I-Junior Awardee: Ngaba imiqondiso ye-cortical ingasixelela ukuba kwenzeka ntoni engqondweni enzulu? | Congress 2026

Ngoko ke olu luphononongo oluguqulela ulwazi oluhlola ukuba ingaba umsebenzi orekhodwe kwi-cortex yengqondo ungasetyenziselwa na ukwakha kwakhona okwenzekayo kwizakhiwo zobuchopho ezinzulu, kwaye oku kunokujongana neminye yemida yokuva kwiinkqubo zangoku zokuvuselela ubuchopho obunzulu obuguquguqukayo.

Zixiao, ndiyavuyisana nawe ngebhaso, kwaye wamkelekile kwi-podcast.

UDkt. Zixiao Yin: Enkosi. Enkosi ngokundibona.

UGqr. Michele Matarazzo: Ndingathanda ukuqala [00:01:00] ngengxaki yezonyango efunwa ngulo msebenzi. I-Adaptive DBS iya isishukumisa ngakumbi ekubeni sinikezele nje ngokulinganisa ukuya kwinkqubo enokumamela ingqondo kwaye iphendule ingqondo kunye nomsebenzi wayo, ewe. Kodwa uphando lwakho luqala kwingcamango yokuba ukurekhoda ngokuthe ngqo kwiindawo ezinzulu zobuchopho akusoloko kuthembekile njengoko besingathanda.

Ngowuphi umda ophambili we-subcortical sensing kwiinkqubo ze-DBS zangoku, kwaye yintoni ekwenze wacinga ukuba umsebenzi we-cortical unokubonelela ngenye indlela yokujonga okwenzekayo nakwiindawo ezinzulu zobuchopho?

UGqr. Zixiao Yin: Ewe, ngumbuzo olungileyo kakhulu lowo. Ngoko ke njengoko ubutshilo, ukuvuselela ingqondo nzulu kule mihla kutshintshela kwindlela yokusoloko usenza izinto eziguquguqukayo. Ngoko ke isixhobo sinokumamela ingqondo size silungise ukuvuselela ngexesha langempela. Kodwa, umsebenzi wethu uzama ukujongana nengxaki yokumamela.

 Ngoko ke ukumamela [00:02:00] kunokuba yinxalenye ebuthathaka. I-electrode efanayo njengoko siyazi enikezela ukuvuselelwa kwe-130 hertz nayo kufuneka irekhode ezo zibonakaliso zincinci kakhulu ze-microvolt, kwaye izinto ezivuselelayo zinokuziphazamisa kwaye i-hardware ngamanye amaxesha inokungaphumeleli. Inkqubo yangoku ikunika itshaneli enye okanye ezimbini ze-bipolar ukurekhoda imiqondiso evela kwingqondo enzulu.

Ngoko ke zonke ezo ziimida zeenkqubo zokuva zangoku. Yiyo loo nto sifuna ukuyiphucula, ukuyenza yomelele ngakumbi, ukwenza inkqubo ye-closed loop. Kwaye i-cortical sensing yenye yeendlela esifuna ukuyicebisa.

UGqr. Michele Matarazzo: Kwaye ubuncinane kumava akho, ngokwemigaqo esebenzayo, zingaphi izihlandlo ezi zinto zisetyenziswayo okanye ezi zinto zikhutshwayo, okanye umda, zi-kangaki ziyingxaki ebalulekileyo kwi-adaptive DBS okwangoku?

UGqr. Zixiao Yin: Ndicinga ukuba ekliniki, njengoko, kwi-adaptive DBS isixhobo seMedtronic Percept singena ekliniki njengesixhobo sorhwebo ngokusesikweni. Kwaye [00:03:00] eTshayina, sinezinye iimpawu ezifana nePINS. Ndingathi ukuvuselela kunokubangela ukuphazamiseka komgangatho wesiginali kwaye ezinye izinto ezifana nokuhamba nazo zinokuba yingxaki.

Kwaye ndicinga ukuba ezo ziingxaki zonke eziya kunciphisa umgangatho wesiginali. 

UGqr. Michele Matarazzo: Ngoku, enye inkalo enomdla kuphando lwakho ikwayidatha, ubungakanani bedatha. Sithetha ngeeyure ezingama-723 zokurekhodwa ngaxeshanye kwe-cortical kunye ne-subcortical. Ukuba andiphazamanga, yayizizigulane ezingama-49 kwiindawo ezintathu zaseBerlin, eBeijing, naseSan Francisco.

Kwaye okubalulekileyo kukuba ubuncinane oko ndikufundileyo kwisishwankathelo sakho ezi rekhodi zigubungela iimeko ezahlukeneyo zokuziphatha, imeko yonyango, iinjongo zobuchopho, kunye neengxaki ezahlukeneyo zokunyakaza. Ungasichazela ngale datha? Yintoni kanye kanye obuyirekhoda kwi-cortex, kwaye ubuyirekhoda ntoni kwingqondo enzulu, kwaye phantsi [00:04:00] zeziphi iimeko?

UGqr. Zixiao Yin: Ewe, kulungile oko. Okokuqala, ndingathanda ukubulela umcebisi wam, uWolf-Julian Neumann, okwiSwitzerland ngoku, kodwa wayeseBerlin, kwaye waququzelela iseti yedatha enkulu kangaka kuquka izigulana kumazwekazi amathathu, njengaseMelika, eYurophu, naseTshayina. Ukuba ngokwenene sifuna le cortical decoding brain signals isebenze, le decoder isebenze, ngokwenene sifuna zonke iintlobo zokuziphatha kunye nazo zonke iintlobo zezigulana, kwanokuxilongwa okwahlukileyo.

Ngoko ke apha kule seti yedatha, sirekhoda izigulane ezinesifo sikaParkinson kunye nezigulane ezine-dystonia, kwanezigulane ezine-Tourette syndrome, kwaye sirekhoda xa besenza ujikelezo lwezandla, oko kukuthi intshukumo, kwaye ngelixa belele. Ngoko ke sinemeko yokuvuka nokulala. Kwaye kwakhona, ekubeni izigulane ezinesixhobo sokuvuselela ingqondo esinzulu zineendlela zonyango, umzekelo, amayeza kunye nokuvuselela.

Ngoko ke sikwahlolisise ukuba ingaba i-decoder yethu ingasebenza na [00:05:00] kwiimeko ezahlukeneyo zonyango ezifana nokuvala/ukuvala i-STN kunye nokuvala/ukuvala amayeza. Ngoko ke le yindlela idatha yethu egubungela ngayo.

UGqr. Michele Matarazzo: Iyamangalisa. Ngoku, ngalo lonke olu lwazi kunye nale datha, uphuhlise inethiwekhi ye-convolutional neural ebizwa ngokuba yi-CTXNet ethatha imiqondiso ye-cortical kwaye iqikelele umsebenzi we-subcortical. Ungacacisa ukuze sonke sikuqonde oku, sisazi ukuba asingochwepheshe kulo hlobo lokubonisa imodeli?

Loluphi ulwazi olufunyanwa yimodeli, kwaye ucela ukuba imodeli iqikelele ntoni?

UGqr. Zixiao Yin: Le modeli ifumana ulwazi oluvela kwi-cortex. Sisebenzisa le ECoG, i-electrocorticography, ukuze siqonde imiqondiso ye-cortical. Ngokusisiseko ezo zii-brain waveforms ezibalwa emva kokucubungula kwangaphambili, size emva koko sifumane zonke iintlobo zolwazi ezifana ne-spectral band kunye ne-waveform features esinokuzondla ezo mpawu kule decoder.

Kwaye [00:06:00] njengoko ubutshilo kwicandelo lenethiwekhi ye-convolutional neural sizama ukwakha kwakhona iimpawu ze-spectrum zobuchopho obunzulu. Ngoko ke sikwinqanaba elo, asiyi kuphinda sivelise isignali epheleleyo yobuchopho obunzulu, kodwa sifuna nje ukwakha kwakhona, umzekelo, amandla e-beta ngokuhamba kwexesha kwingqondo enzulu, kwi-nucleus engaphantsi kwe-thalamic.

Yiloo nto ke eyayiyingeniso nemveliso yaloo datha.

UGqr. Michele Matarazzo: Kwaye uxela ukuba le ndlela ye-convolutional neural network isebenze ngcono malunga ne-40% kuneendlela ze-spectral feature eziqhelekileyo, ubuncinane ekuchazeni umsebenzi we-STN band. Oku kusixelela ntoni? Yintoni le nkqubela ye-40%? Ngaba kwanele oko? Ngaba kuchanekile oko?

UGqr. Zixiao Yin: Ewe, ngumbuzo onomdla lowo. Ngoko ke, i-40%, into esiyithelekisayo ziimodeli zokufunda koomatshini eziqhelekileyo, kuba okokuqala sifuna ukuzama ukwakha ubudlelwane bokumakisha phakathi kwe-cortex kunye nengqondo enzulu. Ngoko ke kuqala sisebenzisa ezo modeli zokufunda koomatshini [00:07:00], sithatha uthotho lwexesha oluqhelekileyo okanye iimpawu ze-spectrum kwimodeli size sizame ukuqikelela amandla e-beta kwingqondo enzulu.

Kodwa ingxaki ngaloo modeli kukuba ayinguye wonke umntu ochwephesha kweli candelo, ngoko ke asisoloko sazi ukuba zeziphi iimpawu ezinika ulwazi ezifunekayo kwimodeli. Ngoko ke amandla enethiwekhi ye-convolutional neural kukuba ithatha imiqondiso eluhlaza kwi-cortex, ize emva koko ikhethe ngokuzenzekelayo iimpawu ezinika ulwazi eziveliswe kwezo zimpawu ziluhlaza ize ilungise ubunzima bazo kwinethiwekhi ye-neural ize ikhuphe imveliso.

Ngoko ke, ndicinga ukuba yindlela olu phuculo lwe-40% oluphuma ngayo. Kwaye malunga nokuba luphezulu kangakanani kuba ngokubanzi kufuneka ndivume ukuba ulwalamano olupheleleyo lwePearson phakathi kwamandla e-beta akhiwe ngokutsha kunye namandla e-beta okwenyani, i-coefficient yolwalamano imalunga noluhlu lwe-0.3 ukuya kwi-0.5.

Oko akuphezulu kangako, kodwa thina [00:08:00] singaqinisekisa ukuba kolu seti lokuchaneka, sinokubona ngokuthembekileyo imisebenzi yokuqhuma kwe-pathological engqondweni enzulu yesifo sikaParkinson, sithi malunga ne-70% yokuchaneka. Kusekho iindawo zokuphucula, kodwa oku kufana nobungqina bengcamango.

UGqr. Michele Matarazzo: Kuhle kakhulu. Enye into oyenzileyo kuphando lwakho, kuba, ukuqikelela i-biomarker enye efana nomsebenzi we-beta yinto enye, kodwa usebenzise neemodeli ze-denoising diffusion probabilistic ukuze wakhe kwakhona isignali yobuchopho enzulu engacacanga. Ungachaza ukuba oku kuthetha ukuthini kwaye kusebenza njani oku?

UDkt. Zixiao Yin: Ndicinga ukuba lelona candelo linomdla kakhulu kulo msebenzi. Ngaphambili besinqwenela ukuba ukuba singakha kwakhona i-beta, loo nto sele imangalisa. Loo nto sele inamandla kakhulu. Kwaye ke sifuna inyathelo elinye eliya phambili ukuze sibone ukuba ingakha kwakhona na isignali epheleleyo erekhodwe kwingqondo enzulu.

Ingxaki ye [00:09:00] yokubhala ikhowudi yebhendi enye kuphela kukuba sifuna idekhoda enye eyahlukileyo yebhendi enye. Oko kuthetha ukuba ukuba sifuna ukubhala ikhowudi ye-beta, ngoko ke sifuna idekhoda ye-beta. Ukuba sifuna ukubhala ikhowudi ye-alpha, ngoko ke sifuna idekhoda ye-alpha. Kodwa ukuba singakha kwakhona isignali eluhlaza, ngoko ke singayithatha loo signal size sibale naziphi na iibhendi esizifunayo.

Ngoko ke i-DDPM, imodeli ye-denoising diffusion probabilistic oyikhankanyileyo ifana nemodeli engasemva kwexesha langoku lokuvelisa imifanekiso. Ngoko ke ayithandi ukuba imodeli ye-regression ivelise inqaku elinye emva kwelinye. Ivelisa uthotho lwamanani, lwamaxabiso.

Ngoko ke yimiqondiso ye-STN engavuthwanga. 

UGqr. Michele Matarazzo: Kuhle kakhulu. Njengoko ubukhankanya, uye wahambela phambili ngengqiqo yokuba ubungathelekisi nje oku nomsebenzi wobuchopho obunzulu, kodwa uphinde wafunda nokuba oku kungaxela ubukhali bentshukumo yomzimba, akunjalo?

UGqr. Zixiao Yin: Njengokuba sithatha amandla e-STN aqikelelweyo, isignali eluhlaza ye-STN, [00:10:00] size sibale amandla e-beta avela kwezo zimpawu ezifakwe kwi-im-imputed. Emva koko sisebenzisa ezo mpawu ukuqikelela amanqaku obunzima be-motor yesigulana se-UPDRS, kwaye sinokubona ukuba ixabiso lokwenyani le-UPDRS xa lithelekiswa nexabiso lethu le-UPDRS eliqikelelweyo, ulwalamano lumalunga ne-0.7.

Oko kuphezulu kakhulu. Ndifuna ukucacisa into apha, njengoko benditshilo, ukuba unxibelelwano lwamandla e-beta phakathi kwesignali yokwenyani kunye neyokubalelwa lumalunga ne-0.3 ukuya kwi-0.5, kodwa oko kungaphakathi kwesigulana. Ngoko ke amandla e-beta esigulana ayatshintshatshintsha, kodwa asemalunga nenqanaba elithile, kwinqanaba elithile lomntu ochaphazelekayo.

Kodwa ukuba senza oku kuzo zonke izifundo, inqanaba elisisiseko lesifundo ngasinye liphantse lahluke. Kwaye iimodeli ze-DDPM kunye ne-neural network zinokubamba ngokuchanekileyo loo mahluko phakathi kwezifundo ezahlukeneyo ngcono kunomahluko ongaphakathi kwesifundo. Yiyo loo nto.

UGqr. Michele Matarazzo: Kulungile. [00:11:00] Ngoku sibuyela umva ukususela ekuqaleni kolu dliwanondlebe sithetha ngengxaki yezonyango esifuna ukuyisombulula. Ngoku ekubeni sisazi ngakumbi ngedatha yakho masizame ukucinga ngesicelo seklinikhi esisebenzayo. Kwisishwankathelo, ucebisa ukuba oku ekugqibeleni kunokubangela i-neuromodulation ye-closed loop engangenisi kusetyenziswa i-scalp EEG.

Kwaye kukho amanyathelo amabini ahlukeneyo okuguqulela apha. Okokuqala, ukusebenzisa ukurekhoda kwe-cortical okufakelweyo ukwandisa okanye ukulinganisa, okanye mhlawumbi ukusindisa ukuqonda okungaphantsi kwe-cortical, kwaye okwesibini, okunokuthi kuthathe indawo yokuqonda kwe-cortical okufakelweyo ngento engangenisi kakhulu, into, masithi, onokuba nayo ngaphandle kwentloko.

Ngowuphi kula manyathelo mabini ocinga ukuba usondele kakhulu ekusetyenzisweni kwezonyango okanye ngowuphi kwezi ndlela zimbini ezahlukeneyo ocinga ukuba unokuba luncedo ngakumbi kwindlela yokusebenza kwezonyango?

UGqr. Zixiao Yin: Ewe ngumbuzo olungileyo kakhulu lowo. Ukuba ndithetha inyani, ndicinga ukuba zombini zixabiseke kakhulu. Kwaye malunga nokuba yeyiphi ekufutshane nesicelo seklinikhi se-[00:12:00], ndicinga ukuba yeyokuqala ukusindisa ukuqonda kwengqondo enzulu. Kuba umzekelo, ukuba singaba ne-alg-algorithm kuqala, ifumanisa izinto ezikwingqondo enzulu, kwaye xa ivakalelwa kukuba kukho into enamandla enokuthintela ukuqonda kwengqondo enzulu njengesignali yempendulo yenkqubo ye-closed loop, ngoko inokutshintsha ukusuka kwi-deep brain sensing ukuya kwi-cortical sensing, kwaye emva koko ukuqonda kwengqondo ngokwayo sele kuxabisekile, kwaye emva koko inokubhekisa kwiimpawu zobuchopho ezinzulu kwaye yongeze oko njengomqondiso wempendulo.

Kwaye xa iyonke, oku kuya kwenza inkqubo ye-closed loop ibe namandla ngakumbi. Kwaye ingongoma yakho yesibini kukuba mhlawumbi singasebenzisa umzekelo, i-scalp EEG ukukhokela ezo ndlela zonyango zokuvuselela ezingangenisi ntsholongwane. Ndicinga ukuba nayo leyo ingongoma ilungileyo. Umzekelo, ngoku sinazo ezo ziphazamisi zexesha okanye i-ultrasound egxile kwi-focused ultrasound ezinokusebenzisa ukuvuselela kwiindawo ezinzulu zobuchopho.

Ukuba [00:13:00] sifuna ngokwenene ukwenza ezo ndlela zonyango zibe yi-closed loop, sidinga nesignali yempendulo. Kwaye kwiimeko ezininzi, umzekelo, kwisifo sikaParkinson, i-pathological oscillation, eyona iqinileyo, njengoko siyenzile ngoku ngamandla e-beta kwi-STN. Ngoko ke ukuba singabeka, kungekhona i-scalp EEG, kodwa singaba yi-ECoG, size siva imiqondiso ye-cortical size siyisebenzise ukuze siphinde sinike imiqondiso yokufumana ulwazi malunga nemeko yengqondo enzulu, size ke siyisebenzise ukukhokela ukuhanjiswa kokuvuselelwa okungenalo uhlaselo.

Emva koko akufuneki sifake i-electrode engqondweni enzulu ngenxa yesizathu sokuziva kuphela.

UDkt. Michele Matarazzo: Kuhle kakhulu. Zixiao, mandigqibe ngombuzo wokugqibela, othi: Uceba ukwenza ntoni ngoku? Ndicinga ukuba usasebenza kule ndawo. Uceba ukufunda ngakumbi indlela yokusebenzisa oku kwiklinikhi. Lithini ikamva-- ikamva lakho njengomphandi oselula, kwaye lithini ikamva lale teknoloji?

UGqr. Zixiao Yin: Ewe, enkosi ngombuzo. Ndicinga ukuba kusekho imida emininzi yale ndlela. Njengoko benditshilo, ndicinga ukuba okwangoku, inqanaba layo langoku lifana nobungqina bengcamango, kodwa ukuba sifuna ngokwenene ukuyakha kwinkqubo yexesha langempela enokulibaziseka okuphantsi kakhulu, kwaye xa isixhobo esifakelwayo, nazo zinememori encinci kakhulu. Asinakufaka imodeli enkulu kuyo, ngoko ke kufuneka sinciphise imodeli kwaye sinciphise ukulibaziseka kokubala.

Ndicinga ukuba yiloo manyathelo alandelayo esiza kuzama ukuwasebenzisa xa sijonga le ndlela yonyango. Ukongeza, umsebenzi wam wangoku nawo uyaqhubeka-- Ngoko ke le yindlela yokuva. Njengoko benditshilo, le ndlela ingasetyenziselwa ukunyusa ukufunyanwa kwesimo. Umzekelo, sifumanise ukuba ngale ndlela yokucinga evela kwi-cortex, imeko yokuhamba okanye imeko yokulala kwezinye izigulane, ukuchaneka kokubhala ikhowudi kunokuphuculwa ukusuka malunga ne-60% ukuya malunga ne-80% umzekelo, kwingxaki yokuhlelwa kwesigaba sokulala.

Ngoko ke ndicinga ukuba eli licala elinye [00:15:00] lenkqubo ye-closed loop, icala lokumamela, kwaye okwangoku sisebenza kwicala lokuvuselela, elo liyindlela engcono yokuvuselela ingqondo xa sibona imeko ethile. Ewe, njengokuba singayisebenzisa njani ngcono i-stimulation xa isigulane silele, silele ngaphandle kwe-REM.

Kodwa ke elo libali elahlukileyo. 

UDkt. Michele Matarazzo: Siza kulindela ukumamela loo nxalenye yebali kwixesha elizayo. Zixiao, ndiyakuhalalisela kwakhona ngokufumana iMbasa yabaNtwana, kwaye enkosi ngokusihambisa kulo msebenzi unomdla kakhulu.

UGqr. Zixiao Yin: Enkosi. Enkosi.

UDkt. Michele Matarazzo: Ngoko ke kwabo bazimase iNkomfa ye-MDS eSeoul, ndinikhuthaza ukuba niyijonge le powusta kwaye le powusta iza kuboniswa kwiseshoni epheleleyo. Khumbula ukuba le powusta inesihloko esithi "Ukuchaza uMsebenzi woBuchopho kwiiSignali zeCortical Usebenzisa iDeep Learning Across Movement Disorders."

Enkosi nonke ngokujoyina.

 [00:16:00]