The labelled junctions image do not seem to be correct. There are some places where the junction has been discontinued.
Also there is only 1 junction in your 2nd image: it is a single connected component.
From the top image alone, you should be able to know which neighbours each cell has using morphological operations.
First, get all the XStart YStart coordinates of the cells withe the Particle Analyzer, then label the connected components, so each cell has a unique colour/label. Since now you know which cell has which colour (check the image colour at the XStart YStart pixel), do a binary reconstruction of each cell using the XStart, YStart pixel as the seed. That produces only a given cell shown in the reconstructed image. Dilate it twice. The dilated region will now overlap with the neighbours, so if you investigate which colours exist in the intersection (e.g. AND) of the dilated image and the labelled image, you know which cells are the neighbours. Then repeat the same for all the other cells.
It sounds complicated but it is doable with a macro.
Alternatively you could use region connection calculus (RCC)
You want the NC (neighbour connection) relation between objects in that image.
We developed a new (very) fast algorithm to do that:
Landini G, Galton A, Randell D, Fouad S. Novel applications of Discrete Mereotopology to Mathematical Morphology. Signal Processing: Image Communication 76:109-117, 2019. PDF
I will release the fast algorithm plugin very soon but if you are in a hurry and are willing to wait in front of the screen, the old plugin in that page can do it… although slowly.
Hope this helps.