Ib Tandem Segmentation-kev faib tawm txoj hauv kev rau Lub Zos ntawm Morphological Predictors Ntawm C. Elegans Lifespan Thiab Motility
Sep 26, 2022
Thov hu rauoscar.xiao@wecistanche.comyog xav paub ntxiv
Abstract
C. elegans yog ib hom kab mob tsim los rau kev kawm txog noob caj noob ces thiab tshuaj cuam tshuam rau kev laus, ntau yam uas tau khaws cia hauv tib neeg. Nws kuj yog ib qho qauv tseem ceeb rau kev tshawb fawb hauv paus, thiab C. elegans pathologies yog qhov chaw tshiab. Ntawm no peb tsim cov pov thawj-ntawm-tus thawj xibfwb convolutional neural network-raws li platform rau ntu C. elegans thiab rho tawm cov yam ntxwv uas yuav pab tau rau kev kwv yees lifespan. Peb siv cov ntaub ntawv ntawm 734 worms taug qab thoob plaws hauv lawv lub neej thiab faib cov worms rau hauv lub neej ntev thiab luv luv. Peb tsim WormNet- convolutional neural network (CNN) los kwv yees cov kab mob hauv lub neej raws li cov tub ntxhais hluas cov duab (hnub 1-hnub 3 cov neeg laus)) thiab pom tias WormNet, ib yam nkaus, InceptionV3 CNN tuaj yeem ua tiav kev faib tawm txoj sia. Raws li U-Net architecture peb tsim HydraNet CNNs uas tso cai rau segmenting ntawm cov cab kom raug rau hauv lub cev, nruab nrab ntawm lub cev, thiab tom qab. Peb muab cov segmentation HydraNet, WormNet twv ua ntej, thiab daim ntawv qhia kev ua haujlwm hauv chav kawm los txiav txim siab cov ntu tseem ceeb tshaj plaws rau kev faib tawm lub neej. Xws li tandem segmentation-classification mus kom ze qhia tau hais tias tom qab ib feem ntawm cov cab kuj tseem ceeb dua rau kev faib cov kab mob ntev ntev. Peb txoj hauv kev tuaj yeem muaj txiaj ntsig zoo rau kev nrawm ntawm kev tshawb nrhiav tshuaj tiv thaiv kev laus thiab kev kawm C. elegans pathologies.
Thov nias ntawm no kom paub ntxiv
Taw qhia
Lub nematode Caenorhabditis elegans (C. elegans) yog ib qho qauv tsim los kawm txog ntau yam kev cuam tshuam rau hauv cov txheej txheem kev laus, uas tso cai rau pom ntau cov noob thiab tshuaj cuam tshuam nrog kev laus. 5 tawm ntawm 7 Tier 1 thiab 4 tawm ntawm 6 Tier 2 cov tshuaj tiv thaiv kev laus raug txiav txim siab rau tib neeg kev sim ua kom muaj sia nyob hauv C. elegans qauv.cistanche แอมเวMuaj ntau txoj kev laus uas tau khaws cia ntawm cov tsiaj thiab cov kab mob yuav tsum tau siv dav tsis yog hauv kev tshawb fawb ntev xwb tab sis kuj tseem tshwm sim hauv kev lag luam los tiv thaiv kev laus [1]. Tsis tas li ntawd, humanized worms yog tam sim no siv los tsim cov qauv cog lus rau neurodegeneration [2]. Txawm li cas los xij, tsis zoo li cov noob caj noob ces ntawm kev ua neej ntev, C. elegans phenotypes ntawm kev laus tseem tsis tau kawm zoo. Tshwj xeeb, peb paub me ntsis txog cov hnub nyoog ntsig txog pathologies thiab lawv txoj kev loj hlob, nrog rau, uas pathologies txiav txim siab txog kev ua neej thiab yuav ua li cas lawv ua rau tuag [3]. Ob peb pathologies suav nrog plab atrophy, uterine qog thiab pharyngeal kab mob tau piav qhia tsis ntev los no [4-6]. Nyob rau hauv lub teeb no, nrhiav pom tshiab C. elegans pathologies, tshwj xeeb tshaj yog txiav txim siab lifespan, yog ib qho nyuaj heev. Kawm txog pathologies hauv C. elegans yuav pab kom nkag siab zoo txog cov txheej txheem kev laus, nrog rau, cov txheej txheem thiab cov teebmeem ntawm cov tshuaj tiv thaiv kev laus.
cistanche tuaj yeem tiv thaiv kev laus
Kev nce qib tsis ntev los no hauv kev kawm tshuab (ML) thiab kev kawm sib sib zog nqus (DL)[7] tuaj yeem pab cov kev tshawb fawb kev laus ua haujlwm C.elegans los ntawm kev nthuav tawm thiab sau cov ntsiab lus yav dhau los tsis pom kev coj tus cwj pwm thiab cov qauv morphological hauv cov ntaub ntawv sim loj. Piv txwv li, nyob rau hauv ib tug tsis ntev los no kev ua hauj lwm ob peb physiological tsis tau ntsuas longitudinally thiab ib daim ntawv thov ntawm kev txhawb vector regression tso cai rau piav qhia txog qhov sib txawv ntawm qhov sib txawv ntawm C. elegans lifespan los ntawm: txav mus los (57 feem pua ), cross-sectionalautofluorescence (52 feem pua), oocyte nteg tus nqi (28 feem pua) [8]. Interestingly, nws tau pom tias qhov loj me me sib txuas nrog lifespan hauv mated hermaphrodites (r=0.28)[9]. Tsis tas li ntawd, cov kev tshawb fawb ywj pheej paub meej tias cov leeg nqaij ua haujlwm yog qhov zoo tshaj plaws kev kwv yees lub cev muaj zog: ceev pharyngeal pumping span (r=0.49), thiab pharyngeal pumping span (r=0.83) tau pom tias yog muaj feem cuam tshuam nrog lub neej ntev [10]. Tsis tas li ntawd, qhov siab tshaj plaws tshaj plaws nyob rau hnub 9[11] thiab tus nqi nrawm nrawm (hnub 3-9)[12] kwv yees 71 feem pua thiab 91 feem pua ntawm cov kev hloov pauv hauv lub neej raws li. Cellular thiab molecular kwv yees ntawm C. elegans lifespan ntev kuj tau pom. Kev nthuav qhia ntawm hsp-16.2 tshwm sim los ntawm kev poob siab thaum tshav kub kub hauv hnub 1 cov neeg laus tau pom tias muaj feem cuam tshuam nrog kev ua neej nyob [13]. Tsis muaj kev cuam tshuam tsis zoo ntawm kev cuam tshuam xws li kev poob siab thaum tshav kub kub, basal qhia ntawm sod -3 nyob rau hnub 9 kuj cuam tshuam nrog lifespan (r=0.57), uas tej zaum yuav qhia txog cov lus teb rau cov zaub mov pathogenic [14]. Mir-71 kev qhia txij hnub 4 mus txog rau tuaj yeem kwv yees tau zoo heev thiab piav qhia txog 47 feem pua ntawm kev hloov pauv hauv lub neej [15]. Strikingly, ib tug muaj zog inverse correlation (r =-0.93) nruab nrab nucleolar loj (tseem rau hnub 1) thiab lub neej ntev qhia tau hais tias deregulated protein synthesis raws li ib tug tseem ceeb feem ntawm kev laus[16]. Qhov tseem ceeb, thaum ntxov ntawm Machine Vision mus kom ze kuj tau siv los faib cov phenotypes kev laus hauv C. elegans. Tshwj xeeb, linear discriminant classifier tau siv los cais cov duab ntawm pharynxes ntawm cov hnub nyoog sib txawv rau cov yam ntxwv ntawm molecular tom ntej [17].
Ntawm lwm txoj hauv kev, ib qho ntawm cov tshuab muaj zog tshaj plaws kev kawm, tshwj xeeb tshaj yog rau kev txheeb xyuas duab, yog kev siv cov kev sib txuas lus neural (CNN) [18], uas tau tshwm sim los ntawm lub koom haum pom cortex neural network. CNN tau tso cai kom ua tiav cov txiaj ntsig zoo hauv kev lees paub cov duab, nrog rau tib neeg kev ua tau zoo ntawm MNIST dataset thiab ua tau zoo dua tib neeg ntawm kev paub txog kev tsheb khiav los ntawm ib qho ntawm ob [19]. CNN rov pom qhov ua tau zoo tshaj plaws thaum lub sijhawm "ImageNet Loj Scale Visual Recognition Challenge" hauv kev faib cov duab [20,21].npaum li cas cistanche cojKev taw qhia txog kev hla kev sib txuas rau CNN tau txhim kho lawv qhov nrawm thiab raug, thiab cov CNN seem tam sim no yog lub xeev-ntawm-tus-kos duab rau kev faib duab [22, 23]. Encoder-decoder residual networks xws li U-Net [24], V-Net, thiab Tiramisu kuj ua tau zoo tshaj qhov qub ciam teb rho tawm, qhov pib, thiab cheeb tsam raws li txoj hauv kev siv hauv cov duab kho mob segmentation teb [25]. Txawm hais tias muaj txiaj ntsig zoo nrog DL txoj hauv kev, ib qho ntawm cov teeb meem tseem ceeb yog DL tes hauj lwm yog lub thawv dub yog li nws nyuaj kom tau txais cov yam ntxwv tseem ceeb rau kev txiav txim siab los ntawm lub network [26]. Txhawm rau tiv thaiv qhov tsis txaus no, ntau cov txheej txheem saliency tau thov [27-29]. Ib qho txheej txheem zoo li no yog siv cov txheej txheem thoob ntiaj teb nruab nrab los tsim ib daim ntawv qhia kev ua haujlwm hauv chav kawm (CAM) thiab ua kom pom cov duab tshwj xeeb hauv chav kawm hauv ib qho kev tsis saib xyuas [30]. Cov yam ntxwv uas tsim muaj hauv zos sib sib zog nqus tuaj yeem pab cov kws tshawb fawb kom nkag siab txog lub hauv paus ntawm kev ntxub ntxaug siv los ntawm CNNs rau lawv txoj haujlwm. Txawm li cas los xij, txog tam sim no, tsis muaj txoj hauv kev los ua ke cov duab muaj txiaj ntsig zoo nkauj thiab kev faib tawm kom muaj txiaj ntsig los pab txhawb kev tshawb pom phenotype los ntawm kev txhais lus tau tsim.
Zoo kawg li, CNN tsis ntev los no tau siv los kwv yees lub neej nyob hauv cov kab mob. Hauv thawj daim ntawv, cov ntaub ntawv ntawm 913 cov duab ntawm C. elegans tau siv. Txhua lub sij hawm taw tes (hnub) muaj tsawg kawg yog 30 worms, thiab tag nrho cov ntawm lawv tau tshuaj loog ua ntej imaging. InceptionResNetV2-raws li architecture ua tiav qhov kev ua yuam kev tsis raug (MAE) ntawm 0.96 hnub nyob rau hauv hom regression, thiab qhov tseeb ntawm 57.6 feem pua hauv hom kev faib tawm [31]. Hauv lwm txoj haujlwm, cov kws sau ntawv siv lub tshuab ntsuas tsis siv neeg uas muaj peev xwm taug qab tib tus kab mob hauv lub neej tag nrho, yog li lawv muaj cov ntaub ntawv rau 734 tus kab mob uas cov duab tau coj txhua txhua 3.5 teev. Lawv siv U-Net los faib cov kab cab los ntawm keeb kwm yav dhau los thiab tom qab ntawd ua cov cab lub cev sib koom ua ke los tsim cov kab sawv cev ncaj nraim. Tom qab ntawd lawv tau siv qhov hloov kho ResNet34 thiab tswj kom rov qab muaj hnub nyoog worm nrog tsawg MAE ntawm 0.6 hnub rau cov duab nyoos [32].
Ntawm no peb siv cov ntaub ntawv tib yam li hauv [8, 32], txawm li cas los xij tsis yog kev kwv yees hnub nyoog ntawm txhua tus cab, peb tsim CNN-raws li platform peb hu ua WormNet muaj peev xwm faib cov tub ntxhais hluas (hnub 1-3) rau hauv luv luv. thiab ua neej nyob ntev, thiab tseem tsim txoj hauv kev rau kev rho tawm cov yam ntxwv tseem ceeb rau kev faib tawm zoo li no. Ib yam li ntawd, peb tau thov WormNet los faib C. elegans txav. Txhawm rau txhais cov txiaj ntsig kev faib tawm hauv ib qho kev tsim qauv, peb tau nrog kev faib tawm CNN nrog kev sib tw segmentation CNN. Rau qhov no, peb tsim ib qho tshiab U-Net-raws li architecture (HydraNet) rau segmenting worms los ntawm keeb kwm yav dhau thiab tseem segmenting tus cab lub cev mus rau hauv anterior, nruab nrab-lub cev thiab posterior qhov chaw. Kev txhais lus ntawm kev faib cov txiaj ntsig tau ua tiav los ntawm kev sib koom ua ke ntawm HydraNet segmentation thiab cov ntawv qhia ua haujlwm hauv chav kawm uas tsim los siv WormNet. Cov chav kawm ua kom daim duab qhia kev txheeb xyuas ua ke nrog lub cev segmentation nyob rau hauv xws li tandem zam tso cai rau peb rho tawm nta lub luag hauj lwm rau lifespan kwv yees. Thaum kawg, siv qhov kev daws teeb meem siab dua segmented version ntawm C. elegans cov duab, peb tau txheeb xyuas peb cov txiaj ntsig hauv qhov muaj peev xwm ntau dua qhov seem CNN InceptionV3 nrog kev txhais lus txhais lus.
TSEEM CEEB
Lub sij hawm-lapse cov ntaub ntawv rau 734 C. elegans ntes txij hnub 1 ntawm kev laus mus txog rau thaum tuag tau siv los tsim peb cov qauv qauv [8,15]. Txhawm rau tsim txoj hauv kev rau kev txhais cov duab ntawm cov duab no peb tau hais txog qhov teeb meem ntawm segmenting worms los ntawm lawv keeb kwm yav dhau, nrog rau qhov txawv ntawm cov kab mob morphological (Daim duab 1). Rau qhov no, peb tau sau tseg 130 cov duab ntawm cov neeg laus cov kab mob nrog lub qhov ncauj qhov ntswg rau sab hauv, nruab nrab ntawm lub cev, tom qab ntawm tus cab thiab suav nrog rau tag nrho cov kab npog npog (Daim duab 1F-1H). Cov ntaub ntawv no tau muab faib rau hauv lub tsheb ciav hlau (90) thiab xeem (40) feem raws li cov ntaub ntawv ID ntawm tus cab cab kom ntseeg tau tias tus kab mob tus kab mob yuav tsis xau mus rau qhov ntsuas- tawm. Ua ntej, los daws qhov teeb meem tag nrho cov kab segmentation peb tau tsim ib qho chaw ntiav ntiav zoo ib yam li U-Net [24] nrog rau lub taub hau sigmoid rau kev faib binary. Kom meej meej, qhov encoding thiab decoding qhov chaw ntawm U-Net yog qhia nyob rau hauv daim duab lA raws li ib tug thiab . Cov duab nyoos tau scaled rau 96 × 96 pixels rau xam efficiency. Peb siv Dice poob muaj nuj nqi thiab saib xyuas Jaccard Performance index los ntsuas qhov segmentation zoo. Ntawm qhov teeb meem segmentation yooj yim no Jaccard index mus txog 0.97 ntawm ob lub tsheb ciav hlau thiab kev xeem feem (Daim duab 1A, 1B, saib Cov Khoom Siv thiab Cov Txheej Txheem rau cov ncauj lus kom ntxaws hyperparameters). Tom ntej no, txhawm rau txuas ntxiv txoj hauv kev no mus rau ntu ntu ntawm tus kheej lub cev ntawm C. elegans peb tau hloov kho qhov teeb meem raws li kev faib ntau chav nrog ib-kub encoded lub qhov ncauj qhov ntswg thiab zoo li U-Net-zoo li architecture (Daim duab 1C, 11). Unsurprisingly, txij li kev faib ntau chav kawm yog ib qho teeb meem nyuaj, qhov no ua rau kev ua haujlwm tsis zoo ntawm 0.92 thiab 0.91 Jaccard index ntawm lub tsheb ciav hlau thiab kev xeem feem feem qhia txog qhov kev ua haujlwm me me.
Qhov zoo kawg li, ib feem ntawm ntau chav kawm U-Net tsis ua tau zoo yog qhov txawv ntawm sab hauv thiab tom qab ntawm tus cab uas ua rau tsim cov qhov ncauj qhov ntswg sib tshooj (Daim duab lI). Txhawm rau hla qhov kev txwv no, peb tau tsim ib qho kev tsim qauv siv U-Net a thiab qhov chaw, nrog rau ntau qhov chaw rau txhua qhov rau nws tus kheej binary segmentation teeb meem (Daim duab 1D, 1E), uas peb hu ua HydraNet. Xws li txoj hauv kev tsim kom muaj kev sib koom ua ke nrog cov txheej txheem nkag thiab cov txheej txheem rau txhua qhov morphological ntawm tus cab, uas tso cai rau kom muaj tus qauv kawg-rau-kawg, thaum daws qhov teeb meem sib faib binary yooj yim. HydraNet3 tau nruab nrog 3 qhov tshwj xeeb rau sab hauv, nruab nrab lub cev, thiab tom qab qhov chaw ntawm lub cev cab. HydraNet4, nyob rau hauv lem, tau nruab nrog 4 qhov chaw mob siab rau lub anterior, nruab nrab-lub cev, posterior qhov chaw raws li zoo raws li tag nrho cov cab lub cev. Txhawm rau kwv yees kev sib koom ua haujlwm ntawm HydraNet peb ntsuas Jaccard index rau txhua ntu ntawm tus kheej thiab thaum kawg ntsuas qhov nruab nrab Jaccard index. Zoo kawg li, ob qho tib si HydraNet3 thiab HydraNet4 tau ua tiav qhov nruab nrab Jaccard Performance index 0.97 ntawm ob lub tsheb ciav hlau thiab cov kev xeem feem ntau qhia tau hais tias zoo generalization (Daim duab 1D, IE, 1J, 1K). Qhov tsim nyog, HydraNet4 tau ua tiav kev hloov dua siab tshiab ua ntej HydraNet3 (Daim duab 1D, IE insets) qhia txog qhov muaj txiaj ntsig zoo los ntawm kev ua ke nrog cov architecture nrog cov chav kawm semantic ntau dua.
Tom ntej no, kom tau txais classifiers rau C. elegans txav los yog lifespan, peb faib tag nrho 734 worms rau hauv 2 tag nrho cov nqi zog ntawm cov chav kawm: qis los yog siab zog kwv yees li motility saum toj no los yog qis dua qhov nruab nrab qhov deb nkag mus rau hauv lub neej-lub sij hawm; thiab 2 chav kawm lifespan: 'short-lived'with lifespan 7 hnub los yog tsawg dua, thiab 'long-lived' nrog lifespan 8 hnub thiab ntau dua. Txoj haujlwm yog los kwv yees cov chav kawm raws hnub l, hnub 2 lossis hnub 3 duab. Raws li cov ntaub ntawv me me, kev siv cov khoom siv muaj peev xwm nthuav dav tuaj yeem ua rau overfitting. Yog li ntawd, peb tsim ib qho ntiav CNN peb hu ua WormNet. Qhov no architecture muaj 5 convolutional khaubncaws sab nraud povtseg, txhua tus ua raws li ib tug max pooling txheej. Kev tso tawm thiab batch normalization tau ua tiav rau txhua txheej txheej hauv lub neural network txhawm rau txhim kho kev ua kom dav dav.dab tsi yog cistancheQhov kawg max pooling txheej yog flattened thiab txuas mus rau ib tug tag nrho kev cob cog rua txheej ua raws li los ntawm ib tug softmax txheej. Peb siv binary cross-entropy raws li kev poob haujlwm. Tag nrho cov txheej txheem, tshwj tsis yog tom kawg, siv cov kab hluav taws xob hloov kho (ReLU) ua haujlwm ua haujlwm (Daim duab 2A, saib Cov Khoom Siv thiab Cov Txheej Txheem rau cov ncauj lus kom ntxaws hyperparameters). WormNet tau siv los ua kom tau ob qho kev txav mus los thiab kev ua neej nyob (Daim duab 2 thiab 3). Txhawm rau txo qhov muaj peev xwm overfitting, peb tau ua ib qho 30-fold cov ntaub ntawv augmentation siv Keras duab generators. Tshwj xeeb tshaj yog, cov duab yuav raug rau random kab rov tav thiab ntsug flipping, kab rov tav thiab ntsug hloov nyob rau hauv 10 feem pua, nrog rau random rotations nyob rau hauv 90 degrees ntawm qhov qub. Cov khoob hauv cov duab hloov pauv tau sau los ntawm kev siv lub tswv yim tus nqi ze tshaj plaws.
Lub WormNet tau pom qhov ua tau zoo ntawm tag nrho cov kev faib tawm mus txog 88 feem pua qhov raug (precision 0.86, nco qab 0.86, thaj tsam hauv qab nkhaus rau cov neeg txais kev ua haujlwm - AUC ROC - yog 0.56) ntawm cov ntaub ntawv xeem rau hnub 3 cov neeg laus feem. Qhov kev ua tau zoo rau hnub 1 thiab hnub 2 cov duab tau qis dua me ntsis (Daim duab 2B-2D) nrog ROC AUC ntawm 0.51 thiab 0.55 raws. Txhawm rau kom ntseeg tau tias peb qhov kev twv ua ntej tau cuam tshuam feem ntau los ntawm cov kab mob morphology ntau dua li nws qhov chaw nyob ib puag ncig, peb tau tsim cov ntaub ntawv ntawm cov duab hluavtaws tom qab uas C. elegans raug tshem tawm los ntawm segmentation. Txhawm rau txo cov kab mob silhouette ntawm kev cob qhia, peb tau sau cov xoom pixels ntxiv nrog cov suab nrov nrov (Cov duab ntxiv 1). Peb cov txiaj ntsig tau pom tias tus qauv kev ua tau zoo yog feem ntau ntaus nqi rau C. elegans morphology es tsis yog keeb kwm ntawm cov duab. Txhawm rau ntsuas seb lub cev twg yuav yog lub luag haujlwm rau kev txiav txim siab WormNet, siv peb txoj hauv kev tandem segmentation-classification txoj hauv kev peb tau txais CAMs rau cov kab mob qis qis (Daim duab 2E,2F) thiab cov kab mob siab zog (Daim duab 2G, 2H) los ntawm WormNet.bioflavonoidsTom ntej no, txhua daim duab tau segmented siv HydraNet4 thiab lub union ntawm WormNet sab sauv quartile CAM nrog morphological ib feem segmentation los ntawm HydraNet4 tau txais. Rau kev txhais lus, peb tau suav cov feem pua ntawm CAMs uas yog ib feem ntawm morphological ntu rau txhua tus kab mob uas muaj nyob rau qib siab lossis qis zog. Tsis tas li ntawd, peb tau soj ntsuam qhov tseem ceeb ntawm qhov kev txhais lus los ntawm kev tsim qauv siv ib-txoj kev ANOVA nrog Tukey qhov kev ncaj ncees tseem ceeb sib txawv (HSD) kho (Daim duab 2F-tsawg zog worms, Daim duab 2H - siab zog worms). Qhov kev sib piv tau pom tias qhov chaw sab nrauv tau them tsawg dua (31 feem pua) dua li nruab nrab lub cev (34 feem pua) thiab cov tom qab (34 feem pua) rau ob qho tib si qis thiab siab zog. Tsis muaj qhov sib txawv tseem ceeb ntawm nruab nrab lub cev thiab lub cev tom qab.
Tom ntej no, peb siv WormNet los faib cov kab ntev ntev thiab luv luv. Ib yam li kev faib tawm ntawm kev txav mus los, WormNet ua tau zoo dua rau hnub 3 cov neeg laus cov qauv mus txog qhov raug ntawm 72 feem pua (precision 0.73, nco qab 0.71, AUC ROC 0.61) ntawm cov ntaub ntawv xeem, piv rau AUC ROC ntawm 0.53 thiab 0.52 rau hnub 2 thiab 1 raws li.muab cistancheQhov tsis meej pem matrix tsom xam pom tias CNN ua tsis tau zoo nyob rau hauv cov kab mob luv luv (daim duab 3A-3C). Tom ntej no, peb tau txhais cov kev faib tawm uas siv cov tandem ntawm HydraNet4 thiab WormNet nrog rau ib-txoj kev ANOVA statistical test. Nyob rau hauv cov ntaub ntawv ntawm lifespan classification, los ntawm-design txhais tau hais tias ntawm 32 feem pua ntawm cov anterior feem yog tsis tshua muaj pronounced nyob rau hauv CAMs piv rau nruab nrab-lub cev thiab lub posterior seem (Daim duab 3D, 3E-short lifespan, daim duab 3F, 3G-ntev. lifespan). Qhov sib txawv no tsis tshua muaj txiaj ntsig rau lub neej ntev dua li lub neej luv luv. Tsis muaj qhov sib txawv tseem ceeb ntawm nruab nrab lub cev thiab lub cev tom qab.
Txhawm rau txheeb xyuas cov kev tshawb pom no nyob rau hauv ib qho kev ywj pheej peb tau kawm lwm yam kev ua neej nyob uas siv cov seem InceptionV3 architecture [3] nrog rau kev txhais phau ntawv (Daim duab 4). Tsis tas li ntawd, nyob rau hauv cov ntaub ntawv no kom ntseeg tau siab daws teeb meem ntawm CAMs es tsis txhob scaling rau 96 × 96 pixels, tag nrho cov daws teeb meem 900 × 900 dluab cropped rau 800 × 800 pixels (516 × 516 μm) tau siv. Raws li qhov muaj peev xwm hais tau ntau dua CNN, InceptionV3 yog qhov ua rau overfitting ntawm peb cov ntaub ntawv me me (Daim duab 4C, 4D). Txhawm rau tiv thaiv qhov no, peb tau ua tiav kev nres thaum ntxov thaum kev cob qhia. Tsis tas li ntawd, peb segmented cov kab cab los ntawm lawv keeb kwm yav dhau los kom ntseeg tau tias InceptionV3 tsuas yog nthuav tawm nrog rau qhov cuam tshuam ntawm daim duab. InceptionV3 ua tau zoo ib yam li WormNet nrog qhov tseeb txog 70 feem pua ntawm cov ntaub ntawv xeem rau kev faib tawm lub neej (Daim duab 4A). Raws li qhov tandem HydraNet4-WormNet txoj hauv kev los txhais lus, nyob rau hauv rooj plaub ntawm phau ntawv txhais lus, qhov anterior ntawm tus cab tau pom los ntawm InceptionV3 CAM tsawg dua. Qhov tseem ceeb, txawm li cas los xij, vim tias qhov kev daws teeb meem siab dua ntawm cov duab nkag, CAMs tam sim no tau teeb tsa lub cev qhov chaw zoo dua, tso cai rau muab ib feem ntawm lub cev ua tus neeg muaj peev xwm cais tawm hauv txhua kis (Daim duab 4B). Interestingly, qhov kev faib tawm ntawm lub cev qhov tseem ceeb los ntawm CAM qhov kev tshuaj xyuas qhia tau hais tias qhov chaw tom qab yog qhov tseem ceeb tshaj rau kev ua neej nyob ntev worms'classification, qhia tias cov yam ntxwv uas kwv yees lub neej ntev tuaj yeem nyob rau hauv qhov chaw tom qab ntawm lub cev cab.
Kev sib tham
Txawm hais tias C. elegans yog tus qauv classical hauv kev tshawb fawb kev laus nrog ntau tshaj 4000 cov ntaub ntawv luam tawm mus txog hnub tim, thiab kev nce qib hauv cov neeg hlau, cov txheej txheem ntawm kev ntsuas C. elegans lifespan tseem yog phau ntawv thiab ua haujlwm hnyav. Txawm li cas los xij, txoj hauv kev tshiab tau tshwm sim zoo li lub tshuab lifespan siv cov tshuab luam ntawv flatbed rau ib txhij ntsuas qhov muaj peev xwm ntawm cov neeg coob ntawm cov kab mob ntawm daim hlau [34]. Lwm txoj hauv kev yog Worm corals-ib txoj kev siv tshuaj tua kab mob uas tso cai rau taug qab cov kab mob thoob plaws hauv lawv lub neej nrog kev ntsuas kom ntxaws ntau dua [8]. Cov ntaub ntawv ntxaws txog lub cev tsim tawm ntawm Worm corals tau pom tias kev txav mus los, autofluorescence thiab cov ntaub ntawv degradation yog qhov kev kwv yees zoo tshaj plaws ntawm lub neej. Txawm li cas los xij, nws tseem tsis tau paub meej tias qhov tseeb morphological nta cuam tshuam txog pathologies thiab txiav txim siab lub neej ntev. Nws kuj tau pom tias kev ntsuas lub cev ua ntej hnub 3 lossis 4 ntawm cov neeg laus thiab ib leeg GFP sau npe biomarkers tsis tuaj yeem paub qhov txawv ntawm cov kab mob luv luv thiab ntev ntev [8,15]. Nucleolar-raws li kev kwv yees ua rau hnub l cov neeg laus tau ua tiav siv 100 × magnification ntawm cov kab mob ruaj khov, uas tsis tuaj yeem ua tiav rau txhua qhov kev tshuaj ntsuam xyuas lub platform.
Ntawm no peb tau ua haujlwm nrog cov ntaub ntawv tsim tawm hauv Pincus lab [8,15], thiab pom tias daim ntawv thov ntawm WormNet tau tsim tshiab tuaj yeem ua tiav kev ntxub ntxaug ntawm cov kab mob luv thiab ntev ntev txawm tias cov duab coj los ntawm hnub 1 lossis hnub 2; qhov tseem ceeb, rau hnub 3 lub CNN tau pom qhov ua tau zoo tshaj plaws (Daim duab 2A-2C). WormNet tseem zoo dua ntawm kev faib cov kab mob nrog siab thiab qis tag nrho kev txav mus los, ua tiav 88 feem pua qhov raug rau hnub 1 cov neeg laus (Daim duab 3). Peb cia siab tias yuav tsim cov ntaub ntawv ntau ntxiv thiab txhim kho CNN kwv yees lifespan [32]. Raws li tau hais ua ntej lawm, cov kws sau ntawv tau faib cov worms thiab tsim cov kab sawv cev ncaj nraim, uas tau siv rau kev cob qhia CNN [32]. Ntau tus qauv tau txhim kho qhov kev rov qab los ntawm kev kwv yees ntawm lub hnub nyoog worm. Interestingly, cov kws sau ntawv tau txwv tsis pub pom t silhouette ib leeg tias cov ntaub ntawv kab mob rau hnub nyoog kwv yees, whereas cov ntaub ntawv keeb kwm yav dhau los ntawm kev txhim kho qhov muaj peev xwm ua tau qhov tseeb, txawm tias tus nqi kwv yees ntawm keeb kwm yav dhau yog ib qho khoom qub ntawm kev sim. Yog li ntawd, tej zaum nws yuav ua tau tias qhov kev kwv yees qhov tseeb ntawm WormNet hauv peb qhov kev sim tuaj yeem piav qhia ib feem los ntawm cov ntaub ntawv keeb kwm yav dhau. Txawm li cas los xij, raws li peb qhov kev sim qhia (Cov duab ntxiv 1), WormNet kev ua haujlwm feem ntau yog nyob ntawm C. elegans morphology es tsis yog keeb kwm ntawm cov duab. Qhov tseem ceeb, kev qhia ua ntej ntawm lub cev-kev sib koom tes sawv cev hauv [32] txhim kho qhov tseeb ntawm cov duab nyoos uas qhia tias cov kab mob cab thiab kev ntxhib los mos yog qhov tseem ceeb rau kev kwv yees hnub nyoog.
Ntxiv nrog rau kev ua neej nyob lossis kev txav chaw raws li cov tub ntxhais hluas cov duab, peb kuj tseem tsom mus nrhiav cov yam ntxwv tseem ceeb rau kev twv ua ntej. Raws li txoj haujlwm qauv peb tau txiav txim siab los txiav txim siab seb lub cev ib feem ntawm sab xub ntiag, nruab nrab ntawm lub cev lossis sab hauv qab muaj cov yam ntxwv cuam tshuam rau lub neej ntev tshaj plaws. Peb tsim HydraNet 3 thiab 4, tshiab architectures raws li U-Net thiab pom tau tias lawv muaj peev xwm ua tau zoo segment cab lub cev qhov chaw ua tau zoo Jaccard Performance index qhov tseem ceeb. Qhov tseem ceeb, txhawm rau txhim kho kev txhais lus los ntawm kev tsim qauv peb ua haujlwm tandem ntawm kev faib cov ntsiab lus lom neeg (lifespan thiab txav) yielding saliency los ntawm chav kawm ua kom cov duab qhia [30, 35] thiab morphological segmentation (anterior, mid-lub cev thiab posterior cheeb tsam) kom pom qhov twg. lub cev ib feem yog pab tau rau kev faib. Tsis tas li ntawd, txawm tias tsawg dua, qhov kev tshawb pom tau los ntawm txoj hauv kev tandem tau zoo ib yam nrog tus neeg ua haujlwm ntawm tus kheej. Qhov no binary classifier yog raws li InveptionV3 CNN. Nws tau kawm txog 800 × 800 pixels tag nrho cov duab kho qhov muag nrog cov kab mob segmented los ntawm lawv keeb kwm yav dhau thiab ua tiav cov txiaj ntsig piv rau WormNet, txawm hais tias tus qauv tsis tshua muaj dav vim muaj ntau dua (Daim duab 4). Txawm li cas los xij, nyob rau hauv cov ntaub ntawv ntawm InceptionV3, qhov sib txawv ntawm lub cev tuaj yeem nyob rau hauv CAMs, thiab cov kev soj ntsuam qhia tau hais tias cov yam ntxwv nyob rau hauv lub posterior ntawm tus cab kuj tseem ceeb dua rau kev faib cov kab mob ntev ntev. Txoj hauv kev no muab txoj hauv kev rau kev tshawb pom ntawm cov hnub nyoog tseem ceeb tshiab biomarkers hauv C. elegans hauv qhov chaw tsis siv neeg, muab qhov tseem ceeb hauv cov duab daws teeb meem thiab kev siv lub cev-kev sib koom tes sawv cev. Cov kab mob uas tsis muaj npe xws li pharynx lossis GFP-daim ntawv teev cov chaw tuaj yeem faib ua ke siv HydraNets thiab ntsuas rau lawv lub neej kev muaj peev xwm kwv yees siv CAM mus kom ze thiab WormNet. Nws yog kev ntxias kom xav tias zoo ib yam li kev sib koom ua yeeb yam sib txawv [36], yav tom ntej kev siv los ntawm kev tsim kev txhais lus los ntawm kev sib faib thiab kev faib tawm yuav raug cob qhia kawg-rau-kawg thiab ua haujlwm rau niaj hnub kev tshawb pom. Cov ntaub ntawv pov thawj-ntawm-txoj cai automated analytical platform yuav muaj txiaj ntsig rau kev tshawb pom tsis muaj kev laus biomarkers, tshwj xeeb tshaj yog hnub nyoog hluas 1-3 cov neeg laus C. elegans. Qhov no muaj peev xwm ua tau zoo los ua kom cov tshuaj ntsuam xyuas tshuaj tiv thaiv kev laus. Txoj kev loj hlob ntawm cov txheej txheem kuj tseem yuav pab tau kom pom thiab ua tus yam ntxwv tshiab pathologies hauv C. elegans tseem ceeb rau kev tshawb fawb kev laus. Ua kom cov cai muaj nyob rau hauv kev tshawb fawb zej zog peb tau tso.
Kab lus no yog muab rho tawm los ntawm www.aging-us.com AGING 2022, Vol. 14, nr 4
