Skip to main content
. 2026 May 30;12(6):80. doi: 10.3390/tomography12060080
Algorithm 1 TASC-SwinMT Forward Propagation
  • Require: Input frames X1,X3∈RB×1×512×512

  • Ensure: Interpolated frame X^2, segmentation masks Y^1,Y^2,Y^3

  • 1:

    Concatenate input frames: X=Concat(X1,X3)∈RB×2×512×512

  • 2:

    Patch embedding: F0=PatchEmbed(X)∈RB×96×128×128

  • 3:

    Extract shared features and record skip connections S1,S2,S3,S4:

  • 4:

    for i=1 to 4 do

  • 5:

        Fi=SwinBlock(Fi−1)

  • 6:

        Fi=TALA(Fi)

  • 7:

        Si=Fi

  • 8:

        if i<4 then

  • 9:

            Fi=PatchMerging(Fi)

  • 10:

      end if

  • 11:

    end for

  • 12:

    Initialize dual decoders:

  • 13:

    Finterp=F4

  • 14:

    Fseg=F4

  • 15:

    for i=1 to 3 do

  • 16:

        Finterp=PatchExpand(Finterp)

  • 17:

        Finterp=MSTAF(Finterp,S4−i)

  • 18:

        Finterp=SwinBlock(Finterp)

  • 19:

        Fseg=PatchExpand(Fseg)

  • 20:

        Fseg=MSTAF(Fseg,S4−i)

  • 21:

        Fseg=SwinBlock(Fseg)

  • 22:

        Finterp,Fseg=CTCI(Finterp,Fseg)

  • 23:

    end for

  • 24:

    Generate final outputs: X^2=InterpHead(Finterp), Y^=SegHead(Fseg)

  • 25:

    Split multi-frame segmentation masks: Y^1,Y^2,Y^3=Split(Y^,dim=1)

  • 26:

    return X^2,Y^1,Y^2,Y^3