A U-Net Transformer for magnetic resonance elastography

{"Abstract":[0],"Purpose":[1],"Magnetic":[2],"resonance":[3],"elastography":[4],"(MRE)":[5],"is":[6,19],"a":[7,47,84,98,149,178,187,197],"critical":[8],"non-invasive":[9],"modality":[10],"for":[11,181,191],"quantifying":[12],"tissue":[13],"stiffness,":[14],"but":[15],"its":[16],"clinical":[17,159,203],"utility":[18],"limited":[20],"by":[21],"the":[22,74,94,103,112,155,173],"high":[23],"computational":[24,166],"cost":[25],"and":[26,39,54,70,172],"noise":[27],"sensitivity":[28],"of":[29,107,115],"traditional":[30],"inversion":[31,37],"techniques,":[32],"such":[33],"as":[34],"algebraic":[35],"Helmholtz":[36],"(AHI)":[38],"finite":[40],"element":[41],"methods":[42],"(FEM).":[43],"This":[44],"study":[45],"proposes":[46],"high-performance":[48],"deep":[49],"learning":[50,144],"alternative":[51],"to":[52,82,157],"accelerate":[53],"stabilize":[55],"MRE":[56,183],"reconstruction.":[57],"Methods":[58],"We":[59],"developed":[60],"an":[61],"open-source,":[62],"physics-based":[63],"simulation":[64,175],"framework":[65,176],"that":[66,101,121,153],"pairs":[67],"realistic":[68],"liver":[69],"tumor":[71],"geometries":[72],"from":[73],"LiTS":[75],"dataset":[76],"with":[77,111,164],"clinically":[78],"informed":[79],"stiffness":[80],"priors":[81],"generate":[83],"large-scale":[85],"benchmark":[86],"dataset.":[87],"To":[88],"improve":[89],"reconstruction":[90,138],"accuracy,":[91],"we":[92,147],"introduce":[93],"U-Net":[95],"Transformer":[96],"(UNT),":[97],"novel":[99],"architecture":[100,134,171],"hybridizes":[102],"local":[104],"inductive":[105],"bias":[106],"convolutional":[108],"neural":[109,192],"networks":[110],"global":[113],"context-awareness":[114],"Transformers.":[116],"Results":[117],"Numerical":[118],"experiments":[119],"indicate":[120],"data-driven":[122],"training":[123,127],"consistently":[124],"outperforms":[125],"physics-informed":[126,150],"in":[128,206],"prediction":[129],"accuracy.":[130],"The":[131,169],"proposed":[132],"UNT":[133,170],"demonstrates":[135],"significantly":[136],"higher":[137],"accuracy":[139,163],"than":[140],"current":[141],"state-of-the-art":[142],"machine":[143],"models.":[145],"Additionally,":[146],"validated":[148,198],"fine-tuning":[151],"strategy":[152],"adapts":[154],"model":[156],"“sparse-data\\"":[158],"scenarios,":[160],"enhancing":[161],"patient-specific":[162],"minimal":[165],"overhead.":[167],"Conclusion":[168],"accompanying":[174],"establish":[177],"standardized":[179],"foundation":[180],"AI-driven":[182],"inversion.":[184],"By":[185],"providing":[186],"rigorous":[188],"performance":[189],"baseline":[190],"operators,":[193],"this":[194],"work":[195],"offers":[196],"pathway":[199],"toward":[200],"real-time,":[201],"AI-assisted":[202],"decision":[204],"support":[205],"radiological":[207],"diagnostics.":[208]}