鄭金晶,董海波,李明,王超超,衛(wèi)雨果
MRI增強(qiáng)序列列線圖預(yù)測腦膠質(zhì)瘤IDH1基因突變的研究
鄭金晶1,董海波1,李明1,王超超1,衛(wèi)雨果2
1.寧波大學(xué)附屬李惠利醫(yī)院 寧波市醫(yī)療中心李惠利醫(yī)院放射科,浙江寧波 315046;2.通用電氣藥業(yè)有限公司,浙江杭州 310000
探討T1WI MRI增強(qiáng)序列的臨床–影像組學(xué)列線圖預(yù)測腦膠質(zhì)瘤異檸檬酸脫氫酶1(isocitrate dehydrogenase 1,IDH1)基因突變的價(jià)值。回顧性分析2016年2月至2022年2月寧波市醫(yī)療中心李惠利醫(yī)院收治的98例經(jīng)手術(shù)病理證實(shí)的腦膠質(zhì)瘤(Ⅱ~Ⅳ級(jí))T1WI MRI增強(qiáng)圖像。其中,26例IDH1基因突變型(IDH1-M)、72例IDH1基因野生型(IDH1-W),以7∶3比例劃分為訓(xùn)練集(=69)和測試集(=29),使用邏輯回歸方法篩選特征并建立臨床模型。勾畫并測量腦膠質(zhì)瘤的實(shí)質(zhì)區(qū)及壞死區(qū)參數(shù),建立Logistic回歸影像組學(xué)模型,計(jì)算Radscore,生成列線圖。采用校準(zhǔn)曲線和受試者操作特征(receiver operating characteristic,ROC)曲線評(píng)價(jià)3個(gè)模型對(duì)腦膠質(zhì)瘤IDH1基因突變狀態(tài)的預(yù)測效能,進(jìn)行決策曲線分析,評(píng)估列線圖臨床實(shí)用性。訓(xùn)練集經(jīng)過特征篩選,最終選擇6個(gè)影像組學(xué)特征和2個(gè)臨床特征用于構(gòu)建列線圖。臨床模型在訓(xùn)練集和測試集中的曲線下面積(area under the curve,AUC)分別為0.834和0.718,影像組學(xué)模型分別為0.902和0.831,臨床–影像組學(xué)組合模型最高,分別為0.906和0.857。校準(zhǔn)曲線表明,臨床–影像組學(xué)列線圖在訓(xùn)練集中IDH1基因型的預(yù)測值和觀察值之間具有較好的一致性(=0.751)。決策分析曲線表明,組合模型的凈收益在幾乎整個(gè)Pt值范圍內(nèi)均高于臨床模型及影像組學(xué)模型?;贛RI T1WI增強(qiáng)序列的臨床-影像組學(xué)列線圖能較精準(zhǔn)地預(yù)測IDH1基因突變狀態(tài)。
腦膠質(zhì)瘤;磁共振成像;異檸檬酸脫氫酶;影像組學(xué);基因型
腦膠質(zhì)瘤是最常見的腦部原發(fā)性惡性腫瘤,約占中樞系統(tǒng)惡性腫瘤的80%,男性的發(fā)病率略高于女性,并以老年人多見。異檸檬酸脫氫酶1(isocitrate dehydrogenase 1,IDH1)是腦膠質(zhì)瘤的一種重要基因分子生物學(xué)標(biāo)志物,分為野生型(IDH1-W)和突變型(IDH1-M)兩種表達(dá)狀態(tài)。IDH1基因突變是膠質(zhì)細(xì)胞瘤中最早可檢測到的基因改變,突變的IDH蛋白被認(rèn)為是通過競爭性抑制參與組蛋白和DNA去甲基化,從而阻斷了細(xì)胞的分化,降低腫瘤細(xì)胞的增殖,因此具有相對(duì)良好的預(yù)后[1-3]。2016年世界衛(wèi)生組織(World Health Organization,WHO)中樞神經(jīng)系統(tǒng)腫瘤分類中新增了分子分型,它將IDH突變型彌漫性星形細(xì)胞瘤又根據(jù)組織學(xué)確定為3種不同的分類,即彌漫性星形細(xì)胞瘤、間變性星形細(xì)胞瘤和膠質(zhì)母細(xì)胞瘤。而2021年第5版進(jìn)一步推進(jìn)了分子診斷的作用,它將所有IDH突變型彌漫性星形細(xì)胞瘤歸為一種類型,再分為2~4級(jí),與此同時(shí),分子特征可以獨(dú)立于組織學(xué)特征來確定是否歸為4級(jí)。因此,治療前精準(zhǔn)預(yù)測腦膠質(zhì)瘤IDH1是否突變,對(duì)指導(dǎo)個(gè)體化治療和預(yù)后評(píng)估具有重要意義,成為近年來影像基因組學(xué)研究熱點(diǎn)[4-5]。但是既往相關(guān)研究多以腫瘤整體和(或)水腫區(qū)作為感興趣區(qū)提取影像組學(xué)特征,沒有全面、完整地反映腫瘤異質(zhì)性的特征,準(zhǔn)確性有待于提高[6-7]。本研究分別勾畫增強(qiáng)MRI腫瘤實(shí)質(zhì)區(qū)域(region of contrast-enhanced tumor,rCET)、壞死區(qū)域(region of necrosis,rNec),并將臨床特征、MRI影像學(xué)特征與影像組學(xué)標(biāo)簽相結(jié)合,構(gòu)建預(yù)測模型、繪制列線圖,評(píng)估基于T1WI MRI增強(qiáng)序列的臨床–影像組學(xué)列線圖預(yù)測腦膠質(zhì)瘤IDH1基因突變的價(jià)值。
選取2016年2月至2022年2月寧波市醫(yī)療中心李惠利醫(yī)院收治的符合以下標(biāo)準(zhǔn)的患者:①經(jīng)手術(shù)病理證實(shí)為腦膠質(zhì)瘤;②獲得患者的臨床特征及IDH1免疫組織化學(xué)結(jié)果;③術(shù)前行常規(guī)T1增強(qiáng)MRI序列檢查,圖像無明顯偽影;④術(shù)前無放化療史。排除標(biāo)準(zhǔn):①腫瘤內(nèi)有出血或鈣化者;②腫瘤無壞死區(qū)或強(qiáng)化程度較弱者。共98例腦膠質(zhì)瘤患者入組,其中,男57例,女41例;IDH1突變型26例,IDH1野生型72例。按7∶3比例隨機(jī)分為訓(xùn)練集(69例,其中IDH1-M型19例,IDH1-W型50例)和測試集(29例,其中IDH1-M型7例,IDH1-W型22例)。本研究經(jīng)寧波市醫(yī)療中心李惠利醫(yī)院倫理委員會(huì)批準(zhǔn)(倫理審批號(hào):KY2022PJ184),并獲得免除知情同意書許可。
采用GE Discovery 3.0T MRI磁共振掃描儀(美國GE Health Care公司)和8通道頭線圈。①常規(guī)MRI平掃序列:T1WI序列:TE 24ms,TR 1850ms,TI 780ms;T2WI序列:TE 105ms,TR 6656ms;矩陣288×224,F(xiàn)OV 240mm×240mm,層間距1mm,層厚4mm,掃描層24層;②DWI序列:TR 4500ms,TE為最小值,b值設(shè)定1000s/mm2,矩陣160×160,F(xiàn)OV 240mm×240mm,層間距1.5mm,層厚5mm,掃描層數(shù)24層;③T1WI增強(qiáng)MRI序列:注入對(duì)比劑釓雙胺注射液(德國Bayer Schering Pharma公司)0.2ml/kg后,橫斷面T1WI序列掃描,參數(shù)同平掃T1WI序列。
收集醫(yī)院影像歸檔和通信系統(tǒng)(picture archiving and communication system,PACS)中的腦膠質(zhì)瘤MRI影像后,應(yīng)用美國GE公司AI-Kit(Artificial Intelligence Kit,Version:3.3)軟件進(jìn)行圖像去噪、去骨、偏差矯正等預(yù)處理。由2位神經(jīng)影像醫(yī)師在開源軟件ITK-SNAP(Version 3.4.0)上對(duì)T1WI增強(qiáng)圖像逐層勾畫,并由1名資深神經(jīng)影像學(xué)專家復(fù)核,選擇勾畫感興趣區(qū)(volume of interest,VOI)。實(shí)質(zhì)區(qū)為T1增強(qiáng)圖像呈強(qiáng)化的區(qū)域;壞死區(qū)為T1WI上低信號(hào)、T2WI上高信號(hào)且T1增強(qiáng)圖像未強(qiáng)化的區(qū)域(圖1)。
應(yīng)用AI-Kit軟件對(duì)MRI圖像特征提取,提取了396個(gè)影像組學(xué)特征,包括直方圖特征(histogram features,HF)、灰度共生矩陣(grey level co-occurrence matrix,GLCM)、灰度游程矩陣(grey level runlength matrix,GLRLM)等。每例患者包括實(shí)質(zhì)區(qū)和壞死區(qū)2個(gè)VOI,共采集792個(gè)有效組學(xué)特征參數(shù)。
在R軟件(v.4.1.0)中完成特征選擇和模型構(gòu)建。選擇用于構(gòu)建模型的非零系數(shù)特征參數(shù)用于預(yù)測IDH1基因表達(dá)類型,將具有非零系數(shù)的特征組合起來構(gòu)建一個(gè)公式,該公式可用于計(jì)算每例腦膠質(zhì)瘤的Radscore,使用Wilcoxon檢驗(yàn)比較不同IDH1基因狀態(tài)Radscore的差別。
圖1 VOI示意圖
A.實(shí)質(zhì)區(qū)的最大切面圖;C.其對(duì)應(yīng)的3D可視化圖;B.壞死區(qū)最大切面圖;D.代表其對(duì)應(yīng)的3D可視化圖
在R軟件中將Radscore與篩選出的臨床預(yù)測因子進(jìn)行多元Logistic回歸分析,構(gòu)建預(yù)測IDH1基因表達(dá)類型的組合模型,利用受試者操作特征(receiver operating characteristic,ROC)曲線對(duì)以上3種預(yù)測模型診斷效能進(jìn)行評(píng)估。校準(zhǔn)曲線和Hosmer- Lemeshow檢驗(yàn)用于評(píng)估IDH1基因型的列線圖預(yù)測概率與實(shí)際結(jié)果之間的一致性,若校準(zhǔn)曲線的擬合度較好,且>0.05,說明模型的預(yù)測性能較好。決策曲線用于評(píng)價(jià)臨床決策的凈獲益情況,并與臨床模型進(jìn)行比較。
訓(xùn)練集和測試集的IDH1基因狀態(tài)、WHO分級(jí)情況及臨床特征比較,差異均無統(tǒng)計(jì)學(xué)意義(表1)。經(jīng)單因素Logistic回歸分析,訓(xùn)練集中IDH1-M組和IDH1-W組的年齡(=0.023)、腫瘤位置2(=0.032)比較,差異有統(tǒng)計(jì)學(xué)意義。經(jīng)多因素Logistic回歸分析后,仍選擇年齡(=0.007)和腫瘤位置2(=0.012)構(gòu)建臨床模型。臨床模型在訓(xùn)練集和測試集中預(yù)測IDH1狀態(tài)的曲線下面積(area under the curve,AUC)分別為0.834和0.718(圖2);訓(xùn)練集的敏感度、特異性和準(zhǔn)確率分別為68.4%、88.0%及82.6%,測試集分別為57.1%、81.8%及75.9%。
對(duì)792個(gè)有效組學(xué)特征參數(shù)進(jìn)行最小冗余最大相關(guān)性(mRMR)算法,得到30個(gè)特征,再進(jìn)行最小絕對(duì)收縮和選擇算子(leastabsolute shrinkage and selection operator,LASSO)邏輯回歸(圖3),得到6個(gè)非零系數(shù)的組學(xué)特征來構(gòu)建影像組學(xué)模型,分別計(jì)算訓(xùn)練集及測試集每個(gè)患者的Radscore,結(jié)果顯示,IDH1-M組的Radsore低于IDH1-W,差異有統(tǒng)計(jì)學(xué)意義(圖3D)。影像組學(xué)模型在訓(xùn)練集的敏感度、特異性和準(zhǔn)確率分別為89.5%、82.0%、84.0%,測試集為71.4%、63.6%、65.5%。模型在訓(xùn)練集中的AUC值為0.902,測試集的AUC值為0.831。
在臨床–影像組學(xué)組合模型的構(gòu)建過程中,Logistic回歸分析將Radscore、年齡和腫瘤位置2確定為獨(dú)立預(yù)測因子,在此基礎(chǔ)上構(gòu)建可視化臨床–影像組學(xué)列線圖(圖4)。與影像組學(xué)模型和臨床模型相比,臨床–影像組學(xué)組合模型達(dá)到最高AUC[訓(xùn)練集:0.906(95%:0.834~0.991);測試集:0.857(95%:0.707~0.995)],敏感度、特異性和準(zhǔn)確率訓(xùn)練集分別為84.2%、90.0%、88.4%;測試集為60.0%、94.7%、82.8%。影像組學(xué)–臨床列線圖的校準(zhǔn)曲線表明,訓(xùn)練集和測試集中IDH1基因突變的預(yù)測和實(shí)際測得值之間都具有較好一致性(分別為0.751、0.196,圖5)。3個(gè)模型的決策分析曲線表明,組合模型的凈收益在幾乎整個(gè)Pt值范圍內(nèi)都高于其他2個(gè)模型(圖6)。
表1 訓(xùn)練集與測試集臨床特征比較
注:*采用Fisher確切概率法
圖2 訓(xùn)練集(A)和測試集(B)中3種模型的ROC曲線
圖3 影像組學(xué)特征LASSO降維圖
A.利用十折交叉驗(yàn)證算法求得最佳懲罰系數(shù)λ,使其對(duì)應(yīng)的特征集合具有最佳分類效能;B.系數(shù)收縮圖,橫軸是log λ,縱軸表示系數(shù);C.最終篩選出的6個(gè)非零系數(shù)特征及其對(duì)應(yīng)系數(shù);D.Radscore在不同IDH1突變狀態(tài)下的區(qū)別,lable0代表野生型,lable1代表突變型,左邊為訓(xùn)練組,右邊為測試組。Radscore的截止值為–1.15
圖5 臨床-影像組學(xué)列線圖的校準(zhǔn)曲線
近年來,對(duì)腦膠質(zhì)瘤分子結(jié)構(gòu)的深入研究揭示了其特有的遺傳特征和表觀遺傳學(xué)表現(xiàn),并據(jù)此劃分為不同的分子亞型,表觀遺傳與DNA的甲基化有關(guān),IDH1基因能誘導(dǎo)去甲基化,因此,表觀遺傳調(diào)節(jié)IDH1基因成為腫瘤分類的關(guān)鍵生物學(xué)標(biāo)志物,在腦膠質(zhì)瘤進(jìn)化和生物學(xué)表現(xiàn)起著關(guān)鍵的作用[8]。Yao等[9]關(guān)于IDH1-R132H突變與神經(jīng)膠質(zhì)瘤干細(xì)胞(glioma stem cells,GSC)的研究發(fā)現(xiàn),IDH1-R132H的過表達(dá)會(huì)導(dǎo)致GSC增殖、遷移和侵襲減少,誘導(dǎo)細(xì)胞凋亡并改善GSC分化,而該位點(diǎn)的突變發(fā)生于80%以上的IDH1-M腦膠質(zhì)瘤。因此,IDH1-M腦膠質(zhì)瘤患者預(yù)后較好。Yan等[10]研究顯示,IDH1-M腦膠質(zhì)瘤患者的中位生存期明顯高于IDH1-W。Su等[11]研究發(fā)現(xiàn),IDH1-M腦膠質(zhì)瘤患者平均年齡低于IDH1-W。本研究中,年齡特征也顯示出了較強(qiáng)的預(yù)測強(qiáng)度,意味著結(jié)合年齡特征的列線圖具有更好的預(yù)測能力[12]。Qi等[13]研究發(fā)現(xiàn),IDH1-M腦膠質(zhì)瘤主要位于額葉,本研究亦有類似發(fā)現(xiàn)。因此,本研究把年齡和位置加入到組合模型中來,期望提高其預(yù)測能力。
Yu等[14]關(guān)于Ⅱ級(jí)腦膠質(zhì)瘤IDH1基因突變的影像組學(xué)研究,將腦膠質(zhì)瘤分級(jí)作為重要特征納入研究,構(gòu)建的影像組學(xué)模型AUC值為0.860,但是腦膠質(zhì)瘤的分級(jí)因素需要經(jīng)過手術(shù)或活檢后病理獲取,本研究是基于無創(chuàng)的影像組學(xué),因此未將膠質(zhì)瘤分級(jí)作為臨床特征納入研究。
一部分相關(guān)研究是基于勾畫腦膠質(zhì)瘤的整體區(qū)域用于構(gòu)建影像組學(xué)模型預(yù)測IDH1基因突變[14-15]。筆者認(rèn)為,腫瘤rCET和rNEC包含細(xì)胞密度、微血管增殖和局部微環(huán)境的信息存在較大差異,分別提取rCET和rNEC影像組學(xué)特征可以更好地量化腫瘤的綜合信息,表征腦膠質(zhì)瘤的異質(zhì)性。本研究中,提取的6個(gè)組學(xué)特征中各有3個(gè)來自于rCET和rNEC,來自rCET的特征系數(shù)總和為1.399,rNEC的系數(shù)總和為1.177,特征系數(shù)的絕對(duì)值可以看作是某個(gè)特征對(duì)預(yù)測IDH1基因突變風(fēng)險(xiǎn)的重要性,絕對(duì)值越大,對(duì)分類效果的影響越顯著,筆者認(rèn)為,rCET是腫瘤細(xì)胞密度較高、生長較活躍及異質(zhì)性較明顯的區(qū)域,因此該區(qū)域?qū)τ贗DH1基因突變預(yù)測效能較rNEC貢獻(xiàn)更大。在這6個(gè)特征中,包括了4個(gè)GLRLM特征和2個(gè)GLCM特征,是影像組學(xué)較常使用的特征類型,從圖像的精細(xì)程度和均勻程度反映了腫瘤的影像信息,定量地描述腫瘤的內(nèi)部特征。GLRLM特征表示紋理在預(yù)設(shè)方向上的粗糙程度,其特征系數(shù)總和達(dá)1.665,筆者推測其占比較高的原因可能是這一類型下多種特征的互補(bǔ)從不同灰度值及方向上反映rCET和rNEC的信號(hào)強(qiáng)度和腫瘤的不均質(zhì)性,從而更好地實(shí)現(xiàn)模型的預(yù)測效能。GLCM特征系數(shù)總和0.911,均為相關(guān)性特征,該特征可以顯示灰度值與GLCM中各自的體素之間的線性相關(guān)性,相關(guān)度愈大,矩陣元素值愈均勻,在本研究中,這意味著IDH1-M較IDH1-W的內(nèi)部結(jié)構(gòu)更加均勻,說明IDH1-M異質(zhì)性低,而IDH1-W較高。
本研究構(gòu)建的臨床–影像組學(xué)組合模型在訓(xùn)練集和測試集中的AUC值分別為0.906和0.857,校準(zhǔn)曲線的擬合度較好(=0.751),表明組合模型的預(yù)測能力好于單獨(dú)的模型[11-12]。決策曲線分析顯示,組合模型的臨床凈收益優(yōu)于單一模型。因此,筆者認(rèn)為基于分別勾畫rCET和rNEC建立的臨床–影像組學(xué)組合模型能夠較全面、完整地反映膠質(zhì)瘤的異質(zhì)性,在預(yù)測IDH1基因狀態(tài)方面可能具有較高的準(zhǔn)確度。
本研究尚有一些局限性,首先,本研究為單中心研究,樣本量還不夠大,模型的效率和穩(wěn)定性還有待于提高,下一步將加大樣變量繼續(xù)研究;其次,勾畫VOI無法避免各種組織類型的重疊,例如,一些沒有強(qiáng)化的腫瘤組織可能與水腫組織重疊,對(duì)結(jié)果可能造成影響。
綜上所述,基于MRI T1WI增強(qiáng)序列的臨床–影像組學(xué)列線圖能較精準(zhǔn)地預(yù)測IDH1基因突變狀態(tài)。
[1] OHGAKI H, KLEIHUES P. The definition of primary and secondary glioblastoma[J]. Clin Cancer Res, 2013, 19(4): 764–772.
[2] TURCAN S, ROHLE D, GOENKA A, et al. IDH1 mutation is sufficient to establish the glioma hypermethylator phenotype[J]. Nature, 2012, 483(7390): 479–483.
[3] WAITKUS M S, DIPLAS B H, YAN H. Biological role and therapeutic potential of IDH mutations in cancer[J]. Cancer Cell, 2018, 34(2): 186–195.
[4] CHOI Y S, BAE S, CHANG J H, et al. Fully automated hybrid approach to predict the IDH mutation status of gliomas via deep learning and radiomics[J]. Neuro Oncol, 2021, 23(2): 304–313.
[5] WANG K, WANG Y, FAN X, et al. Radiological features combined with IDH1 status for predicting the survival outcome of glioblastoma patients[J]. Neuro Oncol, 2016, 18(4): 589–597.
[6] LAO J, CHEN Y, LI Z C, et al. A deep learning-based radiomics model for prediction of survival in glioblastoma multiforme[J]. Sci Rep, 2017, 7(1): 10353.
[7] NIU L, FENG W H, DUAN C F, et al. The value of enhanced MR radiomics in estimating the IDH1 genotype in high-grade gliomas[J]. Biomed Res Int, 2020, 2020: 4630218.
[8] GUSYATINER O, HEGI M E. Glioma epigenetics: from subclassification to novel treatment options[J]. Semin Cancer Biol, 2018, 51: 50–58.
[9] YAO Q, CAI G, YU Q, et al. IDH1 mutation diminishes aggressive phenotype in glioma stem cells[J]. Int J Oncol, 2018, 52(1): 270–278.
[10] YAN H, PARSONS D W, JIN G, et al. IDH1 and IDH2 mutations in gliomas[J]. N Engl J Med, 2009, 360(8): 765–773.
[11] SU X, SUN H, CHEN N, et al. A radiomics-clinical nomogram for preoperative prediction of IDH1 mutation in primary glioblastoma multiforme[J]. Clin Radiol, 2020, 75(12): 963.e7–963.e15.
[12] TAN Y, ZHANG S T, WEI J W, et al. A radiomics nomogram may improve the prediction of IDH genotype for astrocytoma before surgery[J]. Eur Radiol, 2019, 29(7): 3325–3337.
[13] QI S, YU L, LI H, et al. Isocitrate dehydrogenase mutation is associated with tumor location and magnetic resonance imaging characteristics in astrocytic neoplasms[J]. Oncol Lett, 2014, 7(6): 1895–1902.
[14] YU J, SHI Z, LIAN Y, et al. Noninvasive IDH1 mutationestimation based on a quantitative radiomics approach for grade Ⅱglioma[J]. Eur Radiol, 2017, 27(8): 3509–3522.
[15] HSIEH K L, CHEN C Y, LO C M. Radiomic model for predicting mutations in the isocitrate dehydrogenase gene in glioblastomas[J]. Oncotarget, 2017, 8(28): 45888–45897.
Prediction of IDH1 gene mutation in glioma using nomogram based on MRI enhanced sequence
ZHENG Jinjing, DONG Haibo, LI Ming, WANG Chaochao, WEI Yuguo
1.Department of Radiology, Ningbo Medical Center Lihuili Hospital, Ningbo University, Ningbo 315046, Zhejiang, China; 2.General Electric Pharmaceutical Co., Ltd, Hangzhou 310000, Zhejiang, China
To investigate the value of clinical-radiomics nomogram based on T1WI MRI enhanced sequence in predicting isocitrate dehydrogenase 1 (IDH1) gene mutation in glioma.T1WI MRI enhanced images of 98 cases with glioma (grade Ⅱ-Ⅳ) which were received and treated by Ningbo Medical Center Lihuili Hospital from February 2016 to February 2022, confirmed by operation and pathology were analyzed retrospectively, among them, 26 cases of IDH1 gene mutants type (IDH1-M) and 72 cases of IDH1 gene wild type (IDH1-W) were divided into training set (=69) and test set (=29) in a 7∶3 ratio, Logistic regression methods were used for screening the features and the clinical model was established. The parenchymal area and necrotic area of glioma were delineated and parameters were measured to establish a Logistic regression radiomics model, calculate Radscore, and generate the nomogram. Receiver operating characteristic (ROC) curve was used to evaluate the prediction efficiency of the three models for IDH1 gene mutation status, and the decision curve was analyzed to evaluate the clinical usefulness of the nomogram.After feature screening, six imaging features and two clinical features were selected to construct the nomogram. Area under the curve (AUC) of clinical model in training set and test set were 0.834 and 0.718 respectively. AUC of radiomics model were 0.902 and 0.831, respectively. AUC of clinical-radiomics combined model were 0.906 and 0.857, respectively. The calibration curve proved that there was a good agreement between the predicted and observed values of IDH1 genotypes in the training set (=0.751). The decision curve analysis curve showed that the net benefit of the combined model was higher than that of the clinical model and the radiomics model in almost the entire Pt range.Clinical-radiomics nomogram based on MRI T1WI enhanced sequence can accurately predict IDH1 gene mutation status.
Glioma of the brain; Magnetic resonance imaging; Isocitrate dehydrogenase; Radiomics; Genotype
R445.2
A
10.3969/j.issn.1673-9701.2023.29.001
浙江省醫(yī)藥衛(wèi)生科技項(xiàng)目(2023KY1047,2017KY572);寧波市科技惠民項(xiàng)目(2016C51017)
董海波,電子信箱:donghb18@sina.com
(2023–04–07)
(2023–04–29)