kunye nale mifuziselo e-China, apho ulwimi kunye nenkcubeko yahluke kakhulu kwezinye iindawo ezisetyenziswayo (umzekelo, iFransi, i-Netherlands, ne-United States) [7, 8, 27], igxininisa ngakumbi amandla ukwamkelwa kwehlabathi jikelele kunye nexabiso leklinikhi yeqonga elisekelwe kwi-MemTrax. Lo ngumzekelo obonakalayo wokuzabalazela ukulungelelaniswa kwedatha kunye nokuphuhlisa izithethe ezisebenzayo zamazwe ngamazwe kunye nezixhobo zokubonisa imodeli yokuhlolwa kwengqondo esemgangathweni kwaye ilungele ukusetyenziswa ngokulula kwihlabathi jikelele.

Amanyathelo alandelayo ekunciphiseni ukuqonda imodeli kunye nokusetyenziswa

Ukungasebenzi kwengqondo kwi-AD kwenzeka ngokuqhubekayo, kungekhona kwizigaba ezicacileyo okanye amanyathelo [28, 29]. Nangona kunjalo, ngeli nqanaba lokuqala, injongo yethu yayikukuqala ukuseka amandla ethu okwakha imodeli ebandakanya i-MemTrax enokwahlula ngokwesiseko "eqhelekileyo" ukusuka "esingeyosiqhelo". Idata ebandakanyayo engakumbi yobungqina (umzekelo, ukucinga kwengqondo, iimpawu zemfuza, iimpawu zebhayoloji, izinto ezihambelanayo, kunye neziphawuli ezisebenzayo zobunzima. imisebenzi efuna ingqiqo ulawulo) [30] kuyo yonke imimandla eyahlukeneyo yehlabathi, abantu, kunye namaqela eminyaka ubudala ukuqeqesha kunye nokuphuhlisa ubuchwephesha (kubandakanywa ne-aptly weighted ensemble) iimodeli zokufunda ngoomatshini ziya kuxhasa iqondo elikhulu lokuhlelwa okuphuculweyo, oko kukuthi, amandla okuhlula amaqela ezigulana ezinobunzima. I-MCI ibe yi-subsets ezincinci kunye nezicacileyo ngakumbi kunye nokuncipha kwengqondo okuqhubekayo. Ngaphezu koko, uxilongo lweklinikhi oluhambelanayo lwabantu ngabanye kwingingqi ezahlukeneyo zezigulane lubalulekile qeqesha ngempumelelo ezi modeli zibandakanya ngakumbi kwaye zomelele ngokuqikelelwayo. Oku kuya kuququzelela ulawulo lwamatyala acwangcisiweyo ngakumbi kwabo banemvelaphi efanayo, iimpembelelo, kunye neeprofayili zengcinga ezichazwe ngokumxinwa kwaye ngaloo ndlela kunyuswe inkxaso yesigqibo sezonyango kunye nokhathalelo lwesigulane.

Uninzi lophando olufanelekileyo lweklinikhi ukuza kuthi ga ngoku luye lwajongana nezigulane ezinobunzima obuncinci be-dementia; kwaye, ekusebenzeni, kaninzi ukungenelela kwesigulane kuzanywa kuphela kwizigaba eziphambili. Nangona kunjalo, ngenxa yokuba ukwehla kwengqondo kuqalisa kakuhle phambi kokuba kuhlangatyezwane nemigaqo yeklinikhi yokuphazamiseka kwengqondo, isikrini sakwangoko esekwe kwiMemTrax sinokukhuthaza imfundo efanelekileyo yabantu malunga nesi sifo kunye nokuqhubeka kwaso kunye nokungenelela kwangethuba nangexesha. Ke, ukubonwa kwangethuba kunokuxhasa ukubandakanyeka okufanelekileyo ukusuka kukuzivocavoca, ukutya, inkxaso yeemvakalelo, kunye nokuphuculwa kwentlalontle ukuya kungenelelo lwezonyango kunye nokuqinisa utshintsho olunxulumene nesigulana ekuziphatheni kunye nokuqonda ukuba ngokukodwa okanye ngokuhlangeneyo kunokunciphisa okanye kunokunqanda ukuqhubeka kwengqondo [31, 32] . Ngaphezu koko, ngempumelelo ukuhlolwa kwangoko, abantu kunye neentsapho zabo banokukhuthazwa ukuba baqwalasele izilingo zeklinikhi okanye bafumane iingcebiso kunye nezinye iinkonzo zenkxaso yeenkonzo zentlalo ukunceda ukucacisa ukulindela kunye neenjongo kunye nokulawula imisebenzi yemihla ngemihla. Ukuqinisekiswa okungaphaya kunye nosetyenziso olusebenzayo oluxhaphakileyo ngezi ndlela kunokuba sisixhobo ekudambiseni okanye ekumiseni ukuqhubeka kwe-MCI, AD, kunye ne-ADRD kubantu abaninzi.

Enyanisweni, isiphelo esiphantsi soluhlu lweminyaka yesigulana kwisifundo sethu asimeleli inani leenkxalabo zendabuko kunye ne-AD. Nangona kunjalo, umndilili weminyaka kwiqela ngalinye elisetyenzisiweyo kwiiskim zemodeli yokuhlelwa ngokusekwe kumanqaku e-MoCA/umqobo kunye nobukhali bokuxilongwa (Itheyibhile 3) igxininisa uninzi olucacileyo (ngaphezu kwama-80%) ubuncinci beminyaka engama-50 ubudala. Olu nikezelo lufaneleke kakhulu ukwenziwa ngokubanzi, luxhasa ukusetyenziswa kwezi modeli kuluntu olubonakalisa abo bachaphazeleka ngokwesiqhelo ukuqala kwangoko kunye nesifo se-neurocognitive esikhulayo ngenxa ye-AD kunye ne-VaD. Kwakhona, ubungqina bamva nje kunye noxinzelelo lwembono ezo zinto zaziwayo (umzekelo, uxinzelelo lwegazi, ukutyeba, isifo seswekile, kunye nokutshaya) ezinokuba negalelo ekunyukeni kwangoko. abantu abadala kunye namanqaku omngcipheko we-vascular kunye nesiphumo sokwenzakala kwengqondo efihlakeleyo yemithambo ekhula ngokufihlakeleyo kunye neziphumo ezicacileyo nakulutsha. abantu abadala [33–35]. Ngako oko, elona thuba lilungileyo lokuhlola lokuqala lokubona kwangoko Inqanaba lokusilela kwengqondo kunye nokuqaliswa kothintelo olusebenzayo kunye nezicwangciso zokungenelela ekujonganeni ngempumelelo ne-dementia iya kuvela kuphononongo lwemiba enegalelo kunye nezalathi ezingaphambili kuwo wonke umda wobudala, kuqukwa ukuba mdala kwaye okunokwenzeka nokuba sebuntwaneni (kuphawula ukufaneleka kweemeko zemfuza ezifana ne-apolipoprotein E ukusuka ekukhulelweni kokuqala).

Ngokwesiqhelo, uxilongo olusebenzayo lweklinikhi kunye neenkqubo ezixabisa kakhulu zokucinga kwangaphambili, iprofayili yemfuza, kunye nokulinganisa i-biomarkers ethembisayo ayisoloko ifumaneka ngokulula okanye inokwenzeka kubaboneleli abaninzi. Ke, kwiimeko ezininzi, ulwahlulo lokuqala lwemeko yempilo yengqondo lunokufuneka luthatyathwe kwiimodeli kusetyenziswa ezinye iimetrikhi ezilula ezibonelelwa sisigulana (umz., ukuzixela. iingxaki zememori, amayeza angoku, kunye nokunciphisa imisebenzi yesiqhelo) kunye neempawu eziqhelekileyo zabantu [7]. Iirejista ezifana neYunivesithi yaseCalifornia Impilo yeBongo Irejistri (https://www.brainhealthregistry.org/) [27] kunye nabanye abanobubanzi obukhulu beempawu ozixelayo, amanyathelo omgangatho (umzekelo, ukulala kunye nokuqonda yonke imihla), amayeza, isimo sempilo, kunye nembali, kunye ulwazi oluthe kratya lwamanani abantu luya kuba luncedo ekuphuhliseni nasekuqinisekiseni ukusetyenziswa kwezi modeli zamandulo eklinikhi. Ngaphaya koko, uvavanyo olunje ngeMemTrax, olubonise ukuba luluncedo ekuvavanyeni umsebenzi wememori, enyanisweni lunokubonelela ngoqikelelo olungcono kakhulu lwe-AD pathology kuneziphawuli zebhayoloji. Ngenxa yokuba eyona nto iphambili ye-AD pathology kukuphazamiseka kwe-neuroplasticity kunye nelahleko entsonkothileyo ye-synapses, ebonakala njengeepisodic. ukungasebenzi kakuhle kwememori, umlinganiselo ovavanya inkumbulo ye-episodic unokuthi enyanisweni bonelela ngoqikelelo olungcono lwe-AD yomthwalo we-pathological kunamakishi ebhayoloji kwisigulane esiphilayo [36].

Ngazo zonke iimodeli eziqikelelwayo-nokuba zincediswa yidatha entsonkothileyo kunye nebandakanyayo evela kwitekhnoloji ye-state-of-the-art kunye nokuqonda okucokisekileyo kweklinikhi kwiindawo ezininzi okanye ezo zithintelwe kulwazi olusisiseko nolufumaneka ngokulula lweprofayili ekhoyo yesigulane-inzuzo eyaziwayo yobukrelekrele bokwenziwa. kunye nokufunda koomatshini kukuba iimodeli ezinesiphumo ziyakwazi ukudibanisa kunye nokunyanzelwa "ukufunda" kwidatha entsha efanelekileyo kunye nombono obonelelwa ngokusetyenziswa kwesicelo esiqhubekayo. Ukulandela ukuhanjiswa kwethekhnoloji esebenzayo, njengoko iimodeli apha (kunye nokuphuhliswa) zisetyenziswa kwaye zityetyiswe ngamatyala amaninzi kunye neenkcukacha ezifanelekileyo (kubandakanywa nezigulane ezine-comorbidities ezinokuthi zibonise ngokuhla kokuqonda), ukusebenza kwangaphambili kunye nokuhlelwa kwempilo yokuqonda kuya kuba namandla ngakumbi, okukhokelela kuncedo olusebenzayo lwenkxaso yezigqibo zeklinikhi. Olu tshintsho luya kufezekiswa ngokupheleleyo kwaye lufezekiswe ngokufakela i-MemTrax kwisiko (ejoliswe kwizakhono ezikhoyo) amaqonga anokuthi ababoneleli bezempilo basebenzise ngexesha lokwenyani ekliniki.

Okuyimfuneko ekuqinisekiseni nasekusebenziseni imodeli ye-MemTrax yenkxaso yokuxilonga kunye nokhathalelo lwesigulane zifunwa kakhulu emva kwedatha enentsingiselo yelongitudinal. Ngokujonga kunye nokurekhoda utshintsho oluhambelanayo (ukuba lukhona) kwimeko yeklinikhi kulo lonke uluhlu olwaneleyo lwesiqhelo nge-MCI yasekuqaleni, iimodeli zovavanyo oluqhubekayo olufanelekileyo kunye nokuhlelwa kunokuqeqeshwa kunye nokuguqulwa njengezigulane ezineminyaka yobudala kwaye ziphathwe. Oko kukuthi, ukusetyenziswa okuphindaphindiweyo kunokuncedisa ngokulandelelwa kwexesha elide lotshintsho oluncinci lwengqondo, ukungenelela okusebenzayo, kunye nokugcina unonophelo olunolwazi lwe-stratified. Le ndlela ihambelana ngokusondeleyo kunye nokusebenza kweklinikhi kunye nesigulane kunye nokulawulwa kwamatyala.

Imida

Siyawuxabisa umngeni kunye nexabiso lokuqokelela idatha yeklinikhi ecocekileyo kwiklinikhi elawulwayo / indawo yesibhedlele. Nangona kunjalo, bekuya komeleza imodeli yethu ukuba iiseti zethu zedatha ziquka izigulane ezininzi ezineempawu eziqhelekileyo. Ngaphezu koko, ngokuthe ngqo kwimodeli yethu yokuxilongwa, bekuya kuba yinto enqwenelekayo kwaye kufanelekile ukuba kwenziwe uvavanyo olufanayo lwezonyango kwizigulane ezihambelanayo ezinempilo yengqondo eqhelekileyo ukuqeqesha abafundi. Kwaye njengoko kugxininiswe kuhlelo oluphezulu lokusebenza usebenzisa isethi yedatha ehluziweyo (kuphela amanqaku aphezulu aphezulu), ngokubanzi kunye Imilinganiselo yempilo yengqondo/izalathisi kusenokwenzeka ukuba ziphucukile imodeli yokusebenza enenani elikhulu leempawu eziqhelekileyo kuzo zonke izigulana.

Abanye abathathi-nxaxheba kusenokwenzeka ukuba baye bafumana ezinye izigulo ezinokuthi zibangele ukusilela kwengqondo okudlulayo okanye okungapheliyo. Ngaphandle kwe-sub-dataset ye-XL apho izigulane zichazwe njenge-AD okanye i-VaD, idatha ye-comorbidity ayizange iqokelelwe / ixelwe kwi-pool yesigulane se-YH, kwaye i-comorbidity echazwe kakhulu ngokude kwi-sub-dataset ye-KM yayiyisifo sikashukela. Kuyaphikiswa, nangona kunjalo, ukuba kubandakanywa izigulane kwiiskim zethu zokumodareyitha ezine-comorbidities ezinokuthi zibangele okanye zandise izinga lokunqongophala kwengqondo kunye nesiphumo esisezantsi sokusebenza kwe-MemTrax kuya kubonisa ngakumbi izigulane ezijoliswe kwihlabathi lokwenyani kolu vavanyo lwengqondo ngokubanzi. kunye nendlela yokwenza imodeli. Ukuqhubela phambili, ukuxilongwa ngokuchanekileyo kwee-comorbidities ezinokuchaphazela ukusebenza kwengqondo kunenzuzo ngokubanzi ekuphuculeni iimodeli kunye neziphumo zokunakekelwa kwezigulane.

Ekugqibeleni, izigulane ze-YH kunye ne-KM ze-sub-dataset zisebenzisa i-smartphone ukuthatha uvavanyo lwe-MemTrax, ngelixa inani elilinganiselweyo lezigulane ze-XL ezisezantsi zisebenzisa i-iPad kwaye abanye basebenzisa i-smartphone. Oku kunokuba kwazisa umahluko omncinci onxulumene nesixhobo ekusebenzeni kwe-MemTrax kumzekelo wohlelo lwe-MoCA. Nangona kunjalo, iiyantlukwano (ukuba zikhona) kwi-MTx-RT, umzekelo, phakathi kwezixhobo zinokungahoywa, ngakumbi xa umthathi-nxaxheba ngamnye enikwa uvavanyo “lokuziqhelanisa” kanye phambi kovavanyo olurekhodiweyo. Nangona kunjalo, ukusetyenziswa kwezi zixhobo zimbini ziphathwa ngesandla kunokubanakho ukubeka esichengeni uthelekiso oluthe ngqo kunye/okanye ukudityaniswa nezinye iziphumo zeMemTrax apho abasebenzisi baphendule ukuphinda imifanekiso ngokuchukumisa ibar yendawo kwikhibhodi yekhompyuter.

Amanqaku aphambili kwi-MemTrax yokuqikelela usetyenziso lwemodeli

  • • Imifuziselo yethu yengqikelelo eqhuba kakuhle equka iimetrics ezikhethiweyo zeMemTrax zinokuhlela ngokuthembekileyo imeko yempilo yengqondo (impilo yesiqhelo yengqondo okanye i-MCI) njengoko iya kuboniswa luvavanyo olwamkelwa ngokubanzi lwe-MoCA.
  • • Ezi ziphumo zixhasa ukudityaniswa kweemetrics ezikhethiweyo zeMemTrax kuhlelo lwenkqubo yokuhlola imodeli yolwahlulo lwakwangoko.
  • • Imodeli yethu yokuhlelwa kwakhona ibonise amandla okusebenzisa ukusebenza kwe-MemTrax kwizicelo zokwahlula ubuzaza bokuxilongwa kwesifo sengqondo esixhalabisayo.

Ezi ziphumo zenoveli ziseka ubungqina obuqinisekileyo obuxhasa ukusetyenziswa komatshini wokufunda ekwakhiweni kweemodeli zolwahlulo ezisekelwe kwi-MemTrax ezomeleleyo zenkxaso yoxilongo kulawulo olusebenzayo lweemeko zeklinikhi kunye nokhathalelo lwesigulane kubantu abanengxaki yokuqonda.

AMAKHODI

Siyawuqaphela umsebenzi kaJ. Wesson Ashford, uCurtis B. Ashford, kunye noogxa bakhe ekuphuhliseni nasekuqinisekiseni umsebenzi oqhubekayo wokuqatshelwa kwe-intanethi kunye nesixhobo (i-MemTrax) esisetyenzisiweyo apha kwaye sibulela izigulane ezininzi ezinesifo sengqondo esixhalabisayo eziye zanegalelo kuphando olusisiseko olubalulekileyo. . Siphinde sibulele u-Xianbo Zhou kunye noogxa bakhe kwi-SJN Biomed LTD, oogxa bakhe kunye nabasebenzisana nabo kwizibhedlele / iindawo zeekliniki, ngakumbi uDkt. M. Luo noM. Zhong, abancedise ekuqeshweni kwabathathi-nxaxheba, iimvavanyo zokucwangcisa, kunye nokuqokelela, ukurekhoda, kunye nokulawula idatha, kunye nabathathi-nxaxheba abavolontiya abanikela ngexesha labo elixabisekileyo kwaye bazibophezela ekuthatheni iimvavanyo kunye nokubonelela. idatha exabisekileyo ukuba siyivavanye kolu phononongo. Oku uphononongo luxhaswe ngokuyinxenye yi-MD Scientific Research Inkqubo yeYunivesithi yezoNyango yaseKunming (iNombolo yeNkxaso ye-2017BS028 ukuya kwi-XL) kunye neNkqubo yoPhando yeSebe leSayensi kunye neTekhnoloji yaseYunnan (i-Grant no. 2019FE001 (-222) ukuya kwi-XL).

J. Wesson Ashford ufake isicelo selungelo elilodwa lomenzi wechiza ukuze kusetyenziswe iparadigm ethile eqhubekayo yokuqaphela echazwe kweli phepha ngokubanzi. uvavanyo lwenkumbulo.

I-MemTrax, LLC yinkampani ephethwe nguCurtis Ashford, kwaye le nkampani ilawula uvavanyo lwenkumbulo inkqubo echazwe kweli phepha.

Ukubhengezwa kwababhali kufumaneka kwi-intanethi (https://www.j-alz.com/manuscript-disclosures/19-1340r2).

Uvavanyo lwenkumbulo yovavanyo lwengqondo yovavanyo lwenkumbulo yokulahleka kovavanyo lwexesha elifutshane ukulahleka kwenkumbulo uvavanyo lwenqama yovavanyo lwengqondo yezidlo ezahlukeneyo zeencwadi uvavanyo lwengqondo kwi-Intanethi
UCurtis Ashford – uMnxibelelanisi woPhando lweNgcaciso

IZALATHISO

[1] I-Alzheimer's Association (2016) i-2016 i-Alzheimer's disease facts kunye namanani. I-Alzheimers Dement 12, 459-509.
[2] Gresenz CR , Mitchell JM , Marrone J , Federoff HJ (2019) Isiphumo senqanaba lokuqala Isifo se-Alzheimer kwiziphumo zemali zekhaya. Health Econ 29, 18-29.
[3] Foster NL , Bondi MW , Das R , Foss M , Hershey LA , Koh S , Logan R , Poole C , Shega JW , Sood A , Thothala N , Wicklund M , Yu M , Bennett A , Wang D (2019) Ukuphucuka komgangatho ku i-neurology: Iseti yomgangatho wothintelo lwengqondo ephakathi. I-Neurology 93, 705-713.
[4] Tong T , Thokala P , McMillan B , Ghosh R , Brazier J (2017) Ukusebenza kweendleko zokusebenzisa Iimvavanyo zovavanyo lokuqonda ukufumanisa isifo sengqondo esixhalabisayo kunye nokuphazamiseka okuncinci kwengqondo kukhathalelo lokuqala. Int J Geriatr Psychiatry 32, 1392-1400.
[5] Ashford JW , Gere E , Bayley PJ (2011) Imemori yokulinganisa kwiisethingi zeqela elikhulu usebenzisa uvavanyo oluqhubekayo lokuqaphela. J Alzheimers Dis 27, 885-895.
[6] I-Ashford JW , Tarpin-Bernard F , Ashford CB , Ashford MT (2019) Umsebenzi oqhubekayo wekhompyutha wokuqaphela umlinganiselo wememori ye-episodic. J Alzheimers Dis 69, 385-399.
[7] Bergeron MF , Landset S , Tarpin-Bernard F , Ashford CB , Khoshgoftaar TM , Ashford JW (2019) I-Episodic-memory performance kumatshini wokufunda imodeli yokuqikelela ulwahlulo lwemeko yezempilo yokuqonda. J Alzheimers Dis 70, 277-286.
[8] van der Hoek MD , Nieuwenhuizen A , Keijer J , Ashford JW (2019) Uvavanyo lweMemTrax xa kuthelekiswa noqikelelo lovavanyo lwengqondo lwaseMontreal lophazamiseko olungephi lwengqiqo. J Alzheimers Dis 67, 1045-1054.
[9] U-Falcone M , Yadav N , Poellabauer C , Flynn P (2013) Ukusebenzisa izandi zezikhamiso ezizimeleyo zokuhlelwa kokulimala kancinci kwengqondo. Kwi-2013 IEEE International Conference on Acoustics, Speech and Signal Processing, Vancouver, BC, iphe. 7577-7581.
[10] Dabek F , Caban JJ (2015) Ukusebenzisa idatha enkulu ukulinganisa amathuba okuphuhlisa iimeko zengqondo emva kokuphazamiseka. I-Procedia Comput Sci 53, 265-273.
[11] Climent MT , Pardo J , Munoz-Almaraz FJ , Guerrero MD , Moreno L (2018) Umthi wesigqibo sokubonwa kwangaphambili kokuphazamiseka kwengqondo ngabakhemisti basekuhlaleni. Ngaphambili Pharmacol 9, 1232.
[12] Nasreddine ZS , Phillips NA , Bedirian V , Charbonneau S , Whitehead V , Collin I , Cummings JL , Chertkow H (2005) UVavanyo lweNgcaciso lwaseMontreal, i-MoCA: Isixhobo esifutshane sokuhlola ukukhubazeka okuncinci kwengqondo. J Am Geriatr Soc 53, 695–699.
[13] Yu J, Li J, Huang X (2012) Uguqulelo lwaseBeijing lovavanyo lwengqondo yaseMontreal njengesixhobo esifutshane sokuhlola ukukhubazeka okuncinci kwengqondo: Uphononongo olusekelwe kuluntu. BMC Psychiatry 12, 156.
[14] I-Chen KL, i-Xu Y, i-Chu AQ, i-Ding D, i-Liang XN, i-Nasreddine ZS, i-Dong Q, i-Hong Z, i-Zhao QH, i-Guo QH (i-2016) Ukuqinisekiswa kwenguqu yaseTshayina yaseMontreal yovavanyo lwengqiqo olusisiseko sokuhlola ukuphazamiseka kwengqondo encinci. J Am Geriatr Soc 64, e285–e290.
[15] Carson N , Leach L , Murphy KJ (2018) Ukuphononongwa kwakhona kweMontreal Cognitive Assessment (MoCA) amanqaku anqanyuliweyo. Int J Geriatr Psychiatry 33, 379-388.
[16] I-American Psychiatric Association (2013) i-Task Force Diagnostic kunye ne-statistical manual yokuphazamiseka kwengqondo: DSM-5™, American Psychiatric Publishing, Inc., Washington, DC.
[17] Inhlwathi. IPython Software Foundation, http://www.python.org, Ifikelelwe ngoNovemba 15, 2019.
[18] Iqela le-R Core, R: Ulwimi kunye nokusingqongileyo kwi-computing yamanani R Isiseko se-Statistical Computing, eVienna, e-Austria. https://www.R-project.org/, 2018, Ifikelelwe ngoNovemba 15, 2019.
[19] Benavoli A , Corani G , Demšar J , Zaffalon M (2017) Ixesha lokutshintsha: Isifundo sokuthelekisa amacandelo amaninzi ngohlalutyo lwaseBayesi. J Mach Funda Res 18, 1–36.
[20] UFrank E , IHolo MA , Witten IH (2016) I-WEKA Workbench. Kwi Ukuchithwa kwedatha: Izixhobo zokufunda zoomatshini kunye nobuChwephesha obusebenzayo, Frank E, Hall MA, Witten IH, Pal CJ, ed. UMorgan Kaufmann https://www.cs.waikato.ac.nz/ml/weka/Witten_et_al_2016_appendix.pdf
[21] Bergeron MF , Landset S , Maugans TA , Williams VB , Collins CL , Wasserman EB , Khoshgoftaar TM (2019) Ukufundwa komatshini kumzekelo we-sports high school concussion symptom. IMed Sci Sports Exerc 51, 1362-1371.
[22] Van Hulse J , Khoshgoftaar TM , Napolitano A (2007) Iimbono zovavanyo lokufunda kwiidatha ezingalinganiyo. Kwi Iinkqubo zeNkomfa yaMazwe ngaMazwe yama-24 ngokuFunda ngoomatshini, Corvalis, Oregon, USA, iphepha 935-942.
[23] Ashford JW , Kolm P , Colliver JA , Bekian C , Hsu LN (1989) Uvavanyo lwesigulane se-Alzheimer kunye nesimo sengqondo esincinci: Uhlalutyo lweempawu ze-Item.P. J Gerontol 44, 139–146.
[24] Ashford JW , Jarvik L (1985) Isifo se-Alzheimer: Ngaba iplastiki ye-neuron ibeka phambili kwi-axonal neurofibrillary degeneration? N Engl J Med 313, 388–389.
[25] Jack CR Jr , Therneau TM , Weigand SD , ​​Wiste HJ , Knopman DS , Vemuri P , Lowe VJ , Mielke MM , Roberts RO , Machulda MM , Graff-Radford J , Jones DT ML , Schwarz CG , Senjem JL . , Rocca WA , Petersen RC (2019) Ukuxhaphaka kwebhayoloji kunye neklinikhi echazwe kwi-Alzheimer spectrum entities usebenzisa iNational Institute on Aging-Alzheimer's UPhando loMbutho isakhelo. JAMA Neurol 76, 1174-1183.
[26] Zhou X , Ashford JW (2019) Ukuqhubela phambili kwizixhobo zokuhlola Isifo se-Alzheimer. Ukuguga Med 2, 88-93.
[27] Weiner MW , Nosheny R , Camacho M , Truran-Sacrey D , Mackin RS , Flenniken D , Ulbricht A , Insel P , Finley S , Fockler J , Veitch D (2018) The Impilo yeBongo Irejistri: Iqonga elisekelwe kwi-intanethi lokugaya, uvavanyo, kunye nokubeka iliso ixesha elide labathathi-nxaxheba kwizifundo ze-neuroscience. I-Alzheimers Dement 14, 1063-1076.
[28] Ashford JW , Schmitt FA (2001) Ukulinganisa ixesha lekhosi ye I-Alzheimer's dementia. Curr Psychiatry Rep 3, 20-28.
[29] Li X , Wang X , Su L , Hu X , Han Y (2019) Sino Longitudinal Study on Cognitive Decline (SILCODE): Iprothokholi yophononongo lwaseTshayina lwexesha elide lokuphonononga ukuphuhlisa imifuziselo yoqikelelo lwengozi yoguqulelo ukuya kuphazamiseko olungephi lwengqondo kubantu abanengqondo evumayo. ukuhla. BMJ Vula 9, e028188.
[30] Tarnanas I , Tsolaki A , Wiederhold M , Wiederhold B , Tsolaki M (2015) iminyaka emihlanu ye biomarker progression variability for Isifo sika-Alzheimer's dementia Uqikelelo: Ngaba imisebenzi yesixhobo entsonkothileyo yesiphawuli semihla ngemihla singazalisa izikhewu? I-Alzheimers Dement (Amst) 1, 521-532.
[31] McGurran H , Glenn JM , Madero EN , Bott NT (2019) Uthintelo kunye nonyango lwe-Alzheimer's: Iindlela zebhayoloji zokuzilolonga. J Alzheimers Dis 69, 311-338.
[32] Mendiola-Precoma J , Berumen LC , Padilla K , Garcia-Alcocer G (2016) Unyango uthintelo kunye nonyango lwesifo sika-Alzheimer. Biomed Res Int 2016, 2589276.
[33] Lane CA , Barnes J , Nicholas JM , Sudre CH , Cash DM , Malone IB , Parker TD , Keshavan A , Buchanan SM , Keuss SE , James SN , Lu K , Murray-Smith H , Wong A , Gordon E , Coath W , Modat M , Thomas D , Richards M , Fox NC , Schott JM (2020) Imibutho phakathi komngcipheko we-vascular kuwo wonke umntu omdala kunye ne-brain pathology ekupheleni kobomi: Ubungqina obuvela kwi-British cohort yokuzalwa. JAMA Neurol 77, 175-183.
[34] Seshadri S (2020) Uthintelo lwe-dementia-ingcinga ngaphaya kweminyaka yobudala kunye neebhokisi ze-amyloid. JAMA Neurol 77, 160-161.
[35] Maillard P , Seshadri S , Beiser A , Himali JJ , Au R , Fletcher E , Carmichael O , Wolf PA , DeCarli C (2012) Iziphumo ze-systolic blood pressure kwi-white-matter integrity in young people in the Framingham Heart Study: A cross -ufundo lwamacandelo. Lancet Neurol 11, 1039-1047.
[36] Fink HA , Linskens EJ , Silverman PC , McCarten JR , Hemmy LS , Ouellette JM , Greer NL , Wilt TJ , Butler M (2020) Ukuchaneka kovavanyo lwe-biomarker ye-neuropathologically defined Isifo se-Alzheimer kubantu abadala abane-dementia. U-Ann Intern Med 172, 669-677.

Affiliations: [a] SIVOTEC Analytics, Boca Raton, FL, USA | [b] iSebe leKhompyutha kunye nobuNjineli boMbane kunye neNzululwazi yeKhompyutha, iYunivesithi yaseFlorida yaseAtlantic, iBoca Raton, FL, USA | [c] SJN Biomed LTD, Kunming, Yunnan, China | [d] Iziko le Uphando lwe-Alzheimer, Iziko laseWashington loPhando lweKlinikhi, eWashington, DC, eU.SA | [e] iSebe lezoNyango loBuyiselo, iSibhedlele sokuQala esiManyano neYunivesithi yaseKunming Medical, Kunming, Yunnan, China | [f] iSebe leNeurology, iSibhedlele sabantu baseDehong, iDehong, Yunnan, China | [g] iSebe le-Neurology, iSibhedlele sokuQala esiManyeneyo saseKunming Medical University, iSithili saseWuhua, iKunming, iPhondo laseYunnan, eChina | [h] Iziko leSigulo esiNxulumene neMfazwe kunye neNzakalo yokuFunda, i-VA Palo Alto Care Health Inkqubo, Palo Alto, CA, USA | [i] iSebe leNzululwazi ngeNgqondo kunye neNzululwazi yokuziphatha, kwiSikolo seYunivesithi yaseStanford, ePalo Alto, CA, eMelika.

Imbalelwano: [*] Imbalelwano eya ku: Michael F. Bergeron, PhD, FACSM, SIVOTEC Analytics, Boca Raton Innovation Campus, 4800 T-Rex Avenue, Suite 315, Boca Raton, FL 33431, USA. I-imeyile: mbergeron@sivotecanalytics.com.; U-Xiaolei Liu, MD, iSebe le-Neurology, iSibhedlele sokuQala esiBambiseneyo seYunivesithi yezoNyango yaseKunming, i-295 Xichang Road, iSithili saseWuhua, i-Kunming, iPhondo laseYunnan 650032, eChina. I-imeyile: ring@vip.163.com.

Amagama angundoqo: Ukuguga, Isifo se-Alzheimer, isifo sengqondo esixhalabisayo, ukuvavanywa kobunzima