https://www.ebay.com/itm/125966951416
http://www.ke5fx.com/r7400u.htm
http://www.sciencemadness.org/talk/viewthread.php?tid=159366
With a bit of creative scrounging you could put something together for less than US $100, I imagine. The budget would probably be driven by the question of whether the surplus pager scintillators are sensitive enough to observe the effect.
What would probably be better is to see what design limitation crop up if you try doing it for say $1000 without any surplus parts.
Digitization is another question -- they used a fast FPGA-based digitizer but it's not clear why it was necessary given the duration of the events being recorded. If it really is needed, then driving the cost down on the digitizer will be as big a challenge as the scintillator itself.
A few hundred for a combo already assembled together.
This makes it harder to copy a documented design, of course, but most of the time a scientist with a knack for gear can adapt things as easily as copying them.
Also, people did photon counting for a long time without 50 MHz ADC's, just saying. ;-)
Perhaps polystyrene or PMMA or another transparent plastic could serve as a cheaper alternative, but probably at the cost of energy resolution, although this study binned the energies anyway.
The arxiv Paper says they use an R1924A PMT : https://www.hamamatsu.com/us/en/product/optical-sensors/pmt/...
The parent article describes a more recent iteration using silicon multi-pixel photon counters (MPPCs aka Si"PhotoMultipliers" SiMPs).
Hamamatsu makes more than conventional PMT's:
https://www.hamamatsu.com/jp/en/product/optical-sensors/mppc...
When I said $5 to $10 range I was referring to some of these for example (did not thoroughly search for the lowest priced across distributors right now):
https://www.mouser.com/c/optoelectronics/optical-detectors-a...
Most of this system is very fast readout circuitry - fast ADC. You need an FPGA to get the data off at 12-bit 50MSPs.
The crystal is a few hundred bucks: http://www.epic-scintillator.com/CsI-crystal-scintillator/Cs... (not quite the right dimensions) but for other experiments you could go smaller.
The expensive stuff is probably the PMTs from Hamamatsu and the power supply from Matsusada. They will certainly give you a quote, but their sensors can be pretty pricey. Probably hundreds each for the PMTs and that again for the power supply. Hamamatsu are the experts and they have a good monopoly. Edmund sell some pre-packaged tubes for example: https://www.edmundoptics.com/f/hamamatsu-photomultiplier-tub... (but Edmund are always $$$)
Everything else is glue really, with a custom PCB for the data capture. Though some of the components like the FPGA and the ADC are $50 each (and there are two ADCs). I don't know if they would release this open source, but I suspect not (which is a shame).
As is typical with science, the authors emphasise how they designed this to be a low cost system, but never actually say how much it cost. I would hazard a guess that you could do this for under $5k BOM cost (ignoring design and labour) if you planned it well. Let's say $500-1k for a crystal with some provenance, $2-3k for optics and $1k for the circuitry and housing. Might be well off on the crystal if you have to buy it from somewhere reputable though. You could probably MacGuyver something for a lot less if you could get away with bits from eBay.
See: https://academist-cf.com/projects/16?lang=en